PHYSMOD2011 Proceedings

PHYSMOD2011
International Workshop on
PHYSICAL MODELLING OF FLOW AND DISPERSION PHENOMENA
22-24 August 2011
Joint Japanese-German Symposium on
URBAN AREAS IN A CHANGING CLIMATE
25 August 2011
KlimaCampus, University of Hamburg, Germany
Conference Proceedings, 15 August 2011
Dear colleagues,
Welcome to PHYSMOD2011 hosted by the KlimaCampus at Hamburg University and organized by the
Environmental Wind Tunnel Laboratory (EWTL) group. We are excited to have you in Hamburg and
hope you enjoy your stay. Some helpful information is provided in the back of the conference proceedings,
including local contact information. Please do not hesitate to contact us if you have any questions while
you are in Hamburg.
The objective of the PHYSMOD international workshop is to bring together the community active in
physical modelling of atmospheric ow and dispersion in uid modelling facilities such as wind tunnels
and water umes. The biennial PHYSMOD was re-established in 2001 and is attracting a growing number
of experts, young scientists and students active in the eld. In an open-minded atmosphere the most
recent advances in uid modelling, in the state-of-the-art of experimental work as well as new emerging
research areas are discussed and assessed.
Key topics of the workshop are:
ow and dispersion in urban areas and the eect of buildings on transport and diusion
transient ow and dispersion phenomena in the atmospheric boundary layer
ow and dispersion within and above idealized surface roughness
boundary layer modelling for wind technology and wind energy research
test data and validation of numerical and analytical modelling tools
quality assurance and improvement of boundary layer modelling techniques
In 2011 the three-day PHYSMOD2011 workshop will be extended by a one-day Japanese-German Sym-
posium focusing on eects of climate change in urban areas. This symposium is organized in the course
of the 150th anniversary of the treaty of friendship and trade between Japan and former Prussia.
On behalf of the local organizing team,
Scientic Committee:
Dr. Sandrine Aubrun, University of Orleans (France)
Dr. Matteo Carpentieri, University of Surrey (UK)
Dr. Daniele Contini, ISAC-CNR, Lecce (Italy)
Dr. David Hall, Envirobods Ltd (UK)
Dr. David Heist, US Environmental Protection Agency (USA)
Dr. Kyosuke Hiyama, University of Tokyo (Japan)
Dr. Klara Jurcakova, Czech Academy of Science (Czech Republic)
Prof. Bernd Leitl, University of Hamburg (Germany)
Prof. Alan Robins, University of Surrey (UK)
Dr. Pietro Salizzoni, Ecole Centrale de Lyon (France)
Prof. Eric Savory, University of Western Ontario (Canada)
Prof. Michael Schatzmann, University of Hamburg (Germany)
Dr. Lionel Soulhac, Ecole Centrale de Lyon (France)
Ir. Eddy Willemsen, DNW German-Dutch Wind Tunnels (The Netherlands)
Conference Sponsors:
Advanced Flow Measurement
Non invasive, point, planar and volumetric techniques
Applications include:
Aerodynamic investigations
Internal flow field investigation
Mixing phenomena
Flow Visualization
Technologies:
Particle Image Velocimetry
Laser Doppler Anemometry
Hot wire Anemometry
Laser Induced Fluorescence
www.dantecdynamics.com

Dantec Dynamics laser optical measurement systems are used in a wide range of applications in fluid
mechanics and solid mechanics. You can obtain data about physical properties such as velocity, turbulence
temperature, concentration, particle size, strain, stress and vibration.

Fluid Mechanics

A complete range of fluid flow measurement instrumentation systems
designed for diagnostics and research guarantees solutions for any kind of
application into flow, turbulence, micro fluids, spray characteristics and
combustion processes. Measurements include flow velocities, turbulence
distributions, mass transport (concentration) heat transfer (temperature),
combustion species (e.g. OH, NO, CH) and particle
sizing. The range of measurement systems include Constant Temperature
Anemometry (CTA), Laser Doppler Anemometry (LDA), Particle Dynamics
Analyzer (PDA), Particle Image Velocimetry (PIV) and Laser Induced
Fluorescence (LIF), Interferometric Particle Imaging (IPI), Shadow Sizing
and Flow Visualisation. Our highly accurate products are applied in fields
such as automotive, aerospace, wind engineering, hydrodynamics,
combustion technology, process engineering, biomedical, pharmaceutical,
micro technology, spray research, electronics and indoor climate.

Solid Mechanics

Dantec Dynamics is the specialist for laseroptical strain and stress
measurement on every type of material and component as well as for
measuring and analysing harmonic and transient vibrations. Dantec
Dynamics also produces shearography systems for non-destructive testing
on components, particularly on composites. Our systems are applied to
product development and material science in the automotive, aerospace and
electronics industries.
Non-destructive testing (NDT) is an important element in the quality process
for all composite matrial manufacturers and users while the Digital Image
Correlation (DIC) is a non-contact, full-field optical Technique that can
measure 3D deformation and surface strain and almost any material

Partners for Progress

Many of Dantec Dynamics customers are involved in cutting edge research and development that
occasionally require special purpose-designed system elements. Dyntec Dynamics customer design
engineers, who have many years of experience, work closely with customers on the development of special
designed measurement systems.
Dantec Dynamics contract measurement service provides easy access to both our leading edge
instrumentation and the know-ledge of our high skilled engineers. It delivers high-quality measurement
results that are specific to each individual application.
Dantec Dynamics has established partnerships with some of the worlds leading universities and companies
in order to develop new measurement technologies. If you are exploring novel techniques that could have
coomercial potential, you are encouraged to contact us.

www.dantecdynamics.com
Using a flexible camera arrangement, three cameras, each with 4
million pixel resolution, can now be taken out of the V3V-9000
probe and used in any arrangement needed to meet a wide array
of flow measurement applications, including:
Applications:
3-dimensional flow in a volume (V3V)
3-dimensional flow in a plane (Stereo PIV)
2-dimensional flow in a plane (2D PIV)
Scalar measurements in a plane (PLIF)
Stereo PIV + PLIF
microPIV
F L U I D M E C H A N I C S
V3V
9000
New, Flexible Camera Arrangement
Provides the Most Versatile 3D3C
Flow Measurement System Available
Model Description Quantity
PIV-V3V-UP
Upgrade system from PIV to V3V,
including V3V Precision Mount,
Calibration system, Hyperstreaming
system and Insight V3V software
1
630059-ST Powerview 4MP camera 1
RP-630159-FIRM Firmware and camera upgrade 2
Package for StereoPIV upgrade to V3V
3D3C System:
Model Description Quantity
PIV-V3V-UP
Upgrade system from PIV to V3V,
including V3V Precision Mount,
Calibration system, Hyperstreaming
system and Insight V3V software
1
630059-ST Powerview 4MP camera 2
RP-630159-FIRM Firmware and camera upgrade 1
Package for 2DPIV upgrade to V3V
3D3C System:
Upgrading to the V3V-9000
If you have a Stereo PIV or 2D PIV systems based on the PowerView Plus 4MP camera, your system can be easily upgraded to the V3V-9000 for 3D3C
measurements. To upgrade, check out TSIs new packages:
Call 800-874-2811 to contact your TSI Sales managers,
TSI Representatives or visit our Website tsi.com for details
When three cameras are inserted
into to the V3V
-9000 probe, robust

volumetric 3D measurements (3D3C)
can be achieved.
V3V-9000:5001261 5/17/2011 10:05 AM Page 1
TSI Incorporated - 500 Cardigan Road, Shoreview, MN 55126-3996 USA
USA Tel: +1 800 874 2811 E-mail: info@tsi.com Website: www.tsi.com
UK Tel: +44 149 4 459200 E-mail: tsiuk@tsi.com Website: www.tsiinc.co.uk
France Tel: +33 491 11 87 64 E-mail: tsifrance@tsi.com Website: www.tsiinc.fr
Germany Tel: +49 241 523030 E-mail: tsigmbh@tsi.com Website: www.tsiinc.de
India Tel: +91 80 41132470 E-mail: tsi-india@tsi.com
China Tel: +86 10 8251 6588 E-mail: tsibeijing@tsi.com
Singapore Tel: +65 6595 6388 E-mail: tsi-singapore@tsi.com
Contact your local TSI Distributor or visit our website www.tsi.com for more detailed specifications.
Copyright 2010 by TSI Incorporated Printed in U.S.A.
V3V measurements of a vortex ring generated from a piston with an inclined
exit. Ref Troolin D.; Longmire E.
Simultaneous PIV-PLIF measurements of intersecting jets
PLIF concentration measurements of a jet
New, Flexible ArrangementSame Incredible Results!
X=0.75R
X=1.85R
X=0.59R
StereoPIV downstream of a propellar. Courtesy of INSEAN
V3V-9000:5001261 5/17/2011 10:05 AM Page 2
Table of Contents
PHYSMOD2011 Schedule 3
Monday, August 22, 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Tuesday, August 23, 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
Wednesday, August 24, 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
Joint Japanese-German Symposium Schedule 9
Thursday, August 25, 2011 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Session 1 Papers, Basic Studies 11
S-I.1 Modelling the ow regime in a simple urban-type street canyon . . . . . . . . . . . . . . . 15
S-I.2 Studying the eect of evaporative cooling on local conditions in a street canyon . . . . . . 23
S-I.3 Flow and concentration measurements in the wake of reduced scale models of cars for
developing nanoparticle dispersion models . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
S-I.4 Ventilation of an urban intersection characterized by advective and turbulent scalar uxes 39
S-I.5 Wind tunnel study on pollutant dispersion in street canyon with changing aspect ratio and
wind direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
S-I.6 Modelling of near-eld pollutant dispersion in the built environment: Methods and challenges 55
S-I.7 Wind tunnel measurements of the aerodynamic behavior of tree branches . . . . . . . . . 63
S-I.9 Wind tunnel study of the atmospheric boundary layer over vegetation-urban roughness
transition: Canopy modelling and preliminary results . . . . . . . . . . . . . . . . . . . . . 74
S-I.10 Investigation on transport of 2nd order moment of concentration in block arrays . . . . . 82
S-I.11 Proper orthogonal decomposition of velocity and vorticity eld within the street canyon
and wavelet analysis of expansion coecients . . . . . . . . . . . . . . . . . . . . . . . . . 90
S-I.12 Assessment of urban wind environments based on exceedance probability . . . . . . . . . 99
S-I.13 Simultaneous measurement of velocity and temperature in unstable turbulent boundary
layer and numerical analysis by LES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
S-I.14 New wind tunnel facility for agricultural research . . . . . . . . . . . . . . . . . . . . . . 118
S-I.15 Wind tunnel study of concentration uctuations in two merging plumes in dierent geo-
metrical congurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
S-I.16 Inuence of urban roughness on mean and turbulent wind elds in the city of Hamburg . 134
S-I.17 Comparisons of turbulence structures over various types of surface geometries . . . . . . 142
Session 2 Papers, Validation 149
S-II.1 Buoyant ow in a street canyon: Comparison of CFD simulations and wind tunnel mea-
surements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
S-II.2 Validation study of ow and concentration elds in a semi-idealized city . . . . . . . . . 161
S-II.3 LES of ow and plume dispersion within and over various obstacle arrays . . . . . . . . . 169
S-II.4 Turbulent ow around a surface-mounted cube: Time-resolved and time-averaged PIV
measurements and comparison with numerical simulations . . . . . . . . . . . . . . . . . . 177
S-II.6 Characterization of transient dispersion processes in an urban environment . . . . . . . . 185
S-II.8 LES of ow and plume dispersion in actual urban area in a spatially-developing boundary
layer ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
S-II.9 On aspects of LES validation for urban ow elds . . . . . . . . . . . . . . . . . . . . . . 201
S-II.10 Compiling a LES-specic validation data base from systematic wind tunnel modeling of
ow and dispersion phenomena within the lower atmospheric boundary layer . . . . . . . 209
S-II.11 Comparison between WRF calculations and observations of vertical proles of wind
velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
S-II.12 Models in wind energy assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
S-II.13 Model evaluation methodology for wind energy applications . . . . . . . . . . . . . . . . 237
Session 3 Papers, Applied Physical Modelling 245
S-III.1 Atmospheric dispersion simulation: The Dutch round-robin wind tunnel test . . . . . . . 249
S-III.2 Atmospheric dispersion wind tunnel simulation: The rst Dutch round robin-test at
Putten - Lessons learned in wind tunnel simulation and eld data analysis . . . . . . . . . 258
S-III.3 Atmospheric dispersion of trac exhaust emissions - A proposal for a theoretical model
and parameter estimation through eld data analysis . . . . . . . . . . . . . . . . . . . . . 266
1
S-III.5 Experimental study of the inuence of an idealized upstream ridge on the ow characte-
ristics over Alaiz mountain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
S-III.7 An experimental study on the wake of a wind turbine . . . . . . . . . . . . . . . . . . . 280
S-III.8 Measurements in the wake of an aerodynamically scaled wind turbine . . . . . . . . . . 288
S-III.10 Study on air pollutant dispersion under a sky train station in Bangkok . . . . . . . . . 296
S-III.11 Wind tunnel experiments on ow eld in real urban canyons . . . . . . . . . . . . . . . 304
S-III.12 Optimization of urban structures due to city ventilation and pedestrian level wind eld 312
S-III.13 LES of dispersion in a street canyon with tree planting . . . . . . . . . . . . . . . . . . 320
S-III.14 Scalar dispersion within a turbulent boundary layer (TBL) and investigation of the
TBL structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
S-III.15 Study of urban eects on the dispersion process in a wind tunnel . . . . . . . . . . . . 336
S-III.16 Investigation on characteristics of building wake dispersion for the nearby source by
using wind tunnel experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
S-III.17 Pressure measurements of the detachment bubble on the Bolund island . . . . . . . . . 351
Helpful Information 361
Conference Dinner Information 363
2
International Workshop on
Physical Modelling of Flow and Dispersion Phenomena
Monday, August 22, 2011
7.309.30 Registration
9.309.40 Welcome and general announcements
9.4010.20 Keynote Lecture
M. Schatzmann*
*Meteorological Institute, KlimaCampus, Hamburg University, Hamburg, Germany
10.2010.40 COFFEE BREAK
Basic Studies part 1
Session Chair: Alan Robins, University of Surrey, UK
10.4011.00 [S-I.1] Modelling the ow regime in a simple urban-type street canyon
E. Savory*, L. Perret, C. Rivet
*University of Western Ontario, London, Ontario, Canada
11.0011.20 [S-I.2] Studying the eect of evaporative cooling on local conditions
in a street canyon
S. Saneinejad*, P. Moonen, J. Carmeliet
*ETHZ, Zurich, Switzerland
11.2011.40 [S-I.3] Flow and concentration measurements in the wake of reduced scale
models of cars for developing nanoparticle dispersion models
M. Carpentieri*, P. Kumar, A. Robins
*Faculty of Engineering and Physical Sciences, University of Surrey, UK
11.4012.00 [S-I.4] Ventilation of an urban intersection characterized by advective
and turbulent scalar uxes
L. Kukacka*, R. Kellnerova, K. Jurcakov a, Z. Ja nour
*Institute of Thermomechanics AS CR, Prague, Czech Republic
*Charles University, Faculty of Mathematics and Physics, Prague, Czech Republic
12.0013.20 LUNCH BREAK
Session Chair: Eric Savory, University of Western Ontario, Canada
13.2013.40 [S-I.5] Wind tunnel study on pollutant dispersion in street canyon with
changing aspect ratio and wind direction
M. F. Yassin*, M. Ohba
*Department of Environmental Technology and Management, Kuwait Univ., Kuwait
*Faculty of Engineering, Assiut University, Assiut, Egypt
PHYSMOD2011 Schedule 3 Monday, 22 August
13.4014.00 [S-I.6] Modelling of near-eld pollutant dispersion in the built environment:
Methods and challenges
B. Hajra*, M. Chavez, T. Stathopoulos, A. Bahloul
*Department of Building, Civil and Environmental Engineering, Concordia
University, Montreal, Canada
14.0014.20 [S-I.7] Wind tunnel measurements of the aerodynamic behavior of tree branches
A. Schron*, B. Leitl, M. Schatzmann
14.2014.40 [S-I.8] Measurement on the ow and dispersion for a turbulent boundary layer
ow over cubic building array of in-line and staggered congurations
B.-S. Shiau*, F.-J. Tsai
*Institute of Physics, Academia Sinica, Taipei, Taiwan
*Department of Harbor and River Engineering, National Taiwan Ocean University,
Keelung, Taiwan
14.4015.00 [S-I.9] Wind tunnel study of the atmospheric boundary layer over vegetation-
urban roughness transition: Canopy modelling and preliminary results
L. Perret*, T. Ruiz
*Laboratoire de Mecanique des Fluides,

Ecole Centrale de Nantes, Nantes, France
Session Chair: Daniele Contini, Istituto di Scienze dellAtmosfera e del Clima, Italy
15.2015.40 [S-I.10] Investigation on transport of 2nd order moment of concentration
in block arrays
K. Nakao*, S. Kato, T. Takahashi
*University of Tokyo, Japan
15.4016.00 [S-I.11] Proper orthogonal decomposition of velocity and vorticity eld within
the street canyon and wavelet analysis of expansion coecients
R. Kellnerova*, L. Kukacka, Z. Ja nour
*Institute of Thermomechanics AS CR, Prague, Czech Republic
*Department of Meteorology and Environment Protection, Charles University,
Prague, Czech Republic
16.0016.20 [S-I.12] Assessment of urban wind environments based on exceedance probability
S. Kato*, Z. Bu, K. Hiyama
16.2016.40 [S-I.13] Simultaneous measurement of velocity and temperature in unstable
turbulent boundary layer and numerical analysis by LES
K. Katada*, R. Yoshie
*Tokyo Polytechnic University, Kanagawa, Japan
17.3020.00 WELCOME RECEPTION at EWTL
Ice breaker and laboratory tour
PHYSMOD2011 Schedule 4 Monday, 22 August
Tuesday, August 23, 2011
8.509.00 General announcements
Session Chair: Michael Schatzmann, University of Hamburg, Germany
9.009.20 [S-I.14] New wind tunnel facility for agricultural research
M. Fiedler*, K. v. Bobrutzki, M. Samer, W. Berg
*Leibniz-Institute for Agricultural Engineering, Potsdam, Germany
9.209.40 [S-I.15] Wind tunnel study of concentration uctuations in two merging
plumes in dierent geometrical congurations
D. Contini*, C. Elefante, F. M. Grasso, A. G. Robins
*ISAC-CNR, Istituto di Scienze dellAtmosfera e del Clima, Lecce, Italy
9.4010.00 [S-I.16] Inuence of urban roughness on mean and turbulent wind elds
in the city of Hamburg
C. Peeck*, B. Leitl, M. Schatzmann
10.0010.20 [S-I.17] Comparisons of turbulence structures over various types of surface geometries
H. Takimoto*, A. Sato, T. Michioka, A. Inagaki, M. Kanda
*Tokyo Institute of Technology, Tokyo, Japan
Validation part 1
Session Chair: Bert Blocken, Eindhoven University of Technology, The Netherlands
10.4011.00 [S-II.1] Buoyant ow in a street canyon: Comparison of CFD simulations
and wind tunnel measurements
J. Allegrini*, V. Dorer, J. Carmeliet
*Laboratory for Building Science and Technology, EMPA, D ubendorf, Switzerland
11.0011.20 [S-II.2] Validation study of ow and concentration elds in a semi-idealized city
G. C. Efthimiou*, D. Hertwig, F. Harms, J. G. Bartzis, B. Leitl
*Department of Mechanical Engineering, Univ. of West Macedonia, Kozani, Greece
11.2011.40 [S-II.3] LES of ow and plume dispersion within and over various obstacle arrays
H. Nakayama*, K. Jurcakova, H. Nagai
*Japan Atomic Energy Agency, Ibaraki, Japan
11.4012.00 [S-II.4] Turbulent ow around a surface-mounted cube: Time-resolved and time-
averaged PIV measurements and comparison with numerical simulations
E. Paterna*, S. S. Chikatamarla, P. Moonen,
V. Dorer, J. Carmeliet, I. V. Karlin
*Institute of Technology in Architecture, ETHZ, Zurich, Switzerland
PHYSMOD2011 Schedule 5 Tuesday, 23 August
Validation part 2
Session Chair: Kyosuke Hiyama, University of Tokyo, Japan
13.2013.40 [S-II.5] Wind tunnel evaluation of inverse modelling techniques for
emergency response application
A. Robins*, P. Hayden, A. Rudd, S. Belcher
*EnFlo Laboratory, University of Surrey, UK
13.4014.00 [S-II.6] Characterization of transient dispersion processes in an urban environment
F. Harms*, D. Hertwig, B. Leitl, M. Schatzmann
14.0014.20 [S-II.7] Modeling of convective heat transfer coecient from urban canyon surfaces
by experimental and numerical simulation
S. S. Pillai*, R. Yoshie, J. Chung
14.2014.40 [S-II.8] LES of ow and plume dispersion in actual urban area in a
spatially-developing boundary layer ow
H. Nakayama*, B. Leitl, H. Nagai, F. Harms
*Japan Atomic Energy Agency, Ibaraki, Japan
14.4015.00 [S-II.9] On aspects of LES validation for urban ow elds
D. Hertwig*, F. Harms, G. Patnaik, M.-Y. Obenschain, B. Leitl
M. Schatzmann
Validation part 3
Session Chair: Sandrine Aubrun, University of Orleans, France
15.2015.40 [S-II.10] Compiling a LES-specic validation data base from systematic
wind tunnel modeling of ow and dispersion phenomena within
the lower atmospheric boundary layer
R. Fischer*, I. Bastigkeit, B. Leitl, M. Schatzmann
15.4016.00 [S-II.11] Comparison between WRF calculations and observations of vertical proles
of wind velocity
M. Mochizuki*, R. Yoshie, J. Chung
16.0016.20 [S-II.12] Models in wind energy assessment
G. Petersen*, U. Gahde, M. Homann, B. Leitl, M. Schatzmann
16.2016.40 [S-II.13] Model evaluation methodology for wind energy applications
H. A. Holmes*, M. Schatzmann, B. Leitl
18.0022.00 WORKSHOP DINNER
PHYSMOD2011 Schedule 6 Tuesday, 23 August
Wednesday, August 24, 2011
8.509.00 General announcements
Applied Studies part 1
Special Session
Session Chair: Jeroen van Beeck, Von Karman Institute for Fluid Dynamics, Belgium
9.009.20 [S-III.1] Atmospheric dispersion simulation: The Dutch round-robin wind tunnel test
S. W. van Ratingen*, N. L. S. Moonen, K. Artois
*TNO, Utrecht, The Netherlands
9.209.40 [S-III.2] Atmospheric dispersion wind tunnel simulation: The rst Dutch round
robin-test at Putten Lessons learned in wind tunnel simulation and
eld data analysis
N. L. S. Moonen*, S. P. M. van den Akker, J. F. W. Koopmans
*Peutz bv, The Netherlands
9.4010.00 [S-III.3] Atmospheric dispersion of trac exhaust emissions - A proposal for a
theoretical model and parameter estimation through eld data analysis
S. P. M. van den Akker*, J. F. W. Koopmans
10.0010.20 [S-III.4] Atmospheric dispersion wind tunnel simulation: The second Dutch round
robin-test urban situation Overschie
N. L. S. Moonen*, S. P. M. van den Akker, J. F. W. Koopmans
Session Chair: Shinsuke Kato, University of Tokyo, Japan
10.4011.00 [S-III.5] Experimental study of the inuence of an idealized upstream ridge on the
ow characteristics over Alaiz mountain
B. Conan*, J. van Beeck, S. Aubrun
*Von Karman Institute for Fluid Dynamics (VKI), Rhode-St-Gen`ese, Belgium
11.0011.20 [S-III.6] Wind turbine wakes in oshore neutral and stratied atmospheric
boundary layer conditions
S. Zhang, P. E. Hancock*
*EnFlo Laboratory, University of Surrey, UK
11.2011.40 [S-III.7] Design of an aerodynamically scaled wind turbine
F. Cuzzola*, M. Dorenkamper, B. Leitl, M. Schatzmann
11.4012.00 [S-III.8] Measurements in the wake of an aerodynamically scaled wind turbine
M. Dorenkamper*, F. Cuzzola, B. Leitl, M. Schatzmann
PHYSMOD2011 Schedule 7 Wednesday, 24 August
Session Chair: Matteo Carpentieri, University of Surrey, UK
13.2013.40 [S-III.9] Ventilation and air quality investigation of an urban square
M. Balczo*, T. Lajos
*Theodore von Karm an Wind Tunnel Laboratory, Department of Fluid Mechanics
Budapest University of Technology and Economics, Hungary
13.4014.00 [S-III.10] Study on air pollutant dispersion under a sky train station in Bangkok
K. Hiyama*, T. Hoshiko, S. Kato, T. Prueksasit
*Institute of Industrial Science, University of Tokyo, Japan
14.0014.20 [S-III.11] Wind tunnel experiments on ow eld in real urban canyons
T. Katsuki*, A. Sato, T. Michioka, A. Hagishima
*Central Research Institute of Electric Power Industry, Japan
14.2014.40 [S-III.12] Optimization of urban structures due to city ventilation
and pedestrian level wind eld
F. Kipsch*, B. Leitl, F. Harms, S. Werk
14.4015.00 [S-III.13] LES of dispersion in a street canyon with tree planting
P. Moonen*, C. Gromke, V. Dorer, J. Carmeliet
Session Chair: Bernd Leitl, University of Hamburg, Germany
15.2015.40 [S-III.14] Scalar dispersion within a turbulent boundary layer (TBL)
and investigation of the TBL structure
C. Nironi*, P. Salizzoni, P. Mejean, N. Grosjean, L. Soulhac, F. X. Cierco
*Laboratoire de Mecanique des Fluides et dAcoustique, Universite de Lyon, France
15.4016.00 [S-III.15] Study of urban eects on the dispersion process in a wind tunnel
A. R. Wittwer*, A. M. Loredo-Souza, E. B. Cama no Schettini
*Facultad de Ingeniera, UNNE, Resistencia, Argentina
16.0016.20 [S-III.16] Investigation on characteristics of building wake dispersion for
the nearby source by using wind tunnel experiment
J. Yoneda*, K. Okabayashi, E. Hori
*Mitsubishi Heavy Industries, Ltd., Japan
16.2016.40 [S-III.17] Pressure measurements of the detachment bubble on the Bolund island
T. S. Yeow*, A. Cuerva, J, Perez
*Instituto Ignacio Da Riva, Universidad Politecnica de Madrid, Madrid, Spain
16.4017.00 Closing remarks
PHYSMOD2011 Schedule 8 Wednesday, 24 August
Joint Japanese-German Symposium on
Urban Areas in a Changing Climate
Thursday, August 25, 2011
10.0010.30 Welcome addresses
Symposium part 1
10.3011.15 Towards climate change adaptation in civil engineering
Kazuyoshi Nishijima*
*Department of Civil Engineering, Technical University of Denmark, Denmark
11.1512.00 Changes in European storm climate of the past decades
Frauke Feser*
*Institute for Coastal Research, Helmholtz-Zentrum Geesthacht, Germany
Symposium part 2
12.4513.30 Ventilating cities
Shinsuke Kato*
13.3014.15 Approaches for investigating the current and future urban climate
Heinke Schl unzen*
14.3016.00 COFFEE in the wind-tunnel laboratory
Joint Japanese-German Symposium 9 Thursday, 25 August
Session 1 Papers
S-I. Basic Studies
Monday, August 22
10.4011.00, S-I.1
11.0011.20, S-I.2
11.2011.40, S-I.3
11.4012.00, S-I.4
13.2013.40, S-I.5
13.4014.00, S-I.6
14.0014.20, S-I.7
14.2014.40, S-I.8
14.4015.00, S-I.9
15.2015.40, S-I.10
15.4016.00, S-I.11
16.0016.20, S-I.12
16.2016.40, S-I.13
Tuesday, August 23
9.009.20, S-I.14
9.209.40, S-I.15
9.4010.00, S-I.16
10.0010.20, S-I.17
11
NOTES AND COMMENTS:
12
NOTES AND COMMENTS:
13
NOTES AND COMMENTS:
14
PHYSMOD2011 International Workshop on Physical Modeling of Flow and Dispersion Phenomena
KlimaCampus, University of Hamburg, Germany August 22-24, 2011
Modelling the flow regime in a simple
urban-type street canyon
Eric Savory
a
, Laurent Perret
b
, Cdric Rivet
b

a
Univ of Western Ontario, London, Ontario, Canada, esavory@eng.uwo
b
Laboratoire de Mecanique des Fluides, Ecole Centrale de Nantes (ECN)
Nantes, France, laurent.perret@ec-nantes.fr, cedric.rivet@ec-nantes.fr
ABSTRACT: The present work compares the mean flow within a model street canyon, of
width / height (W/h) = 1 and variable length L/h = 1 20, immersed in an array of cubic
roughness elements of the same height, with results from other studies where the upstream
roughness elements are much smaller than the canyon buildings. Then, for the L/h = 14 case
the dynamics of the canyon flow are examined using two point spatial correlations and Proper
Orthogonal Decomposition, with the results revealing a sequence of penetration and
flushing events that ventilate the canyon.
1 INTRODUCTION
The aims of the present work are; (1) to compare the flow in a simple street canyon
(streamwise width/height = 1, but of varying length/height from 1 to 20), when subjected to a
simulated boundary layer generated over an array of obstacles of similar height to the canyon,
to previous results where the canyon is much larger than the upstream roughness (e.g.
Kastner-Klein et al., 2004, Simoens et al., 2007) and (2) to use this arrangement to examine
the influence of the large turbulent structures in the shear layer on the intermittent flow
behaviour within the canyon (utilizing two point correlations and proper orthogonal
decomposition of the velocity field). Little is known about the unsteady flow dynamics which
is believed to be as important as the mean flow due to the high levels of turbulence occurring
in urban areas. Recent studies (e.g. Castro et al., 2006, Barlow and Leitl, 2007, Coceal et al.,
2007a,b) highlight the strong unsteadiness of the flow developing at the building roof and its
role in generating intermittent coherent turbulent structures which penetrate downwards,
causing mixing of air in the street.
The dynamics of the flow inside and at the top of a 2-D aerodynamic cavity (that is, a surface
cut-out) has been shown to be strongly influenced by the level of upstream turbulence
(Chang et al., 2006; Haigermoser et al., 2008; Kang and Sung, 2009). In particular, the ratio
between the depth of the cavity and the momentum thickness of the oncoming boundary layer
has been found to be a key parameter in the dynamics of the shear layer at the top of the
cavity (Kang and Sung, 2009). In the case of the urban canyon, the typical scale of the
coherent structures of the oncoming boundary layer are larger than the depth of the canyon
(Castro et al., 2006) and so the dynamics of the flow are expected to be quite different. The
present work is an experimental study of the turbulent flow within and above an urban
canyon, whose main goal is to quantify the dynamics of the shear layer emanating from the
top of the upstream obstacle and the occurrence of unsteady penetration and ejection of fluid
inside the canyon.

15
2 EXPERIMENTAL DETAILS
The tests were conducted in the low-speed, boundary layer wind tunnel at Ecole Centrale de
Nantes, which has working section dimensions of 2 m (width) x 2 m (height) x 24 m length,
and an empty-tunnel freestream turbulence intensity of 0.5% with good spanwise uniformity
to within 5%. The best match to ESDU data gave a 1:200 scale suburban boundary layer
simulation (roughness length z
0
= 0.20 m) at the test location when using 3 vertical, tapered
spires over the entire height of the inlet to the working section, followed by an 18.5 m fetch
of staggered 100 mm cube roughness elements (plan area density of 6.25%) and a 300 mm
high solid fence across the working section 1.5 m downstream of the inlet. A freestream
velocity of 5.9 m/s was used in all of the experiments (see Table 1) and the boundary layer
mean velocity and turbulence profiles were measured using crossed hot-wire anemometry,
with the results shown in Figure 1 in full-scale units, together with corresponding ESDU
profiles, ESDU (1982, 1985). The experimental data are vertical profiles averaged over five
lateral positions that encompass one period of the block spacing within the array. The vertical
profile of the streamwise turbulence integral length scale (L
u
), together with the ESDU
(1985) profile, (Fig. 1) shows that except in the lower part of the boundary layer the
experimental set-up does not precisely model the larger scales. In general, comparison of the
spectral density (not shown here) at different full-scale heights with ESDU (1985) show that
the turbulence is modeled well in the boundary layer. The Jensen number scaling is not exact
since the model-scale roughness length of z
0
= 2.2 mm would imply a rougher value of 0.44
m at full-scale than the 0.20 m achieved here. Nevertheless, the agreement is within the
acceptable 2-3 factor between model and full-scale. Thus, the scaling with ESDU suggests a
full-scale canyon height of 20 m, which is somewhat large but not too unreasonable, whereas
Jensen number scaling would give a height of 9.1 m.
Figure 2 shows the canyon geometries examined within the array of roughness elements that
had the same height as the two blocks forming the canyon. The streamwise width / height
was W/h = 1 whilst the lateral length was varied from L/h = 1 to 20 (the largest value
representing a nominally 2D canyon that spanned the working section width). The velocity
fields were measured in vertical planes aligned with the freestream flow direction by a two-
component Dantec Dynamics Particle Image Velocimetry (PIV) system, as discussed in
Savory et al. (2011). The final spatial resolution of the vector field was 1.5 - 2 mm (that is
0.015 - 0.02h) depending on the specific experiment.
The present results are compared with the Laser Doppler Velocimetry data of Kastner-Klein
et al. (2004) and Barlow and Leitl (2007) and the PIV data of Simoens et al. (2007) and Sato
et al. (2009), with the parameters of the different experimental arrangements as summarised
in Table 1 below. Key differences between the present work and the most extensive study,
that of Kastner-Klein et al. (2004), are the larger roughness elements, much higher turbulence
length scales throughout the boundary layer and significantly greater values for the boundary
layer thickness (/h) and roughness length (z
0
/h) in the present work. In addition, some of the
experiments of Kastner-Klein et al. (2004) had very high model area blockage ratios of 12-
16% which are likely to influence the flow regime, whereas the maximum blockage in the
present study was 5%. However, it will be seen that the Reynolds number is similar in both
cases as are the maximum turbulence intensity (defined using u/u
*
) and the normalised
friction velocity (u
*
/U
e
). It should be noted that the integral length scale profile of Kastner-
Klein et al. (2004) does not show any increase in scale with height, due to the smaller
roughness elements and the absence of full-height spires. It is stated that their best-fit scale
ratio is 1:500, implying full-scale value of z
0
= 0.4m, giving a poor scaling with the canyon
(that would be 60m high). However, acceptable agreement is claimed if 1:150 is used,
giving a z
0
of 0.1m. The stated scale of Barlow and Leitl (2007) is 1:400.
16
Table 1: Comparison of boundary layer parameters for the different studies (all with canyon width W/h = 1)
Present Kastner-
Klein et al.
(2004)
Simoens et
al.
(2007)
Sato et al.
(2009)
Barlow
and Leitl
(2007)
Method Full-height spires,
300mm fence, 100
mm cube roughness
Part-height
spires, small
saw-tooth
fence and
roughness
Smooth wall,
sandpaper
and 5mm
diameter rod
trip
75mm
roughness,
25%,
regular
cube array
Spires
and, Lego
roughness
Thickness, (mm) 1165 480 70 250 300
Displacement ht., d, (mm) 54.8 2.0 -- 68.3 --
Roughness lgth., z
0
(mm) 2.2 0.8 -- 3.0 4.2
Ref. velocity, U
e
(m/s) 5.90 7.00 2.30 0.65 6.10
Friction velocity, u
*
(m/s) 0.345 0.385 0.111 0.065 0.440
u
*
/ U
e
0.058 0.055 0.048 0.100 0.072
Velocity at canyon roof
height, U
h
(m/s)
2.7 4.9 1.6 (est.) 0.46 3.2
Max. turb int. u / u
*
2.3 2.4 2.8 1.5 2.2
Maximum integral length
scale, L
u
(m)
0.76 0.29 -- -- --
L
u
(m) at z /h = 1 0.37 0.25 -- -- --
Re (using h and U
e
) 4.0 x 10
4
5.6 x 10
4
1.5 x 10
3
5.2 x 10
3
2.6 x 10
4

Canyon height, h (mm) 100 120 10 120 63
/ h 11.6 4.0 7.0 2.1 4.76
z
0
/ h 0.022 0.007 n/a 0.025 0.067
Canyon length, L / h 1,5,7,9,11,13,14, 20 5,10,15 50 6 24
3 RESULTS AND DISCUSSION
3.1 Comparison of canyon mean flow for different canyon lengths (L/h)
The mean flow data of Kastner-Klein et al. (2004), for L/h = 5, 10 and 15 are very interesting
(Savory et al., 2011). The mean flow vortex for the shortest model (L/h = 5) is located
towards the downstream lower corner of the canyon, whereas the other two cases show a
more centred vortex. The L/h = 5 and 15 cases show a very short separation region on top of
the upstream block (i.e. the flow is attached over the block width, whereas the flow for L/h =
10 is separated over the upstream block, giving a stronger canyon recirculation. Their data for
sloped roof obstacles does not show such variations over this L/h range, suggesting that when
the upstream roughness heights are << canyon block height the canyon flow is sensitive to
the experimental set-up due to its influence on the canyon leading edge separation.
Figure 3 shows the maximum mean streamfunction (), associated with the canyon flow
recirculation, normalised by the undisturbed velocity at canyon roof height (h). It is seen that
this integrated parameter is largely independent of the arrangement of the last row of
roughness in the present work since the normal and shifted (S) (where the order of
elements in the final row of roughness before the canyon is reversed) block results are
essentially the same. The data show a change in flow regime between L/h = 1 and 5 but
relatively little change after L/h = 9 is reached. This is confirmed by the maximum
streamfunction location (representing the centre of the recirculation) which also remains
fairly constant after L/h = 9, as shown in figure 4. It is not possible to collapse the data from
the different studies using any velocity scale (e.g. u
*
, U
e
), although when using U
h
the data of
Simoens et al. (2007) appear to represent an asymptotic condition in relation to the present
17
results, with /U
h
h 0.1 being the limiting value. The data of Kastner-Klein et al. (2004)
show dramatic changes from L/h = 5 to L/h = 10 and then again to L/h = 15, resulting in no
clear trend in maximum streamfunction value. There is also no trend in their maximum
streamfunction location data. Only at L/h = 10 do their data show a well-centred recirculation
(and reasonably close to the present L/h = 9 and 11 cases) and it is only for that canyon
length where there is agreement in maximum streamfunction value when U
h
is used for the
normalisation. The 2-D case of Barlow and Leitl (2007) shows a more centred vortex than
in the present work (for the L/h = 20 case) and both are different from the smooth-wall, large
aspect ratio (L/h = 50) data of Simoens et al., (2007), suggesting that a 2-D case (and,
hence, a set of profiles) cannot be defined for canyon flows. Until these discrepancies are
resolved and understood such tests cannot be used as a reliable tool to define even just the
mean canyon flow regime where the specific application is to full-scale urban canyons.
3.2 The dynamics of the canyon with L/h = 14
The two-point coefficient correlation of the streamwise velocity components is defined as:
( )
( ) ( )
( ) ( )
) 1 (
z , x ' u z , x ' u
z , x ' u z , x ' u
z , x , z , x R
2
ref ref
2
ref ref
ref ref uu
=

where (x
ref
, z
ref
) are the coordinates of the fixed point. Vertical (w) correlations are computed
in a similar way. Correlations within the canyon (not shown here) show that it is decoupled
from the shear layer and the flow above it, in agreement with the skimming flow regime.
Figure 5 shows a correlation map obtained in the shear layer (z
ref
/h = 1.2). Both R
uu
and R
ww

are mostly positive with some negative values (black solid lines correspond to a correlation of
zero) inside the canyon, consistent with the clockwise recirculation. R
ww
shows negative
levels above the upstream obstacle, interpreted as the footprint of large-scale vortical
structures developing inside the shear layer emanating from the upstream corner of the
upstream building. The origin of the shear layer is the main difference between cavity flow
configurations where the shear layer starts from the downstream corner (Kang and Sung,
2009) and the present case. This is confirmed by the correlations near the top of the obstacles
(not shown here) that indicate that, contrary to many cavity flows (Kang and Sung, 2009),
large-scale vertical motion of fluid exists in this region rather than well-organised vortex
shedding. Correlations computed above the shear layer have a shape and length scales in
agreement with those in flows developing above urban-like roughness (Coceal et al., 2007b).
The correlation maps have two distinct features: shape and inclination. R
uu
is more elongated
in the streamwise direction and exhibits a preferential inclination; R
ww
is more isotropic, with
shorter length scales. To quantify the variation of these characteristics with the location
within the flow, ellipses were fitted to a given contour extracted from R
uu
and R
ww
calculated
for different values of x
ref
and z
ref
, as shown in Figure 6. The solid lines correspond to the
limits of the shear layers developing from the upstream obstacle, identified from the local
maxima of the 2-D TKE gradient, TKE/z (Cardwell et al., 2011). In the shear layer region,
the inclination and the length scales of R
uu
decrease and increase, respectively, with the x-
distance from the upstream edge of the upstream obstacle, following the growth of the shear
layer. Within the shear layer, at a given x-location, length scales are constant with z. Above
the shear layer, the characteristic length scales progressively increase with z, in accordance
with length scale evolution in a boundary layer over large roughness (Coceal et al., 2007b).
Thus, on average, three distinct regions exist, delimited by the shear layers from the upstream
building and characterized by different length scales and orientation of the structures.
To investigate the shear layer flapping, the dynamics of the vertical component w at z/h = 1
was examined using Proper Orthogonal Decomposition (POD) (Lumley, 1967). Using
18
R
ww
(x,x,z=h), a 1-D analysis was performed, decomposing the vertical velocity component
into a set of spatial and temporal eigenmodes
w
n
(x) and a
n
(t), respectively, such that:
( ) ) 2 ( ) x ( t a ) t , x ( w
n
w
N
1 n
n
=
=
The spatial modes
w
n
(x) form an orthonormal basis of the flow and the temporal coefficients
a
n
(t) represent the dynamics of the corresponding modes. The eigenvalue
n
represents the
contribution of the n
th
mode to the total turbulent energy. The energy content of the first two
POD modes, given by the first two eigenvalues (Fig. 7), represents more than 60% of the
total energy associated with the w-fluctuations. The spatial shape of these two modes (Fig. 8,
left) corresponds to strong penetration or ejection through the canyon opening. Analysis of
higher order modes (not shown here) did not suggest the existence of organized vortex
shedding. These two most energetic modes are also found to contain most of the energy
during 70% of the time (e.g. the maximum mode amplitude is either a
1
(t) or a
2
(t), 70% of the
time), such that the flow dynamics are driven by these two modes. The phase portrait of a
1
(t)
and a
2
(t) when either is at a maximum reveals that they are essentially located around a circle
(Fig. 8, right), such that a strong dynamical relationship exists between the two modes,
confirming the temporal coherence of the penetration and ejection motions observed from the
instantaneous velocity fields. The phase angle between a
1
(t) and a
2
(t) is defined by:
) 3 (
) t ( a
) t ( a
arctan
1
2
2
1
a a
2 1

Phase averaged velocity fields are computed by averaging instantaneous velocity fields which
correspond to the instantaneous angle
a1-a2
falling into a given range. The whole range of
possible angles [0, 2] was divided into 18 sub-intervals. Analysis of the 18 phase-averaged
velocity fields shows that a strong flapping of the flow exists above the canyon, which
corresponds to penetration (Figs. 9a,b) and ejection (Figs. 9c,d) of fluid through the canyon
opening. In order to investigate further the organisation of the flow, vortical structures,
identified by a positive value of the swirling strength
ci
were extracted from the
instantaneous velocity fields and phase averaged. The
ci
distributions show that the vertical
fluid motions throughout the canyon opening are accompanied by vortical structures (figure
9). Based on the occurrence of ejection (Q2) or penetration (Q4) events, conditionally
averaged velocity field and swirling strength signed by the vorticity sign
ci
s
were computed
from the PIV data measured above the canyon. Q2 or Q4 events occurring in the shear layer
region, along the canyon centre-line (Figs. 10, left and right, respectively), turned out to be
strongly correlated to a clockwise rotating vortical motion (
ci
s
< 0) existing in the shear layer
above the top of the obstacles, in agreement with the previous POD analysis. The flapping
motion of the shear layer is also clearly evidenced here. Figure 10, left, also suggests that the
vortical structures associated with strong Q2 events in the shear layer are convected
downstream and may participate in the generation of vortical structures populating the
boundary layer. When based on a Q2 or Q4 event occurring well above the shear layer, the
conditionally-averaged swirling strength distribution is found to be almost independent of the
nature of the conditional event, showing that the dynamics of the flow just above the canyon
is primarily driven by the dynamics of the shear layer separating from the upstream building.
4 CONCLUSIONS
The finer details of the approach flow conditions have a strong effect on the canyon flow,
even when comparing cases where similar geometries are fully immersed in thick turbulent
19
boundary layers. These effects are not only in the magnitudes of the mean and turbulent
velocities but also in the shape of the mean flow pattern.
The shear layer that develops over the top of the two-obstacle canyon generates, as in a
classical mixing layer, spanwise vorticity, the length scale of which increases with
downstream distance. In addition, the shear layer region is animated by a flapping motion that
is responsible for the unsteady flow exchanges between the canyon and the outer flow. The
strong Q2 and Q4 events above the canyon opening (heights up to z = 2h), are strongly
correlated to the shear layer. These should be considered more as penetration and
flushing events rather than the sweeps and ejections related to the hairpin vortex
packet model. The presence of the shear layer tends to decouple the flow within the canyon
from the dynamics of the upstream boundary layer.
5 ACKNOWLEDGEMENTS
The work was carried out whilst E Savory was on sabbatical at ECN, funded by the Region
Pays de la Loire. Thanks are due to P Klein, S Simoens, J Barlow and A Sato (and
colleagues) for very kindly providing the data files from their experiments.
6 REFERENCES
Barlow, J. F., Leitl, B., 2007. Effect of roof shapes on unsteady flow dynamics in street canyons. International
Workshop on Physical Modelling of Flow and Dispersion Phenomena (PHYSMOD 2007), Orlans, France.
Cardwell, N. D., Vlachos, P. P., Thole, K. A., 2011. Developing and fully developed turbulent flow in ribbed
channels. Experiments in Fluids 50, 13571371.
Castro, I.P., Cheng, H., Reynolds, R., 2006. Turbulence over urban-type roughness: deductions from wind-
tunnel measurements, Boundary-Layer Meteorology, 118, 109131.
Chang, K, Constantinescu, G & Park, S 2006 Analysis of the flow and mass transfer processes for the
incompressible flow past an open cavity with a laminar and a fully turbulent incoming boundary layer.
Journal of Fluids Mechanics 561, 113145.
Coceal, O., Dobre, A., Thomas, T. G. 2007a. Unsteady dynamics and organized structures from DNS over an
idealized building canopy. International Journal of Climatology 27, 19431953.
Coceal, O., Dobre, A., Thomas, T. G., Belcher, S. E., 2007b. Structure of turbulent flow over regular arrays of
cubical roughness, Journal of Fluids Mechanics, 589, 375409.
ESDU, 1982. Strong winds in the atmospheric boundary layer. Part I: mean-hourly wind speeds. Data Item
82026 (amended 1993), Engineering Sciences Data Unit International.
ESDU, 1985. Characteristics of atmospheric turbulence near the ground. Part II: single point data for strong
winds (neutral atmosphere). Item 85020 (amended 1993), Engineering Sciences Data Unit International.
Haigermoser, C., Vesely, L., Novara, M., Onorato, M. 2008. A time-resolved particle image velocimetry
investigation of a cavity flow with a thick incoming turbulent boundary layer. Physics of Fluids 20, 114.
Kang,W., Sung, H. J. 2009. Large-scale structures of turbulent flows over an open cavity. Journal of Fluids and
Structures 25, 13181333.
Kastner-Klein, P., Berkowicz, R., Britter, R., 2004. The influence of street architecture on flow and dispersion in
street canyons, Meteorology and Atmospheric Physics, 87, 121131.
Lumley, J. L., 1967. The structure of inhomogeneous turbulence. Atmosphere, Turbulence and Radio Wave
Propagation. Nauka, Moscow.
Sato, A., Takimoto, H, Michioka, T., 2009. Impact of wall heating on air flow in urban street canyons.
International Workshop on Physical Modelling of Flow and Dispersion Phenomena (PHYSMOD 2009),
Brussels, Belgium, E4.1-E4.7.
Savory, E., Perret, L., Rivet, C. 2011. Modelling considerations for examining the mean and unsteady flow in a
simple urban-type street canyon. Proc 13
th
International Conference on Wind Engineering, Amsterdam, July.
Simoens, S., Ayrault, M., Wallace, J. M., 2007. The flow across a street canyon of variable width - Part 1:
Kinematic description, Atmospheric Environment, 41, 90029017.
20
z
x
W
h
Roughness
cube
Canyon
block
Canyon
block
Flow
Elevation
Plan
L
Tunnel wall
Canyon length L / h
0 10 20 30 40 50 60

m
a
x

/

U
h

h
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
Present data
Present data (blocks shifted)
Kastner-Klein et al (2004)
"2D" case - Simoens et al (2007)
Sato et al (2009)
x / h
-0.5 0.0 0.5
z

/

h
0.0
0.5
1.0
Present
Kastner-Klein et al (2004)
Simoens et al (2007)
Sato et al (2009)
Barlow and Leitl (2007)
1
1S
5
7
9
14,14S
11,13
5
10
15
50
20
6
24
x / h x / h
z

/

h
z

/

h

Figure 1: Mean velocity profile (left), turbulence intensity (centre), integral length scale (right) comparisons

Figure 2: Roughness array, canyon Figure 3: Variation with L/h of max. Figure 4: Location of max stream
geometry (LMF at ECN, Nantes) streamfunction on centre-line function (labels are L/h values)

Figure 5: Two-point correlation maps for reference at Figure 6: Shape of the two-point correlation for
x
ref
/h = 0, z
ref
/h = 1.2. Left: R
uu
(x
ref
,z
ref
,x,z), right: different locations of the reference point. Left:
R
ww
(x
ref
,z
ref
,x,z). Black solid line, zero-contour level, R
uu
, right: R
ww
. For clarity, ellipses drawn half
white solid line, level used for ellipse fitting actual size. Solid line: shear layer boundaries
Mean velocity U / U
100
0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4
z

(
m
)
0
50
100
150
200
250
U / U
100
(Expt, 1:200)
U / U
100
(ESDU, z
0
=0.2m)
Normalised turbulence intensity
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
z

(
m
)
0
50
100
150
200
250
u'/U (Expt, 1:200)
u'/U (ESDU, z
0
=0.2m)
w'/U (Expt, 1:200)
w'/U (ESDU, z
0
=0.2m)
-uw/U
2
(Expt, 1:200)
-uw/U
2
(ESDU, z
0
=0.2m)
L
u
(m)
0 50 100 150 200 250
z

(
m
)
0
20
40
60
80
100
120
140
ESDU (1985) (z
0
=0.2m)
Experiment (1:200)
x / h x / h
z

/

h
z

/

h
21
Figure 7: Energy contribution of POD Figure 8: First two POD modes
w
1
(x) and
w
2
(x) (left) and phase
eigenvalues
n
portrait of POD coefficients a
1
(t) and a
2
(t) (right)

Figure 9: Examples of phase averaged velocity fields (every 3
rd
vector shown) corresponding to, a) and b),
penetration of fluid and, c) and d), ejection of fluid. The mean velocity field has been subtracted for clarity.
Contours: phase-averaged swirling strength
ci
; blue corresponds to zero level, red to high value of
ci

Figure 10: Conditional average of signed swirling strength distribution
ci
s
based on, left, a Q2 event and, right,
a Q4 event, occurring at x/h = 0, z/h = 1.37
x / h a
1
(t)
a
2
(
t
)
x / h x / h
x / h x / h
z

/

h
z

/

h
z

/

h
z

/

h
x / h x / h
z

/

h
z

/

h
22
Introducing a coupled model to simulate drying of a
porous wall in a street canyon and the resulting
evaporative cooling
Saba Saneinejad
a
, Peter Moonen
b
, Jan Carmeliet
b

a
Chair of Building Physics, ETH, Zrich, Switzerland,
Saneinejad@arch.ethz.ch
b
Laboratory for Building Science and Technology, Empa, Dbendorf,
Switzerland, Peter.Moonen@empa.ch
a
Chair of Building Physics, ETH, Zrich, Switzerland,Carmeliet@arch.ethz.ch
a
Laboratory for Building Science and Technology, Empa, Dbendorf,
Switzerland, Jan.Carmeliet@empa.ch

ABSTRACT: The urban heat island affects the energy use for cooling in an urban
environment, as well as human comfort and health. Water evaporation from moist surfaces
could potentially reduce the local temperature in urban areas. We have studied the effect of
evaporative cooling on the temperature conditions in an urban street canyon by introducing a
computational model for determining convective heat and mass exchange between the
canyon walls and the air. The model couples two sub-models: (i) a Computational Fluid
Mechanics (CFD) model, which solves heat and moisture transfer in the air, and (ii) a
Building Envelope Heat, Air and Moisture (BE-HAM) transport model which solves heat and
moisture transfer within the porous building walls. Using this new coupled method, drying of
a wet windward wall of a street canyon is studied. The effect of evaporation on the
temperature reduction in a street canyon is analyzed and is shown to be important.
1 INTRODUCTION
A heat island is an urban area which can be several degrees warmer than its surrounding rural
hinterland. The urban heat island affects the energy use for cooling in an urban environment,
as well as human comfort and health. One way to mitigate the excess heat is to make use of
evaporative cooling, for example from ground-level ponds (Krger and Pearlmutter, 2008)
and roof ponds (Runsheng et al., 2003; Tiwari et al., 1982), from surfaces wetted by wind-
driven rain (Blocken et al., 2007), or from vegetated surfaces (Alexandri and Jones, 2008).
Despite the vast amount of research conducted to date, the interaction between the wind flow
pattern in the urban environment, and the evaporation or drying rate of the urban surfaces, as
well as the resulting cooling potential, is not yet fully understood.

Two key parameters in studying the hygrothermal behavior of porous material are the
convective heat and mass transfer coefficients (CHTC and CMTC respectively). CHTC
relates the heat flux normal to a surface (q
c,h,w
, expressed in W/m
2
) to the temperature
difference between the surface and a reference location, while CMTC relates the mass flux
normal to the surface (q
c,m,w
, expressed in kg/m
2
s) to a difference in vapor pressure:

, , c h w
w ref
q
CHTC
T T
=
;
, , c m w
w ref
q
CMTC
p p
=
(1)
23

with T (K) the absolute temperature and p (Pa) the vapor pressure. Subscript c relates to
convective, h to heat, m to moisture, w to (wall) surface values and ref to values at the
reference location.

Based on the Richardson number, thermal convection can be subdivided into regimes of
natural and forced convection. For the forced convective flow regime, the CHTC is usually
correlated to the wind speed at a reference location by means of a power-law correlation
derived from wind-tunnel experiments on flat plates or wall-mounted bluff bodies, or from
full-scale experiments on specific building surfaces. An overview is given in Defraeye et al.
(2011a) and Palyvos (2008). The choice of the reference location is very critical in these
correlations. Also normally a single CHTC value for each surface is considered and the
spatial distribution over the surface is not taken into account.

The spatial distribution of convective transfer coefficients (CTCs) have been studied by
means of Computational Fluid Dynamics (CFD) (Defraeye et al., 2010; Emmel et al., 2007;
Saneinejad et al., 2011a). Figure 1 shows the distribution of CTCs along the walls of a street
canyon. It can be seen that CTCs vary depending on the considered wall and on the location
(height) on the wall. Adopting a single CTC value for the entire wall is therefore less
appropriate.

Figure 1: Vertical distribution of (a) CHTC and (b) CMTC along the windward (WW) and leeward wall (LW)
of a street canyon. Inlet: 20C, 20% RH, U
10
= 5 m/s; wall under study: 30C, capillary saturated with water;
other surfaces: adiabatic and impermeable to moisture.

To show the varying nature of the CTCs, Figure 2 depicts the CHTCs along the walls of a
street canyon for mixed conditions, including forced and buoyant cases. The windward wall
of the street canyon is heated while considering eight temperature differences between
incoming air and the canyon wall. In Figure 2a, the temperature of the incoming air is used as
reference temperature (see Eq. 1), while in Figure 2b, the average temperature in the street
canyon is selected. The dashed curve shows the limit case for purely forced convection (i.e.
no buoyancy). It is seen that forced convection is dominant for temperature differences below
10C. For larger temperature differences, buoyant effects become important. A number of
observations can be made:

1) For the forced convective regime (temperature differences below 10C), the CHTC curves
for the heated wall have a similar distribution and we can establish a single correlation for
CHTC.
0
1
2
3
4
5
6
7
8
9
10
0 2 4 6 8 10 12 14 16
l
o
c
a
t
i
o
n

(
m
)
CHTC (W/m
2
k)
CHTC-LW
CHTC-WW
a
0
1
2
3
4
5
6
7
8
9
10
0.0E+00 2.0E-08 4.0E-08 6.0E-08 8.0E-08 1.0E-07 1.2E-07
l
o
c
a
t
i
o
n

(
m
)
CMTC (s/m)
CMTC-LW
CMTC-WW
b
24
2) It is not possible to achieve a single correlation for CHTC for the non-heated surfaces
(here being the ground and leeward wall).
3) When choosing a more local reference location, e.g. the average temperature in the street
canyon, the spread in the CHTC profiles does not get smaller.

These findings imply that the CHTC profiles change continuously in time, for example when
a wall is heated by the sun. Given the unsteady nature of the flow, especially in the street
canyon, and knowing that, depending on the boundary conditions, the heat and moisture
distribution in a street canyon can vary over time, choosing one surface averaged constant in
time as the reference CHTC profile applicable to all cases and all times is unrealistic.
Defraeye (2011b) draws the same conclusion for a stand-alone building.
Figure 2: Profiles of CHTC on the walls of a street canyon with heated windward wall. The reference for
calculating CHTC is (a) inlet temperature, and (b) average temperature in the street canyon

Therefore we propose a coupled-model for accurately modeling the drying behavior of a
porous surface, such as a wall in a street canyon. In this model, use of a single constant-with-
time CTC or constant-with-time CTC location dependent profile is avoided by exchanging
CTC profiles at the surface between the air and the solid domain at every time-step. The
details of the coupled-model will be explained in the next section. Afterwards, the
capabilities of the model are illustrated for a case study.
2 COUPLED CFD & BE-HAM MODEL
2.1 Description of the coupled model
The coupled model consists of two sub-models: a CFD model which solves the convective
heat, air and moisture transport in the air, and a Building Envelope Heat and Moisture (BE-
HAM) model which solves heat and moisture transport within the porous material. As an
example, we consider the calculation domain shown on Figure 3. The air domain consists of
the space between two buildings in a street canyon and the air above it. The solid domain
consists of the brick cladding of the windward wall of the street canyon. In this section, we
first describe the two sub-models in detail. Afterwards the coupling strategy is discussed.

0 10 20
Reference (inlet)
T = 1C
T = 2C
T = 5C
T = 10C
T = 20C
T = 30C
T = 50C
T = 60C
Forced
0 5 10
Reference
(average)
0 10 20
T = 1C
T = 2C
T = 5C
T = 10C
T = 20C
T = 30C
T = 50C
T = 60C
Forced
a
b
25

Figure 3: Model for numerical analysis with boundary conditions (AD = adiabatic; IP = impermeable to vapor).

2.1.1 CFD model
The fluid domain consists of a 2D 1:1 street canyon in an Atmospheric Boundary Layer
(ABL). The domain extends 4H in front of and 15H behind the canyon, where H is the
canyon height. A domain height of 4H was selected. These dimensions are equal to or larger
than the dimensions recommended by Franke et al. (2007). The CFD simulations are
conducted using Ansys-Fluent 12.0 which uses the control volume method (Ansys Inc.,
2009). Based on the validation work of Blocken et al. (2009), Defraeye et al. (2010) and Xie
et al. (2006), steady RANS in combination with the realizable k- turbulence model and Low
Reynolds number modeling (LRNM) was selected. Second-order discretization schemes as
well as the SIMPLE algorithm for pressure-velocity coupling are employed. Pressure
interpolation is second order. At the inlet of the domain, vertical profiles of the mean
horizontal wind speed (U), turbulent kinetic energy (k) and turbulent dissipation rate () are
imposed:

( )
*
0
0
ln
ABL
z z U
U z
z
+
=

;
* 2
ABL
U
k
C
= ;
* 3
0
( )
ABL
U
z z
=
+
(2)

where z
0
is the aerodynamic roughness length of the terrain, assumed to be 0.03 m (grass-
covered terrain according to Wieringa (1992)), is the von Karman constant (0.41), C
is a
model constant (0.09), z is the vertical coordinate, and U
*
ABL
is the friction velocity, given by:

* 1 0
10
0
10
ln
ABL
z
U U
z

+
=

(3)

with U
10
the horizontal wind speed at 10 m height. The temperature and relative humidity of
the incoming air is initialized at the inlet. Symmetry is imposed at the top of the
computational domain, which implies that the velocity component normal to the surface is
zero, as well as the normal gradients of all quantities. Zero static pressure is imposed at the
domain outlet. The remaining surfaces, i.e. the street canyon walls and the ground surface,
are modeled as no-slip boundaries with zero roughness. Since the focus of this study is on the
windward wall, this wall is initialized with a temperature and water mass fraction, while the
leeward wall and the ground are modeled as adiabatic and impermeable to vapor. A 2D
structured grid was constructed based on grid sensitivity analysis. The y
+
values of the wall-
adjacent cells are below 1, which is required for boundary-layer modeling with LRNM
(Ansys Inc., 2009).

windward wall
4H
15H
U10=5 m/s
Tin=21C
RHin=20%
4H
AD & IP
AD & IP
10m
z
x
10m
10m
26

2.1.2 BE-HAM model
Drying of the windward wall is simulated using a Building Envelope Heat, Air and Moisture
(BE-HAM) transport model. A detailed description of the basic assumptions and governing
equations can be found in Defraeye (2011b) and Janssen et al. (2007).

A 2D structured grid consisting of 2120 finite elements (20 in x-direction and 106 in z-
direction) was constructed based on grid sensitivity analysis. The exterior wall surface is
exposed to climatic boundary conditions. Solar radiation is not considered in these
simulations.

The wall is assumed to be 10 m high, built of masonry with a thickness of 0.09 m and
insulated on the interior side. Because of the specific wall construction, we can limit the
study to the transport in the masonry outer leaf (Fig. 3), considering the interior surface as
well as top and bottom to be adiabatic and impermeable to moisture. The material properties
for the considered brick can be found in Janssen et al. (2007).
2.1.3 Coupling strategy
At every global time step, information (i.e. CHTC and CMTC, or temperature and mass
fraction) is exchanged between the two models. Since the dominant time-scales in the air
domain are much shorter than in the porous domain, we can assume that the air flow
immediately accommodates to changes in boundary conditions. The transport in the porous
material is characterized by larger time-scales and hence unsteady behavior has to be
considered. Furthermore, the properties of common porous materials exhibit severe non-
linearity, especially those governing moisture transport. Therefore, steady-state CFD
simulations are conducted for the air domain at every global time step, while the BE-HAM
simulations are transient and employ adaptive time stepping. The size of the global time step,
governing data exchange between the two models, is determined based on a sensitivity study
and was chosen to be 112 sec (please refer to Saneinejad et al., 2011b for further details). The
diagram below (Figure 4) shows the schematic of the coupled model.

CFD
BE-HAM
t t t
T
,

P
v
C
H
T
C
C
M
T
C
T
,

P
v
C
H
T
C
C
M
T
C
T
,

P
v
C
H
T
C
C
M
T
C

Figure 4: Schematic of the coupled model. The large circles denote the global time steps at which information
is exchanged between the two sub-models. The small circles denote the time steps of the transient BE-HAM
model.

27
0
500
1000
1500
2000
2500
15
20
25
30
35
40
0 8 16 24 32 40 48 56 64 72
v
a
p
o
r

p
r
e
s
s
u
r
e

(
p
a
)
t
e
m
p
e
r
a
t
u
r
e

(
C
)
time (hr)
9.5 m
6.5 m
3.5 m 0.5 m
0.5 m
3.5 m
6.5 m
9.5 m
temperature
vapor pressure
3 CASE STUDY
3.1 Drying of the windward wall of a street canyon
To illustrate the applicability of the proposed approach, the coupled model is used to analyze
evaporative cooling from a wet wall in an urban street canyon. We consider the same
configuration as described in the previous section. The windward wall of the street canyon is
assumed to be at a constant temperature of 20C and 100% RH and the inflowing air is
chosen to be at 30C and 20% RH. The reference wind speed at 10 m height is assumed to be
5 m/s. The leeward wall and the ground are assumed to be adiabatic and impermeable to
vapor. For simplicity, buoyancy and radiation are not considered in this study.

We monitor the drying behavior of the windward wall for a period of 3 days. Figures 5a and
5b show the mass flow rate, temperature and vapor pressure as a function of time at several
locations on the outer surface of the windward wall. A high drying rate is observed at the wall
surface experiences and the temperature drops from 20C to 15C, at the beginning of the
drying process. During the first couple of days the drying rate decreases rapidly until it
reaches a quite constant level the duration of which depends on the location on the wall. This
regime is characterized by a constant vapor pressure at the surface, a constant drying rate and
a constant material temperature, namely the wet bulb. Local drying of the material at the
interface results in a decrease in the drying rate. This is caused since the dry outer-layer
forms a resistance to liquid water removal from the bulk towards the interface, in addition to
the boundary-layer resistance. This phase is called the second drying phase, marked by
sudden drop of the vapor pressure and an increase of the local temperature since less heat is
required for the evaporation process.

Figure 5: (a) Mass flow rate, and (b) temperature and vapor pressure as a function of time for various locations
on the windward canyon wall. (U
10
= 5 m/s)

3.2 Effect of evaporative cooling on conditions in the canyon
In addition to local reduction of the surface temperature, evaporation from the porous wall
surface influences the local micro-climate in the canyon. In the studied case, we look at the
temperature at location representing pedestrian height (1.7 m high), 1 m away from the
windward wall. At this location, temperature drops by approximately 0.7C compared to the
initial condition (shown by the solid line in Figure 6a). For the same location, temperature
evolution is quite different if the windward wall is considered dry (dotted line in Figure 6a).
In the studied case, evaporative cooling causes a temperature reduction of up to 1.7C.

0.0E+00
2.0E-05
4.0E-05
6.0E-05
8.0E-05
1.0E-04
1.2E-04
1.4E-04
0 8 16 24 32 40 48 56 64 72
d
r
y
i
n
g

r
a
t
e

(
k
g
/
m
2
s
)
time (hr)
9.5 m
6.5 m
3.5 m
0.5 m
b
a
28
Comparing the temperature at the point at pedestrian height, 1m away from windward wall,
with another point, also at pedestrian height, but 5m away from the windward wall (center of
the street canyon), shows a slightly larger temperature drop at the point in the middle of the
street (Figure 6b). Comparing the temperature of the same two points, but without having the
effect of evaporative cooling, shows almost no difference (Figure 6c).

Figure 6: (a) temperature at the pedestrian location for cases with and without moisture, (b) temperature at the
pedestrian height, 1m and 5m from WW wall (with moisture), (d) temperature at pedestrian height, 1m and 5m
from WW wall (without moisture), (d) average temperature in the street canyon achieved with coupled and
uncoupled approaches. (U
10
= 5 m/s)

3.3 Comparison of coupled and uncouple approach
In order to evaluate the relevance of the proposed coupled model, the same example was
studied using an uncoupled approach. Herein, the initial height-dependent profiles of CHTC
and CMTC are used for the entire duration of the simulation. Figure 6d depicts the average
temperature in the street canyon. Temperature drop due to evaporative cooling is much less
pronounced in the uncoupled approach. This shows that employing a coupled modeling
strategy is crucial to arrive at reliable predictions of the urban micro-climate.
4 CONCLUSIONS
In this paper, a coupled model for simulating drying processes is proposed. The model
couples a Computational Fluid Mechanics (CFD) model to a Building Envelope Heat and
Moisture (BE-HAM) transport model. By exchanging information (i.e. CHTC and CMTC, or
temperature and mass fraction) between the two models at every global time step, use of
constant-with-time CTC profiles is eliminated. Using the proposed model, drying of a
27.5
28.0
28.5
29.0
29.5
30.0
0 8 16 24 32 40 48 56 64 72
t
e
m
p
e
r
a
t
u
r
e

(
C
)
time (hr)
1m from ww
5m from ww
27.5
28.0
28.5
29.0
29.5
30.0
0 8 16 24 32 40 48 56 64 72
t
e
m
p
e
r
a
t
u
r
e

(
C
)
time (hr)
1m from ww
5m from ww
27
28
29
30
0 8 16 24 32 40 48 56 64 72
t
e
m
p
e
r
a
t
u
r
e

(
C
)
time (hr)
with moisture
without moisture
evaporative
cooling
a
27
28
29
30
0 8 16 24 32 40 48 56 64 72
a
v
e
r
a
g
e

t
e
m
p
e
r
a
t
u
r
e

(
C
)
time(hr)
uncoupled
coupled
d
c
b
29
windward wall of a street canyon was studied and its influence on reducing the local micro-
climate temperature was investigated. By comparing the results with an uncoupled approach,
It was shown that the coupled model could successfully be employed to study the drying
process of a porous surface and the resulting evaporative cooling in the street canyon.
5 REFERENCES
Alexandri, E., Jones, P., 2008. Temperature Decreases in an Urban Canyon Due to Green Walls and Green
Roofs in Diverse Climates. Building and Environment, 43, 480.
Ansys Inc., Ansys Fluent 12.0 Users Guide, 2009.
Blocken, B., Roels, S., Carmeliet, J., 2007. A combined CFDHAM approach for wind-driven rain on building
facades. J. Wind Eng. Ind. Aerodyn. 95(7), 585-607.
Blocken, B., Defraeye, T., Derome, D., Carmeliet, J., 2009. High-resolution CFD simulations of forced
convective heat transfer coefficients at the facade of a low-rise building. Building and environment 44(12),
2396-2412.
Defraeye, T., Blocken, B., Carmeliet, J., 2010. CFD analysis of convective heat transfer at the surfaces of a cube
immersed in a turbulent boundary layer. Int. J. Heat Mass Trans., 53(13) 297308.
Defraeye, T., Blocken, B., Carmeliet, J., 2011a. Convective heat transfer coefficients for exterior building
surfaces: Existing correlations and CFD modelling, Energy Conversion & Management, 52(1), 512-522.
Defraeye, T., 2011b, Convective heat and mass transfer at exterior building surfaces, PhD thesis, Katholieke
Universiteit Leuven, Belgium.
Emmel, M.G., Abadie, M.O., Mendes, N., 2007. New external convective heat transfer coefficient correlations
for isolated low-rise buildings. Energy Build. 39 (3), 335342.
Franke, J., Hellsten, A., Schlnzen, H., Carissimo, B., 2007. Best practice guideline for the CFD simulation of
flows in the urban environment. COST Action 732: Quality Assurance and Improvement of Microscale
Meteorological Models.
Janssen, H., Blocken, B., Carmeliet, J., 2007. Conservative modeling of the moisture and heat transfer in
building components under atmospheric excitation, Int. J. Heat Mass Trans., 50, 1128-1140.
Krger, E.L., Pearlmutter, D., 2008. The effect of urban evaporation on building energy demand in an arid
environment. Energ. Build. 40, 2090-2098.
Runsheng, T., Etzion, Y. and Erell, E., 2003. Experimental studies on a novel roof pond configuration for the
cooling of buildings. Renew. Energy, 28, 1513-1522.
Tiwari, G.N., Kumar, A., Sodha, M.S., 1982. A review: Cooling by water evaporation over roof. Energy
Conversion & Management, 22, 143-153.
Saneinejad, S., Moonen, P., Defraeye, T., Carmeliet, J., 2011a. Analysis of convective heat and mass transfer at
vertical walls of a street canyon. Journal of Wind Engineering & Industrial Aerodynamics: in press,
doi:10.1016/j.jweia.2010.12.014
Saneinejad, S., Moonen, P., Carmeliet, J., 2011b. Studying the effect of evaporative cooling on local conditions
in a street canyon. 13
th
International Conference on Wind Engineering, Amsterdam.
Wieringa, J., 1992. Updating the Davenport roughness classification. J. Wind Eng. Ind. Aerodyn. 41-44, 357-
368.
Xie, X., Liu, C., Leung, D.Y.C., Leung, M.K.H., 2006. Characteristics of air exchange in a street canyon with
ground heating. Atmospheric Environment, 50, 6396-6409.

30
Flow and concentration measurements in the wake of
reduced scale models of cars for developing
nanoparticle dispersion models
Matteo Carpentieri, Prashant Kumar, Alan Robins
University of Surrey, Faculty of Engineering and Physical Sciences,
Guildford, UK, m.carpentieri@surrey.ac.uk
ABSTRACT: The present study is part of an EPSRC-funded project aimed at understanding
the transformation processes acting on nanoparticles emitted from road vehicles. A better
understanding of these phenomena is essential for developing appropriate methods for
including nanoparticle dynamics in dispersion models. Very short time scales are usually
associated with nanoparticle transformations in the first stages after their emission, as
highlighted by previous field measurements and modelling studies. The interaction between
these transformations and the flow and turbulence fields immediately behind the vehicle is
complex and it strongly affects nanoparticle dispersion at local scale, hence the need of
characterising in detail the mixing processes in the vehicle wake. The experiments were
carried out in the EnFlo (Environmental Flow Research Centre) wind tunnel at the University
of Surrey, with reduced scale models of the diesel car used in several field campaigns. The
flow and turbulence fields were characterised both in the near and the far wake of the
modelled vehicle by using a 2component LDA, while concentration measurements were
obtained by using a fast response flame ionisation detector. The high resolution experimental
database obtained from these experiments, in conjunction with a large nanoparticle
concentration data set obtained from field measurements, will be used for deriving
mathematical parameterisations for operational nanoparticle dispersion models.
1 INTRODUCTION
Transient and shortterm concentration fluctuations at local scale from pollutants emitted by
ground vehicles in cities can cause severe damage to human health. Physical and chemical
processes involved in nanoparticles dispersion modelling are still poorly understood (Kumar
et al., 2011), especially at fine spatial and temporal scales, where the impact of the single
vehicle wake on the dispersion process can become important (Baker, 2001). Vehicle wakes,
for example, can strongly affect the turbulence field in street canyons (Di Sabatino et al.,
2003), and their effects on pollutant dilution cause complex interactions with other
transformation processes affecting number and size distributions of ultrafine particles
(Carpentieri et al., 2011). Operational dispersion models rarely acknowledge these effects.
This study involves a series of experiments performed in the wind tunnel laboratory at the
University of Surrey. Flow and dispersion characteristics were investigated both in the near
and farwake regions downwind of reduced scale car models. The aims included the
experimental characterisation of the flow, turbulence and concentration fields in both wake
regions, and the preliminary assessment of existing wake models using the experimental data
base. Compared to previous studies, the research presented here benefits from the integrated
approach used, which includes previous field measurements.
31
The work has been carried out in the framework of the project Understanding dispersion of
nanoparticles in vehicle wake combining fast response measurements and wind tunnel
simulations, funded by the UK EPSRC (Engineering and Physical Sciences Research
Council). The project is aimed at characterising the dispersion of ultrafine particles in the
wake of moving vehicles through field measurements and laboratory experiments, in order to
derive simple mathematical parameterisations (Carpentieri et al., 2011; Carpentieri and
Kumar, 2011).
2 WIND TUNNEL EXPERIMENTS
The core of the experimental work was carried out at the boundary layer wind tunnel of the
Environmental Flow Research Centre (EnFlo) at the University of Surrey, UK (referred to
hereafter as the EnFlo tunnel). The geometrical characteristics of the reducedscale models
were derived from the diesel car used (2004 Vauxhall AstraVan) during related field
measurements (Carpentieri and Kumar, 2011). Two different scales were investigated: a 1:5
model in order to focus on the nearwake characterisation, and a smaller 1:20 model in order
to extend our measurements well into the farwake.
Models were placed approximately at 11 m from the inlet to the wind tunnel working section,
where a relatively deep boundary layer would develop on the smooth tunnel floor, even
without the use of turbulence generators or roughness elements. In order to remove the
unrealistic effects of such a boundary layer, all models were placed near the edge of a raised
false floor. Velocity and turbulence fields in the wake of the models were measured by means
of a two component LDA. A fibreoptic system was used, while concentration measurements
were obtained by using a Fast response Flame Ionisation Detector (FFID).
Threedimensional LDA measurements were performed over most of the false floor using a
reference wind speed of 2.5 m s
1
. Initial evaluations were carried out without the vehicle
model to assess the development of the boundary layer above the floor. To further assess this
particular aspect, additional FFID tests were also carried out in a smaller closedcircuit wind
tunnel designed for aerodynamic tests on ground-vehicles. This wind tunnel (hereafter
referred to as the Aero tunnel) has a rolling road section with a boundary layer suction
system. Unrealistic boundary layer effects were removed by running the rolling road at the
same speed as the wind flow and adjusting the boundary layer suction fans speed
accordingly.
The preliminary tests in the Aero tunnel were conducted on the 1:20 model, as the 1:5 model
was too large to fit the wind tunnel without causing excessive blockage effects. In order to
measure tracer concentrations at a high resolution within the near wake, an intermediate size
model (i.e. 1:8 scale) was constructed and used. The rolling road belt was 1.47 m long and
0.60 m wide, sufficient to cover at least the near wake of the 1:20 scale model, and the
recirculation region of the wake of the 1:8 scale model. A wind speed of 10 m s
1
was set for
these tests as this was the lowest operating speed at which the rolling road could be used
satisfactorily. Given the higher wind speed in the Aero tunnel, a shorter averaging time (20 s)
was used. Experiments in the Aero tunnel were conducted both with and without operation of
the rolling road, to assess the significance of boundary layer effects.

32
3 RESULTS
This section reports the results of the experimental campaign carried out using the EnFlo
wind tunnel. Velocities have been normalised by using the reference wind speed U
ref
, which
was measured by the fixed ultrasonic anemometer. Hereafter, non-dimensional mean velocity
components (longitudinal, lateral and vertical) are referred to as U
*
, V
*
and W
*
, respectively.
Similarly, nondimensional root mean square (rms) velocity components are referred to as u
*
,
v
*
and w
*
. Concentrations have been normalised by using the standard expression: C
*
= (C
U
ref
h
2
)/Q, where C
*
is the nondimensional concentration, and Q is the source volumetric
flow rate. Distances were normalised using the vehicle height, h.
3.1 Velocity measurements
Figure 1 shows the longitudinal evolution of the measured vertical profiles along the centre
line, both for the 1:5 scale (Fig. 1a) and 1:20 scale (Fig. 1b) models. The wake effect on the
measured longitudinal velocities is clearly visible in both cases. However, part of the cause
for lower speeds, especially at the lowest levels, could be due to the boundary-layer
developing on the false floor, particularly at larger distances from the vehicle.

Figure 1: Selected vertical profiles of longitudinal non-dimensional mean speed (U
*
) along the centre line (y =
0) for (a) the 1:5 model, and (b) the 1:20 model.
Figure 2 shows the velocity vectors and rms velocity contour plots for the 1:5 model at
selected vertical planes. In particular, the Y=0 plane (centre line, Figs. 2a, b and c) and
Y=0.33 plane (in line with tailpipe, Figs. 2d, e and f) are shown. A recirculating flow region
can be observed, particularly along the centre line (Fig. 2a), having an approximate length of
about 2h. While an initial high speed zone exists at the lowest levels, corresponding to the air
flow coming from below the vehicle, the wake begins to affect this region at a distance of
about 1h because of downward spread of the wake region. As observed earlier, part of this
effect could be due to the interaction with the developing boundary.
Longitudinal turbulence (i.e. along x) is particularly high within the recirculation zone, and
maximum levels can be found at heights corresponding to the upper and lower surfaces of the
car (Fig. 2b). These shear layers are caused by the flow separation on the upper and lower
edges of the vehicle. Similar patterns (see Figs. 3b and e) are caused by the lateral surfaces of
the car. Vertical turbulence (along z) is mainly concentrated within the recirculation wake
(Figs. 2c and f).
(a) (b)
33

Figure 2: LDA measurements on the 1:5 model, vertical planes: (a) U
*
-W
*
vector plot at Y=0, (b) u
*2
velocity
variance contour plot at Y=0, (c) w
*2
velocity variance contour plot at Y=0, (d) U
*
-W
*
vector plot at Y= 0.33,
(e) u
*2
velocity variance contour plot at Y=0.33, (f) w
*2
velocity variance contour plot at Y=0.33. Y=0.33 is
approximately in line with the tailpipe.
Horizontal sections from the LDA measurements behind the 1:5 model are presented in
Figure 3. In particular, results at the Z=0.07 plane (lowest measured level, Figs. 3a, b and c)
and Z=0.23 plane (at tailpipe height, Figs. 3d, e and f) are reported. Regions of high
longitudinal variance extend in line with the vehicle sides (Figs. 3b and e), while a large
lateral turbulence zone is observed within the recirculation region (Figs. 3c and f),
approximately overlapping the vertical turbulence field observed in Figures 2c and f. These
results are consistent with the measurements by Kanda et al. (2006) in the wake of a
passenger car and a truck.
(a)
u* , w*
2 2
Y=0
Y=0.33
(b)
Y=0
u*
2
(c)
Y=0
w*
2
(d)
(e)
Y=0.33
u*
2
(f)
w*
2
Y=0.33
34

Figure 3: LDA measurements on the 1:5 model, horizontal planes: (a) U
*
-V
*
vector plot at Z=0.07, (b) u
*2

velocity variance contour plot at Z=0.07, (c) v
*2
velocity variance contour plot at Z=0.07, (d) U
*
-V
*
vector plot
at Z=0.23, (e) u
*2
velocity variance contour plot at Z=0.23, (f) v
*2
velocity variance contour plot at Z=0.23.
Z=0.07 is the lowest height for the measurements, while Z=0.23 is approximately at the same height as the
tailpipe.
3.2 Concentration measurements
Figure 4 shows the results with the 1:5 scale model, presented in terms of vertical planes
(Figs. 4a, b and c) and horizontal sections (Figs. 4d and e). As in the case of velocity
measurements, the Y=0 plane (centre line, Fig. 4a) and Y=0.33 plane (in line with tailpipe,
Fig. 4b) are shown, as well as the Z=0.07 (lowest section, Fig. 4c), Z=0.23 (approximately at
tailpipe height, Fig. 4d) and Z=0.60 (approximately at the middle of the car body, Fig. 4e)
horizontal sections. The vertical section at Y=0 (Fig. 4a) highlights the effect of the
recirculation region, with an accumulation of tracer close to the vehicle rear driven by
embedded vertically aligned vortices. This is also evident in the horizontal sections at heights
between the vehicle base and the roof, and at Z=0.60 in particular (Fig. 4e). The shape of the
(a)
(d)
Z=0.07
Z=0.23
2
v*
(b)
(f)
Z=0.07
Z=0.23
2
v*
2
u*
(c) Z=0.07
2
v*
(e) Z=0.23
2
u*
2
u* ,
35
wake and the flow field reported in Section 3.1 greatly affects the development of the plume,
as shown by the contour plots in Figures 6b, c and d (compare them, for instance, with Figs. 2
and 3). In general, the measured plume is far from the classical Gaussian plume, often
assumed in operational mathematical models. This aspect will be further analysed in Section
4.

Figure 4: FFID measurements on the 1:5 model; C
*
contour plots at: (a) Y=0 vertical plane, (b) Y=-0.33 vertical
plane, (c) Z=0.07 horizontal plane, (d) Z=0.23 vertical plane, (e) Z=0.60 vertical plane. Z=0.07 is the lowest
height for the measurements, Z=0.23 and Y=-0.33 are approximately in line with the tailpipe, Y=0 is the centre
line and Z=0.60 is approximately at the middle of the car body.
Similar contour plots for FFID measurements were found on the 1:20 scale model, so they are
not reported here.
3.3 Boundary layer effects
Despite the precautions taken in designing the wind tunnel tests (see Section 2), some
unwanted effects due to the development of the boundary-layer on the false floor must affect
the experimental results (Sections 3.1 and 3.2). A preliminary experimental campaign was
carried out in the Aero wind tunnel, which features a rolling road with a boundary layer
suction system, in order to assess the influence of this phenomenon on the data. A direct
comparison of the velocity profiles was not feasible as no velocity measurements were
available but concentration measurements in the wake of the vehicle models allowed us to
perform an analysis of any effects of boundary layer development.
(a)
Y=0
(b)
Y=-0.33
(c)
Z=0.07
[vertical planes]
(d)
Z=0.23
(e)
Z=0.60
[horizontal
36
Results are not reported here for brevity. The comparison between the results in the Aero
tunnel with and without the rolling road, however, highlights the different behaviour in tracer
dispersion. In particular, we have generally lower concentrations close to the car when the
rolling road and boundary layer suction system are in operation. Further from the model, the
concentration profiles show higher concentrations (at least close to the ground) than during
the tests with the rolling road stationary. This could be due to the increased turbulence in the
latter case, which in turns increases the spreading of the plume as it travels downwind. This
effect is also evident in the horizontal (lateral) profiles, highlighting the fact that the
increased mixing occurs both in the vertical and lateral senses. In contrast, the higher velocity
of the air flow below the car (with the rolling road is running) caused an enhanced mass
exchange through the flow separation surface, decreasing the concentration of tracer within
the recirculation region (as explained earlier, the emission source is within the recirculation
area).
Comparison of EnFlo and Aero tunnel results, the latter with a stationary floor, is not very
satisfactory. As we gathered from the experiments, high concentration gradients can be found
in the vehicle wake. In this situation, even a very small variation in the positioning of the
probe may lead to large differences. The different conditions in the two wind tunnels (the
Aero tunnel being much narrower than the EnFlo tunnel) could also account for a different
behaviour of the plume on the horizontal plane.
The measurements of velocity profiles above the false floor in the EnFlo tunnel, performed
without the model, allowed us to estimate the boundary-layer growth. The depth of the
boundary-layer was approximately 15-20 mm at 450 mm from the floor leading edge,
reaching 50-60 mm towards the downwind end of the floor. The comparisons discussed
above involved measurement points taken within the first metre or so of the floor, so that the
boundary-layer depth was never more than 20-30 mm. Despite this fact, important effects in
the concentration field close to the vehicle were observed, influencing it well beyond the
extent of the boundary-layer itself. The explanation for this lies in the fact that the
concentration field within the recirculation zone is controlled by the turbulent exchange with
the external flow, and a major part of the interface is located within the boundary-layer and
affected by boundary layer turbulence.
Further downwind the effects on the concentration levels seem to decrease and the actual
matching of the exit velocity for the emission source becomes more important than the
presence of the boundary layer. The extension of the rolling road did not allow us to measure
further away from the models. It is expected that the presence of the boundary layer would
start significantly affecting the concentration field at these distances, especially for the
smaller model, where the vertical dimensions of the boundary layer will be comparable to
those of the wake.
4 CONCLUSIONS
Wind tunnel experiments were carried out in the wake of small scale models of passenger
cars. Both velocity and concentration fields were measured by means of a laser Doppler
anemometer and a fast response flame ionisation detector.
The results showed the influence of the vehicle wake on the dispersion process, while this
aspect is usually neglected by standard mathematical dispersion models. Since the effects of
the wake on pollutant dilution cause complex interactions with other transformation
processes affecting ultrafine particles (Carpentieri et al., 2011), operational dispersion models
for nanoparticles will need to acknowledge these effects, in order to correctly reproduce the
physical phenomena.
37
The high resolution experimental database obtained from these experiments, in conjunction
with a large nanoparticle concentration data set obtained from field measurements, will be
used for deriving mathematical parameterisations for operational nanoparticle dispersion
models. The first step will be to compare the experimental results with the well established
building wake models (e.g., ADMS-BUILD, Robins et al., 1997) in order to gather
information about the different parameters involved and formulate new parameterisations.
The model by Eskridge and Hunt (1979) will also be tested, even if it is only valid in the far
wake. Further research on nanoparticle dispersion will also be necessary to understand the
complex interactions between dilution and other transformation processes and, eventually,
develop new mathematical models.
5 ACKNOWLEDGEMENTS
This work has been carried out as a part of EPSRC funded grant EP/H026290/1.
6 REFERENCES
Baker C. J., 2001. Flow and dispersion in ground vehicle wakes. Journal of Fluids and Structures, 15(7), 1031-
1060.
Carpentieri M., Kumar P., 2011. Groundfixed and onboard measurements of nanoparticles in the wake of a
moving vehicle. Atmospheric Environment, in press, doi: 10.1016/j.atmosenv.2011.1006.1079.
Carpentieri M., Kumar P., and Robins A., 2011. An overview of experimental results and dispersion modelling
of nanoparticles in the wake of moving vehicles. Environmental Pollution, 159(3), 685-693.
Di Sabatino S., Kastner-Klein P., Berkowicz R., Britter R. E., and Fedorovich E., 2003. The modelling of
turbulence from traffic in urban dispersion models Part I: Theoretical considerations. Environmental Fluid
Mechanics, 3(2), 129-143.
Eskridge, R.E., Hunt, J.C.R., 1979. Highway modelling Part I: prediction of velocity and turbulence fields in the
wake of vehicles. Journal of Applied Meteorology, 18, 387-400.
Kanda I., Uehara K., Yamao Y., Yoshiwara Y., and Morikawa T., 2006. A wind-tunnel study on exhaust gas
dispersion from road vehicles Part I: Velocity and concentration fields behind single vehicles. Journal of
Wind Engineering and Industrial Aerodynamics 94(9), 639-658.
Kumar, P., Ketzel, M., Vardoulakis, S., Pirjola, L., and Britter, R., 2011. Dynamics and dispersion modelling of
nanoparticles from road traffic in the urban atmsopheric environment a review. Journal of Aerosol Science
42, 580-603.
Robins, A.G., McHugh, C.A., and Carruthers, D.J., 1997. Testing and evaluating the ADMS building effects
module. International Journal of Environment and Pollution, 8, 708-717.

38
Ventilation of an urban intersection characterized by
advective and turbulent scalar fluxes
Libor Kukaka
a,b
, tpn Nosek
c
, Radka Kellnerov
a,d
, Klra Jurkov
e
,
Zbynk Jaour
f

a
Charles University in Prague, Faculty of Mathematics and Physics, Czech
Republic

b
Institute of Thermomechanics AS CR, Prague, Czech Republic,
kukacka@it.cas.cz
c
nosek@it.cas.cz

d
kellnerova@it.cas.cz
e
jurcakova@it.cas.cz
f
janour@it.cas.cz
ABSTRACT: The objective of this study is to determine processes of vertical ventilation
above the X-shaped intersection in an idealised urban area in several approach flow
directions. An experimental set-up for simultaneous measurement of the flow velocity and
the tracer gas concentration was assembled. Vertical advective and turbulent fluxes of passive
contaminant were computed from synchronised measured velocity and concentration signals
in a regular grid placed at the roof top level above the studied intersection. A significant
positive contribution of the turbulent transport to ventilation of the area was determined
especially higher approach flow angles. Results had also shown considerably different
distribution of vertical turbulent and advective pollution transport in the measured grid.
1 INTRODUCTION
Vehicle traffic pollutants emitted directly to street-canyons represent serious health hazard
for people in large cities. As shown in Wang and McNamara (2007), geometry of street
intersections plays an important role in pollutant dispersion and ventilation in urban areas.
Wind tunnel and field studies for relatively symmetrical and regular street canyons
arrangements express influence of geometry of streets and intersections in pollutant
dispersion and hence ventilation of urban areas, e.g. Dabbert et al. (1995) and Brown et al.
(2004). Mixing and transport processes in a simple street and its ventilation were elaborated
by Belcher (2005). In this work ventilation fluxes were determined for estimation of the mean
scalar transport within the urban street network. Barlow and Belcher (2002) focused on
studying the ventilation characteristics of a street canyon for the simple case of wind
perpendicular to the street. Wind tunnel experiments published by Robins (2008) show that
the mass exchange between street canyons may be significantly changed due to small
variations of the building geometry. These results were obtained from computing scalar
fluxes determining pollution transport. Results from numerical simulation published by
Scaperdas and Colvile (1999) show a very complex behaviour of the flow in an urban area.
39
This work shows that configuration of the street canyon
exchange between alongwind and crosswind streets is reversed. Numerical and wind
simulation of the flow and dispersion near regular and
studied by Wang and McNamara (20
and dispersion processes to the intersection geometry and wind direction
connected with ventilation of an urban
The objective of this study is to
shaped intersection in an idealised urban area in several approach flow directions
2 EXPERIMENTAL SET
2.1 Wind tunnel and urban area model
The experiment was carried out at the low
Thermomechanics AS CR in Nov Knn. The suck
long generating section equipped by spires and covered by 10 cm high roughness elements.
The test section of the tunnel is 2.0 m long, 1.5 m wide and 1.5 m tall.
tunnel can be found in Kellnerov
The model of idealised urban area with apartment houses was designed after the common
European inner-city area in scal
roofs formed a perpendicular arrangement of
the buildings was H = 12 cm and the width of streets was
in full scale). The aspect ratio
placed on a turntable that allowed an
The data were measured in the grid contained 76 points in horizontal plane
(22x22 m in full scale) placed above the studied X
see Figure 1. Results were obtained for 5 different values of the angle of approach fl
5, 15, 30 and 45.

Figure 1: Scheme of the idealised symmetric urban
photograph of the model placed in the wind tunnel.
2.2 Measuring instruments and data processing
The flow characteristics were measured with a
Anemometry (LDA), based on DANTEC BSA F
International Workshop on Physical Modeling of Flow and Dispersion Phenomena
configuration of the street canyons and the wind direction when air
exchange between alongwind and crosswind streets is reversed. Numerical and wind
simulation of the flow and dispersion near regular and irregular street inte
and McNamara (2007). These papers demonstrate high sensitivity of flow
and dispersion processes to the intersection geometry and wind direction, which
connected with ventilation of an urban area.
The objective of this study is to determine processes of vertical ventilation above the X
shaped intersection in an idealised urban area in several approach flow directions
EXPERIMENTAL SET-UP
urban area model
s carried out at the low-speed wind tunnel of Institute of
Thermomechanics AS CR in Nov Knn. The suck-trough open-circuit tunnel has 20.5 m
The test section of the tunnel is 2.0 m long, 1.5 m wide and 1.5 m tall. More details about the
Kellnerov (2009).
in scale 1:200. Regular blocks of apartment houses with pitched
a perpendicular arrangement of street canyons and intersections
and the width of streets was S = 10 cm (H = 24 m and
he aspect ratio of street canyons was H/S = 1.2. The urban area model was
that allowed an approach flow direction change.
The data were measured in the grid contained 76 points in horizontal plane
(22x22 m in full scale) placed above the studied X-shaped intersection at the roof top level,
Results were obtained for 5 different values of the angle of approach fl

ed symmetric urban area model, the studied X-shaped intersection and the
photograph of the model placed in the wind tunnel.
Measuring instruments and data processing
characteristics were measured with a two-dimensional optical fibre
, based on DANTEC BSA F-60 burst processor.
and the wind direction when air
exchange between alongwind and crosswind streets is reversed. Numerical and wind tunnel
irregular street intersections were
. These papers demonstrate high sensitivity of flow
which are naturally
determine processes of vertical ventilation above the X-
shaped intersection in an idealised urban area in several approach flow directions.
tunnel of Institute of
circuit tunnel has 20.5 m
More details about the
. Regular blocks of apartment houses with pitched
and intersections. The height of
= 24 m and S = 20 m
1.2. The urban area model was
The data were measured in the grid contained 76 points in horizontal plane of size 11x11 cm
shaped intersection at the roof top level,
Results were obtained for 5 different values of the angle of approach flow : 0,
shaped intersection and the
dimensional optical fibre Laser Doppler
. Tracing particles
40
(glycerine droplets) were produced by a commercial haze generator placed at the beginning
of the tunnel generating section in front of spires. After running the haze generato
tunnel for several minutes, we got the
particles. An average data rate of the LDA system reached
Point concentration measurements of tracer gas
Ionisation Detector (FFID), type
ethane as the tracer gas which was emitted from a point source placed at the bottom of street
canyon in front of the studied
data rate of 1 KHz.
For simultaneous velocity and concentration measurement,
mounted on the traverse system
the intake to the FFID sampling tube. The sampling tube intake was placed 1
mm behind and 1 mm beside the centre of the LDA measuring volume.
measurements with different positions of both probes demonstrated a negligible influence of
FFID sampling tube placed close to the LDA measuring volume on the flow.
configuration of probes is captured in the Fig

Figure 2: The configuration of the LDA and FFID probes mounted on the traverse system in the wind tunnel.
For simultaneous measurement by FFID and LDA, t
obtained using clean air (air
filled by seeding particles and two span gases of known hydrocarbons concentrations.
was calibrated approx. every
caused mainly by changes of temperature of atmospheric air sucked into the wind tunnel,
reached up to 5% through the measuring campaign.
computed from measured voltage signal using linear interpolated values from two FFID
calibrations realised before and after
As expected, the presence of the seeding particles in the
At first we got isolated spikes in the recorded concentration
combustible aerosol particles from air into the FFID probe. The problem was mentioned in
Hall and Emmont (1991) and in Contini et al. (2006)
similar count of spikes in time series obtained from measurements
contained seeding particles in most cases
because the frequency of isolated spikes was about
The second influence of seeding particles on the measured concentration data was an almost
constant shift of recorded concentration values caused obviously by sucking seeding particles
by FFID probe. This shift reached about
corrected by the mentioned calibration
of the tunnel generating section in front of spires. After running the haze generato
we got the air flow in the test section equally filled by
particles. An average data rate of the LDA system reached about 1 KHz in the studied area
Point concentration measurements of tracer gas were realised by Fast
, type Cambustion Ltd. HFR400 Atmospheric Fast FID. We used
canyon in front of the studied intersection, see Figure 1. The FFID was set to acquire data at a
For simultaneous velocity and concentration measurement, LDA and FFID probe
the traverse system in a way that the measuring volume of the LDA was close to
to the FFID sampling tube. The sampling tube intake was placed 1
mm behind and 1 mm beside the centre of the LDA measuring volume.
ampling tube placed close to the LDA measuring volume on the flow.
onfiguration of probes is captured in the Figure 2.

The configuration of the LDA and FFID probes mounted on the traverse system in the wind tunnel.
measurement by FFID and LDA, the four point FFID calibrations
sucked into the wind tunnel from the atmosphere)
filled by seeding particles and two span gases of known hydrocarbons concentrations.
ibrated approx. every four hours of measurement. The differences in output voltage,
through the measuring campaign. All the concentration values were
omputed from measured voltage signal using linear interpolated values from two FFID
calibrations realised before and after the recorded data set.
As expected, the presence of the seeding particles in the air influenced FFID measurement.
lated spikes in the recorded concentration signal probably due to suction of
Hall and Emmont (1991) and in Contini et al. (2006). Unlike these published results,
time series obtained from measurements in clean
in most cases. We neglected the influence of spikes on the results
he frequency of isolated spikes was about 0.006% of used sampling data rate.
by FFID probe. This shift reached about 0.5 % of the FFID measuring range. This shift was
calibration sequence.
of the tunnel generating section in front of spires. After running the haze generator inside the
in the test section equally filled by seeding
in the studied area.
by Fast-response Flame
Cambustion Ltd. HFR400 Atmospheric Fast FID. We used
The FFID was set to acquire data at a
LDA and FFID probes were
in a way that the measuring volume of the LDA was close to
to the FFID sampling tube. The sampling tube intake was placed 1.5 mm above, 1
mm behind and 1 mm beside the centre of the LDA measuring volume. Several test
ampling tube placed close to the LDA measuring volume on the flow. The
The configuration of the LDA and FFID probes mounted on the traverse system in the wind tunnel.
FFID calibrations were
sucked into the wind tunnel from the atmosphere), air equally
filled by seeding particles and two span gases of known hydrocarbons concentrations. FFID
of measurement. The differences in output voltage,
All the concentration values were
omputed from measured voltage signal using linear interpolated values from two FFID
air influenced FFID measurement.
probably due to suction of
published results, we got
clean air and in air
We neglected the influence of spikes on the results
pling data rate.
FFID measuring range. This shift was
41
All the recorded time series were processed by post processing
equidistant velocity signal records
using moving-average filter to equidistant time series.
ppm were recalculated to the dimensionless concentration given by

where C means measured concentration,
height z = 2H and Q is a source
Then the software synchronised velocity and concentration data
correlation between both signals. The synchronised time series were shifted by
15 ms. This shift expressed the delay between
FFID probe tube and the moment of the sample analysing in
agrees with very similar experimental set up published by
Dimensionless vertical advective
contaminant were computed from synchronised signal
Arya (1999) or Stull (1988).
velocity and average dimensionless concentration,
indicates fluctuations of vertical velocity and dimensionless concentration.
to turbulent transport computing was published by Jur
2.3 Modelled boundary layer
Fully turbulent boundary layer was developed by spires and roughn
tunnel. The characteristics of the m
dimensional LDA in four vertical profiles
and downstream from it, see Figure

Figure 3: Wind profile measurement location
The vertical profiles of mean longitudinal velocity and momentum flux
4, the vertical profiles of longitudinal and vertical
The high above the surface is expressed in full scale.
Vertical profiles of measured turbulent approach flow characteristics data were fitted by the
logarithmic and the power law. Parameters
displacement obtained from logarithmic fit, power exponent obtained from power fit, friction
velocity (alias square-root of constant value of Reynolds stress within the inertial sublayer),
respectively. All the parameters are listed in Table 1.
, *
2
2
Q
H CU
C
H
=
All the recorded time series were processed by post processing Matlab script.
signal records (obtained in burst measurement mode
average filter to equidistant time series. The calibrated concentration records in
ppm were recalculated to the dimensionless concentration given by

means measured concentration, U
2H
means a reference velocity measured at the
is a source ux.
Then the software synchronised velocity and concentration data using
correlation between both signals. The synchronised time series were shifted by
15 ms. This shift expressed the delay between a suck of the sample into the intake of the
FFID probe tube and the moment of the sample analysing in the probe. The value of the shift
agrees with very similar experimental set up published by Contini et al. (2006).
ertical advective C*W/U
2H
and turbulent <c*w>/U
2H
were computed from synchronised signals using eddy-correlation method,
We used obvious symbols: W and C* mean average vertical
velocity and average dimensionless concentration, < > means a time average,
indicates fluctuations of vertical velocity and dimensionless concentration.
to turbulent transport computing was published by Jurkov (2009).
Modelled boundary layer
Fully turbulent boundary layer was developed by spires and roughness elements placed it the
The characteristics of the modelled boundary layer were measured with a two
four vertical profiles placed above the studied intersection and upstream
t, see Figure 3.
Wind profile measurement locations.
longitudinal velocity and momentum flux are
longitudinal and vertical turbulent intensity are depicted in Fig
The high above the surface is expressed in full scale.
logarithmic and the power law. Parameters z
0
, d
0
, a, u* mean roughness length and
om logarithmic fit, power exponent obtained from power fit, friction
root of constant value of Reynolds stress within the inertial sublayer),
respectively. All the parameters are listed in Table 1.
Matlab script. All the non-
mode) were interpolated
alibrated concentration records in
(1)
means a reference velocity measured at the
using the maximum of
correlation between both signals. The synchronised time series were shifted by an average of
a suck of the sample into the intake of the
the probe. The value of the shift
Contini et al. (2006).
2H
fluxes of passive
correlation method, see
mean average vertical
< > means a time average, w and c*
indicates fluctuations of vertical velocity and dimensionless concentration. Similar approach
ess elements placed it the
were measured with a two-
above the studied intersection and upstream
are depicted in Figure
are depicted in Figure 5.
* mean roughness length and
om logarithmic fit, power exponent obtained from power fit, friction
root of constant value of Reynolds stress within the inertial sublayer),
42
Figure 4: The vertical profiles of mean longitudinal velocity and momentum flux.
Figure 5: The vertical profiles of longitudinal and vertical turbulent intensity

Table 1: Parameters of the modelled boundary layer

Categories of boundary layer are defined according to classification in VDI Guideline (2000).
Measured parameters corresponded to a
densely built-up area without much obstacle height variation.
To verify requirements for the
was found, see Townsend (1976)

where is kinematical viscosity. The experiment was carried out by building Reynolds
number Re
B
~ 21 000 that lies on the lower edge of interval for valid Townsend hypothesis.
The experiment was carried out by building Reynolds number in the interval for valid
Townsend hypothesis. Free stream velocity was

z
0
[m] d
0
[m]
a
0.83 13.40 0.24
, Re
2
H U
H
B
=

of mean longitudinal velocity and momentum flux.

longitudinal and vertical turbulent intensity.
Parameters of the modelled boundary layer in full scale.

corresponded to a neutrally stratified boundary layer flow
up area without much obstacle height variation.
the Townsend hypothesis the critical Reynolds building number
, see Townsend (1976). Building Reynolds number is given by

is kinematical viscosity. The experiment was carried out by building Reynolds
000 that lies on the lower edge of interval for valid Townsend hypothesis.
ownsend hypothesis. Free stream velocity was approx. 4 m.s
1
.
u* [m/s]
0.27
neutrally stratified boundary layer flow above a
Townsend hypothesis the critical Reynolds building number
(2)
is kinematical viscosity. The experiment was carried out by building Reynolds
000 that lies on the lower edge of interval for valid Townsend hypothesis.
43
3 RESULTS
We measured dimensionless vertical advective
of passive contaminant in the horizontal plane above the studied intersection to determine
processes of vertical ventilation of the area
Robins (2009). Computed fluxes express
advective and turbulent transport
flux outwards (positive ventilation of the are
intersection (contaminating of the area). Results were obtained for 5 values of the approach
flow angle = 0, 5, 15, 30 and 45.
3.1 Vertical advective flux field
Values of computed advective fluxes for three approach flow directions are plotted in Figures
6a) 6c). As you see from the first Figure 6a), we got rather uniform
values of vertical advective flux for appro
= 30, there is an area of the
to the flow direction), see Figure 6b)
by approach flow angle = 45
fluxes in the right and the left corner of the area (relative to the flow direction)
6c).

Figure 6: Vertical dimensionless advective flux of passive contaminant for 3 approach flow directions
the studied intersection.
3.2 Vertical turbulent flux field
Values of computed advective and turbulent fluxes for three approach flow directions are
plotted in Figures 7a) 7c). We measured relatively flat turbulent flux field by angle
but, compared with the advective flux,
case of angle = 15 there are significantly positive values on the upwind side o
see Figure 7b). The observed phenomenon became stronger in case
We estimated a significant turbulent transport of pollution near the leeward side of the
buildings, see the upper part of Figures 7a) and 7b). In compariso
the turbulent fluxes are positive in every case. The
almost two times the magnitude of
We measured dimensionless vertical advective C*W/U
2H
and turbulent <
in the horizontal plane above the studied intersection to determine
processes of vertical ventilation of the area, see similar approach in Belcher
Computed fluxes express a rate of emissions spreading through a unit area by
advective and turbulent transport with following convention of signs: the positive sign means
flux outwards (positive ventilation of the area) and the negative sign means flux inwards the
= 0, 5, 15, 30 and 45.
dvective flux field
6c). As you see from the first Figure 6a), we got rather uniform and slightly negative
values of vertical advective flux for approach flow angle = 0. In the second case for angle
here is an area of the moderately positive flux on the right side of the
, see Figure 6b). There is mostly positive flux on the top of the crossing
= 45, but in there are places with significantly negative advective
fluxes in the right and the left corner of the area (relative to the flow direction)
dimensionless advective flux of passive contaminant for 3 approach flow directions
Vertical turbulent flux field
We measured relatively flat turbulent flux field by angle
with the advective flux, there is a positive turbulent transport of pollution. In
there are significantly positive values on the upwind side o
The observed phenomenon became stronger in case = 45, see Figure 7c).
buildings, see the upper part of Figures 7a) and 7b). In comparison with advective transport,
the turbulent fluxes are positive in every case. The magnitude of mean values
magnitude of mean advective fluxes.
and turbulent <c*w>/U
2H
fluxes
in the horizontal plane above the studied intersection to determine
Belcher (2005) and
a rate of emissions spreading through a unit area by
positive sign means
) and the negative sign means flux inwards the
and slightly negative
= 0. In the second case for angle
positive flux on the right side of the area (relative
. There is mostly positive flux on the top of the crossing
, but in there are places with significantly negative advective
fluxes in the right and the left corner of the area (relative to the flow direction), see Figure

dimensionless advective flux of passive contaminant for 3 approach flow directions above
We measured relatively flat turbulent flux field by angle = 0,
e turbulent transport of pollution. In
there are significantly positive values on the upwind side of the area,
= 45, see Figure 7c).
n with advective transport,
of mean values achieved
44
Figure 7: Vertical dimensionless turbulent flux of passive contaminant for 3
studied intersection.
4 CONCLUSIONS
The described wind tunnel experiment quantified vertical
intersection in an idealised symmetrical urban area
flow. A unique experimental set
the tracer gas concentration was assembled
laser Doppler anemometer. To determine ventilation character, a
fluxes of passive contaminant were computed from obtained synchronised signals.
In cases of small angles of approach flow direction we obtained almost flat distribution of
flux values in the studied area. In these cases, we
positive turbulent transport.
The comparison of the advective and turbulent fluxes above the intersection has shown a
significant positive contribution of the turbulent transport
contribution became stronger with higher angles of approach flow direction.
approach flow angles we also
turbulent and advective pollution transport in the area from the shape of flux field
magnitude of the average turbulent fluxes in individual grid points reach up to two times
higher values than the magnitude of
The realised experiment was a follow
(2010). Compared with former measured
directly in the street canyons within the same urban area model, the determined vertical
turbulent transport magnitude achieved only approx.
transport magnitude.
7 ACKNOWLEDGEMENT
Authors acknowledge financial support of project
Education, Sports and Youth of the Czech Republic
Academy of Sciences of the Czech Republic.
Environmental Wind Tunnel Laboratory working group
to Hamburg.
dimensionless turbulent flux of passive contaminant for 3 approach flow directions above the
The described wind tunnel experiment quantified vertical ventilation
d symmetrical urban area for several direction
nique experimental set-up for simultaneous measurement of the
concentration was assembled, based on fast-response ionisation detector and
To determine ventilation character, advective and turbulent scalar
flux values in the studied area. In these cases, we found out a negative advective but
contribution of the turbulent transport to the ventilation of the area. This
became stronger with higher angles of approach flow direction.
also determined considerably different distribution of vertical
turbulent and advective pollution transport in the area from the shape of flux field
the magnitude of the average advective fluxes.
The realised experiment was a follow-up to a former study of the author, see Kuka
former measured values of horizontal advective fluxes measured
turbulent transport magnitude achieved only approx. one fifth of the horizontal ad
ACKNOWLEDGEMENT
acknowledge financial support of project AVOZ20760514 of
Education, Sports and Youth of the Czech Republic and project M100760901 of
Academy of Sciences of the Czech Republic. Author also would like to thanks to
Environmental Wind Tunnel Laboratory working group, Meteorological Institute, University

approach flow directions above the
ventilation of the X-shaped
directions of the approach
the flow velocity and
response ionisation detector and
dvective and turbulent scalar
negative advective but a
ventilation of the area. This
became stronger with higher angles of approach flow direction. For higher
determined considerably different distribution of vertical
turbulent and advective pollution transport in the area from the shape of flux fields. The
up to a former study of the author, see Kukaka et al.
values of horizontal advective fluxes measured
one fifth of the horizontal advective
of the Ministry of
M100760901 of the
Author also would like to thanks to
, Meteorological Institute, University
45
6 REFERENCES
Arya, S. P., 1999. Air pollution meteorology and dispersion. Oxford University Press, New York.
Barlow J. F., Belcher, E.S., 2002. A wind tunnel model pro quantifying fluxes in the urban boundary layer.
Boundary-Layer Meteorology, 104, 131-150.
Britter, R. E., Hanna, S.R., 2003. Flow and Dispersion in Urban Areas. Annu. Rev. Fluid Mech., 35, 469-96.
Brown, M. J., Khalsa, H., Nelson, M., Boswell, D., 2004. Street canyon flow patterns in a horizontal plane,
measurements from the joint URBAN 2003 field experiments, in: Proceedings of the AMS symposium on
the urban environment, Vancouver, B.C.
Belcher, E. S., 2005. Mixing and transport in urban areas. Phil. Trans. R. Soc., A 363, 2947-2968.
Dabberdt, W., Hoydysh, W., Schoring, M., Yang, F., Holynskyi, O., 1995. Dispersion modelling at urban
intersections. Sci. Total Env., 196, 93-102.
Contini, D., Hayden, P., Robins, A., 2006. Concentration field and turbulent fluxes during the mixing of two
buoyant plumes. Atmospheric environment, 40, 7842-7857.
Hall. D. J., Emmont M.A., 1991. Avoiding aerosol sampling problems in fast response flame ionisation
detectors. Experiments in Fluids, 10, 237-240.
Jurkov, K., Masaaki O., Jaour, Z., 2009. Contribution of advective and turbulent mass transfers to the
ventilation of urban canopy, in: Proceeding of the 7
th
Asia-Pacific conference on wind engineering, Taipei,
Taiwan.
Kellnerov, R., 2009. Troubles with symmetry in wind-tunnle modeling, in: Proceedings of the International
Workshop on Physical Modelling of Flow and Dispersion Phenomena, Rhode-St-Genese, Belgium.
Kukaka, L., Kellnerov, R., Jurkov, K., Jaour Z., 2010. Estimation of scalar fluxes within modelled
intersection depending on the approach flow direction. In proceedings of the 13th International Conference
on Harmonisation within Atmospheric Dispersion Modelling for Regulatory Purposes, Paris.
Robins, A., 2008. DAPPLE (dispersion of air pollution and its penetration into the local environment)
experiments and modelling, HPA Chemical Hazards and Poisons Report, 13, 24-28.
Robins, A., Savory, E., Scaperdas, A., Grigoriadis, D., 2002. Spatial variability and Source-receptor relations at
a street intersection. Water, Air and Soil Pollution, 2, 381-393.
Scaperdas, A., Colvile, R.N., 1999. Assessing the representativeness of monitoring data from an urban
intersection site in central London, UK, Atmos. Env., 33, 661-674.
Stull, R. B., 1988. An Introdution to Boundary Layer Meteorology. Kluwer Aademic Publishers, Dordrecht.
Townsend, A.A., 1976. A Structure of Turbulent Shear Flow. Cambridge University Press.
VDI Verein Deutcher Ingenieure, 2000. Physical modelling of flow and dispersion processes in the atmospheric
boundary layer application of wind tunnels. Beuth Verlag, Berlin.
Wang, X., McNamara, K.F., 2007. Effects of street orientation on dispersion at or near urban street
intersections. J. Wind Eng. Ind. Aerodyn., 95, 1526-1540.
46
Wind Tunnel Study on Pollutant
Dispersion in Street Canyon with
Changing Aspect Ratio and Wind Direction
Mohamed F. Yassin
a
, Masaake Ohba
b
,
a
Department of Environmental Technology and Management, Kuwait
University, Kuwiat, mohamed_f_yassin@hotmail.com
b
Department of Architecture Engineering, Tokyo Polytechnic University, Kanagawa, Japan

ABSTRACT: Pollutant dispersion in street canyon was investigated using wind-tunnel as a
research tool under neutral atmospheric conditions. It was conducted using tracer gas
techniques from a line source without buoyancy. The street canyon model was formed of six
parallel building rows of the same length. The flow and dispersion field were analyzed and
measured using a hot wire anemometer with a split-fiber probe and a fast flame ionization
detector (FID). The diffusion flow field in the boundary layer inside the urban street canyon
was examined at different locations of varying geometry of the street canyon (aspect ratio=
width to height; W/H = , & 1) and wind directions (NNN, NNE, NE and ENE) in the
downwind distance of the leeward side of the street canyon model. The results show that the
pollutant concentration decreases exponentially in the vertical direction. The pollutant
concentration decreases as the wind direction increases from NNN=0
o
here.
1 INTRODUCTION
Pollutant dispersion inside the urban street canyon is of great importance given its direct
implications on the health of human beings living in the urban environment. The pollutants
emitted into a street canyon tend to disperse less than those emitted in an open area, and
therefore, the air quality becomes a serious problem in most large cities, especially because of
the emissions from the motor vehicles. To understand pollutant dispersion in urban street
canyons and to help urban planners to take into account urban geometry and wind direction
with optimal dispersion, investigation of the street-canyon flows is the first essential step.
Several experimental simulation studies in wind tunnels have been conducted to investigate
the flow and pollutant dispersion within the street canyon (Hoydysh et al. 1974, Kennedy and
Kent 1977, Wedding et al. 1977, Builtjes 1984, Hosker and Pendergrass 1987, Hoydysh and
Dabberdt 1988, Davidson et al. 1992). Hoydysh and Dabberdt (1994) and Hoydysh et al.
(1995) carried out wind tunnel experiments for a grid of orthogonal streets, measuring the
concentrations of a tracer gas within the intersections. These studies demonstrated that the
concentrations varied significantly within the intersection; with maximum values consistently
located at street corners. The study also showed that the street aspect ratio had an important
influence on conditions within the intersection. Meroney et al. (1996) studied the line source
characteristics at different street canyons widths. Pavageau and Schatzmann (1999) examined
the pollutant concentration distribution in street canyon in a wind tunnel with urban
47
roughness. Kastner-Klein and Plate (1999) investigated the influence of roof shape on the
distribution of pollutants within the canyon. Gardes and Olivari (1999) also investigated the
pollutant contamination of an urban street in a wind tunnel. They studied the influence of the
following parameters: the landscape upstream of the canyon, the ratio between the height of
upstream and downstream canyon walls and the spacing between the canyon walls.
Recently, there have been several studies of pollutant dispersion in urban street using wind
tunnel experiments. Robins et al. (2002) performed wind tunnel experiments to investigate
the dispersal of pollutants in an urban street intersection. Their results showed that the flow
was very sensitive to the wind direction relative to the intersection, so that even small
asymmetries in the configuration could lead to very different dispersion patterns. Park et al.
(2004) characterized the dispersion of vehicle emissions by conducting wind tunnel tests. The
aspect ratio of a street canyon (i.e. the ratio of the width of a street and the average height of
buildings) and the direction of external wind are the major test parameters. Ahmed et al.
(2005) extensively reviewed the various wind tunnel studies on pollutant dispersion at urban
street canyons and intersections. Simoens et al. (2007, 2008) made measurements of the
scalar dispersion of smoke released from a two-dimensional slot in the wall perpendicular to
a boundary layer flow and located parallel to and midway between two square obstacles
placed on the wall. Gromke and Ruck (2007, 2009) carried out wind tunnel studies on the
impact of avenue-like tree planting on flow fields and dispersion of traffic exhausts in urban
street canyons.
The aim of the present work is to investigate the impact of aspect ratio and wind direction on
dispersion of vehicle emission inside street canyon in a boundary layer wind tunnel
experiments. For this purpose, mean pollutant concentration fields are analyzed and discussed
at different locations in the downwind distance of the leeward side of the street canyon under
various configurations of the aspect ratio of the street canyon (W/H) and wind directions.

Fig.1: A photo of the street canyon with different wind
2 EXPERIMENTAL METHOD
Diffusion experiments are performed in the boundary layer wind tunnel at the Tokyo
Polytechnic University, Japan, under neutral atmospheric conditions. A simulated
atmospheric boundary layer is obtained by using a combination of barrier wall, elliptic vortex
generators and roughness elements on the floor of the tunnel. This kind of combination
produced a simulated atmospheric boundary layer with a normal depth, , of 0.6 m and a free
stream wind speed,
!
U of 5 ms
-1
. The power law U Z
n
was applied to the vertical wind
profile. The typical value of for n in an urban area is employed. More details of the wind
tunnel experiments can be found in Yassin and Ohba (2008).
The model of urban street canyon used in the wind tunnel experiments is shown in Fig. 1.
The modeled street canyon consists of six idealized long streets. These are aligned
perpendicular to the prevailing flow direction in the working test section to form the street
canyons. All the idealized street canyons have the following values respectively; H=74 mm
NNN=0
0

NNE=22.5
0
NE=45
0
NEE=67.5
0

48
height of the street, L = 1000 mm length of the street, and B =95 mm the width of the
buildings. The scale of the street canon model is taken as 1:400 to the full scale construction.
The widths of streets vary with respect to the street aspect ratio; W/H is , & 1. Four wind
directions; NNN= 0
0
(wind perpendicular to street), NNE=22.5
0
, NE=45
0
and ENE=67.5
0

were studied of the three street aspect ratio.
A line source is used to simulate emissions from vehicles along the centerline of the street
canyon model. Ethylene, C
2
H
4
is used as the tracer gas. The tracer gas has been emitted
from line sources co-located with the traffic lines. The detection of tracer concentration is
achieved with a high response Flame Ionization Detector (FID). Each measurement is carried
out for 90 s and produced an average value. In the present study, the emission velocity from
the line source is smaller than that of the free stream velocity. Therefore, the effluent
velocity of the pollutant is assumed to be negligible. Since a density of C
2
H
4
gas is almost
the same as that of the air, the density of the pollutant gas can be thought to have the same
density at the height of the pollutant effluent in the boundary layer. The mean wind velocity
at model height U
H
was set to 3.459 m/s. As a result, the Reynolds number Re based on
street canyon height is about 2 x 10
4
.

Fig.2: Measurement locations in street canyon model
To evaluate the effects of the structural geometry and wind direction on the pollutant
dispersion characteristics in urban street canyon, wind tunnel experiments are examined by
studying twelve cases of street configurations under neutral atmospheric conditions. The
measurements are made in the center of the long street canyon of the leeward side in the
downwind distance at four different locations; X/H =0.3 (leeward), 0.4, 0.6 (middle) and 0.7
(windward), which is shown in Fig. 2. The characteristics of the pollutant dispersion included
are the mean concentration and concentration fluctuation intensity. Figs. 3-5 show contour
lines of the normalized mean concentration inside the street canyon of the three street aspect
ratios, W/H for different wind directions. These figures show that the pollutant is carried to
the upwind side from the line source and dispersed further in the street canyon. The pollutant
concentration on the upwind side is higher than that on the downwind side building at the
lower region of the street canyon as shown in Figs. 3-5. This is because the horizontal
velocity of the lower region of street canyon was negative, where the vortex is clockwise.
Pollutant dispersion is solely removed vertically from the street canyons to the free surface
layer by the vertical mean flux and vertical turbulent flux. The pollutant concentration at the
upwind side of the aspect ratios W/H= 0.75 and 1.0 is higher than that in the aspect ratios
W/H= 0.5 because the main vortex is distorted along the street roof. The pollutant
concentration is increased from top to bottom of the aspect ratios W/H= 0.75 and 1.0, and the
Wind
H
Up-roof Down-roof
Up-building
Down-building
U
p
w
i
n
d

w
a
l
l

D
o
w
n
w
i
n
d

w
a
l
l

B W B

X
/
H
=
0
.
3

X
/
H
=
0
.
4

X
/
H
=
0
.
6

X
/
H
=
0
.
7

49

Fig.3 Contour lines of mean concentration, K for W/H=0.5
Fig.4 Contour lines of mean concentration, K for W/H=0.75
NNN
NNE NE NEE
NNN
NNE NE NEE
NNN
NNE NE NEE
Fig.5 Contour lines of mean concentration, K for W/H=1
50

Fig.6 Vertical profiles of mean concentration, K
51
highest concentration is found in the corner of the down upwind side building. The highest
concentration is found in the corner of the down downwind building of the aspect ratio W/H=
0.5 for the wind direction NNN. The lowest pollutant concentration is appeared in the corner
of the down leeside building of the aspect ratios W/H= 0.75 and 1.0.
The vertical profiles of the normalized mean concentration are measured through the neutral
atmospheric conditions at different locations inside the street canyon of the three aspect
ratios, W/H and four wind directions, which are shown in Fig. 6. The pollutant concentration
of the leeward side is observed to be higher than that of the windward side. This is because
the longitudinal velocity is lower in the leeward side. The pollutant concentration inside the
street canyon increased towards the ground surface and decreased up to the roof canyon. The
pollutant concentration of the aspect ratio W/H= 0.5 demonstrates to be higher than that of
the aspect ratios W/H= 0.75 and 1.0 due to the lower vertical velocity. The pollutant
concentration decreases with the downwind distance of aspect ratios W/H= 0.75 and 1.0 in
the middle region at X/H=0.4 & 0.6. It can be seen that the plume spreads very slow in the
leeward side. Whereas, a clockwise vortex is generated inside the street canyon because the
longitudinal velocity near the ground surface is negative. Therefore, the pollutant discharged
from the line source transportes by the clockwise vortex via the negative longitudinal velocity
to the leeward side gave rise to maximum concentrations at the lower locations in this side.
Then, the lower clockwise vortex carried the pollutant to the windward side, which led to
pollutant accumulation at the lower location in this side. Since, the lower clockwise vortex is
very weak, it is difficult to flush the pollutant out the street canyon. These results confirms
the previous results by Huang et al. (2009)
On the other hand, it shows clearly that the dispersion pollutant transports to the windward
side, when the longitudinal velocity is positive. The maximum concentration of the aspect
ratio W/H= 0.5 with wind direction NNN occurred in the middle region at X/H=0.4 & 0.6
and windward side due to the lower turbulent energy in the middle region. The minimum
pollutant concentration of the aspect ratio W/H= 1.0 for wind directions; NE and NEE
occurred in the windward side due to the higher turbulent energy in this side. The pollutant
concentration in the cavity region is found to be higher for wind direction NNN than that for
the other wind directions. The value of pollutant concentration is not much different in the
leeward side, except near the ground surface. The results show that the pollutant
concentration increases as the aspect ratio decreases. The concentration decreases
exponentially in the vertical direction. The pollutant concentration decreases as the wind
direction increases from NNN=0
o
, confirming the previous results obtained using wind tunnel
(e.g. Hoydysh and Dabberdt, 1988).
3 CONCLUSION
The impact of the variability in the structural geometry, and wind direction to the
pollutant dispersion inside urban street canyon was studied using a wind tunnel. It is
indicated that the structural geometry, and wind direction had strong influence on the wind
flow and pollutant behavior. The results show that the pollutant concentration decreases as
the wind direction increases from NNN=0
o
. The concentration fluctuation intensities of the
aspect ratio W/H= 1.0 is found lower than that of the aspect ratios W/H= 0.5 and 0.75. The
pollutant concentration at the leeward side of the aspect ratios W/H= 0.75 and 1.0 is higher
than that in the aspect ratios W/H= 0.5. The highest concentration is found in the corner of
the down windward side building of the aspect ratio W/H= 0.5 for the wind direction NNN.
The lowest pollutant concentration is appeared in the corner of the down leeward side
52
building of the aspect ratios W/H= 0.75 and 1.0. The pollutant concentration in the cavity
region is found to be higher for wind direction NNN than that for the other wind directions.
4 REFERENCES
Ahmed K., Khare M. and Chaudhry K.K., 2005. Wind tunnel simulation studies on
dispersion at urban street canyon and intersections-A review. J. Wind Engrg. & Aero. 93:
697-717.
Boerner, T., Leutheusser, H.J, 1984. Calibration of split fiber probe for use in bubby two-
phase flow. DISA Info., 29, 10-13.
Builtjes P.J.H., 1984. Determination of flow- and concentration filed in a street canyon by
means of wind tunnel experiments. TNO Report No. 84-02616, Apeldoorn, The Netherlands.
Dabberdt W.F. and Hoydysh W.G., 1991. Street canyon dispersion: Sensitivity to block shape
and entrainment. Atmospheric Environment 25A: 1143-1153.
Gerdes F. and Olivari D., 1999 Analysis of pollutant dispersion in an urban street canyon. J.
Wind Engrg. & Aero. 82: 105-124.
Davidson M.J., Snyder W.H., Lawson J.R. and Myline K.R., 1992. Wind tunnel and field
investigations into plume dispersion through an array of obstacle. Proc. 11
th
. Australasian
Fluid Mechanics Conf., Hobart, Australia.
Gromke C, Ruck B., 2007. Influence of trees on the dispersion of pollutants in an urban
street canyon experimental investigation of the flow and concentration field. Atmospheric
Environment, 41, 3287302.
Gromke C, Ruck B., 2009. On the impact of trees on dispersion processes of traffic emissions
in street canyons. Boundary Layer Meteorology, 131, 1934.
Huang Y., Hu X. and Zeng N., 2009. Impact of wedge-shaped roofs on airflow and pollutant
dispersion inside urban street canyons. Building and Environment, 44, 2335-2347.
Hosker R.P., Pendergrass W.R., 1987. Flow and dispersion near clusters of buildings.
NOAA Technical Memorandom ERL-ARL-153. Department of Commerce, USA.
Hoydysh W.G., Griffith R. A. and Ogawa Y., 1974. A scale model study of the dispersion of
pollution in street canyons. 67
th
Annual Meeting of the Air Pollution control Association.
Denver, Colorado, June 9-13.
Hoydysh W.G. and Dabberdt W.F., 1988. Kinematics and dispersion characteristics of flows
in asymmetric street canyons. Atmospheric Environment 22: 2667-2689.
Hoydysh W.G. and Dabberdt W.F., 1994. Concentration fields at urban intersections: Fluid
modeling studies. Atmospheric Environment 28(11): 1849-1860.
Hoydysh,W.G., Dabberdt,W.F., Schorling, M., Yang, F., Holynskyj, O.,1995. Dispersion
modeling at urban intersections. Science of the Total Environment 169, 93102.
Kastner-Klein P. and Plate E.J., 1999. Wind Tunnel study of concentration fields in street
canyon. Atmospheric Environment 33: 3973-3979.
Kastner-Klein P., Fedorovich E. and Rotach M.W., 2001 A wind tunnel study of organized
and turbulent sir motions in urban street canyons Journal of Wind Engineering &
Aerodynamics, 89, 849-861.
Kennedy L.M. and Kent J.H., 1977. Wind tunnel modeling of CO dispersal in city streets
Atmospheric Environment 11: 541-547.
Kiya, M., Saski, K., 1983. Structure of turbulent separation bubble. J. Fluid Mech. 137, 83-
113.
Li X, Dennis Y., Leung C., Liu C., 2008. Physical Modeling of flow inside urban street
canyon Journal Meteorology & Climatology, V.47, pp.2058-2066.
53
Meroney R.N., Pavageau M., Rafailidis and Schatzmann M., 1996. Study of line source
characteristics for 2-d physical modeling of pollutant dispersion in street canyons. J. Wind
Engrg. & Aero. 62: 37-56.
Mfula, A.M., Kukadia, V., Griffiths, R.F. and Hall, D.J., 2005. Wind tunnel modeling of
urban building exposure to outdoor pollution. Atmospheric Environment, 39, 2737-2745.
Oke T.O., 1988. Street design and urban canopy layer climate, Energy Buildings 11, PP.
103113.
Pavageau M. and Schatzmann M., 1999. Wind tunnel measurements of concentration
fluctuation in an urban street canopy. Atmospheric Environment 33: 3961-3971.
Park S., Kim S. and Lee H., 2004. Dispersion characteristics of vehicle emission in an urban
street canyon. Science of the Total Environment 323: 263-271.
Robins A., Savory E., Scaperdas A. and Grigoriadis D., 2002. Spatial variability and source-
receptor relations at a street intersection Water, Air, and Soil Pollution: Focus 2:381-393.
Rotach M.W., 1995. Profiles of turbulence statistics in and above an urban street canyon
Atmospheric Environment 29, 1473-1486.
Simoens S., Ayrault M., and Wallace J., 2007. The flow across a street canyon of variable
widthPart 1: Kinematic description, Atmospheric Environment 41, 90029017
Simoens S., Ayrault M., and Wallace J., 2008. The flow across a street canyon of variable
widthPart 1: Scalar dispersion from a street level line source, Atmospheric Environment
42, 24892503.

54
Modelling of near-field pollutant dispersion in the built
environment: methods and challenges
B. Hajra
a
, M. Chavez
a
, T. Stathopoulos
a*
, A.Bahloul
b

a
Department of Building, Civil and Environmental Engineering, Concordia
University, Montreal, Canada
b
Institut de recherche Robert-Sauv en sant et en scurit du travail,
Montreal, Canada
*Corresponding email: statho@bcee.concordia.ca
ABSTRACT: Airborne pollutants released from rooftop stacks can re-enter the building from
which they are released and affect an adjacent building causing potential health hazards.
Therefore, an accurate assessment of plume dilution in the urban environment is necessary.
Physical modelling through wind tunnel studies has been widely used due to its greater
reliability over numerical simulations. Although, Computational Fluid Dynamics (CFD) and
Gaussian-based dispersion models have also been widely used to simulate near-field plume
dilution, validations with experimental data are required due to the CFDs questionable
ability to accurately represent airflow and dispersion in an urban environment. This paper
discusses key aspects of wind tunnel simulation criteria, recirculation length estimation in
dispersion models and some problems in CFD modelling. In order to accurately simulate
urban dispersion characteristics, wind tunnel simulations must take account of various
pertinent issues such as: scaling considerations, effect of background concentrations and
others. The study shows that although wind tunnel modelling is a good representation of full-
scale measurements, certain simulation criteria such as stack Reynolds number and scaling
considerations must be revisited. Dispersion models must estimate the recirculation length of
a building based on local topography and turbulence besides just building dimensions. In
general, no fixed approach such as the use of a particular turbulence model, turbulent
Schmidt number (Sc
t
) or grid resolution can be adopted. Rather, one must consider the flow
characteristics at the micro-scale which are largely governed by turbulence due to buildings,
before deciding to apply a particular CFD approach. Indeed, computational and experimental
techniques must be used in parallel for a better understanding of urban pollutant dispersion.
KEYWORDS: Building; CFD; Dispersion; Physical modelling; Reynolds number; Urban
environment; Wind
1 INTRODUCTION
Air pollution is ubiquitous and a major cause of concern especially for occupants of a
laboratory building who are exposed to particulate matter for long durations. This can be
harmful to human health, and although in the past mortality rates have been linked even to
short term exposure to air pollution (Schwartz, 1994; Dockery et al., 1993). This greatly
stresses the importance of accurate dispersion modelling in the built environment by which
pollutant concentrations in the vicinity of the source can be estimated.
55
Estimating pollutant concentration in the near-field is important and challenging because of
the complexity of air flow due to the turbulence generated by the buildings and local
topography (Saathoff et al., 2009). The term near-field has been defined by various
researchers in different ways. For instance, studies carried out by Upadhay et al., 2004 within
the urban canopy through large groups of obstacles and similarly, wind tunnel experiments by
Li and Meroney, 1983 have defined the near-wake region as x/H < 5, where x is the
distance of the receptor from the source and H is the height of the building. Wilson et al.,
1998 performed water channel studies to assess plume behaviour in the presence of adjacent
buildings and defined near-field as a region within the zone of recirculation of a building
which is estimated from the windward wall dimensions of the emitting building. Therefore,
most studies have defined near-field as the distance within close proximity of the emitting
building (Hajra et al., 2010).
Currently, near field pollutant dispersion can be assessed by physical modelling, which
consists of wind tunnel and water channel experiments whilst computational modelling can
be performed by CFD and dispersions models. This paper primarily focuses on the challenges
faced by physical modelling techniques, inappropriate calculations of recirculation length in
dispersion models and a few deficiencies in CFD modelling that need to be addressed.

2 CRITERIA IN PHYSICAL MODELLING
In this section wind tunnel and water channel modelling criteria of pollutant dispersion is
discussed. The focus is on Snyder, 1981 criteria and the averaging time used in the
experiments.
2.1 Snyder, 1981 criteria
In general, to model non-buoyant plume dispersion in a wind tunnel, the criteria suggested by
Snyder, 1981, need to be satisfied. These include:
Geometric similarity
This means that the buildings in a full scale model (prototype) should bear the same shape as
the model tested in the wind tunnel, at a reduced scale. However, Saathoff et al., 1995
showed that changes in geometric scales could produce differences in measured
concentrations, as shown in Figure 1.

Figure 1. The influence of the model scale on the concentration coefficient measured on the model
centerline in suburban terrain (Saathoff et al., 1995)

The investigations consisted of wind tunnel tests of an isolated building model constructed at
three different scales: 1:500, 1:250 and 1:125. Pollutants released from rooftop vents were
estimated at various rooftop receptors and were also compared to field studies performed by
56
Ogawa et al., 1983. It was reported that downstream of the vent at x/h ~ 0.25 and x/h ~ 0.45,
concentrations obtained with the small model are approximately twice as large as those
obtained with the large model. However, the effect of scale was mostly confined to rooftop
locations. This discrepancy occurs possibly because Snyders criterion was based on wind
tunnel measurements for far-field studies, in which turbulence due to buildings is reduced.
Most tracer studies in the wind tunnel apply this criterion irrespective of whether it is a
micro-scale or far-field pollutant dispersion problem.

The other criteria that take account of building and stack turbulence are:
Building Reynolds Number > 11000
Stack Reynolds Number > 2000

It is generally necessary to match the Reynolds number measured in the wind tunnel and full
scale because the turbulence generated in the vicinity of the source is greatly responsible for
dispersing the plume in the field. However, due to smaller sized models used in the wind
tunnel this is generally not possible. For near-field pollutant dispersion studies, Saathoff et
al., 1995 suggested that it is generally not possible to satisfy the stack Reynolds number for
small diameter stacks and it is also difficult to trip the flow for such stacks. Since Snyders
criteria are not defined for a particular scale, smaller models tested in the wind tunnel may
not satisfy them. Additionally, using larger models could result in higher blockage ratio. For
instance, ASCE 1999 states that blockage ratio, which is defined as A
m
/A
o
, should not exceed
5%, where
A
m
is the cross-sectional area of the model (m
2
),
A
o
is the cross sectional area of the wind tunnel (m
2
)
Additionally, no criterion is set for Roof Top Structures (RTS) that may be located on a
building. Studies by Saathoff et al., 2009 have shown that an RTS can increase rooftop
concentrations significantly compared to an isolated building. Therefore, this makes it
necessary to re-define this criterion taking into consideration the model scale and other
elements of a building (RTS, balconies), which may also contribute to additional turbulence.

Equivalent exhaust momentum ratio.

Exhaust momentum ratio (M) is defined as:
) / ( ) / (
5 . 0
H e a e
U V M = (1)
where
e
and
a
are the densities of exhaust and air respectively (kg/m
3
),
V
e
is the exhaust velocity (m/s),
U
H
is the wind velocity at building height (m/s)

However, in most wind tunnel studies the densities of air and exhaust are nearly equal
because the concentration of the tracer is mixed with sufficient quantities of air (see
Stathopoulos et al., 2008). Hence, equation 1 reduces to a ratio of velocities with the density
term cancelling out:

M = V
e
/U
H
(2)

Previous wind tunnel studies have shown that equation 2 holds good for non-buoyant plumes.
While this may be true for models constructed at a particular scale, a change in scale could
produce different results (Saathoff et al., 1995). Additionally, in wind tunnel measurements
sulphur hexafluoride (SF
6
) or helium (He) is used due to their chemically inert nature; SF
6
is
57
heavier than He owing to the latters low molecular weight (Chui and Wilson, 1988).
Therefore, applying Snyder, 1981 criteria for lighter (buoyant) gas could result in an incorrect
simulation.
Similar criteria are applied for water channel measurements. For instance, Wilson et al., 1998
performed water channel experiments using light scattering technique to assess plume
dilution for tracer released from rooftop stacks in the presence of upstream buildings. In this
case it may be easier to satisfy the building and stack Reynolds number criteria owing to the
greater density of water over air. However, discrepancies between water channel and wind
tunnel measurements have been reported in the past (Gupta, 2009). One of the main reasons
could be the application of Reynolds number criteria, which were based on wind tunnel
measurements by Snyder. Therefore, an alternate set of criteria must be introduced for
laboratory simulations of fluids (medium of transport) other than air flowing past rooftop
emissions.

2.2 Averaging time
Another major issue in physical modelling is proper simulation of field averaging time in the
wind tunnel. The averaging time for collecting samples in the field is not the same as in the
wind tunnel. This is because in the field the plume meanders and owing to the changes in
wind direction and speed, pollutant transport occurs mostly in longitudinal and transverse
directions; these are difficult to accurately reproduce in a wind tunnel. Sampling times in the
field are generally larger than in laboratory measurements due to the reduced scales used in
the wind tunnel (Kothary et al., 1981). In fact, some studies also report that when the stack
and receptor are in close proximity to each other, the averaging time effects are expected to
reduce (Hajra et al., 2011).
There is no precise definition available in literature for the appropriate use of averaging time.
For instance, ASHRAE 2007 considers the 60 minute field averaging time to an equivalent 2
minutes in the wind tunnel. Studies by Saathoff et al., 2009 have shown that wind tunnel
measurements carried out at an averaging time of about one minute is sufficiently accurate in
simulating hourly averaging time. Most dispersion models such as Industrial Source Complex
(ISC) also adopt a 2-minute averaging time in the wind tunnel representing an hourly field
averaging time. Although it is understood that hourly sampling times in the field reduce to a
few minutes in the lab, it is necessary to define clearly the averaging time to be used in wind
tunnel measurements since this could affect the accuracy of the simulation. For instance, a 2-
minute averaging time could allow greater accumulation of the plume within the lab
increasing the possibility of background concentrations compared to a 1-minute sampling
time.

3 RECIRCULATION LENGTHS IN DISPERSION MODELS
Numerical approaches consist of CFD and dispersion models, of which an important issue of
calculating recirculation length in the latter is discussed in this section. Recirculation length
refers to the dimension of a region that forms in the wake of a building whose size is mostly
estimated from the dimensions of the building (Wilson et al., 1998). This region causes the
plume released from rooftop stack to re-enter the building through the leeward wall.
Dispersion models such as ADMS and ISC are based on Gaussian equations that estimate
pollutant concentrations at rooftop receptors using computer programs. Accurate assessment
of the recirculation length of a building is important especially for near-field dispersion
problems because the air and pollutant flow in this region is complex and increased
downwash effects at low exhaust speeds result in accumulation of effluents within this region
(Hajra et al., 2011). However, most dispersion models assume a uniform concentration field
within this region which leads to an incorrect assessment of concentrations. Each model
58
follows a different approach in calculating the recirculation length. For instance ADMS uses
the equations by Fackrell and Pearce, 1981 whilst ASHRAE uses the formulation by Wilson,
1979. Recent studies by Hajra et al., 2011 have shown that recirculation lengths predicted
from ADMS and ASHRAE result in dilution estimates that do not compare well with
experimental results. Therefore, it is necessary to investigate this issue in greater detail. More
importantly, the recirculation length should also take account of upstream turbulence, effect
of rooftop structures and local topography besides building dimensions.
Dispersion models are of great importance mainly because their results are used by regulatory
agencies for designing of safe stack and intake locations. Owing to the cost involved in
experimental studies and inaccuracies in CFD modelling, it is imperative to improve the
existing dispersion models particularly for near-field problems since there are plenty of
validations with far-field problems available in literature (Stathopoulos et al., 2008). Some of
the main CFD modelling approaches of near-field dispersion are presented in the subsequent
section.

4 CFD MODELLING
Recently CFD modelling has been widely used due to increased accessibility to model
complex geometries as urban areas with dense high-rise buildings which are characterized by
a large range of turbulence length scales. Different approaches are available depending on
the turbulence treatment that includes Direct Numerical Simulations (DNS), Large-Eddy
Simulations (LES) and Reynolds Averaged Navier Stokes (RANS) models. DNS is a time-
dependent solution of flow equations which are directly computed in a very fine grid without
any turbulence model. Although the physics is well represented with this method, its
computational demand is extremely high (usually prohibitive) to be applied in wind
engineering studies. In LES the equations are filtered so scales of motion larger than the mesh
size are resolved explicitly and the smallest scales are simulated by the so-called subgrid-
scale model (SGS). This is a promising approach due to its natural way to capture the
relevant scales of motion and flow unsteadiness, which are crucial for transport mechanism;
however it still suffers from numerical issues such as its high sensibility on mesh quality and
significant computational requirement (Tominaga and Stathopoulos, 2011). RANS-based
models, which state the flow equations in an Eulerian framework upon control volumes are
the most widely-used methods in industry due to their significantly smaller computational
effort compared to high fidelity approaches mentioned previously. Unfortunately, these
models are generally unsuitable for near-field dispersion problems due to their inability to
accurately simulate the near-field turbulence caused by buildings and neighbouring structures
(Murakami et al., 1990). Nevertheless, a critical evaluation in terms of mass transport
accuracy in urban areas by comparison with experimental data and results from other
accurate methods, e.g. LES, still needs to be addressed in order to place all different
approaches in perspective. In general, CFD results need to be validated with experimental
studies and, hence, cannot be applied by designers directly. In this section only a few issues
that affect the simulations are presented.

4.1 Turbulent Schmidt number (Sc
t
)
Turbulent Schmidt number (Sc
t
) is necessary where the scalar transport equation is solved
using the standard gradient-diffusion hypothesis for turbulent scalar flux. Sc
t
is defined as the
ratio of turbulent momentum diffusivity (eddy viscosity) to the turbulent mass diffusivity. In
most CFD simulations a default value of 0.7 is chosen, a common feature of most
commercially available packages. Although this number may not affect the simulations of
tracer gas several kilometres away from the source, it affects the results markedly within the
near field. Recent studies by Chavez et al., 2011 suggest that lower values of Sc
t
give better
59
comparisons with wind tunnel data obtained on rooftop of isolated buildings as shown in
Figure 2.

Figure 2. Normalised dilution on rooftop of isolated building: a) M = 1; b) M = 3 (Chavez et al., 2011)

In fact, Figure 2 shows good comparisons between CFD and wind tunnel data for Sc
t
= 0.3 at
M = 1 and Sc
t
= 0.1 at M = 3. Therefore, a fixed value of Sc
t
cannot be used for near-field
simulations. In fact, a thorough understanding of the flow field is necessary and due
consideration must be given to the terrain conditions that affect the flow (Tominaga and
Stathopoulos, 2007).

4.2 Mesh resolution
Most CFD simulations use structured hexahedra grids defined by Hefny and Ooka, 2009 due
to good results obtained in the past. It was also reported that although tetrahedral elements
could be constructed much faster in complex geometries, they could affect the simulations by
increasing the level of numerical diffusion. However, Cowan et al., 1997 suggested that a
change in mesh resolution particularly for non-symmetric buildings may produce variations
in concentration distributions around the building.
Figure 3 shows comparisons between Upwind Differencing (UD), Self Filtered Central
Differencing (SFCD) and second order Upwind differencing (LUD) scheme, in the form of
normalised concentration iso-surface for coarse and fine grid resolution for an L-shaped
building with the wind direction parallel to the longer side. The study reported that the
general spread in the results with the various differencing schemes is not reduced much by
using the finer grid. Elsewhere, however, the mesh effects can dominate differencing scheme
effects. Therefore, a proper assessment of grid resolution is necessary to obtain accurate
results.

4.3 Other factors
In addition to grid size and Sc
t
, there are other factors that contribute to incorrect simulations
leading to discrepancies with experimental data. One of the major issues is the use of
different turbulence models. Recent studies have shown that LES predictions, despite being
time consuming, are not only better but also markedly different from RANS-based models for
simulating pollutant dispersion in street canyons (Tominaga and Stathopoulos, 2011). Other
factors such as surface boundary conditions, the inlet turbulence profile and domain size can
also contribute to significant discrepancies when CFD results are compared to experimental
data (Stathopoulos, 1997). Therefore, it is difficult to generalise any given technique for
simulating micro scale dispersion of airborne particles.
60

Figure 3. Spanwise surface normalized concentration profiles: () Coarse grid; (----) Fine grid (Cowan et
al. 1997).

5 CONCLUDING REMARKS
This paper summarises and discusses a few modelling issues encountered in wind tunnel,
dispersion models and CFD for near-field urban pollutant dispersion. Physical modelling
through wind tunnel studies is generally preferred to water channel measurements, although
the Reynolds number criteria are easier to achieve in water. Better measurement techniques
compared to the widely used light scattering, are required for accurate plume dispersion
studies in water channels. Guidelines given by Snyder, 1981 particularly for assessing stack
Reynolds number need to be re-visited because small sized models tested in the wind tunnel
do not adhere easily to this criterion. Additionally, guidelines for Reynolds number of RTS
must also be provided since past studies have shown increased rooftop concentrations in the
presence of RTS. A proper definition of averaging time is also needed because increased
averaging time in wind tunnel could lead to high background concentrations yielding
inaccurate measurements. Past wind tunnel studies have shown that different model scales
alter rooftop dilution; this makes it necessary to perform additional studies to investigate the
suitability of Snyders criteria for such cases.
Proper assessment of recirculation length is necessary to estimate accurate plume
concentrations for smoke released from rooftop stacks. In particular its estimation should be
based on the turbulence generated by the building and surrounding structures and local
topography besides building dimensions.
CFD simulations cannot be generalised for any given case. In fact, further validations with
experimental studies must be performed to assess the plume behaviour in the near field to
check the suitability of a given turbulence model and grid size. Turbulent Schmidt number
must be used with caution considering the near-field flow characteristics in the urban
environment. In the present scenario, CFD and dispersion models should never be used
directly for assessing stack and intake location because these are not fully reliable; more
efforts in this direction are required to alleviate their deficiencies.
In summary, this study shows that some of the assumptions in physical modelling criteria
need to be re-visited to achieve better comparisons with field data. Further validation of CFD
and dispersion models is imperative for isolated building cases prior to its further extent to an
array of buildings.

61

6 REFERENCES
ASCE 1999. Wind Tunnel Studies of Buildings and Structures. Manual of Practice No. 67, American Society of
Civil Engineers, Reston, VA, 20191-4400, USA.
ASHRAE 2007 Building Air Intake and Exhaust Design. ASHRAE Applications Handbook, Chapter 44,
American Society of Heating, Refrigeration and Air-Conditioning Engineers Inc., Atlanta, USA.
Cowan I.R, Castro I P. Robins A.G. 1997. Numerical considerations for simulations of flow and dispersion
around buildings. Journal of Wind Engineering and Industrial Aerodynamics 67&68, 535 545.
Chavez M, Hajra B, Stathopoulos T, Bahloul A 2011. Near-field pollutant dispersion in the built environment
by CFD and wind tunnel simulations. Journal of Wind Engineering and Industrial Aerodynamics, 99, 330-
339.
Chui E.H, Wilson D.J. 1988. Effect of varying wind direction on exhaust gas dilution. Journal of Wind
Engineering and Industrial Aerodynamics, 31, 87-104.
Dockery, D.W., Pope, C.A., Xu, X., Spengler, J.D., Ware, J.H., Fay, M.E., Ferris, B.G., Speizer, F.E. 1993. An
Association between Air Pollution and Mortality in Six U.S. Cities. The New England Journal of Medicine,
329, no. 24, 1753-1759.
Fackrell J.E and Pearce J.E. 1981. Parameters affecting dispersion in the near wake of buildings, CEGB report
RD/M/1179/N81.
Gupta A, 2009. Physical Modelling of the Downwash Effect of Rooftop Structures on Plume Dispersion. PhD
Thesis in the Dept. of BCEE, Concordia University, Canada
Hajra B, Stathopoulos T, Bahloul A, 2010. Assessment of pollutant dispersion from rooftop stacks: ASHRAE,
ADMS and wind tunnel simulation. Building and Environment. 45, 27682777.
Hajra B, Stathopoulos T, Bahloul A, 2011. The effect of upstream buildings on near-field pollutant dispersion in
the built environment. Atmospheric Environment, 45, 4930-4940.
Hefny, M., Ooka, R., 2009. CFD analysis of pollutant dispersion around buildings: effect of cell geometry.
Building and Environment 44 (8), 16991706.
Kothary K.M, Meroney, R.N, Bowmeester, R.J.B. 1981. An algorithm to estimate field concentrations in the
wake of a power plant complex under non steady meteorological conditions from wind tunnel experiments.
Journal of Applied Meteorology, 20, No. 8, 934-943.
Li, W.W and Meroney, R.N 1983. Gas dispersion near a cubical model building Part ii. Concentration
fluctuation measurements. Journal of Wind Engineering and Industrial Aerodynamics, 12, 35-47.
Murakami, S., Mochida, A., Hayashi, Y. 1990. Examining the k- model by means of a wind tunnel test and
large-eddy simulation of the turbulence structure around a cube. Journal of Wind Engineering and Industrial
Aerodynamics, 35, 87-100.
Ogawa Y, Oikawa S, Uehara, K. 1983. Field and wind tunnel study of the flow and diffusion around a model
cube - Part II. Near-field and cube surface flow and concentration patterns, Atmospheric Environment. 17,
No. 6, 1161-1171.
Saathoff P.J, Stathopoulos T, Dobrescu M, 1995. Effects of model scale in estimating pollutant dispersion near
buildings. Journal of Wind Engineering and Industrial Aerodynamics, 54, 549559.
Saathoff, P.J, Gupta, A., Stathopoulos, T., Lazure, L. 2009. Contamination of fresh air intakes due to downwash
from a rooftop structure. Journal of Air & Waste Management Association. 59, 343353.
Schwartz, J. 1994. Air pollution and daily mortality: A review and meta analysis. Journal of Environmental
research, 64, 36-52.
Snyder W. H. 1981. Guidelines for fluid modelling of atmospheric diffusion. EPA office of Air quality,
planning and standards, Research Triangle Park, USA, EPA-600/8-81-009.
Stathopoulos T. 1997. Computational wind engineering: Past achievements and future challenges. Journal of
Wind Engineering and Industrial Aerodynamics, 67&68, 509-532.
Tominaga, Y., Stathopoulos, T. 2007. Turbulent Schmidt numbers for CFD analysis with various types of flow-
field. Atmospheric Environment, 41, 80918099.
Tominaga, Y., Stathopoulos, T. 2011. CFD modelling of pollution dispersion in a street canyon: Comparison
between LES and RANS. Wind Engineering and Industrial Aerodynamics, 99, 340-348.
Upadhay, J.K, Kobayashi, N, Venkatram, A, Klewicki, J. 2004. Study of near-field dispersion through large
groups of obstacles. Journal of Asian Architecture and Building Engineering, 3, No. 2, 305-309.
Wilson, D.J. 1979. Flow patterns over flat roofed buildings and application to exhaust stack design. ASHRAE
Transactions, 85, 284-295.
Wilson D.J, Fabris I, Ackerman M.Y. 1998. Measuring adjacent effects on laboratory exhaust stack design.
ASHRAE Transactions 88(1), 513-533.
62
Wind Tunnel Measurements of the Aerodynamic Behavior
of Tree Branches
Anne Schrn
Meteorological Institute, Hamburg, Germany, anne.schroen@web.de
ABSTRACT: Trees and parks play an important role for the life in cities, especially regarding the
feeling good factor. However it might also have an impact on the air ventilation in cities. To find
out the actual impact of urban greenery on the air exchange in street canyons or cities in general
the aerodynamic behavior of real vegetation must be known. Therefore a special test section has
been build and mounted within the Gttingen-type wind tunnel of the Environmental Wind Tunnel
Laboratory (EWTL) at the Meteorological Institute in Hamburg.
1 INTRODUCTION
Several approaches have been made before to investigate the impact of urban greenery on the wind
field within street canyons. However, these studies used far to high wind velocities or tree models
that are not appropriate to represent real vegetation. An extensive literature research showed that
the collected data is not sufficient as the boundary conditions were not correctly chosen. The
problem in most cases is the lack of comparison studies with living vegetation. To investigate the
true aerodynamic behavior of trees a new approach was chosen. The first step is to study the
behavior of living tree branches and then to translate it into proper physical models. This study
deals mostly with designing a fully functioning test section that can be used to collect the
necessary data of living tree branches. Furthermore first tests with several tree species
representative for urban greenery have been done. Future improvements and approaches for
following studies have been ascertained. Yet the actual test section is absolutely suited for the
examination of real tree branches and the results can be used to improve physical models of urban
greenery.
2 THE DESIGN OF THE TESTSECTION
Preliminary investigations ensued that in spite of former plans to build a vertical wind tunnel
(Physmod 2009) a horizontal test section mounted within the Gttingen-type wind tunnel is just as
appropriate. Only for very small leaves and really low wind velocities the gravitational force is
more important than the wind induced force. Considering that oaks, limes and maples are the most
common trees in Hamburg (due to the BSU of Hamburg) a horizontal setup is an absolutely proper
approach to investigate living tree branches. Hamburg itself counts as a really green European city.
Therefore the results can be used for other studies and cities with similar vegetation.
Figure 1 shows the final setup of the test section. As can be seen it is divided into three parts. The
first section reaches into the wind tunnel. It is fixed and sealed so that no air can escape. The last
part has windows for visualization of the flow within the testing tube. The in-between part is the
actual test section. Inside there are six plates to fix the branches within the test section. Several
densities can be analyzed as well as branches with and without leaves to represent the winter and
the summer case.
63
Two optical windows, one before and one after the test section, were inserted to do LDA
measurements for analyzing the changes within the wind field. Furthermore the pressure can be
measured by six measurement points before and six after the test section. Hereby the average
pressure is taken and the pressure difference over the test section is calculated. This pressure
difference is essential to get to know the drag induced by the branches. An especially contrived
measurement procedure assures most accurate data which ensures a deviation of less than 0.1 Pa.
The actual results are shortly presented in the next chapter and will be presented more detailed at
the congress.
Figure 1: Design of the test section mounted in the Gttingen-type wind tunnel. It consits of three parts: The approach
section, the test section and the after flow section.
3 FIRST RESULTS
As can be seen in Figure 2 the pressure difference over the test section is the most different for the
winter and summer case. Regarding the individual tree species the difference is not as immense as
expected. It seems that the density itself has the most impact. However, due to certain time
problems it has not been possible to investigate living tree branches of all species for lesser
densities and without leaves. Still, the already collected data is conclusive and promises good and
extensive data of the aerodynamic behavior of living trees.
Figure 3 shows that a certain correlation can be seen between the pressure loss over the test section
and the pressure difference between the settling chamber and the normal air pressure. This relation
does not change significantly for the different tree species which proves that there seem to be no
significant differences between the diverse species. Furthermore one can see that the differences
increase with higher wind velocities. As there are mostly lower wind speeds in cities it can be
64
assumed that the individual tree species do not count that much regarding future physical models.
The most important factor seems to be the density of the tree crowns and the number of trees
within a street.
Figure 2: The y-axis represents the pressure difference over the test section, the x-axis shows the rotation speed of the
wind tunnel: 110 1/min stands for 1.2m/s, 160 1/min for 2.0m/s, 220 1/min for 3.0 m/s, 320 1/min for 5.0m/s.
Figure 3: The y-axis represents the pressure difference over the test section, the x-axis shows the pressure difference
between the settling chamber and the actual air pressure.
3 CONCLUSIONS
65
To investigate that the number of trees within a street is of importance further experiments must be
done. For these studies the test section might be changed and widened horizontally so that more
layers behind each other are possible. Furthermore aditional in-between states of density cases
must be studied to confirm the previous results. All in all, these examinations are the next step to
analyze the aerodynamic behavior of urban greenery and to find a way for a proper physically
modeling urban greenery.
4 REFERENCES
[1] Achim, A., Nicoll, B., Mochan, S. and Gardiner, B., Wind Stability of Trees on Slopes.
International Conference Wind Effects on Trees,16th-18th September 2003, Karlsruhe,
Germany.
[2] Amiro, B.D., Drag coefficients and turbulence spectra within three boreal forest canopies;
Boundary Layer Meteorology, Vol. 52 (1990), 227-246.
[3] Angster, W. and Ruck, B., The Influence of Aerodynamic Proportion of Forest Edges and
Stands on the Press Pattern within a Forest, International Conference Wind Effects on Trees,
16th-18th September 2003, Karlsruhe, Germany.
[4] Aubrun, S., Koppmann, R., Leitl, B., Mollmann-Coers, M. and Schaub, A., Physical
modelling of a complex forest area in a wind tunnel-comparison with field data, Agricultural and
Forest Meteorology, Vol. 129 (2005), 121-135.
[5] Aubrun, S. and Leitl, B., Development of an improved physical modelling of a forest area in a
wind tunnel, Atmospheric Environment, Vol. 38 (2003), 2797-2801.
[6] Aubrun, S., Leitl, B., Schatzmann, M., Physical Modelling of a Finite Forest Area-Transport of
biogenic Emissions, International Conference Wind Effects on Trees, 16th-18th September
2003, Karlsruhe, Germany.
[7] Balczo, M., Gromke, C., Ruck, B., Numerical Modeling of Flow and Pollutant Dispersion in
Street Canyons with Tree Planting, Meteorologische Zeitschrift, Vol. 18 (2009), 197-206.
[8] Beckett, K.B., Freer-Smith, P. H. and Taylor, G., Particulate pollution capture by urban trees
effect of species and wind speed, Global Change Biology, Vol. 6 (2000), 995-1003.
[9] Braun, S. and Fluckiger, W., Observations on Wind damage in permanent forest plots,
International Conference Wind Effects on Trees, 16th-18th September 2003, Karlsruhe,
Germany.
[10] Chen, J.M., Black, T.A., Novak, M.D., Adams, R.S., A wind tunnel study of turbulent airflow
in Forest Clearcuts, Cambridge University Press (1995), 71-87.
[11] Cionco, R. M., Intensity of turbulence within canopies with simple and complex roughness
elements, Boundary-Layer Meteorology, Vol. 2 (1972), 453-465.
[12] Coutts, M.P., Grace, J., Wind and Trees, Cambridge University Press (1995) (individual
references are included in this bibliography and labeled
accordingly).
66
[13] D`aniel, V., Models of oscillating trees, Mendel University of Agriculture and Forestry in
Brno, Czech Republic.
[14] De-xin, G., Ting-Yao, Z. and Shi-Jie, H., Wind tunnel experiment of drag of isolated tree
models in surface boundary layer, Journal of Forestry Research, Vol. 11, No. 3 (2000), 156-160.
[15] Donat, J. and Ruck, B., Simulated ground deposition of fine airborne particles in an array of
idealized tree crowns, Meteorologische Zeitschrift, Vol. 10, No. 4 (1999), 229-233.
[16] Finnigan, J., Turbulence in plant canopies, Annual Review of Fluid Mechanics, Vol. 32
(2000), 519-571.
[17] Finnigan, J. J. and Mulhearn, P. J., Modelling waving crops in a wind tunnel, Boundary Layer
Meteorology, Vol. 14 (1977), 253-277.
[18] Frank, C and Ruck, B., U ber den Einfluss der Bestandsdichteauf die Strmungsvorgnge an
Waldkanten, Symposium Lasermethoden in der Strmungsmesstechnik, 9th-11th September
[19] Frank, C and Ruck, B., Windkanalstudie zur Strmung in Waldlichtungen, Symposium
Lasermethoden in der Stromungsmesstechnik, 4th-6th September 2007, Rostock, Germany.
[20] Frank, C., Ruck, B., About the influence of the stand density on the flow characteristics at
forest edges, Proceedings of the 2nd International Conference Wind Effects on Trees, 13th-16
October 2009, Freiburg, Germany.
[21] Frank, C., Ruck, B., Tischmacher, M., About the influence of the windward edge structure on
the flow characteristics at forest edges, Proceedings of the 2nd International Conference Wind
Effects on Trees, 13th-16 October 2009, Freiburg, Germany.
[22] Gardiner, B.A., The Interactions of Wind and Tree Movement in Forest Canopies, Wind and
Trees, Cambridge University Press (1995), 40-59.
[23] Gardiner, B.A., Wind and wind forces in a plantation spruce forest, Boundary Layer
Meteorology, Vol. 67 (1994), 161-186.
[24] Gardiner, B. A., Mathematical modelling of the static and dynamic characteristics of
plantation trees, published in J. Franke and A. Roede, Mathematical Modelling of Forest
Ecosystems, Sauerlanders, Verlag (1992), 40-61.
[25] Gardiner, B., Marshall, B., Achim, A., Belcher R. and Wood, C., The stability of different
silvicultural systems: a wind tunnel investigation, Forestry, Vol. 78, No. 5 (2005), 471-484.
[26] Gayev, Y. A. and Savory, E., Influence of street obstructions on flow processes within urban
canyons, Journal of Wind Engineering and Industrial Aerodynamics, Vol. 82 (1999), 89-103.
[27] Gayev, Y. A., Hunt, J. C. R., Flow and Transport Processes with Complex Obstructions,
Proceedings of the NATO Advanced Study Institute on Flow and Transport Processes in Complex
Obstructions, Springer, 2004.
[28] Gillies, J. A., Drag coefficient and plant form response to wind speed in three plant species,
Journal of Geophysical Research, Vol. 107, No. D24 (2002), 10(1)-10(15).
67
[29] Grant, P. F. and Nickling, W. G., Direct field measurements of wind drag on vegetation for
application to windbreak design and modelling, Land Degradation and Development, Vol. 9
(1997), 57-66.
[30] Gromke, C. and Ruck, B., Aerodynamic modelling of trees for small-scale wind tunnel
studies, Forestry, Special Issue: Wind and Trees, Vol. 81, No. 3 (2008), 243-258.
[31] Gromke, C., Buccolieri, R., Di Sabatino, S. and Ruck, B., Dispersion study in a street canyon
with tree planting by means of wind tunnel and numerical investigations, Atmospheric
Environment, Vol. 42 (2008), 8640-8650.
[32] Gromke, C. and Ruck, B., Ein Ansatz zur Modellierung von Vegetation fur Gebude- und
Umweltaerodynamische Windkanaluntersuchungen, Fachtagung Lasermethoden in der
Strmungsmesstechnik, 9th-11th September 2008, Karlsruhe, Germany.
[33] Gromke, C., Buccolieri, R., Di Sabatino, S. and Ruck, B., Evaluation of numerical flow and
dispersion simulations for street canyons with avenue-like tree planting by comparison with wind
tunnel data, Dissertationsreihe of the Institute for Hydromechanics, University of Karlsruhe, No. 2
(2008), Germany.
[34] Gromke, C. and Ruck, B., On the impact of trees on dispersion processes of traffic emissions
in street canyons, Boundary Layer Meteorology, Vol. 131 (2009), 19-34.
[35] Gromke, C., Denev, J. and Ruck, B., Dispersion of traffic exhausts in urban street canyons
with tree plantings, International Workshop on Physical Modelling of Flow and Dispersion
Phenomena 2007, Orleans, France.
[36] Gromke, C. and Ruck, B., Flow and Dispersion Phenomena in Urban Street Canyons in the
Presence of Trees, Proceedings of the 12th International Conference on Wind Engineering, 2007,
Australia.
[37] Gromke, C. and Ruck, B., Stromungsfelder in Straenschluchten mit und ohne
Baumpflanzungen, Fachtagung Lasermethoden in der Stromungsmesstechnik, 4th-6th
September 2007, Rostock, Germany.
[38] Gromke, C. and Ruck, B., Trees in urban street canyons and their impact on the dispersion of
automobile exhausts, Proceedings of the 6th International Conference on Urban Air Quality, 2007,
Cyprus, Greece.
[39] Gromke, C. and Ruck, B., Der Einfluss von Baumen auf d. Strmungs- und
Konzentrationsfeld in Straenschluchten, Proceedings of the 14th GALA Fachtagung
Lasermethoden in der Strmungsmesstechnik 2006, Braunschweig, 59-110, Germany.
[40] Gromke, C. and Ruck, B., Influence of trees on the dispersion of pollutants in an urban street
canyon, Atmospheric Environment, Vol. 41(2006), 3387-3302.
[41] Gromke, C., Ruck, B., Implications of tree planting on pollutant dispersion in street canyons,
Proceedings of the 2nd International Conference Wind Effects on Trees, 13th-16th October
2009, Freiburg, Germany.
68
[42] Gromke, C., Ruck, B., Effects of trees on the dilution of vehicle exhaust emissions in urban
street canyons, International Journal of Environment and Waste Management, Vol. 4, No. 1/2,
2009, 225-242.
[43] Gross, G., A numerical study of the air flow within and around a single tree, Boundary Layer
Meteorology, Vol. 40 (1987), 311-327.
[44] Hiraoka, H., An investigation of the effect of environmental factors on the budgets of heat,
water vapor, and carbon dioxide within a tree, Energy, Vol. 30 (2004), 281-298.
[45] Hiraoka, H., Modelling of Turbulent Flows within Urban Plant Canopies, Journal of Wind
Engineering and Industrial Aerodynamics, Vol. 46 and 47 (1993), 173-182.
[46] Irvine, M. R., Gardiner, B. A. and Hill, M. K., The Evolution of Turbulence Across A Forest
Edge, Boundary Layer Meteorology, Vol. 84 (1997), 467-496.
[47] James, K., Dynamic Loading of Trees, Journal of Arboriculture, Vol. 29 (2003),165-171.
[48] James, K.R., Haritos, N., Ades, P.K., Mechanical stability of trees under dynamic loads,
American Journal of Botany, Vol. 93 (2006), 1522-1530.
[49] Jiao-jun, Z., Zu-gen, L., Matsuzaki, T., Gonda, Y., Review: effects of wind on trees, Journal
of Forest Research, Vol. 15 (2004), 153-160.
[50] Johnson, R. C., Ramey, R. C. Jr. and O'Hagan, D. S., Wind induced Forces on Trees, Journal
of Fluids Engineering, Vol. 104 (1982), 25-30.
[51] Judd, M. J., Raupach, M. R. and Finnigan, J. J., A wind tunnel study of turbulent flow around
single and multiple windbreaks, Boundary Layer Meteorology, Vol. 80 (1996), 127-165.
[52] Kamiyama, K., Ohashi, M., and Narita, K., Wind tunnel experiment on drag coefficient of
trees with the leaf area density as a reference area, (in Japanese), Journal of Environmental
Engineering, AIJ, Vol. 578 (2004), 71-77.
[53] Kane, B., Smiley, E. T., Drag Coefficients and Crown Area Estimation of Red
Maple,Canadian Journal of Forest Research, Vol. 36 (2006), 1951-1958.
[54] Kouwen, N. and Fathi-Moghadam, M., Friction Factors for Coniferous Trees along Rivers,
Journal of Hydraulic Engineering, Vol. 126, No. 10 (2000), 732-740.
[55] Kruijt, B., Klassen, W., Hutjes, R.W., Edge Effects on Diffusivity in Roughness layer over a
Forest, Wind and Trees, Cambridge University Press (1995),60-70.
[56] Lai, W., Aerodynamic Crown Drag Of Several Broadleaf Tree Species, Interim Technical
Report AFSWP-863, 1955.
[57] Lohou, F., Response of a Homogeneous Forest Canopy to Wind Stress for the Analysis of
Accelerometer Measurement, International Conference Wind Effects on Trees, 16th-18th
September 2003, Karlsruhe, Germany.
[58] Marshall, B. J., Wood, C. J., Gardiner, B. A. and Belcher, R. E., Conditional sampling of
forest canopy gusts, Boundary Layer Meteorology, Vol. 102 (2002), 225-251.
69
[59] Mayer, H., Script Forstliche Meteorologie, October 2004, Meteorological Institute of the
University Freiburg.
[60] Mayer, H., Baumschwingungen und Sturmgefahrdung des Waldes, Munchener Universitts-
Schriften,Wissenschaftliche Mitteilung, No. 51 (1985), University of Munich, Meteorological
Institute, aku-Fotodruck GmbH, Bamberg, Germany.
[61] Mayhead, G. I., Some Drag Coefficients for British Forest Trees Derived from Wind Tunnel
Studies, Agricultural Meteorology, Vol. 12 (1973), 123-130.
[62] Mayhead, G. J., Gardiner, J. B. H. and Durrant, D.W., A Report on the Physical Properties of
Conifers in relation to plantation stability, Forestry Commission, Research and Development
Division,Silviculture Section, Northern Research Station, 1975, Edinburgh, United Kingdom.
[63] Meroney, R. N., Prediction of flow and diffusion over vegetated regions on hills and ridges,
86th Annual Meeting and Exhibition Air and Waste Management Association, 14th-18th June
1993, Denver, Colorado, United States of America.
[64] Meroney, R. N., Characteristics of wind and turbulence in and above model forests, Journal of
Applied Meteorology, Vol. 7 (1968), 780-788.
[65] Mikulowsky, M. and Gil, W., Wind-Induced Damage to Polish Forests and the Methods of
Mitigating its Effects, International Conference Wind Effects on Trees, 16th-18th September
[66] Milne, R., Dynamics of swaying of Picea sitchensis, Tree Physiology, Vol. 9 (1991), 383-399.
[67] Moore, J.R., Maguire, D.A., Simulating the dynamic behaviour of Douglas-fir trees under
applied loads by finite element method, Tree Physiology, Vol. 28 (2007), 75-83.
[68] Morse, A. P., Brunet, Y., Devalance, M., Gamboa-Marrufo, M., Irvine, M. R., Marshall, B. J.,
Paw U, K. T., Shaw, R. H., Wood, C., Yang, B. and Gardiner, B. A., The Role of Forest Edges in
the Patterns of Turbulence Development, International Conference Wind Effects on Trees, 16th-
18th September 2003, Karlsruhe, Germany.
[69] Morse, A. P., Gardiner, B. A., Marshall, B. J., Mechanisms controlling turbulence
development across a forest edge, Boundary Layer Meteorology, Vol. 103 (2002), 227-251.
[70] Musick, H. B., Trujillo, S. M., Truman, C. R., Wind tunnel modelling of the influence of
vegetation structure on saltation threshold, Earth Surface Processes and Landforms, Vol. 21(1995),
589-605.
[71] Neff, D. E. and Meroney, R. N., Wind tunnel modelling of hill and vegetation influence on
wind power availability, 2nd European and African Conference on Wind Engineering (EACWE),
22th-26th July 1997, Genova, Italy.
[72] Peltola, H., Swaying of trees in response to wind and thinning in a stand of Scots pine,
70
[73] Puijalon, S., Adaptations to increasing hydraulic stress: morphology, hydrodynamics and
fitness of two higher aquatic plant species, Journal of Experimental Botany, Vol. 56, No. 412
(2004), 777-786.
[74] Raupach, M. R., Finnigan, J. J. and Brunet, Y., 1996, Coherent eddies and turbulence in
vegetation canopies: The mixing-layer analogy, Boundary Layer Meteorology, Vol. 78, 51-382.
[75] Raupach, M. R., Leys, J. F. and Woods, N., Modelling the effects of riparian vegetation on
spray drift and dust, Technical Report 29/00 for the NSW Departement of Land and Water
Conservation on Project PAT3, 2000, Australia.
[76] Raupach, M. R. and Shaw, R. H., Averaging procedures for flow within vegetation canopies,
[77] Ries, K., Eichhorn, J., Simulation of effects of vegetation on the dispersion of pollutants in
street canyons, Meteorologische Zeitschrift, Vol.10 (2001), 229-233.
[78] Roodbaraky, H. J., Baker, C. J., Dawson, A. R. and Wright, C. J., Experimental observations
of the aerodynamic characteristics of urban trees, Journal of Wind Engineering and Industrial
Aerodynamics, Vol. 52 (1994), 171-184.
[79] Rosenfeld, M., Marom, G., Bitan, A., Numerical Simulation of the Airflow across Trees in a
Windbreak, Conference on Modelling Fluid Flow, 9th -12th September 2009, Budapest, Hungary.
[80] Ruck, B. and Adams, E., Fluid mechanical aspects of the pollutant transport to coniferous
trees, Boundary Layer Meteorology, Vol. 56 (1991), 163-195.
[81] Ruck, B. and Schmitt, F., Das Stromungsfeld der Einzelbaumumstrmung,
Forstwissenschaftliches Centralblatt, Vol. 105 (1986), 178-196.
[82] Ruck, B., U ber die Aerodynamik der Baume, Fachtagung Lasermethoden in der
Strmungsmesstechnik, 6th-8th September 2005, BTU Cottbus, Germany.
[83] Rudnicki, M., Mitchell, S. J. and Novak, M. D., Wind tunnel measurements of crown
streamlining and drag relationships for 3 conifer species, Canadian Journal of Forest Research,
Vol. 34 (2004), 666-676.
[84] Sauer, T. J., Haan, F. L., Tyndall, J. C., Hernandez-Ramirez, G., Trabue, S. L.,
Pfeiffer, R. L. and Singer, J. W., Wind tunnel evaluation of vegetative buffer effects on air flow
near swine production facilities, USDA-ARS, National Soil Tilth Laboratory, Department of
Aerospace Engineering, Department of Natural Resource Ecology and Management, Iowa State
University, 2008, Ames, IA, United States of America.
[85] Schindler, D., Responses of Scots pine tree to dynamic wind loading, Agricultural and Forest
Meteorology, Vol. 148 (2008), 1733-1742.
[86] Schonborn, J., Schindler, D., Mayer, H., Measuring vibrations of a single, solitary broadleaf
tree, Proceedings of the 2nd International Conference Wind Effects on Trees, 13th-16th October
2009, Freiburg, Germany.
71
[87] Schrn, A., Leitl, B., Schatzmann, M., Proper Physical Modeling of Urban Greenery - What
Techniques are reasonable?, Proceedings of the 5
th
International Workshop on Physical Modelling
of Flow and Dispersion Phenomena, 24th-26th October 2009, Sint-Genesius-Rode, Belgium.
[88] Seginer, I., Aerodynamic roughness of vegetated surfaces, Boundary Layer Meteorology,
Vol. 5 (1974), 383-393.
[89] Seginer, I., Mulhearn, P. J., Bradley, E.F. and Finnigan, J. J.,Turbulent flow in a model plant
canopy, Boundary Layer Meteorology, Vol. 10 (1976), 423-453.
[90] Shaw, R. H., Silversides, R.H. and Thurtell, G. W., Some observations of turbulence and
turbulent transport within and above plant canopies, Boundary Layer Meteorology, Vol. 5 (1974),
429-449.
[91] Stacey, G. R., Belcher, R. E. and Wood, C. J., Wind flows and forces in a model spruce forest,
[92] Telewski, F.W., Wind-induced physiological and developmental responses in trees, Wind
and Trees, Cambridge University Press (1995), 40-59.
[93] Torita, H. and Nemoto, M., Wind tunnel experiment on densities and widths of shelterbelt, (in
Japanese), Journal of Japanese the Forestry Society, Vol. 84, No. 2 (2002), 85-90.
[94] Ucar, T., Hall, F. R., Tew, J. E. and Hacker, J. K., Wind tunnel studies on spray deposition on
leaves of tree species used for windbreaks and exposure of honey bees, Pest Management Science,
Vol. 59 (2003), 358-364.
[95] Vogel, S., Drag and Flexibility in Sessile Organisms, American Zoologist, Vol. 24, No. 1
(1984), 37-44.
[96] Vogt, J., Lauerbach, H., Meurer, M. and Langner, M., Interaction of Trees with Boundary
Layer Flow, International Conference Wind Effects on Trees, 16th-18th September 2003,
Karlsruhe, Germany.
[97] Vollsinger, S., Mitchell, S. J., Byrne, K. E., Novak, M. D. and Rudnicki, M., Wind tunnel
measurements of crown streamlining and drag relationship for several hardwood species, Canadian
Journal of Forest Research, Vol. 35 (2005), 1238-1249.
[98] Walter, B., Gromke, C., Lehning, M., The SLF boundary layer wind tunnel - an experimental
facility for aerodynamical investigations of living plants, Proceedings of the 2nd International
Conference Wind Effects on Trees, 13th-16th October 2009, Freiburg, Germany.
[99] Wilson, N. R. and Shaw, R.H., A higher order closure for Canopy flow, Journal of Applied
Meteorology, Vol. 16 (1977), 1197-1205.
[100] Wood, C.J., Understanding Forces on Trees, Wind and Trees, Cambridge University Press
(1995), 133-164.
[101] Ylinen, A., U ber die mechanische Schaftformtheorie der Baume, Holz als Roh- und
Werkstoff, Vol. 6 (1953), 16-17.
72
[102] Zaschke, M., Ruck, B., Stromungsmuster im Bereich von Waldrandern, Symposium
Lasermethoden in der Stromungsmesstechnik, 5th- 7th September 2006, Braunschweig,
Germany.
[103] Zhu, J., Matsuzaki, T., Li, F. and Yutaka, G., Assessment of Effects of Thinning on Wind
Damage in Pinus thunbergii Plantations, International Conference Wind Effects on Trees, 16th-
18th September 2003, Karlsruhe, Germany.
73
PHYSMOD2011 International Workshop on Physical Modeling oI Flow and Dispersion Phenomena
KlimaCampus, University oI Hamburg, Germany August 22-24, 2011
Wind tunnel study oI the atmospheric boundary layer
over vegetation-urban roughness transition: canopy
modelling and preliminary results
Laurent Perret
a
, Tonv Rui:
b
a
Laboratoire de Mecanique des Fluides, UMR CNRS 6598, Ecole Centrale de
Nantes, Nantes, France, laurent.perretec-nantes.Ir
b
Laboratoire de Mecanique des Fluides, UMR CNRS 6598, Ecole Centrale de
Nantes, Nantes, France, tony.ruiz ec-nantes.Ir
ABSTRACT: The goal oI the present study is to elucidate the inIluence oI the transition
between vegetation and urban canopies that can occur when large patches oI vegetation, such
parks, are embedded in a city. To do so, a wind tunnel investigation is perIormed via
stereoscopic Particle Image Velocimetry in diIIerent planes. The experimental setup is
described. First results obtained over a vegetation canopy are then presented. The two-point
correlation tensor oI the three component velocity Iield and the conditional average velocity
Iields show the strong spatio-temporal organization oI the Ilow and its imprint on the
spanwise component.
1 INTRODUCTION
One possible solution to control the urban micro-climate and reduce the heat island eIIect
induced by cities is to integrate vegetation within the built area. Studying and understanding
the role played by the integrated vegetation on the comIort oI inhabitants or on air quality in
cities is the primary goal oI the research project VEGDUD led by the Institute Ior Research
on Urban Sciences and Techniques (IRSTV, France).
The proposed study is carried out within the Iramework oI the VEGDUD research project in
order to elucidate the dynamics oI the Ilow structures that are strongly involved, in the urban
environment, in the mixing and transport oI air and scalars such as heat and pollutants.
Numerous studies have been devoted either to vegetation canopy Ilows (see Finnigan, 2000
Ior a detailed review) or urban-like canopy Ilows (Castro et al. 2006, Coceal et al., 2007,
Macdonald et al., 2002, Ior instance). These studies highlighted common Ieatures between
these two types oI Ilow but also strong diIIerences due to the structural diIIerences oI the
roughness elements. In particular, a mixing layer analogy has been proposed Ior Ilows over
vegetation roughness (Finnigan, 2000), whereas its applicability to urban canopy Ilows is less
than clear (Coceal et al., 2007). Fewer studies are devoted to mixed conIigurations with
vegetation embedded in an urban environment. Gayev and Savory (1999) investigated the
Ilow in a two-dimensional cavity with a regularly spaced vertical array oI cylinders and Iound
a decrease oI the volume Ilow rate oI the recirculating Iluid and a non-negligible average
increase in turbulence intensity. More recently the Ilow in a tree-planted street canyon was
investigated by Gromke and Tuck (2007) via one-point measurements and was Iound to be
inIluenced by the tree crown porosity. Recent numerical studies conducted on rural-
vegetation transition have shown the impact oI the edge on the dynamics oI the Ilow and the
associated coherent structures (Dupont and Brunet, 2009). The sparse literature on these
mixed conIigurations as well as the lack oI details on the spatio-temporal dynamics oI
74
canopy Ilows motivated the present study. The results presented here are the Iirst step oI an
ongoing study oI the transitions between vegetation and urban canopies that can occur when
large patches oI vegetation such parks are embedded in a city. In the present paper, the
experimental setup used to model the Ilow over a vegetation canopy and investigate its
coherent structures is described. One- and two-point statistics oI the velocity Iield are
presented, Iocusing on the spatial structure oI the Ilow and the coherent structures, such as
hairpin vortices, streaks, ejections and sweeps, that have been Iound to exist in turbulent
boundary-layer Ilows and play an important role in the transport mechanism oI momentum
and scalars (Adrian, 2007).
2 EXPERIMENTAL SETUP
2.1 Wind-tunnel
Experiments were conducted in the atmospheric boundary layer wind tunnel oI the
Laboratoire de Mecanique des Fluides (Nantes, France), which has working section
dimensions oI 24 2 2 m (Fig. 1, leIt). A suburban-type atmospheric boundary layer was
generated upstream oI the canopy model using a 13m Ietch oI staggered cube roughness
elements with a plan area density oI 25. The cubes height was h 0.05m. The
characteristics oI the simulated urban boundary layer (Iriction velocity, height oI
displacement and roughness height) were Iound to be in good agreement with the literature
(Cheng et al., 2007) and are describe in details in Rivet et al., 2011. The canopy model,
described below, was located downstream this Ietch oI cubes.
Figure 1: photograph oI, leIt, the wind-tunnel at LMF and, right, the vegetation canopy model.
2.2 Canopy model
In this study, an idealized vegetation canopy model was employed. It consisted oI an array oI
staggered rigid cylinders, vertically standing, arranged in a regular pattern over a Ietch oI 7m
(Fig. 1, right). The height oI the cylinders was h 50mm, and their diameter was d
r
4mm,
giving an aspect ration oI 12.5. The spacing between two cylinders in the same row was
32mm and the distance between two rows was 16mm. These dimensions led to a canopy
density oI n 980 cylinders/m and a Irontal area per unit volume a n.d
r
3.92m
-1
. The
Irontal area index oI this staggered arrangement is 0.39. These geometrical parameters
correspond to a dense canopy, deep enough to allow the existence oI three distinct regions
75
Figure 2: leIt, experimental setup; right: example oI an instantaneous 3-component velocity Iield measured in
a longitudinal plane (contours: spanwise velocity component).
where (1) the Ilow is dominated by the vortex shedding Irom the cylinders close to the
ground, (2) the mixing layer analogy holds near the canopy top and (3) the Ilow resembles a
displaced rough-wall boundary-layer above the canopy (Poggi et al., 2004).
2.3 Stereoscopic PIV setup
In the Iollowing, x, v and : denote the longitudinal, the spanwise and the vertical directions,
respectively (Fig. 2 leIt), and u, v, w, the longitudinal, spanwise and vertical velocity
components, respectively. Measurements oI the instantaneous 3-component velocity were
perIormed independently via stereoscopic particle image velocimetry (SPIV) in two diIIerent
vertical planes: a streamwise aligned (x-z) plane and a spanwise aligned (y-z) plane (Fig. 2,
leIt). The laser was arranged so that the light sheet Iell in the middle between two cylinder
rows. These planes were located 120h downstream the start oI the vegetation canopy model.
At measurement location, the obtained Ilow is believed to be representative oI the Ilow
developing over a vegetation canopy (Dupont and Brunet, 2009). The Ilow was seeded just at
the outlet oI the contraction oI the wind-tunnel with glycol/water droplets (typical size 1m)
using a Iog generator. A double-pulsed Litron DualPower 200-15 Nd:YAG laser (wavelength
oI 532nm) was used to illuminate the Ilow. Pairs oI images were recorded with a sampling
Irequency oI 4Hz, using two Dantec Dynamics FlowSense 4M cameras mounted in a
stereoscopic conIiguration with an angle about 50 between the camera axis and the normal
to the light sheet. These cameras have a resolution oI 2048 2048 pixels and were equipped
with a Nikon AF DC NIKKOR 105 mm I/2D lens. The cameras and the laser were mounted
outside the wind tunnel, on the side and above the ceiling, respectively. The ScheimpIlug
condition was satisIied by rotating the image plane with respect to the lens plane. To ensure
this condition, the cameras were installed on special remote controlled mounts designed to
enable rotation between the lens and the CCD chip. 4000 image pairs were recorded
independently in each plane and processed with an adaptive multi-pass cross-correlation
algorithm using 32 32 pixels Iinal interrogation windows and 50 overlap. The dimensions
oI the obtained velocity Iields are approximately 6h 4h with a spatial resolution oI 2.66 mm
and 1.85 mm in the horizontal and vertical directions, respectively. Given the retained optical
arrangement, measurements were perIormed only above the canopy (1 z/h 5). An example
oI an instantaneous velocity Iield obtained in the x-z plane is shown in Figure 2, right.
76
77
78
79
80
inIluence can be interpreted as the transition Irom an attached to a detached regime when
moving away Irom the canopy. However, the relationships between these two regimes and
their interaction mechanisms remain to be studied.
The present results demonstrate the suitability oI SPIV to investigate turbulent canopy Ilows
and its strong potential to provide valuable inIormation on the spatio-temporal dynamics oI
these Ilows. Based on the present Iindings, the proposed work will be extended to urban
conIigurations and vegetation-urban transitions. Future works will also involve the
modiIication oI the experimental setup in order to provide an optical access within the canopy
to investigate the dynamics oI the Ilow in this highly complex region.
5 ACKNOWLEGMENTS
This work was perIormed in the Iramework oI the project ANR Villes Durables VEGDUD,
Iunded by the French National Research Agency (ANR).
6 REFERENCES
Adrian, R. J., 2007. Hairpin vortex organization in wall turbulence. Physics oI Iluids, 19, 041301-041301-16.
Brunet, Y., Finnigan, J. J. and Raupach M. R., 1994. A wind tunnel study oI air Ilow in waving wheat: single-
point velocity statistics. Boundary-Layer Meteorology, 70, 95-132.
Castro, I. P., Cheng, H., and Reynolds, R., 2006. Turbulence over urban-type roughness: deductions Irom wind-
tunnel measurements. Boundary-Layer Meteorology, 118, 109-131.
Coceal, O., Dobre, A., Thomas, T. G., and Belcher, S. E., 2007. Structure oI turbulent Ilows over regular arrays
oI cubical roughness. Journal oI Fluid Mechanics, 589, 375-409.
Cheng H., Hayden P., Robins A.G. and Castro I.P., 2007. Flow over cube arrays oI diIIerent packing densities,
Journal oI Wind Engineering and Industrial Aerodynamics. 95, 715-740.
Dupont, S., and Brunet, Y., 2009. Coherent structures in canopy edge Ilow: a large-eddy simulation study.
Journal oI Fluid Mechanics, 630, 93-128.
Finnigan, J. J., 2000. Turbulence in plant canopies. Annual Review oI Fluid Mechanics, 32, 519-571.
Finnigan, J. J., Shaw R. H. and Patton E. G., 2009. Turbulence structure over a vegetation canopy. Journal oI
Fluid Mechanics, 637, 387-424.
Gayev, Y. A., Savory, E., 1999. InIluence oI street obstructions on Ilow processes within urban canyons.
Journal oI Wind Engineering and Industrial Aerodynamics, 82, 89-103.
Gromke, C., and Ruck, B., 2007. InIluence oI trees on the dispersion oI pollutants in an urban street canyon-
Experimental investigation oI the Ilow and concentration Iield. Atmospheric Environment, 41, 3287-3302.
Poggi, D., Porporato, A., RidolIi, I., Albertson, J. D. and Katul, G., 2004. The eIIect oI vegetation density on
canopy sub-layer turbulence. Boundary-Layer Meteorology, 111, 565-587.
Rivet, C., Ruiz, T. and Perret, L., 2011. Etude experimentale par SPIV de l'organisation d'une couche limite se
developpant au-dessus d'une canopee urbaine, in Proceedings oI the 20eme Congres Franais de Mecanique,
Strasbourg, France.
81
Investigation of transport of 2nd order moment of
concentration in block arrays
Keisuke Nakao
a
, Shinsuke Kato
b
, Takeo Takahashi
c

a
Grad. School, Univ. Tokyo, Tokyo, Japan, knakao@iis.u-tokyo.ac.jp
b
IIS, Univ. Tokyo, Tokyo, Japan, kato@iis.u-tokyo.ac.jp
c
IIS, Univ. Tokyo, Tokyo, Japan, aa51702@iis.u-tokyo.ac.jp
ABSTRACT: This paper reveals the detailed properties of the transport of concentration
fluctuation in a complicated urban terrain. The formal building shapes were recreated in a
wind tunnel. In a condition of point source contaminant generation, the time history of the
properties of concentration and velocities were acquired by the equipment with high response
frequency. The experiment revealed the properties of transport equations of variance of
concentration. Each property is considered with the tendencies of flow field (circulation) and
of the scalar transport occurring between the obstacles.
1 INTRODUCTION
This paper reports the transport feature of the variance of concentration in an urban street
canyon. In a situation such as a terrorist attack or accidental leakage of city gases in an urban
area, the fluctuation of concentration must be considered. Regarding the condition of point
source on a plane and of an elevated source from a plane, Fackrell and Robins (1982) have
shown the balance of transport equation of concentration variance. For an urban wind
environment, the obstacles that disturb the wind paths have a wide variety of influences on
the transport of concentration, and also on the concentration fluctuation. Although details of
the stochastic properties of velocity and turbulent scalar flux have been reported (Seomens et
al. (2007), Seomens et al. (2008)), the mechanism of transport of fluctuation remains unclear.
By using the constant temperature anemometry (CTA) and flame ionization detector (FID),
the volume distribution of the concentration-velocity cross correlation term was acquired.
The discretized forms of partial differences that exist in the transport equation of scalar
fluctuation (concentration variance) were calculated by the experimental results. Each term
was considered by comparing the transport of average concentration and variance of
concentration.
2 EXPERIMENTAL SET UP
A wind tunnel at the institute of industrial science, the University of Tokyo, 2.2 m wide x
1.8m high and 16.7 m in inlet length was utilized for the experiment. Figure 1 shows a
schematic sketch of the experimental set up in the wind tunnel. The vortex generators were
set upstream of the inlet area and downstream of the contraction area. Wooden made
roughness blocks 0.09 m, 0.06 m, and 0.03m high were aligned in front of the recreated urban
block area. By this configuration, the inlet flow was recreated to fit to 1/4 power law (Figure
2). The thickness of the boundary layer is almost 1.1 m. Velocity and concentration were
82
measured in the same temporal sequence. Constant temperature anemometry (CTA:
DANTEC) was utilized to measure the velocity and a split fiber film probe (55R55) was
employed. Concentration measurement was conducted using a Flame Ionization Detector
(FID: Technica, THC-2A). The sampling tube was 0.210
-3
m in length with a response
frequency of 170 Hz. The tracer was discharged in the wind tunnel via a mass flow controller
(DGMS-6-48: Scanivalve), which permits a constant tracer ejection. Ethylene (C
2
H
4
) was
utilized as the tracer and passively discharged in the amount of Q=0.03610
-6
m
3
/s. The
electrical signal of the tracer was logged by an A-D recorder (NR-2000: Keyence). Data
acquisition lasted for 60 s at 1 KHz. For the velocity measurement, a low pass filter that
passes less than 300 Hz was utilized. For the concentration measurement, a low pass filter
that passes less than 200 Hz was utilized. The instruments were located 2 mm apart.

Inlet length 16.47 m
inlet 5-degree tilting alignment
1.5 m vortex generators 0.09 m, 0.06 m, 0.03 m roughness blocks
Fig.1 schematic picture of the wind tunnel

0
0.5
1
0 1
z
/
[
-
]
U/U
0
0.5
1
0 0.1 0.2
z
/
[
-
]
<u'2> <v'2> <w'2>

Fig.2 Inlet wind velocity and r.m.s

x1/L0
x1/L0
x2/L0
x3/L0
L0

(1) Plane (2) Elevation
Fig.3 Closeup of measurement area

Figure 3 shows a closeup of the measurement area, which corresponds to the grey marked
area shown in Fig. 1. Wooden blocks 0.1 m wide x 0.1 m high (L
0
, reference height)0.2 m
deep were arrayed to recreate the urban area. As shown in Fig. 1, the blocks were inclined at
a 5-degree angle to the inlet flow in order to avoid the influence of large time scale drift,
o U
u
2
'
o U
v
2
'
o U
w
2
'
inlet
inlet
83
which is often observed in a symmetrical block alignment and often causes erroneous
measurement results. The measurement was conducted in dot points in Fig. 3. The tracer was
discharged in the center of the blocks (Fig. 3) and the normalized inner diameter of ejection
point d/L
0
was 6.010
-2
.
The balance of the equation in concentration variance is shown below. The equation is
given as
|
|
.
|
\
|
c
' c
I
c
c
' +
c
c
'
c
' ' c
c
' c
=
c
' c
j j j
j
j
j
j
j
x
c
x
c
x
c
u c
x
c u
x
c
u
t
c
2 ' 2
2 2 2
(1)
where means ensemble averages of properties and the dashed value means displacement
from the ensemble averaged value. I is molecular diffusivity. The 1
st
term of the left hand
side is the time evolution term, the 1
st
term on the right is average transport, the 2
nd
term is
transport by turbulence, the 3
rd
term is the production term, and the last term includes the
molecular diffusion and dissipation of fluctuation. Except for the dissipation, by acquiring the
correlation term of velocity and concentration, each discretized partial difference can be
calculated by the measurement data (the molecular diffusion term is often neglected because
the order of value is small compared to others). In the experiment, the measurements of
vector properties were managed through the coordinates of the wind tunnel (main stream of
approach flow corresponds to the x
1
-direction). In order to show the property in the
coordinates of Fig.3, coordinate conversion was controlled. Consequently, eight measurement
points were used to calculate the partial differences in the center of the control volume, which
is framed by eight measurement points.
The results of measured velocity u
i
[m/s] and concentration C[m
3
/m
3
] are normalized as
follows:

0
*
/ u u u i i = (2)

Q L Cu C /
2
0 0
*
=
(3)

where u
0
is the reference velocity at the height of L
0
, which is 0.971 [m/s].
3 RESULTS
3.1 Average velocity and stochastic properties of concentration
Figure 4_(1) shows the vertical distributions of average velocity and average concentration
in the corresponding location. From the velocity field, we can observe a circulation occurring
in the space (Oke (1988)). The concentration might be transported by the velocity field. As
the path of the tracer proceeds from the source location, the values decrease. The skewness
(Fig.4_(2)) shows a different tendency compared to c . At a height just higher than the
boundary phase x
3
/L
0
=1.0, there is a region of strong value. In the lower area, the peak value
can be observed in the region along with the high value of the average concentration. Figure
5 shows the vertical profiles of stochastic properties at x
1
/L
0
=0, x
2
/L
0
=-0.75~0.5, and
x
3
/L
0
=0.1~1.1. Turbulent kinetic energy shows a similar tendency in every horizontal
location (Fig.5_(1)). A large value is observed in the upper area of the boundary height
(x
3
/L
0
=1.0) and it decreases drastically as the height lowers. The mean concentration shows a
non-symmetric trend. The location x
2
/L
0
<0 showed a comparably large value because of the
velocity of the horizontal direction which is enhanced by the 5-degree inclination. The
84
circulation caused the transport of concentration to migrate to higher locations (x
3
/L
0
=0.9).
The variance of concentration assumed a large value in the region close to the source point
(Fig.5_(3)). In some locations, a slight re-increasing in the value is observed in the location
x
3
/L
0
=0.9 because of the circulation, although its effect might be weak compared to that of
mean concentration.

(1) ( ) 3 1 , u u and c (2) 2
3
2 3
c c ' '

x1/L0
x1/L0
x2/L0
x3/L0
L0

(Corresponding locations)
x
1
/L
0
= -0.40.4, x
2
/L
0
= -0.75, x
3
/L
0
=0.11.05
Fig.4 Velocity, concentration and skewness

0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1
0
0.2
0.4
0.6
0.8
1
1.2
0 5 10
0
0.2
0.4
0.6
0.8
1
1.2
0.1 1 10 100
0
0.2
0.4
0.6
0.8
1
1.2
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
x2/L0=-0.75 x2/L0=-0.5 x2/L0=-0.25

x2/L0=0 x2/L0=0.25 x2/L0=0.5

(1)
|
.
|
\
|
' + ' + '
2
3
2
2
2
2
2
1
2
2
1
u u u
(2) c (3)
2
c'
Fig.5 Vertical profiles of stochastic properties at x
1
/L
0
=0, x
2
/L
0
=-0.75~0.5, x
3
/L
0
=0.1~1.1
3.2 Balance of the transport equation
When the source emission rate is considered to be constant, the 1
st
term of Equ. (1) can be
ignored. Figure 6 shows the balance of transport equation of
2
c' . The residual term was
calculated by subtracting all the terms. The molecular diffusion and dissipation term are
included in the residual term.
As a broad overview, in all the cases, a dominant value was observed in the average
transport term. This means the concentration variance flows in the location by average
transport. The sum of the dissipation term and the molecular diffusion term assume the role
of balancing the budget by taking a large negative value. The dominancy of average transport
has also been shown in another report (Fackrell and Robins (1982)).
x
3
/L
0
x
3
/L
0
x
3
/L
0

85
Focusing on the average advection term, a strong positive contribution can be observed in
the region x
1
/L
0
=-0.3, x
3
/L
0
<0.6. In these regions, the fluctuation which is generated in the
region near the source might inflow by the average advection term. Although the value
decreased as heights increased, it re-increased around x
3
/L
0
=0.9 possibly because the
circulation often transports the concentration in the corresponding location. In x
1
/L
0
=-0.1,
the large value cannot be observed in the lower regions. These areas might be at the edge of
the plume and the term was not affected to any great degree. As the heights increase, the
values increase. The circulation generated in the street canyon raised the fluctuation in the
higher locations. These tendencies are also observed in (3) x
1
/L
0
=0.1, and (4) x
1
/L
0
=0.3. At a
height around x
3
/L
0
=1.0 a steep negative gradient in the vertical direction is observed in all
cases. The concentration might be attenuated by the upper (rather than the obstacles height
L
0
) air flow which does not include the concentration, and then the value of the term in
x
3
/L
0
>1.0 decreased drastically.
0
0.2
0.4
0.6
0.8
1
1.2
-150 -50 50 150
0
0.2
0.4
0.6
0.8
1
1.2
-100 -50 0 50 100
0
0.2
0.4
0.6
0.8
1
1.2
-50 0 50
0
0.2
0.4
0.6
0.8
1
1.2
-10 -5 0 5 10

0
0.2
0.4
0.6
0.8
1
1.2
-150 -50 50 150
0
0.2
0.4
0.6
0.8
1
1.2
-100 -50 0 50 100
0
0.2
0.4
0.6
0.8
1
1.2
-50 0 50
0
0.2
0.4
0.6
0.8
1
1.2
-10 -5 0 5 10

0
0.2
0.4
0.6
0.8
1
1.2
-150 350
average advection turbulent diffusion production dissipation

Fig.6 Balance of transport equation of
2
c'
The contribution of the production term is comparably small. In x
1
/L
0
=-0.3, the lower
locations x
3
/L
0
<0.6 take a characteristically large value. At the foot of the upwind obstacle,
fluctuation might be generated possibly because collision of the eddy transported to the
obstacles wall occurs in these regions. As the heights increase, the value is decreased. The
feature of the peak observed in x
1
/L
0
=-0.3, x
3
/L
0
<0.6 cannot be seen in other cases. Except
for these areas, the characteristic production is observed only in the boundary face of
obstacles although the value was too small to see. In x
1
/L
0
=0.1 and x
1
/L
0
=0.3, while the effect
of the average transport term becomes comparably small, and although still small when it is
seen absolutely, the relative contribution of the production term in the boundary phase
x
3
/L
0
=1.0 is increased.
(3)x
1
/L
0
=0.1 (4)x
1
/L
0
=0.3

(1) x
1
/L
0
=-0.3 (2) x
1
/L
0
=-0.1
86
Finally, the turbulent diffusion term is discussed. As well as the production term, it is
observed that this factor slightly affects the balance of equation compared to the average
transport term. The characteristic positive value can be observed in the lower locations of
x
1
/L
0
=-0.1 and x
1
/L
0
=0.1. In these areas, the average transport takes little part in the balance,
so it might not be in the mean path of scalar transport. In these areas, fluctuation mainly
flows in by turbulent diffusion. In the lower area of x
1
/L
0
=0.1, the fluid flows in the negative
x
1
/L
0
direction because of the circulation. So this location is in the upwind region of the
source location. In these regions, instead of the tracer seldom being transported by the
average advection term, the mixing dominantly affected the transportation. In x
1
/L
0
=0.1 and
x
1
/L
0
=0.3, the dominant negative value can be observed on the boundary phase x
3
/L
0
=1.0.
therefore, the fluctuations might outflow from the area by turbulent diffusion.
3.1 Contributions of three components of the terms
The terms shown in the former session can be decomposed to the components of the three
dimensional coordinate systems. The results of j=13 in Equ. (1) are shown.
Figure 7 shows the allocations of average transport terms in (1) x
1
/L
0
=-0.3, x
2
/L
0
=-0.625
and (2) x
1
/L
0
=0.3, x
2
/L
0
=-0.625. In x
1
/L
0
=-0.3, the j=2 component shows a dominant negative
value in x
3
/L
0
<0.6. In these locations, strong average velocity that flows in the span wise
direction of blocks being generated by the 5-degree inclines contributes to the transport of
fluctuation. These two components decrease near the boundary phase, while on the contrary,
the j=3 component increases. In x
1
/L
0
=0.3, the dominant contribution in the boundary phase
x
3
/L
0
=1.0 is observed in the j=1 component. In the boundary phase of obstacles, the transport
of fluctuation is averagely managed by the component of the main stream in the wind tunnel.
In the lower locations, the contribution of the j=3 component is observed. The circulation
transports the concentration fluctuation from the upper to the lower.
Figure 8 shows the allocations of production terms in (1) x
1
/L
0
=-0.3, x
2
/L
0
=-0.625 and (2)
x
1
/L
0
=0.3, x
2
/L
0
=-0.625. In x
1
/L
0
=-0.3, especially in the lower measurement locations, the
production of fluctuations is mainly due to the j=1 component. Although the figures are not
shown, the mean concentration gradient assumed a large negative value and the correlation
term j u c ' ' took a comparably large positive value. But in the higher locations, that component
showed a small value. In the boundary phase x
3
/L
0
=1.0, the production is mainly due to the
j=3 component. In x
1
/L
0
=0.3, the production terms are negatively dominant only in the higher
locations and this is due to the j = 3 component. The vertical concentration gradient is quite
large in the region. The exchange of the control volume to the upper locations might cause
the large production of fluctuation.
0
0.2
0.4
0.6
0.8
1
1.2
-100 0 100 200

0
0.2
0.4
0.6
0.8
1
1.2
-10 0 10 20

(1) x
1
/L
0
=-0.3, x
2
/L
0
=-0.625 (2) x
1
/L
0
=0.3, x
2
/L
0
=-0.625
0
0.2
0.4
0.6
0.8
1
1.2
-500 0 500

x1/L0 x2/L0 x3/L0

Fig.7 Allocations of the average advection j j x c u c ' c
2

in corresponding locations
87
0
0.2
0.4
0.6
0.8
1
1.2
-20 0 20 40

0
0.2
0.4
0.6
0.8
1
1.2
-5 0 5

0
0.2
0.4
0.6
0.8
1
1.2
-500 0 500

x1/L0 x2/L0 x3/L0

(1) x
1
/L
0
=-0.3, x
2
/L
0
=-0.625 (2) x
1
/L
0
=0.3, x
2
/L
0
=-0.625
Fig.8 Allocations of the production j j x c u c c c ' ' 2

in corresponding locations
4 CONCLUSIONS
Wind tunnel experiments were conducted to obtain the correlation term between the
velocity and concentration of transported matter (C
2
H
4
) in a modeled urban street canyon. By
measuring the properties in the entire volume of the targeted area, each term of the transport
equation of concentration variance was calculated by discretizing the partial difference.
The paper shows a sketch of the plane distributions of the average velocity field, average
concentration, r.m.s of concentration and skewness. The clockwise circulation that is
enhanced by the flow field assumes the role of scalar transportation and of the transportation
of the fluctuation. The skewness shows a large value along with the transport path of mean
concentration. The balances of the transport equation of concentration variance are shown. In
every case, the average transport term shows a characteristic value and the terms that are
balanced with this term were the dissipation term and the molecular diffusion term (although
the molecular diffusion term may be disregarded). The production terms assumed a
comparably smaller value than the advection terms. The values are characteristic at the foot
of the upwind obstacle and the boundary face of obstacles. In the region where the flow
encounters the obstacles wall or where the shear stress is strong, fluctuation might be
generated. The turbulent diffusion term shows a dominant value in the off peak locations of
concentration.
The allocations of the average transport term, production term are shown in specific
measurement locations. The average transport terms are mainly due to the component of the
span wise direction of obstacles in the lower areas. In the higher measurement locations, the
component of the direction of the main stream in the wind tunnel takes part of the value. The
higher production terms at the foot of upwind obstacles are due to the components of the
direction of the main stream in the wind tunnel. In the boundary face of obstacles (the height
x
3
/L
0
=1.0) the vertical exchange of volume caused the production of fluctuations.
4 ACKNOWLEDGEMENT
This research is financially supported by the Japan Institute of Construction Engineering
(JICE)
88
5 REFERENCES
J. E. Fackrell and A. G. Robins, Concentration fluctuations and fluxes in plumes from point sources in a
turbulent boundary layer, Journal of Fluid Mechanics, 117, pp.1-26, 1982

Bu, Z., Kato, S., Ishida, Y. and Huang, H., New criteria for assessing local wind environment at pedestrian
level based on exceedance probability analysis, Building and Environment, Volume 44, Issue 7, July 2009,
Pages 1501-1508

Serge Simoens, James M. Wallace, The ow across a street canyon of variable widthPart 2: Scalar
dispersion from a street level line source, Atmospheric Environment, 42, pp. 2489 2503, 2008, 3

Serge Simoens, Michel Ayrault, James M. Wallace, The flow across a street canyon of variable width
Part 1: Kinematic description, Atmospheric Environment, 41, pp. 9002 9017, 2007

Oke, T. R., 1988. Street design and urban canopy layer climate, Energy and Buildings 11, pp. 103113

89
Proper Orthogonal Decomposition of Velocity and
Vorticity Field Within the Street Canyon and Wavelet
Analysis of the Expansion Coefficients
Radka Kellnerova
a,b
, Libor Kukacka
a,b
, Zbynek Janour
b
a
Department of Fluid Dynamics, Institute of Thermomechanics AV CR,
Prague, Czech Republic, radka.kellnerova@it.cas.cz
b
Department of Meteorology and Environment Protection, Charles University,
Prague, Czech Republic
ABSTRACT: POD analysis helps to reveal coherent structures in the turbulent flow. POD,
when applied on velocity and vorticity, decomposes complex flow into simpler modes and
thus brings a better insight into dynamics of flow. Analysis of velocity carries out the
structures with dimension of obstacle (e.g. vortex behind roof, recirculation zone). The
coherence patterns based on vorticity are rather like an amorphous regions with only a weak
weight factor of individual modes. This paper presents the results from POD analysis tested
on velocity field of turbulent flow generated over very rough urban terrain in wind channel.
1 INTRODUCTION
In recent decade, Particle Image Velocimetry (PIV) has been intensively used in many
windtunnel laboratories. Unlike the LDA, which measures in one point only, PIV can cover
whole 2-D area in the flow and obtain information about two components of velocity. Stereo-
PIV can additionally carried out the third - out of the plane component. LDA usually
provides measurement with sample frequency between a few hundreds up to a few thousands
Hz. Common PIV, on the other hand, had often used the solid-state Nd:YAG laser which was
able to reach typically about 15 Hz ('slow' PIV). However, new time-resolved (TR) PIV with
diode-pumped Nd:YLF laser has increased the sample frequency on several thousands Hz
what allows to record a temporal evolution of the flow (Raffel et al., 2007). This upgrade in
temporal resolution together with high spatial resolution significantly helped in better
understanding of turbulent motions.
To analyze such a large amount of data obtained from TR-PIV, we have to employ some
suitable method. For this purpose, Lumley (1967) designed so-called proper orthogonal
decomposition (POD). Method is called also by several other names like Karhunen-Loeve
decomposition, principal component analysis (PCA), empirical orthogonal functions (EOF)
or singular value decomposition (SVD). However the core procedure stays the same.
Thanks to implicit orthogonality, POD is linear method applied on non-linear problem. To
understand its mathematical background, nice introduction to the PCA can be found in Shlens
(2005) or Smith (2002).
POD can be applied on both experimental data and numerical simulation. The method
actually provides nice tool for comparison of results between numerical and physical
modeling. Input for procedure can be an arbitrary physical quantity (velocity, pressure,
vorticity, brightness, temperature) and POD always carries out the maximum projection of
fluctuations in overall sense. Sometimes, POD results yield up in interesting patterns,
however their physical meaning can be hidden. This paper has focused mainly on the
interpretation of particular POD modes inside the street canyon. Also, test of robustness of
POD is done, where data input are reduced or enhanced.
90
2 EXPERIMENTAL SET-UP
Experiment was performed using Particle Image Velocimetry (PIV) with high repeti-
tion rate (500 Hz) in wind-channel. The channel has the dimension of 0.25 m x 0.25 m in
cross-section and 3 m in longitudinal direction. Reference wind speed in the middle of
channel is 5 m/s. Reynolds building number is Re
2H
= 40000, derived from velocity U
2H
at the
double level of the building H and kinematic viscosity.
Floor is covered with 30 parallel street-canyons. Canyons are arranged in a strictly
perpendicular angle to the approach ow. By this, we intend to reduce an inherently 3-D
turbulent ow into two dimensions as far as possible. The aspect ratio of the canyon equals to
one (Fig. 1a). Longitudinal direction copying the approach flow in channel is labeled as X,
lateral one as Y and vertical one as Z. According to this coordinates, velocity component in
X-direction is labeled as U, in Y-direction as V and in vertical Z-direction as W.
Figure 1: Scheme of the wind-channel covered by street canyons with flat roof.
Two geometries of street canyons roof are used for comparison purpose - triangle and flat
shape of the building roof. Scale in which the model is manufactured is 1:400, so the streets
with height and width of 20 m in reality have 50 mm in dimension. The triangle roof is 20
mm high and makes the wedge angle of 102.6.
Regarding the morphological parameters like a dimensionless plan area
p
and a frontal area
f
suggested by Bitter and Hanna (2003), in our case both
p
and
f
equal to 0.5 and lie in the
interval for urban downtown areas.
According to Cheng & Castro (2002), measurement position has to be located sufficiently
far from the channel mouth in order to provide long fetch for development of fully turbulent
internal boundary layer. This specific distance depends on a roughness length of surface
beneath the internal layer. The rougher the surface is, the further from the channel mouth a
measurement has to be located. For this arrangement of street canyons, the distance is
established to be 2000 mm (20H) for full development of inertial layer up to its upper
boundary. Thus, measurement position was picked up in 21
s t
canyon. To verify, if the
distance is long enough, we also measured the differences between consecutive profiles along
the x-axis. Using the CTA hot-wire Dantec 55 P01 (referred to HWA) with sample frequency
of 25 kHz, we measured mean U-component of velocity in the four consecutive street
canyons. The deviation among these profiles does not exceed 1.5%.
We have to emphasize that proper simulation of experiment should be done in wind-tunnel
with cross-section much larger then the dimension of obstacles. In wind-channel, the flow is
restricted into a small tube and top wall undesirably influences the development of internal
layer. Furthermore, the model itself is large enough and can block the flow inside tube. The
frontal area of model occupies 20% of channel cross-section. From this reason, we performed
an another experiment in wind-tunnel with the same model, where all fundamental
//
//
2 000 mm
Laser
//
//
3050 mm
Wind Channel
5
0

50
50
2
5
0

m
m
2
5
0

m
m
91
requirements are fulfilled. The wind-tunnel has 1.5 x 1.5 m in cross-section and 20 m in
length.
By LDA we measured vertical profiles of U and W-component and inferred turbulent
characteristics. Comparison of mean values of U-component and Reynolds stress between
channel and tunnel for both pitched and flat roof is depicted in Figure 2.
Figure 2: Left: Comparison
of profiles of mean longitudinal velocity (U-component) from wind-tunnel (LDA) and wind-channel (HWA and
PIV). Right: Reynolds stress from tunnel and channel.
The profiles in wind-channel and in wind-tunnel inevitably differ. Especially in the pitched
case, the flow is considered to be non-representative above the critical elevation 1.7 H, where
flow accelerates due to blocking effect of internal boundary layer. However, inside the street
canyon, channel flow is very similar to the tunnel flow. In Figure 2 right, the Reynolds
stress from both facilities collapsed well. More information about boundary layer from the
tunnel and the channel experiment can be found in Kellnerova et al. (2009 and 2010).
In the wind-channel, time-resolved Particle Image Velocimetry with high repetition rate
(500 Hz) is used for the measurement campaign. One run of PIV measurement consists of
1600 snapshots, each of them with more than 4800 velocity vectors. We repeated each run for
three times. In table 1 below, the parameters of PIV set-up are published.
Table 1: Parameters of PIV
Repetition rate 500 Hz
Resolution 1280 x 1024 pxs
Interrogation area 32 x 32 pxs
Overlapping 50% (80 x 64 vectors)
Area 100 x 100 mm
Acquisition time 3.2 s
3 RESULTS
92
3.1 POD modes
POD analysis is applied on velocity data for both roof-shape arrangements. Figure 3 depicts
the first four modes for pitched roof. The modes are reordered from left to right by
significance. On the very left is therefore the first mode that represents vortex behind the
upstream roof. Shape of vortex is derived from all the snapshots involved in computation and
is function purely of space, so it is constant in time. Temporal evolution of mode is captured
in the so-called weight function (e.g. expansion coefficient). In Figure 3, the small black
arrows on the streamlines of spatial mode can point either in one direction if weight function
is positive or in opposite direction if weight function becomes negative.
For each mode can be calculated relative contribution of the energy contained in the mode
with respect to the total TKE in the system (see Figure 5 - left). For example, the most
dominant mode No. 1 contains 32% of total TKE. Such a value is rather high and indicates
that there is substantive degree of organization in the flow.
The second dominant mode displays the recirculation zone in the middle of the canyon
(Fig. 3b). The position of vortex core of the mode No. 2 differs from the position of time-
mean vortex core (not shown). Sometimes, this recirculation produces such a strong motion
that it subsequently causes a strong outflush, which acts like barrier for the approaching flow
above the roofs. Depending on the strength of the outflow, the saddle point can be formed.
Contribution to the TKE of this mode is 7%.
The third mode captures the sweep or ejection event (Fig. 3c) and includes 6% of total
TKE. Finally, the fourth mode shows the lower vortex as the dominant structure (Fig. 3d).
Lower vortex participates on the total energy by 4%.
Figure 3: The four most dominant spatial modes (topos) displayed with streamlines for pitched roof geometry:
from the left: a) fist mode, b) second mode, c) third mode, d) fourth mode. Black arrow on the top shows the
direction of the approaching flow.
Results for the flat case is displayed in the Figure 4. The first three most important modes
show the similar patterns as in the pitched case. The first mode represents the vortex behind
the upstream roof. The energy contribution is generally slightly lower, the mode contains
27% of TKE.
The second mode is a little modified recirculation zone between the walls which captures
about 6% of TKE. Third mode shows the sweep and ejection event, contributing to the flow
system by 4%.
The last fourth mode deviates much more from the analogous mode of the pitched
arrangement. Notwithstanding, the lower recirculation pattern can be still detected.
The comparison between rate of convergence of cumulative contribution of individual modes
suggests that flow generated over streets with pitched roofs involves slightly more coherency
than the flow above canyons with flat roofs (Figure 5 - right).
93
Figure 4: The four most dominant spatial modes (topos) displayed with streamlines for flat roof geometry. From
the left: a) fist mode, b) second mode, c) third mode, d) fourth mode. Black arrow on the top shows the
direction of the approaching flow.
Figure 5: Left: Relative contribution of particular modes to the total TKE. Triangles represent pitched roof,
squares stay for flat roof. Right: Cumulative relative distribution of both arrangements.
3.3 Expansion coefficients
Expansion coefficient provides a weight factor function for each mode. The extreme values
(peaks) along the time-line witness about strong influence of corresponding mode at the
certain time (see Figure 6). The overview of standard deviation and maximum values in
absolute sense from all three pitched-case runs (labelled as Prun1, Prun 2 and Prun 3) is
depicted in Figure 7. Figure shows the values for the first 50 expansion coefficients, since the
both statistics monotonously decrease and higher modes thus have only a minute importance.
All three runs exhibit the similar behavior. Standard deviations (squares) collapsed into
one curve. The maximum values (circles) slightly differ from each other in low mode
number. However this is natural, since some strong events happen rather from time to time.
The acquisition time of 3.2 s can not cover outstanding dynamical situation every time.
Difference between modes from three runs using Euclidean norm is calculated in Table 2.
The more important mode, the smaller deviation occurs. This can be consider as a proof
either for verification if the acquisition time of PIV records is at least of reasonable length or
for robustness of POD methods since it brings similar results for different time-pieces of
flow.
94
Figure 6: Temporal evolution of the first four expansion coefficients. The local extremes are emphasized by
circles (positive by green color, negative by red color).
Figure 7: Standard deviation and maximum values of expansion coefficients for all three runs in pitched case.
Table 2: Difference between POD modes taken from three runs (Pitched case).
Prun 1 vs. 2 [%] Prun 2 vs. 3 [%]
Mode 1 4.1 2.8
Mode 2 12.3 10.2
Mode 3 12.3 8.9
Mode 4 18.1 14.0
3.3 Wavelet analysis of expansion coefficients
Wavelet analysis is usually applied on time series of some physical property (velocity,
pressure). It brings the information about frequency and time of its appearance in the flow.
Great introduction into wavelet methodology can be found in Adisson (2002). POD
coefficients exhibit interesting repetitive pattern as well as the velocity. In this paper, we
compare wavelet analysis applied on time-series of expansion coefficients in order to reveal
the frequency behavior of modes.
When starting with Mode 1, the wavelet analysis of its coefficient is plotted in the Figure 8
upper. However, deep inspection revealed that some spatial locations in the street canyon
provide the same wavelet results, what means they very well capture the dynamics of whole
street canyon expressed in the first mode.
95
Using Frobenius norm, we established the deviation of wavelet scalograms of all positions
in the street canyon from the reference scalogram of first coefficient (labelled as a
1
).
Figure 8: Upper: Wavelet analysis of the first expansion coefficient. Morlet function serves as a mother
wavelet. Lower: Wavelet analysis of the velocity field in the position with lowest declination from reference
upper scalogram (derived from Figure 9).
The level of similarity is displayed in Figure 9. The lower percentage, the closer both
scalograms are to each other. We found an incredible agreement between analysis of a
1
and
positions grouped around the level Z/H=1.2 (see Figure 8 - lower). Area of high similarity
(low deviation in Figure 9) extends over whole width of street canyon. This indicates that
flow dynamics captured in the first POD mode is controlled by flow located slightly above
the roof-top level.
96
Figure 9: Deviation in percentage derived using Frobenius norm between wavelet analysis of coefficient a
1
and
every positions in the street canyon.
4 CONCLUSION
The PIV experiment with via POD analysis based on TKE showed that turbulent flow
generated over very rough terrain contains a high rate of coherency. The individual POD
modes were drafted and than were reliably assigned to the real flow dynamics. Since the
mathematically introduced orthogonality can be non-physical, certain caution when using
POD is legitimate. Sweep an ejection events express themselves in the first and in the third
mode what justifies and confirms the importance of this doublet of specific momentum fluxes
in turbulence.
POD also carried out similar results for similar, but non-identical flows, performed under
the identical boundary conditions. The difference between consecutive runs are 4% for the
first mode and around 12% for the other ones.
Finally, the expansion coefficients were analyzed by Wavelet analysis. For the coefficient
associated with the first mode, the great consistency was found out with the analysis of the
velocity signal within roof-above region. The flow dynamics in this certain region is very
well correlated with dynamics of the first POD mode.
5 REFERENCES
Adisson P. S., 2002. The Illustrated Wavelet Transform Handbook. Institute of Physics
Publishing, Bristol.
Britter R.E., Hanna S.R., 2003. Flow and Dispersion in Urban Areas. Annu. Rev. Fluid Mech., 35, 46996.
Cheng H., Castro I. P., 2002. Near-Wall Development After a Step Change in Surface Roughness. Boundary
Layer Meteorology, 105, 411- 432.
Kellnerova R., Kukacka, L., Janour Z., 2009. Quadrant analysis of boundary layer above pitched
and flat roofs. Acta Technica CSAV, 54, No. 4, 401-413.
Kellnerova R., Kukacka L., Uruba V., Antos P., Odin J., Janour Z., 2010. Wavelet and POD
Analysis of Turbulent Flow Within Street Canyon, in: Experimental Fluid Mechanics,
Conference proceeding, Liberec, Czech Republic, 263-270.
Lumley J. L., 1967. The structure of inhomogeneous turbulent flows, in: A. M. Yaglom
and V. I. Tatarski, editors, Atmospheric Turbulence and Radio Wave Propagation, 166-178.
Ra
ffel M., Willert C. E., Kompenhans J., 2007. Particle Image Velocimetry A Practical Guide. Springer.
97
Shlens J., 2005. A Tutorial on Principal Component Analysis. 'La Jolla, CA 92037: Salk Institute for
Biological Studies'.
Smith L.I., 2002. A tutorial on Principal Components Analysis.
98

Assesment of Urban Wind Environments
Based on Exceedance Probability
Shinsuke Kato
1)
Zhen Bu
2)
Kyosuke Hiyama
1)

1: IIS., the University of Tokyo, 2: Mott MacDonald
E-mail: kato@iis.u-tokyo.ac.jp
Keywords: Wind Environment, Local Purging Flow Rate, Exceedance Probability

INTRODUCTION
Frequency of Weak or Strong Wind for Evaluating Wind Environment
By the strong friction of the ground with buildings, wind velocity decreases rapidly from the
upper region to the grand
[1-3]
. The strung buildings hinder the wind streams as if they act as
wind breaks. They weaken averaged velocity overall and strengthen turbulent motion to the
contrary. Wind environment is a stochastic phenomenon; there is a day with strong wind and
a day with weak wind. People do not recognize the slight changes of annual mean velocity but
the frequency of extreme high velocity wind days or that of extreme low velocity wind days.
To treat this stochastic feature, exceedance probability analysis is introduced in order to
evaluate the properties with the dynamic features of wind
[10]
.

Wind Environment being Evaluated with Ventilation Efficiency at a Finite Space
Wind velocity is vector and does not directly express the ability of diluting and transporting
contaminant. Wind velocity will change rapidly between the buildings close to the ground; it
is difficult to find the representative point where the characteristic wind environment is
represented for its vicinity. We can use kinetic energy representing wind magnitude instead of
velocity vector since it is scalar. People will evaluate the wind environment not at a specific
or representing point but within a finite space they feel. To treat this ability of diluting and
transporting contaminant within the finite space between buildings, ventilation efficiency
index defined within a certain space and averaged kinetic energy for a certain space are
introduced instead of point value of velocity vectors
[4-6]
.

Acceptable Wind Environment with Stochastic Evaluation
Wind environment will be different from city to city. There is a city where wind is relatively
strong and is also a city with weak wind. In the city with relatively strong wind through the
year, building can be arranged crowdedly; density of buildings can be raised since the
possibility of urban wind getting stuffed, and in the city with relatively weak wind through the
year, density of buildings should be limited under the some range in order not to be suffered
from stagnant wind environment. A recommendation of minimum requirement for the wind
environment is introduced here. We believe the urban building density should be controlled
with this recommendation.

INDOOR AND OUTDOOR AIR QUALITY IN CONFINED SPACE
Ten Times Rule for Concentration
Indoor air is ventilated with outdoor air. The indoor air quality depends on the outdoor air
quality, indoor pollutant source, and air cleaning ability of HVAC. People spend over 90% of
their time indoor and it should be controlled not adversely affects human health. From the
indoor environmental control engineering aspects, the outdoor air quality is usually expected
ten times cleaner than that of indoor. If people want to keep some indoor pollutant
concentration with indoor inevitable source under the guideline of pollutant with ventilation
and or other possible measures, it is better that the outdoor pollutant concentration be less
than one tenth of the guideline value so that the control can be attained through only
99

ventilation. The indoor air quality may vary to the some extent with indoor conditions. If the
variation range of outdoor air quality is ten times smaller and ten times cleaner than that of
indoor, ventilation can be effective measures of controlling indoor air quality under the
adequate indoor pollutant source control.

Figure 1 shows the concept of Ten Times Rule. The region A has a source of pollution and is
ventilated with the region B. The region A is enclosed within the region B. The pollutant
concentration in the region A is required of less than 1000 ppm. The region B is ventilated
with the inner region A and also with the outer region C. If the pollutant concentration
control of the region A should be attained with the one digit number accuracy of 10%, the
concentration of the region B should less than 100 ppm, 10 times lower concentration of the
targeted concentration of the region A. If there are no sources of pollutant in the regions B
and C, the concentration of the region A can be controlled with the certain ventilation rate
between regions A and B, the region B should be ventilated with ten times larger ventilation
rate with the region C to keep the concentration less than ten times below that of the region A.
The region A can be a building and then the region B is the vicinity region of the outer
building and the region C can be urban wind environment.

Ten Times Rule for Air Flow Rate
With this argument, we may conclude that the ventilation rate of outside air region where the
building ventilation air intake is faced should be ten times larger than the building indoor
ventilation rate under the condition of no outside pollutant source. We may suppose the void
space between buildings from where building ventilation air is taken. If a building ventilation
rate is, say, 30,000m
3
/h, this outside void space should have air flow rate of 300,000m
3
/h with
no outdoor pollutant source and no pollutant migration from other void spaces. The
correspondence between building and void space can be not so rigid. A building can be
accompanied more than one void space and one void space can be shared with more than one
building. The void space can be larger or smaller. Only important thing is that at least ten
times larger air flow rate than indoor one should be always ensured. Even though the void is
relatively smaller compared with the building scale, it should have larger air flow rate which
can sustain the building ventilation.

Volume of Void Space and Air Change Rate Determining Air Flow Rate
The size of the void space between buildings can be changed largely with the various
arrangements of buildings in an urban block. Even though the air flow rate is essential index
for the outside air quality, the air flow rate divided by the volume of void space which
expresses the air flow rate per unit volume and is called air change rate, will be good index
for the outside air quality. The air change rate expresses how often the air will be exchanged
Figure 1 Ten Times Rule for one digit number accuracy
1000ppm
100ppm
10ppm
Region A
Region B
Region C
contaminant source

Ventilation Rate 10
Ventilation Rate 100
100

with the outer air in a unit volume and the larger one means that such the space is ventilated
better and cleaner. It is rational to determine the minimal value of the air change rate of void
space between the buildings for controlling good air quality within the space.

THERMAL CONFORT WITH WIND
Wind of Modest Temperature Cooling Human Body
Wind with lower temperature than the human skin temperature can cool a human body and is
useful to regulate the thermal comfort in summer. Many people claim the utilization of wind
induced cross ventilation to reduce the cooling energy for buildings. It is apparent that wind
induced ventilation utilizes wind energy of kinetic and potential (pressure) one. In the field of
wind induced ventilation engineering, wall surface static pressure induced with wind is an
important factor for evaluating the wind induced ventilation. Wind static pressure is
transformed from the kinetic energy of the free stream at the place where wind velocity
becomes zero on a wall and kinetic energy becomes also zero. The ability of wind induced
cross ventilation can be represented with the kinetic energy of the void space surrounded by
buildings.

Ten Times Rule for Wind Induced Cross Ventilation
The rising stream velocity around a human body with metabolism is estimated less than
0.3m/s. In other words, wind velocity more than 0.3m/s will cool down the human body more
efficiently than the natural convection with human heat generation does. We can assume the
wind kinetic energy more than 0.05m
2
/s
2
(around 0.3m/s) will be useful for cooling a human
body in the space. We might be modest that we consider the required minimal outdoor wind
energy would be ten times larger than indoor even though the various conditions of wind
induced ventilation. This means that indoor cross ventilation utilizes one tens outdoor wind
kinetic energy. We suggest that the minimal condition of utilization of cross ventilation is
more than 0.5m
2
/s
2
(around 1m/s) of wind energy outside building. With the analogous
discussion of indoor contamination control with ventilation, we may expect the efficient
indoor cross ventilation with the existence of the ten times larger outside wind kinetic energy.

VOID SURROUNDED BY BUILDINGS
Figure 2 illustrates the characteristics of the void spaces in built-up urban areas. It is not an
easy task to evaluate the air change rate and the space averaged kinetic energy of wind in a
void space. The wind tunnel experiment with a scaled urban model or 3-D CFD (Three
Dimensional Computational Fluid Dynamics) can execute this complicate works.

From the view point of thermal comfort of the inside building utilizing cross ventilation, the
kinetic energy of wind in void spaces is desired over 0.5m
2
/s
2
(around 1m/s).If the kinetic
energy is less than the value, we cannot expect cooling effect by cross ventilation generally.
Re-circulating Flows in Void Spaces
Turbulence Diffusion
Figure 2 Void Spaces in Urban Built up Area
Wind Blowing over Buildings
101

Then, how much the air flow rate per unit volume or the air change rate should be in the void
spaces from the view point of contaminant control in buildings? The answer should not be
derived from the normal conditions but from the abnormal or accidental conditions where
hazardous materials are released in a building with an accident and the rapid ventilation for
decreasing the concentration with opened windows, i.e. emergency cross ventilation, is
attempted. The nominal time constant of 6 ACH (air change rate per hour) is 10 minutes and
we can expect twice of the nominal time constant, 20 minutes, for the complete exchange of
room air with outdoor air under the complete room air mixing condition. 10 to 20 minutes are
the first and least time of purging the one shot of the hazardous material release and are
corresponding to the popular emergency response time. Within the time, fire fighters,
ambulance car officers will reach the accidental room and may deal with the accident. To
ensure the 6ACH in the room of emergency cross ventilation with opened windows, the air
flow rate per unit volume or the air change rate should be more than 60 ACH in the void
spaces with the same volume of the room. When the room faces the smaller void space, the
lager air change rate should be ensured for the void.
The nominal time constant of 60 ACH is 1 minute and we can expect 2 minutes, for the
complete exchange of void space air with outside void air. People may hold his breath for a
minute not to inhale the pollutant discharged outside with accident. 60 ACH may correspond
to peoples holding breath time.

VENTILATION EFFICIENCY INDICES FOR VOID SPACE
Local Purging Flow Rate
Local purging flow rate (LPFR) is an index of ventilation efficiency in a void space. It was
originally defined as the effective airflow rate to remove/purge contaminants from the local
domain. The definition is the net ventilation rate by which the domain-averaged concentration
of the local domain. The value of LPFR can be simply calculated from the concentration
simulation based on the scalar transport equation
[7]
.

LPFR = V
p
/ (VFT
p
) = q
p
/ C
p
(1)

where,
LPFR: local purging flow rate [m
3
/s]
q
p
: tracer particles generation rate per unit time [kg/s]
C
p
: local domain-average concentration [kg/m
3
]

EXCEEDANCE PROBABILITY METHODS
[8]

Wind environment is a stochastic phenomenon. To treat this stochastic feature, the
exceedance probability analysis is introduced in order to evaluate the extreme wind properties,
not the mean value. People may not recognize the daily change of averaged velocity, they are
however sure to recognize the frequency of the stagnant wind days with hot thermal
conditions and or highly air polluted atmosphere. They will evaluate wind environment with
the frequency of low velocity wind days when the wind did not help them to reduce the hot
and humid thermal sensation and or air pollution. The wind environment is evaluated with the
feature of probability density function and not with a few moment values such as a mean or a
variance. It is important how often stagnant wind day is. We may consider the probability
density function of instantaneous properties which is evaluated with three seconds response
(0.33 Hz time resolution), but the hourly mean values (0.310
-3
Hz time resolution) are
usually used. The former is used for evaluating the strong wind which affects on pedestrian
since the instantaneous strong wind affect greatly for pedestrian wind safty. In practice, since
it is difficult to predict the instantaneous wind features with a wind tunnel experiment or a
CFD, hourly mean values are used and a gust factor estimating instantaneous features from
102

the mean values are introduced. The latter is used for evaluating stagnant wind which is
important for contamination purge and or controlling thermal environment for human body.
The bottom lines is that in the wind engineering field, hourly mean values are usually used for
both the strong and stagnant wind evaluation.
It is well know that usual anemometers used at weather stations cannot measure low wind
velocity less than 1 m/s. Above 1 m/s wind speed, they have a measurement resolution of 0.1
m/s but below 1 m/s they only show wind velocity is below 1m/s and measurement resolution
is indeed 1 m/s.

Velocity-based EP
In practice, the exceedance probability method has often been used for urban planning and
design to assess the impacts of the proposal on the pedestrian wind environment. In making
such assessment, wind velocity ratios, which are defined as the ratios between velocities at
pedestrian level and velocities at reference point, are required for each wind direction. These
ratios are constant based on the assumption of a linear correlation between the velocities and
can be determined by measurement, wind tunnel experiment or CFD simulation with the local
metrological observatory usually being chosen as the reference point, where wind speed
distributions for 16 azimuths can be statistically described by Weibull function. It is
confirmed by many researches that the exceedance probability of hourly mean wind velocity
in any regions are well described with Weibull function. As a result, the total exceedance
probability for wind speed at ground level, referred to hereafter as velocity-based EP (V-
EP), can be expressed as follows
[10]
:
( ) ( ) ( )
( ) ( )
( )

= =
)
|
|
\
|
= > = >
15
0
0
15
0
0 0
exp
n
n K
V n
n C n R
V
n A n V V P V V P (2)
where P(V>V
0
|n)

is the probability of exceeding a given ground wind velocity (V0) for wind
approaching from each azimuth n; A(n) is the relative frequency of wind direction occurrence;
C(n) and K(n) are Weibull distributions parameters for each azimuth; R
V
(n) is the velocity
ratio between scalar velocity at ground level V
0
and velocity at reference height Vs(n) for each
azimuth, expressed by Equation (2)
[10]
:
( ) ( )
0 s
V
R n V V n = (3)
The Weibull distributions parameters C(n) and K(n) are calculated from the fitting procedure
of the observatory data at the reference height. As mentioned before, since the accuracy of
observatory data of lower velocity than 1 m/s is not assured and the accuracy of the fitting
process for the low velocity region is generally not so good, we should use the Weibull
distributions of low velocity region with considerable care. The exceedance probability of the
observatory data for low wind velocity region requires further researches at this moment.

Local Air Change Rate-based EP
As shown in Equation (3), local air change rate can be expressed by local purging flow rate
(LPFR) per volume of Void space (V
P
), while LPFR represents the effective airflow rate
required to remove/purge pollutant from the specified domain.
( )
P P P P
N LPFR V q C V = = (4)
where C
P
is the local average concentration [kg/m
3
]; and q
P
is the uniform emission rate of
pollutants [kg/s]. By analogy with Equation (1), the exceedance probability can also be
calculated based on the local air change rate by using Equation (4), referred to hereafter as
103

local air change rate-based EP (LACR-EP). Just as velocity-based EP means the
exceedance probability of a given velocity, local air change rate-based EP indicates the
exceedance probability of a given air change rate (N
0
), with the assumption of a linear
correlation between the velocity at the reference point and the calculated local air change rate.
( ) ( ) ( )
( )
( ) ( )
( )

= =
|
|
\
|
= > = >
15
0
0
15
0
0 0
exp
n
n K
g
s
n
n C n N
n V N
n A n N N P N N P (5)
where P(N>N
0
|n), N
g
(n), V
s
(n)

are the probability of exceeding N
0
,

the calculated local air
change rate and the velocity at the reference point for each azimuth respectively
[8]
.

Local Kinetic Energy-based EP
As shown in equation (6), total kinetic energy is used as another index to calculate
exceedance probability, referred to as KE.

( )
2 2 2
1 1
2
P Void
KE U V W k dv
V
| |
= + + +
|
\
(6)
The corresponding EP is referred to as local kinetic energy-based EP (KE-EP) and is
calculated by using equation (7), which takes the similar form as equation (2) and equation
(5).
( ) ( )
( )
( )
( ) ( )
( )
15 15
0
0 0
0 0
exp
K n
s
n n
g
KE V n
P KE KE P KE KE n A n
KE n C n = =

| |

|
> = > =
`
|

\
)

(7)
where P(KE>KE
0
|n),

KE
g
(n), V
s
(n)

are the probability of exceeding KE
0
, the calculated local
average kinetic energy and the velocity at the reference point from each azimuth respectively.
Though the calculation procedures for the three EPs above are similar, their meanings and
applications are somewhat different. The velocity-based EP analysis is suitable for assessing
wind conditions at some pedestrian locations so as to evaluate whether spot wind velocity is
acceptable or not for outdoor human activities, so it is essentially a point-oriented index.
Comparatively, the other two EPs are based on domain-averaged indices, so that makes it
possible to assess the overall ventilation performance for a target domain, void space
[8]
.

EXAMPLES OF WIND ENVIRONMENT ACCESMENT
[10]
With the urbanization process, the built-up areas of big cities are developing with greater
numbers of long, narrow streets flanked by buildings. The problems associated with pollution
within these so-called street canyons have received increasing attention over the past decades.
Here, the wind environment of modeled street canyons in Tokyo are evaluated with the local
air change rate-based EP (LACR-EP) and the local kinetic energy-based EP (KE-EP). The
local air change rate and local kinetic energy at the modeled street canyons are evaluated
with the CFD
[11-12]
. and the exceedance probability of the annual wind are based on the
observatory data of the metrological agency at Tokyo. The reference height in this example is
the height of the anemometer which is set at 74.5m high in the center of Tokyo.

Street Canyon Model and Analysis
Figure 3 shows the configuration of a computational street canyon model. Here, three isolated
street canyons are arranged in parallel along the X coordination at a distance apart of 100 m.
In this model, each canyon is located below ground level, where L = 100 m; L is the canyon
104

length and H = 9 m; H is the height and W is the width varying from 1 m to 6 m. At the
centre-bottom of each canyon, a rectangular domain (W 10 m 3 m in the X, Y, and Z axes
respectively) is defined as the targeted void space for evaluation of wind environment.
As shown in Figure 4, the horizontal plane is divided into 16 azimuths, marked with the
symbol n starting from North-North-East (NNE) in a clockwise direction. The direction of the
short axis of the canyon is defined as its orientation and the canyon orientation angle is
represented by . For the canyon model, the approaching wind is grouped into 16 directions
(i), starting from the canyon orientation clockwise and the wind incidence angle is represented
by .

Calculation Method of EP

The calculation procedure for EPs follows three steps. Firstly, for the model shown in Figure
3, CFD simulations are performed to calculate local air change rate or average kinetic energy
Figure 3. Configuration of model

Figure 4. Definitions for canyon model

Table 1. Analysis Conditions

Figure 5. Wind Rose (Tokyo)
Turbulent
model
Standard k- model
Differential
scheme
Convection terms: MARS*
1

Inflow*
2
(Murakami
1988)
0
4 3 4 1
0
2 1
2 3
2
4 1
0 0
4
1 0 5 1
U Z Z k C L
L k C
i U i k
Z Z U U
/ ) (
/
. , ) ( .
) / (
/ / /
/
/
=
=
= =
=

Side, sky Free slip
Wall Generalized logarithmic law
Other
Void pollutant is assumed to be passive
pollutant with a uniform emission rate of
0.001 kg/(m
3
s)

*
1
MARS: Monotone Advection and Reconstruction Scheme, second-
order scheme (STAR-CD 2001)
*
2
Reference height Z
0
=74.5 m; Reference velocity U
0
=1.0 m/s

105

of the targeted void space for 16 wind directions respectively. Next, the exceedance
probabilities for each azimuth are calculated respectively. Finally, EPs are summarized from
individual directional results by using Equations (7) and (9). All the numerical simulations
were performed using STAR-CD in this study and the detailed analysis conditions are shown
in Table 1
[11-12]
.
The aspect ratio, which is defined as the ratio of the canyon height to its width (H/W), has
been regarded as one of the main factors affecting the interior flow and pollutant transport
behavior. In this study, the height of the canyon model has a constant value of 9 m, and
widths from 1 m to 6 m were investigated, with corresponding relatively high aspect ratios
from 1.5 to 9. The distributions of EPs in the range N
0
= 0~400 h
-1
and KE
0
= 0~0.1 m
2
/s
2
were
calculated and analyzed for each case.
Here, the Weibull distribution parameters and frequency of occurrences of 16 wind directions
in Tokyo were adopted (the Wind Rose is shown in Figure 5). This statistical data is based on
10 years of measurements recorded at the Tokyo Meteorological Observatory from 1995 to
2004 (Japan Association for Wind Engineering 2005). As shown in Figure 5, the annual
prevailing wind in Tokyo is prominent from NNW (20.6%), compared with the other
directions.

Results and Discussion
Local Air Change Rate (N) & Local Kinetic Energy (KE)

(a) Local Air Change Rate (W = 4 m) (b) Local Kinetic Energy (W = 4
m)
Figure 6. Comparison between three canyon Voids
0
0.01
0.02
0.03
0.04
0.0 22.5 45.0 67.5 90.0
Wind Incidence Angle []
L
o
c
a
l

K
i
n
e
t
i
c

E
n
e
r
g
y

[
m
2
/
s
2
]
Void1
Void2
Void3
0
50
100
150
200
0.0 22.5 45.0 67.5 90.0
L
o
c
a
l

A
i
r

C
h
a
n
g
e

R
a
t
e

[
h
-
1
]
Void1
Void2
Void3
(a) Local Air Change Rate (b) Local Kinetic Energy

Figure 7. CFD Calculation Results for Local Air Change Rate and Local Kinetic Energy
0
50
100
150
200
250
0.0 22.5 45.0 67.5 90.0

L
o
c
a
l

A
i
r

C
h
a
n
g
e

R
a
t
e

[
h
-
1
]
W=1
W=2
W=3
W=4
W=5
W=6
0
0.01
0.02
0.03
0.04
0.05
0.0 22.5 45.0 67.5 90.0
L
o
c
a
l

K
i
n
e
t
i
c

E
n
e
r
g
y

[
m
2
/
s
2
]
W=1
W=2
W=3
W=4
W=5
W=6
106

The CFD simulation results of local air change rate and local average kinetic energy of the
Void spaces were firstly compared among the three canyons along the stream-wise direction.
Taking the case of W = 4 m as an illustration, the results of each canyon are almost the same
(see Figure 6) because there is sufficient distance between the three canyons, and similar
conditions can also be found for other canyon widths.
The simulation results for local air change rate and average kinetic energy for all cases are
shown in Figures 7-(a) and 7-(b) respectively. The results for the air change rate indicate that
the minimal value occurs for all widths when the incidence angle is 0.0, in other words when
the wind direction is perpendicular to the canyon, the Void space has the worst ventilation. As
the wind incidence angle increases, air change rate shows an increasing tendency and
increases sharply from 45.0 to 67.5. The peak value is reached at = 67.5 for cases where
W > 2 m and at = 90.0 for the other cases with a narrower width. In all cases, the air
change rate increases with an increase in canyon width. Thus, W has a great influence on the
improvement of the air change rate in void spaces, especially when the width increases from 1
m to 3 m; however this effect becomes insignificant for widths exceeding 4 m. The results for
average kinetic energy show similar distributions in general (see Figure 7-(b)), with the
exception that average kinetic energy remains almost the same value for a given canyon width
where the incidence angle is within 45.0.
The EPs

were calculated for N
0
= 0~400 h
-1
and KE
0
= 0~0.5 m
2
/s
2
respectively for each case
based on the above simulation results. The distributions of EPs for W = 4 m are only
presented in Figure 8. Figure 8-(a) indicates that the distributions of air change rate-based
exceedance probabilities are quite different for canyon orientations. As a whole, the EP has
the highest value when the orientation of the canyon is East (E), while the lowest value is
observed for North-North-East (NNE). From the CFD results, ventilation performance in the
void space becomes higher at the incidence angle 67.5 and 90.0. That is to say, for the
canyon facing E, the predominant wind directions are SSE, S, SSW, NNW, N and NNE.
Figure 8-(b) shows the distribution of average kinetic energy-based exceedance probabilities
within the range of 0~0.5 m
2
/s
2
. The maximum value and the minimum value are also found
at orientation E and orientation NNE respectively.

Influence of Canyon Orientation on EPs

Influence of Canyon Orientation on EPs
Desirable local air change rate in the void space outside building would be 60 ACH for
ensuring 6 ACH emergency cross ventilation with opened windows in a room. Figure 8-(a)
indicates that 60 ACH is attained with 90% probability for the canyon orientation of East (E)
and with 83% for the canyon orientation of North-North-East (NNE) for the W= 4 m and H=
9 m street canyon. We have assumed that the wind kinetic energy more than 0.05m
2
/s
2

(around 0.3m/s) will be useful for cooling a human body with wind induced cross ventilation
(a) Local Air Change Rate-based EP (b) Local Kinetic Energy-based EP

Figure 8. Distributions of EPs according to Void Orientation (W = 4 m)
0
20
40
60
80
100
0 40 80 120 160 200 240 280 320 360 400
Local Air Change Rate [h
-1
]
E
x
c
e
e
d
a
n
c
e

P
r
o
b
a
b
i
l
i
t
y

[
%
]
NNE NE ENE E ESE SE SSE S
NNE
E
0
20
40
60
80
100
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
E
x
c
e
e
d
a
n
c
e

P
r
o
b
a
b
i
l
i
t
y

[
%
]
Local Kinetic Energy [m
2
/s
2
]
NNE NE ENE E ESE SE SSE S
NNE
E
107

and that the minimal condition of utilization of cross ventilation is more than 0.5m
2
/s
2
(around
1m/s) of wind energy outside build. Figure 8-(b) indicates that any W= 4 m, H= 9 m street
canyon in Tokyo cannot obtain such the conditions. The building faced to such the street
canyon should use only air-conditioning and cannot expect the utilization of natural cross
ventilation any time.

CONCLUSIONS
Wind environment is a stochastic phenomenon. To treat this stochastic feature, the
exceedance probability analysis should be introduced in order to evaluate the probability of
the low speed wind properties. The wind environment might be evaluated in the aspect of
ventilation performance and thermal comfort. Two representative indices of wind
environment are chosen here. One is Local Purging Flow Rate (LPRF), and another is space
averaged Kinetic Energy of Wind (KEW). Here, the wind environments of modeled street
canyons in Tokyo are evaluated for the demonstration purpose. The exceedance probabilities
of the annual wind are based on the observatory data of the metrological agency at Tokyo.

REFERENCES

1 Yassin, M.F., Kato, S., Ooka, R., Takahashi, T. and Kouno, R. (2005), Field and wind-
tunnel study of pollutant dispersion in a built-up area under various meteorological
conditions, Journal of Wind Engineering and Industrial Aerodynamics, Volume 93,
Pages 361382
2 Huang, H., Ooka, R. and Kato, S. (2005), Urban thermal environment measurements and
numerical simulation for an actual complex urban area covering a large district heating
and cooling system in summer, Atmospheric Environment, 39, 6362-6375.
3 Huang, H., Ooka, R., Chen, H., Kato, S., Takahashi, T. and Watanabe, T. (2008), CFD
analysis on traffic-induced air pollutant dispersion under non-isothermal condition in a
complex urban area in winter, Journal of Wind Engineering and Industrial
Aerodynamics, Volume 96, Issues 10-11, Pages 1774-1788
4 Huang, H., Kato, S., Ooka, R. and Jiang, T. (2006),CFD analysis of ventilation efficiency
around an elevated highway using visitation frequency and purging flow rate, Wind &
Structures, An International Journal, 9, 4, 297-313.
5 Bady, M., Kato, S. and Huang, H. (2008), Towards the application of indoor ventilation
efficiency indices to evaluate the air quality of urban areas, Building and Environment,
Volume 43, Pages 1991-2004
6 Kato, S., and Huang, H. (2009). Ventilation efficiency of void space surrounded by
buildings with wind blowing over built-up urban area, Journal of Wind Engineering and
Industrial Aerodynamics, Volume 97, Pages 358-367
7 Kato, S. and Murakami, S. (1988), New ventilation efficiency scales based on spatial
distribution of contaminant concentration aided by numerical simulation, ASHRAE
Transaction, Vol. 94, 2, 309-330.
8 Kato, S., Murakami, S. and Kobayashi, H. (1993), New scales for evaluating ventilation
efficiency as affected by supply and exhaust openings based on spatial distribution of
contaminant, Proceedings of ISRACVE Tokyo, ASHRAE.
9 Kato, S. Ito, K. and Murakami, S. (2003), Analysis of Visitation Frequency through
particle tracking method based on LES and model experiment, Indoor Air, 13, 182-193.
10 Bu, Z., Kato, S., Ishida, Y. and Huang, H. (2009), New criteria for assessing local wind
environment at pedestrian level based on exceedance probability analysis, Building and
Environment, Volume 44, Issue 7, July 2009, Pages 1501-1508
11 Launder, B. E. and Spalding, D. B. (1974), The numerical computation of turbulent
flows, Comp. Mech. Eng., 3, 269-289.
108

12 STAR-CD, (2001), Methodology, STAR-CD Version 3.15, Computational Dynamics
Limited.
109
Simultaneous Measurement of Velocity and
Temperature in Unstable Turbulent Boundary Layer
and Numerical Analysis by LES
Koudai Katada
a
, Ryuichiro Yoshie
b
, Guoyi Jiang
c

a
Tokyo Polytechnic University, Atsugi, Kanagawa, Japan, m1064005@st.t-kougei.ac.jp
b
Tokyo Polytechnic University, Atsugi, Kanagawa, Japan, yoshie@arch.t-kougei.ac.jp
c
Tokyo Polytechnic University, Atsugi, Kanagawa, Japan, jiang@arch.t-kougei.ac.jp
ABSTRACT: Wind tunnel experiment of an unstable turbulent thermal boundary layer was
carried out, and mean velocity, mean temperature, the r.m.s. value of velocity and
temperature fluctuations, Reynolds stresses and turbulent heat fluxes were measured. Then
large eddy simulation was conducted to generate the turbulence statistics corresponding to the
experimental result. Two methods (precursor method and recycling method) were examined
for the LES by comparing the simulated results with the wind tunnel experimental data.
1 INTRODUCTION
Urban heat island phenomena and air pollution problems become more and more serious in
weak wind regions such as behind buildings and within street canyons. Computational Fluid
Dynamics (CFD) is expected to be a useful means for predicting and evaluating these
phenomena. However, the prediction results by RANS model are poor in weak wind regions
such as behind buildings and within street canyons (Yoshie et al., 2007). Thus, practical
applications of LES to these problems are anticipated. In the weak wind regions, buoyancy
effect cannot be neglected. However, LES study of non-isothermal flows is quite rare in the
field of wind engineering. When simulating the turbulent atmospheric boundary layers using
LES, a crucial issue is how to impose physically correct inflow fluctuations. The influence of
inflow turbulence on large eddy simulation in a neutral boundary layer has been presented by
Tominaga et al. (2008), and in a non-isothermal boundary layer by Yoshie et al. (2011). It is
confirmed that the inflow turbulence for LES is extremely important. Several techniques have
been proposed for generating inflow turbulence for LES in a neutral boundary layer. Lund et
al. (1998) proposed a rescaling recycling method to generate developing turbulent inflow data
for LES. Kataoka and Mizuno (2002) simplified Lunds method by assuming that the
boundary layer thickness is constant within the driver section. Nozawa and Tamura (2002,
2005) extended Lunds method to a rough-wall boundary layer flow, using a roughness block
arrangement. Now these methods are widely used to generate inflow fluctuations for LES in
the neutral boundary layers. When LES is applied for non-isothermal flow field, not only
inflow velocity fluctuation but also temperature fluctuation is necessary. The important issue
is how to generate physically correct temperature fluctuation for the non-isothermal LES, but
the generation method of the temperature fluctuation has not been sufficiently examined yet
and few reliable experimental data are available to compare with the non-isothermal LES
results. Thus we carried out wind tunnel experiment of the unstable turbulent thermal
boundary layer, and examined two LES methods of generating the velocity and temperature
fluctuations in that boundary layer.

110
Table 1: Nomenclature
X,Y,Z :space coordinates (stream-wise, lateral, and vertical directions in the wind tunnel)
u,v,w :three components of wind velocity [m/s]
f : instantaneous value of a quantity
<f> : time-averaged value of f
f : fluctuation from time-averaged value
f = f<f>
f
: r.m.s. value (standard deviation of variable f)
k: turbulent kinetic energy (m
2
/s
2
)
Ri: gradient Richardson number
: boundary layer thickness (m)
: temperature [C]
U
: generated mean bulk velocity above the

boundary layer height
: generated mean bulk temperature above the

boundary layer height
f
: floor temperature [C]
0
: vertically averaged air temperature along the
boundary layer thickness
: |
f
-
|
2 WIND TUNNEL EXPERIMENT
The experiment was conducted in a thermally stratified wind tunnel (cross section at
measurement part: 1.2 m 1.0 m) of Tokyo Polytechnic University shown in Figure 1. The
wind tunnel consists of fun, temperature profile cart, floor heating and cooling panels and
ambient air conditioner. In this experiment, the surface temperature of the wind tunnel floor
was uniformly controlled to be 46C. Mean wind velocity and air temperature at the inlet of
the wind tunnel were set at about 1.4 m/s and 9C, respectively. The turbulent boundary layer
was generated by 26 thin aluminum plates 9mm high shown in Figures 2 and 3. They were
placed at 200mm pitch on the wind tunnel floor.
A split-film was adopted for wind velocity measurement and a cold-wire was used for
temperature measurement. In order to precisely compensate for the contribution of
temperature fluctuation to the output voltages of the split-film under the low wind speed
condition, it is necessary to obtain precise calibration data under the stable wind speed and
temperature condition. For this purpose we developed a calibrator for a hot-wire anemometer
and a cold-wire thermometer (Figure 5) to improve calibration precision and efficiency.
The horizontal measuring points for the unstable turbulent thermal boundary layer are
shown in figure 2. The split-film and the cold-wire were set adjacent with distance of
Y=2mm to measure fluctuating velocity and temperature simultaneously. The sampling
frequency and the low-pass filters cutoff frequency were set at 1000 Hz and 200 Hz,
respectively, and 60,000 data were sampled in 60 seconds. The measurements were repeated
three times on different days. The repeatability was good and the averaged values of the
repeated measurements are shown in this paper. The characteristics of the generated turbulent
boundary layers at the measuring station (average along the span-wise direction at X=-
400mm) are as follows. The boundary layer thickness = 0.24m which is used as the
reference height in this paper. U
and

are mean stream-wise velocity and mean
temperature at the height of , which are 1.45 m/s and 10C, respectively. The temperature
difference =|
f
-
| is 36 C. This is used as the reference temperature difference in this

paper.
0
is the averaged temperature from the ground to the boundary layer height position
which is 16 C. Gradient Richardson number Ri is the most widely used indicator of stability
in the atmospheric boundary layer, defined as:
2
0
) (
) )( (
z u
z g
Ri
c c
c c O
=
u
(1)
In which, g is the acceleration due to the gravity (9.8 m/s
2
), <u> is the stream-wise mean
velocity, and <> is the mean temperature. The magnitude of Ri gradually increased along
111
the vertical direction, and reached -0.3 near the boundary layer height, which corresponded to
a moderate thermal stratification.

Ambient air conditioner
Temperature profile cart
Floor heating and cooling panels
Fun

Figure 1: Thermally Stratified Wind Tunnel (mm)

:Measurement points :Heating panels
1200 1670 500 5000
7200
100 400 400
1
2
0
0
100
100
300
X=0
X=-400 X=400
Flow
Temperature
profile cart
Y
X
100
Aluminumplates @200mm
500

Figure 2: Setup of experiment (mm)
9mm
9
m
m
1mm
Flow
200mm
Z
X
Y

A - A
3
Z
Y

Figure 3: Aluminum plates as roughness elements Figure 4: Layout of sensors

2
6
5
5,000 428 800 200
Flow
measurement part
Temperature
control part
Calibration
part
Fan
Laminar
flow meter
D=100
Heater
Filter Filter
2
6
5
5,000 800 200
Flow
measurement part
Temperature
control part
Fan
Laminar
flow meter
D= 100
Heater
Filter Filter
Calibration
Part
428

Figure 5: Calibrator for hot-wire and cold-wire (mm)
3 GOVERNING EQUATIONS AND NUMERICAL METHOD
The filtered three dimensional governing equations expressing the conservation of mass,
momentum and energy with the Boussinesq approximation for LES can be written in the
Cartesian coordinates as follows:
112
0 =
c
c
j
j
x
u
(2)
) ( g )]
x
u
x
u
)( [(
x x
p
x
) u u (
t
u
ref i
i
j
j
i
SGS
j i j
j i
i

c
c
+
c
c
+
c
c
+
c
c
=
c
c
+
c
c
(3)
]
x
)
Pr Pr
[(
x x
) u (
t
j SGS
t
j j
j
c
u c v
+
v
c
c
=
c
u c
+
c
u c
(4)
Where,
SGS
is the sub-grid scale eddy viscosity, and
SGS
=
ij ij S
S S C 2 ) (
2
A in the standard
Smagorinsky model. C
S
is the model constant, and a value 0.12 was used. Pr
SGS
is SGS
Prandtl number, and a value 0.71 was given. The last term in the momentum equation
(equation (3)) is called buoyancy term. In this study, if this term was included, then a buoyant
solver was applied. If this term was not included, then the temperature was treated as a
passive scalar, and the existence of temperature doesnt influence the flow field. A second
order backward differencing scheme (using the current and two previous time-step values)
was adopted for time discretization. The second order central difference scheme was adopted
for the convection terms. The PISO algorithm (Pressure Implicit with Splitting of Operators)
was used for pressure-velocity calculation procedure. The PISO involves one predictor step
and two corrector steps. At each time step, velocity and temperature are predicted, and then
pressure and velocity are corrected. The PISO algorithm was adopted, because it has a better
performance in the simulation of unsteady flow (Issa, 1986).
4 GENERATION OF VELOCITY AND TEMPERATURE FLUCTUATIONS
4.1 Case-a: precursor method
A precursor method for generating both velocity and temperature fluctuations simultaneously
in non-isothermal boundary layers was introduced here. In this precursor simulation, the
whole wind tunnel (6.5 meters long) and all the aluminum plates (shown in Figure 6) were
reproduced by LES using a buoyant solver. The plates were treated as zero-thickness plates in
the simulation. The wind velocity and temperature distributions at the wind tunnel inlet were
spatially uniform and turbulence intensity was very small (less than 1%), so a uniform
velocity U=1.42 m/s, and a uniform temperature = 9.4C without turbulence were given to
the inflow boundary of the precursor simulation. A zero gradient condition was used for the
outlet boundary condition. Uniform grids of X=5mm and Y=10mm were used in the
stream-wise and span-wise directions, respectively. The depth of the first fluid cells on the
floor was 0.1mm. A no-slip boundary condition was applied to the wall shear stress on the
floor. The non-dimensional distances from the surfaces to the first fluid cells were below 1.0
for most regions. As the thermal boundary conditions, the surface temperature 45.3C was
prescribed and a heat conduction boundary condition (Fouriers Law) was applied for the
heat flux on the floor surface. A wall function of spading law (1961) was used for the two
lateral boundaries (side walls of the wind tunnel) and the top boundary (ceiling of the wind
tunnel). The total mesh number used in this simulation was 104212079 = 9,878,160. A
unique time steps t= 0.001 s was used to make sure that the maximum courant number was
less than 1 in all positions of the domain. The sampling station to obtain fluctuating velocity
and temperature data was put 0.1m downstream of the last aluminum plate (the same position
as the experimental measuring station at X=-400mm). The Van Driest-type damping function
was used to account for the near wall effect.
113
4.2 Case-b: recycling of velocity fluctuation with passive scalar treatment for temperature
In this case, the velocity fluctuation was generated using Kataokas method (Kataoka and
Mizuno, 2002) with the roughness ground arrangement described by Nozawa and Tamura
(2002). The roughness elements were exactly the same as those used in the precursor
simulation, but a short domain (6) was adopted here. A mean velocity profile of the
experimental was prescribed for the inflow condition, and only the fluctuating part was
recycled between outlet position and inlet position. The velocity components at the inflow
boundary are given as follows:
{ }
recy inlt
inlt
) z , y ( u ) t , z , y ( u ) ( ) z ( u ) t , z , y ( u q | + =
{ }
recy
inlt
) z , y ( v ) t , z , y ( v ) ( ) t , z , y ( v q | = (5)
{ }
recy
inlt
) z , y ( w ) t , z , y ( w ) ( ) t , z , y ( w q | =
The following damping function (Kataoka, 2008) was used to restrain the velocity fluctuation
from developing:
)} 0 . 8 tanh( / ]
82 . 0 4 . 0
) 0 . 1 ( 0 . 8
tanh[ 1 {
2
1
) (
+ q
q
= q | (6)
Where, = z/. A no-slip condition was used for the ground surface, and a wall function was
used for the two sides and the top boundaries, which are same boundary conditions as case-a.
For this case, the temperature was treated as a passive scalar and no buoyancy term was
included in the momentum equation. A mean temperature profile of the experiment was
prescribed at the inflow boundary, and we tried to use the fluctuating velocity field to
generate a fluctuating temperature field. Total mesh number was 15012055 = 990,000,
which was much less than that used for the case-a (precursor simulation). The arrangement
for this case is shown in Figure 7.

1
0
0
0
1
2
0
0

6500 500 200 100
aluminum plates
sampling station
X
Y
900
Inlet
Outlet
Top View

M
e
a
n

t
e
m
p
e
r
a
t
u
r
e

p
r
o
f
i
l
e
Y X
Z
44
40
36
32
28
24
20
16
12
10

Figure 6: Outline of simulation for case-a Figure 7: Recycling method for case-b
5 RESULTS
Figure 8 (a) and (b) shows the stream-wise mean velocity and mean temperature distribution
at the sampling station. All the values have been averaged along the span-wise direction for
both experiment and numerical simulation. Generated (calculated) mean profiles by both
case-a and case-b agreed well with those of the experiment. Figure 8 (c) - (g) show the
vertical distributions of various turbulence statistics. The generated r.m.s. value of stream-
wise velocity and temperature fluctuations,
u
and
, by both case-a and case-b well matched

114
the experimental data, but
w
were somewhat overestimated. The distribution patterns of the
generated Reynolds stress and turbulent heat fluxes were similar to those of the experiment.
For case-b although only a mean temperature profile was given at inlet boundary, the
temperature gradually fluctuated due to the turbulence in velocity as the flow proceeded
downstream. At the sampling station, a target profile of temperature fluctuation was formed.

0
0.2
0.4
0.6
0.8
1
1.2
0 0.2 0.4 0.6 0.8 1 1.2
Z
/
<u>/U
Exp
case-a
case-b

0
0.2
0.4
0.6
0.8
1
1.2
-1.2 -1 -0.8 -0.6 -0.4 -0.2 0
Z
/
(<>-
f
)/
Exp
case-a
case-b

0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1 0.15 0.2
Z
/
/
Exp
case-a
case-b

(a) Mean stream-wise velocity (b) Mean temperature (c) The r.m.s. value of temperature
0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1 0.15 0.2
Z
/
u
/U
Exp
case-a
case-b

0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1 0.15 0.2
Z
/
v
/U
Exp
case-a
case-b

0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1 0.15 0.2
Z
/
w
/U
Exp
case-a
case-b

(d) The r.m.s. value of
stream-wise velocity
(e) The r.m.s. value of
lateral velocity
(f) The r.m.s. value of vertical
velocity
0
0.2
0.4
0.6
0.8
1
1.2
0 0.01 0.02 0.03
Z
/
k/U
2
Exp
case-a
case-b

0
0.2
0.4
0.6
0.8
1
1.2
-0.015 -0.01 -0.005 0
Z
/
<u'w'>/U
2
Exp
case-a

0
0.2
0.4
0.6
0.8
1
1.2
-0.01 -0.005 0 0.005 0.01
Z
/
<w''>,<u''>,<v''>/U
Exp <w''>
case-a <w''>
Exp <v''>
case-a <v''>
Exp <u''>
case-a <u''>

(g) Turbulent kinetic energy (h)Reynolds stress (i) Turbulent heat fluxes
Figure 8: Comparison of vertical profiles at X=-400mm
Figure 9 shows the instantaneous contours of stream-wise velocity and temperature in the
symmetry plane (Y= 0) for case-b. L is the distance between the aluminum plates (200 mm),
115
and X/L= 0 corresponded to the inlet position, the black triangular is the sampling station. As
the recycling procedure was adopted for the velocity field, the velocity was fluctuating in the
whole region (Figure 9a). But only a mean temperature profile was given as inflow condition,
so the temperature fluctuation is very small near the inlet position (Figure 9b). As the flow
proceeded downstream, the temperature field was gradually fluctuated by turbulence in
velocity. Figure 10 shows the vertical distributions of turbulent kinetic energy, the r.m.s.
value of temperature fluctuation in 6 different positions along stream-wise direction (case-b).
The r.m.s. value of temperature fluctuation was very small near the inflow boundary
(X/L=0.5 in Figure10b) because temperature fluctuation was not given to the inflow
boundary. But as the flow proceeded downstream, the temperature fluctuation became larger
and larger until the position of X/L=4.5. After this position, the turbulence reached
equilibrium and the profile that matched the experimental data were formed and maintained
downstream. According to the results of Figure 9 and Figure 10, the method by Tamura et al.
(2003) is considered to be valid if sufficient distance of fetch is ensured before the target
region. The temperature would fluctuate more quickly if the recycling procedure is also
adopted to temperature (Kongs method). But it is too complicated, and it is not necessary to
do this.

0 0.5 1.5 2.5 3.5 4.5 5.5 6.5 7.5
0
1
2
3
4
X/L
0 0.5 1.5 2.5 3.5 4.5 5.5 6.5 7.5
0
1
2
3
4
X/L
(a) Instantaneous contour of stream-wise velocity (b) Instantaneous contour of temperature
Figure 9: Instantaneous contours of stream-wise velocity and temperature in symmetry plane (case-b, Y= 0)

0
0.2
0.4
0.6
0.8
1
1.2
0 0.01 0.02 0.03
Z
/
k/U
2
Exp
X/L=0.5
X/L=1.5
X/L=2.5
X/L=3.5
X/L=4.5
X/L=5.5
X/L=6.5

0
0.2
0.4
0.6
0.8
1
1.2
0 0.05 0.1 0.15
Z
/
/
Exp
X/L=0.5
X/L=1.5
X/L=2.5
X/L=3.5
X/L=4.5
X/L=5.5
X/L=6.5

(a) Turbulent kinetic energy (b) The r.m.s. value of temperature fluctuation
Figure 10: Vertical distributions of turbulence statistics in different positions along stream-wise direction
6 CONCLUSIONS
In this study, wind tunnel experiment of an unstable turbulent thermal boundary layer was
carried out using the simultaneous measuring techniques developed by the present authors.
Then large eddy simulation was conducted to generate the turbulence statistics corresponding
to this experimental result. For LES, two methods were examined, and the results were
116
compared with the wind tunnel experimental data. The main conclusions of this study are as
follows:
(1) The characteristics of the generated flow (mean profiles and fluctuation profiles) by both
precursor simulation and recycling method almost agreed well with those of the wind tunnel
experiment. Because no inflow turbulence is required for precursor method (only a uniform
velocity and temperature are given as inflow condition), this method is the simplest. But it
requires a long computational domain and long simulation time.
(2) In the recycling method, a short domain can be adopted to make turbulence develop. For
the temperature field, giving a mean profile as the inflow condition is the simplest way.
(3) Even when the inflow temperature fluctuation was not given, temperature fluctuation
developed as the flow proceeded downstream in a turbulent boundary layer, but a sufficient
distance should be ensured for it to occur. The recycling procedure is not necessary for
temperature field.
7 ACKNOWLEDGEMENTS
This study was partially supported by the Ministry of Education, Culture, Sports, Science and
Technology, Japan, through the Global Center of Excellence Program of Tokyo Polytechnic
University. Also we would like to express our gratitude to Japan Society for the Promotion of
Science (JSPS) for Grant-in-Aid for Scientific Research (B), (No. 21360283).
8 REFERENCES
Issa, R.I., 1986. Solution of the implicit discretized fluid flow equations by operator-splitting. Journal of
Computational Physics, 62: 40-65.

Kataoka, H., Mizuno, M., 2002. Numerical flow computation around aeroelastic 3D square cylinder using
inflow turbulence. Wind and Structures, 5, 379-392.

Lund, T.S. et al., 1998. Generation of turbulent inflow data for spatially-developing boundary layer simulations.
Journal of Computational Physics, 140, 233-258.

Nozawa, K., Tamura, T., 2002. Large eddy simulation of the flow around a low-rise building immersed in a
rough-wall turbulent boundary layer. Journal of Wind Engineering and Industrial Aerodynamics, 90, 1151
1162.

Nozawa, K., Tamura, T., 2005. Large eddy simulation of wind flows over large roughness elements.
Proceedings of EACWE4, 16.

Tamura, T. et al., 2003. LES of spatially-developing stable/unstable stratified turbulent boundary layers.
DLES5, 6566.

Tominaga,Y. et al., 2008. Comparison of various revised k- models and LES applied to flow around a high-rise
building model with 1:1:2 shape placed within the surface boundary layer. Journal of Wind Engineering and
Industrial Aerodynamics, 96, 389- 411.

Yoshie, R. et al., 2007.Cooperative project for CFD prediction of pedestrian wind environment in the
Architectural Institute of Japan, Journal of Wind Engineering and Industrial Aerodynamics, 95, 15511578.

Yoshie, R. et al., 2011. CFD simulations of gas dispersion around high-rise building in non-isothermal boundary
layer. Journal of Wind Engineering and Industrial Aerodynamics, 99, 279-288.
117

New wind tunnel facility in agricultural research
M. Fiedler
a
, K. v. Bobrutzki
a
, M. Samer
a
, W. Berg
a
a
Leibniz Institute for Agricultural Engineering Potsdam-Bornim,
14469Potsdam, Germany, m.fiedler@atb-potsdam.de
ABSTRACT:
A new wind tunnel facility for agricultural research has been built at the Leibniz Institute for
Agricultural Engineering Potsdam-Bornim. The test section of the new wind tunnel will be 20
m long, 3 m wide and 2 m high and will provide a turn table. The ceiling will be adjustable
up to 2.5 m. An axial fan sucks air through the test section and should reach wind speeds up
to 20 m/s. Two main research applications for the new wind tunnel are exemplarily presented.
One is a dispersion study in complex terrain the other is a ventilation experiments for natural
ventilated livestock buildings. These two applications require different kinds of experiments
with different model scales. These considerations were taken into account during the basic
concept by planning the wind tunnel facility.
1 INTRODUCTION
One of the most important agricultural research areas is the investigation of atmospheric
pollution by agriculture itself. Especially animal husbandry is one major source of gaseous
emissions with either high greenhouse gas potential (e.g. methane and nitrous oxide) or
health and environmental damages (e.g. ammonia). In order to establish effective regulatory
measures of emissions and immissions, dispersion scenarios need to be investigated. Firstly,
this involves understanding the behaviour of the emission source and, secondly, the
dispersion processes from livestock buildings.
Economic reasons lead to an industrialisation of the animal husbandry with the result of an
intense mass animal farming. Often, large livestock building complexes arise which cause a
strong heterogeneous flow field around these facilities and provide many different sources.
Furthermore, many livestock buildings are naturally ventilated and the outgoing emission
flow of pollutant gases is directly dependent on the wind and turbulence fields within and
around the building. For such cases the source characteristics (e.g. emission mass flow) are
highly variable and difficult to estimate. Only few reliable datasets exist which provide the
measured emission streams of naturally ventilated livestock buildings. The results of these
studies have shown a large variability of the estimated emission streams (e.g. Groot
Koerkamp et al., 1998; Samer et al., 2011a, b).
More datasets are needed which provide information on the emission behaviour and
dispersion processes as well as wind and turbulence fields within and around naturally
ventilated buildings and complex livestock building arrangements.
In order to provide such datasets, a new boundary layer wind tunnel facility is currently built
at the Leibniz Institute for Agricultural Engineering Potsdam-Bornim.
118

2 APPLICATIONS OF AGRICULTURAL RESEARCH
Two examples of agricultural research topics are presented that are currently investigated at
the institute. These examples illustrate the character of the research questions that will be
addressed with the new wind tunnel facility.
2.1 Dispersion studies around a broiler farm
Emission and dispersion of atmospheric ammonia (NH
3
) from large livestock facilities have a
direct impact on the environment. Von Bobrutzki et al. (2011a) performed field experiments
around a huge broiler farm that is typical for agricultural facilities in Germany and Europe.
The broiler farm consists of nine small barns (length: 88 m, width: 12 m and height: 3.5 m),
each is housing 21,800 broilers (see a in Fig. 1). There are also three big barns (length: 93
m, width: 29 m and height: 4.5 m) with 62,000 broilers each (see b in Fig. 1). In total,
around 382,200 broilers are raised during a growing cycle. The broiler farm is surrounded by
a forest (further description see von Bobrutzki et al., 2011a).

Figure 1: The investigated broiler farm, where a and b represent the smaller and the bigger broiler barns,
respectively, adopted from von Bobrutzki et al. (2011a).
The field measurements described by von Bobrutzki et al. (2011a) focus on the estimation of
the emission mass flow from the barn as well as the monitoring of Ammonia in the
surrounding of the broiler farm. The Ammonia was sampled with high resolution at one
monitoring point (measurement tower) and spatially more detailed, but with low resolution
with passive samplers.
Von Bobrutzki et al. (2011a) found that the ammonia emission and dispersion are influenced
by various local factors such as NH3-mass-flow, internal and external temperatures, mean
and turbulent wind components in horizontal and vertical directions, atmospheric stability,
and exhaust air height. Thereby, major factors which characterise the atmospheric conditions
as well as the operational parameters of an animal production facility were evaluated. The
emitted NH
3
was strongly influenced by the actual wind direction. The sensor at a monitoring
point detected NH
3
in the air even when the wind blew over an area without sources of NH
3
.
This observation implied that NH
3
-enriched reverse winds blew back in the direction of the
monitoring tower. It also implied that a downward flow of NH
3
-enriched air reached the
119

measuring device and was detected. Von Bobrutzki et al. (2011a) suggest that these processes
need to be considered in any future study focusing on the spatial dispersion of NH
3
in the air.
They further suggests that the location of a monitoring point is essential for a successful and
accurate investigation outcome, especially in terms of applying suitable monitoring methods.

In a second study for the same broiler farm, von Bobrutzki et al. (2011b) quantified NH
3

emissions and the resulting atmospheric NH
3
concentrations in the immediate vicinity.
Additionally, they analysed the impact of NH
3
in the air surrounding the woodland vegetation
with increasing distance from the farm. They found, that the farm as a source of NH
3

emission caused a significant increase in atmospheric NH
3
concentrations in the immediate
vicinity in which the total effect was strongly influenced by the prevailing wind direction.
Von Bobrutzki et al. (2011b) also found rapidly declined atmospheric NH
3
concentrations
within ~400 m distance from the farm, from ~12 g NH
3
m
-3
in 45 m to < 3 g NH
3
m
-3
in
415 m. They concluded that the air pollution was restricted to a small local area by the tree
belt and the adjacent woodland. This reduced the long-range transport of NH
3
. Von Bobrutzki
et al. (2011b) also point out, that it could not be proved that the observed rapid reduction of
NH
3
concentrations is due to the effect of the adjacent woodland.

In future projects dispersion studies of the broiler farm are planned in the wind tunnel in
order to investigate the remaining questions.

2.2 Ventilation rate measurements in natural ventilated livestock buildings
In order to estimate the ventilation rate of a naturally ventilated livestock building several
field studies were performed (Fiedler and Mller, 2011; Samer et al. 2011a, b) in the same
cowshed. The cowshed is located near a small town in the north-east of Germany
(Mecklenburg-Vorpommern). The surrounding area is rural in character and interspersed with
small villages.
The building is 96 m long, 34 m wide and the height varies from the side wall H = 4.2 m to
the gable top of H~10 m (see figure 2). The room volume amounts to 25,499 m. The
cowshed houses ca. 364 dairy cows that are kept in a loose housing system with lying boxes.
The liquid manure system is equipped with a winch drawn dung channel cleaner. The natural
ventilation uses the air draft by a permanent open ridge slot, openings in the gable walls
(space boards) and adjustable openings in the sidewalls (see figure 2). During summer
season, three additional ceiling fans are used to mechanically enhance the natural ventilation.
The fans are mounted at the ceiling along the building centreline.

Figure 2: Cowshed with open sidewalls and open doors during summer time.

120

Figure 2 implies that the investigated livestock building has no clearly defined air outlets. All
existing openings in the building are also very large, which made measurements of the
ventilation rate directly at the openings impossible. Instead, the ventilation rate was
determined indirectly by performing tracer gas experiments with radioactive Kr-85 (Mller
and Mller, 1998). The air exchange rate was estimated by determining the resulting decay
curve of all radiation counters (Samer et al., 2011b). However, the determinations of the air
exchange rate and the air volume stream are associated with uncertainties. Firstly, the mixing
of the tracer gas within the building is non-uniform. Secondly, the ambient wind conditions
have a large influence because of the large openings of the building. Therefore the results of
the tracer gas measurements have only snap shot character for the underlying conditions and
many experiments are needed to get statistically sound representative results (Fiedler and
Mller, 2011).
Because of the uncertainties of the tracer gas method the air volume flow was additionally
estimated by the CO
2
balance method in order to get more information on the air volume
flow. The CO
2
balance method is based on the assumption that the heat and CO
2
produced by
the animals in the livestock building can be calculated according to heat and CO
2
production
models (e.g. DIN EN 18910-1; Samer et al. 2011b). Thus, the estimated air volume flow
depends on the accuracy of these models, additionally to other factors (e.g. ambient
conditions). On the other hand, the CO
2
balance method is most reliable for a continuously
estimation of the air ventilation rate during the whole measurement period.
Table 1 shows the results of estimated air exchange rates and ventilation rates determined by
both methods at the same time and cowshed. The data was obtained during several
measurement campaigns with various ambient conditions. Table 1 shows the high variation
of the estimated air exchange rates and the ventilation rates for the cowshed and is
emphasising the snap shot character of the field measurements. For this reason, systematic
ventilation experiments are planned in the new wind tunnel facility in order to get more
information about the ventilation rate of naturally ventilated livestock buildings.
Table 1: Estimated air exchange rates (AER) and ventilation rates (Q) subject to Kr-85 tracer gas experiments
and CO
2
balance method according to Samer et al. (2011a).
Experiment AER Kr-85 Q Kr-85 (10
5
) AER CO
2
Q CO
2
(10
5
)
Nr. [h
-1
] [m h
-1
] [h
-1
] [m h
-1
]
1 99 25 56 14
2 43 11 28 7
3 34 9 17 4
4 120 31 51 13
5 98 25 24 6
6 70 18 34 9
7 89 23 47 12
8 50 13 25 7
9 35 9 18 5
10 21 5 19 5
3 CONCEPT OF THE NEW WIND TUNNEL FACILITY
On the basis of the two aforementioned examples, two main application areas for the new
wind tunnel are be identified: (1) dispersion studies, and (2) ventilation experiments. These
121

two topics lead to different requirements for the experimental set-up. For dispersion studies
small scale models (1:300 or smaller) are preferred in order to get a large model area in the
wind tunnel. The wind speed for dispersion studies is generally moderate (around 10 m/s).
For ventilation experiments large-scale models are preferable (1:150 or larger) to achieve a
best available monitoring. Ventilation experiments require higher wind speeds to avoid e.g. a
re- laminarization of the flow inside the buildings. These considerations were the basic
concept for planning the test section new wind tunnel.
3.1 Technical data of the new tunnel
A conceptual sketch of the new tunnel is given in figure 3.
The flow establishment and test section together will be 20 m long and 3 m wide and 2 m
high. In order to achieve a constant static pressure gradient within the test section, the tunnel
will be equipped with an adjustable double ceiling which should achieve heights in a range
from 2 to 2.5 m. In addition the test section will be equipped with a 3D probe positioning
system. The ground floor of the tunnel is covered with special model-boards which also host
a turntable.
The axial fan has a diameter of 2.8 m and the driving power is chosen such that wind speed of
the approach flow of 20 m/s could be reached.
The inlet nozzle will be 6 m wide and 4 m high and lead to a contraction ratio of 4. The
nozzle will be equipped with honeycombs in order to achieve a uniform inflow.

Figure 3: Sketch of the new wind tunnel facility at the Leibniz Institute of Agricultural Engineering Potsdam-
Bornim.
Table 5 e Air exchange rates, ventilation rates and gaseous concentrations.
With the dimensions of the test section it is expected that for dispersion studies with 1:300
and smaller a boundary layer of high quality could be modelled. For ventilation experiments
with model scales 1:100 and larger the modelled boundary layers are expected not to have the
full similarity to natural flows because large eddy structures are limited by the chosen cross
section. However, ventilation experiments should be possible with smaller scale models.
Wind speeds of the approach flow up to 20 m/s should ensure sufficient wind speeds of the
flow within building structures in order to avoid thick laminar wall boundary layers inside the
building structures.
4 ACKNOWLEDGEMENTS
122

The authors gratefully acknowledge funding of the European Regional Developing Fund
which allows the invention of the new wind tunnel facility.
The authors would also thank the members of the technical staff, mainly Knut Schrter and
Andreas Reinhardt, who exerts efforts for building up the wind tunnel. Also, Prof. Dr. Bernd
Leitl for advice and planning of the new wind tunnel facility.
5 REFERENCES
DIN 18910-1 2004. Thermal insulation for closed livestock buildings- Thermal insulation and ventilation forced
part1: Principles for planning and design for closed ventilated livestock buildings. Normenausschuss
Bauweses im DIN, Beuth Verlag, Berlin
Fiedler, M., Mller, H.-J. 2011. Emissions of ammonia and methane from a livestock building natural cross
ventilation. Meteorologische Zeitschrift, 20(1), 59-65.
Groot Koerkamp, P.W.G., Metz, J.H.M., Uenk, G.H., Phillips, V.R., Holden, M.R., Sneath, R.W., Short, J.L.,
White, R.P., Hartung, J., Seedorf, J., Schrder, M., Linkert, K.H., Pedersen, S., Takai, H., Johnsen, J.O.,
Wathes, C.M., 1998 concentrations and emissions of ammonia in livestock buildings in northern Europe,
Journal of Agricultural Engineering Reseach, 70, 79-95.
Mller, H. J., & Mller, B. 1998. The further development of airexchange measurement techniques using tracer
gases. Landtechnik, 53(5), 326-327.
Samer, M., Berg, W., Mller, H.-J., Fiedler, M., Ammon, c., Sanftleben, P., Brunsch, R., 2011a. Radioactive
85
Kr and CO
2
balance for ventilation rate measurements and gaseous emissions quantification through
naturally ventilated barns. Transactions of the ASABE, 54 (3), 1137-1148.
Samer, M., Mller, H.-J., Fiedler, M., Ammon, C., Glser, M., Berg, W., Sanftleben, P., Brunsch, R. 2011b.
Developing the
85
Kr tracer gas technique for air exchange rate measurements in naturally ventilated animal
buildings. Biosystems Engineering, 106, 276-287.
Von Bobrutzki, K., Mller, H.-J., Scherer, D 2011a. Factors affecting the ammonia content in the air
surrounding a broiler farm. Biosystems Engineering, 108, 322-333.
Von Bobrutzki, K., Ammon, C., Berg, W., Einert, P., Fiedler, M., Mller, H.-J., Scherer, D. Strohbach, B.
2011b. Ammonia emissions from a broiler farm: spatial variability of airborne concentrations in the vicinity
and impact on adjacent woodland. Environmental Monitoring and Assessment, published online first
TM
.

123
Wind tunnel study of concentration fluctuations in two
merging plumes in different geometrical configurations
D. Contini
a
, C. Elefante
a
, F. M. Grasso
a
, A. G. Robins
b

a
Istituto di Scienze dellAtmosfera e del Clima, ISAC-CNR, Lecce, Italy,
d.contini@isac.cnr.it
b
Enflo, FEPS, University of Surrey, Guildford GU2 7XH, Guildford, UK,
a.robins@surrey.ac.uk
ABSTRACT: Understanding the statistical properties of the fluctuating concentration field
within a plume is a matter of considerable interest in environmental analysis of dispersion of
toxic or inflammable releases as well as of odorous gases. Several studies have been devoted
to analyze concentration fluctuations in single plumes released in the surface layer or at
ground level. In this work an analysis of variance, skewness, kurtsosis, intermittency,
probability density functions and power spectra of the concentration field during the mixing
of two identical plumes is reported. Results show that the normalised variance, skewness and
kurtosis, on the plume centreline, increase with the distance X from the stack for the single
plume over the range studied. However, for the two-plume configurations, a minimum was
found at low X (depending on the angle ); further, skewness and kurtosis in the mixing
plumes were generally smaller, at large distances form the sources, than those observed in the
single plume. An asymmetry was observed between the upper part of the plume and the lower
part, with the upper part having larger intermittency and a smaller occurrence of high
concentration levels. The power spectra show the asymmetry with the lower part of the plume
having larger contribution at high frequencies (i.e. for f= n Z / U > 10) and smaller
contribution at low frequencies (f < 1) with respect to the upper part of the plume. The
asymmetry is present in both single and two-plume configurations, but tends to disappear at
large distances downwind of the stacks. Further, for two-plume configurations, there are
substantial changes in the shape of the concentration power spectra in the intermediate
frequency range (1 < f < 10), at least in the near field (X 30 D) with respect to the shape of
the spectra for single plume.
1 INTRODUCTION
The response of biological systems to toxic gases and vapors and to odors is often nonlinear,
making them susceptible to concentration fluctuations. Short-term concentration fluctuations,
caused by turbulent mixing processes, can be important in estimating the hazard caused by
industrial releases and accidental emissions as well as for understanding the behavior of
reacting plumes. The fluctuating concentration field has been analyzed in full scale
experiments (Yee et al., 1994; Klein and Young, 2011; Klein et al., 2011), using small scale
models in wind tunnel (Fackrell and Robins, 1982; Wilson et al., 1982a) or in water channel
(Bara et al., 1992; Huq and Stewart, 1997), also developing simple models used to evaluate
the intensity of fluctuations and the overcome of specific concentration thresholds (Wilson et
al., 1982b; Hanna, 1984; Yee and Chan, 1997; Mortarini et al., 2009) and using
computational simulations (Mavroidis et al., 2007). Remote-sensing with lidar has been used
to characterize fluctuations of full scale plumes (Levellen and Sykes, 1986; Jorgensen et al.,
124
2010). In general terms, laboratory small scale models suffer from the disadvantage that large
scale fluctuations in wind directions, influencing the meandering of plumes, can not be
adequately simulated so that measurements are representative of relatively short time average
in full scale (from a few minutes to about half an hour). Previous research works have been
performed on single plume for both elevated and ground level plumes. Therefore, in this
work the single plume configuration is discussed as a prerequisite and the goal of the analysis
is to study the changes in the fluctuating concentration field induced by the mixing process of
two identical plumes for several geometrical conditions. This is accomplished by comparing
the concentration probability density functions, power spectra and high order concentration
moments of two-stack configurations with those of the single plume configuration. The two-
stack geometry is often present in industrial sites and in power plants and their interaction
should be taken into account if a correct simulation of dispersion is needed. The mixing of
these plumes generates strong interactions, especially under favorable wind directions that
create distortions of the plume cross-sectional shape and extra-rise or downwash effect
with respect to the single source cases with zero interaction (Contini and Robins, 2004).
2 EXPERIMENTAL METHOD AND EQUIPMENT
The experiments were carried out at the EnFlo (University of Surrey) environmental wind
tunnel facility (20 m long working section, 3.5 m wide and 1.5 m tall). The tunnel was
operated in neutral conditions with a free-stream velocity U
ref
=2 m/s, maintained constant
through a software-controlled feedback system based on a two-dimensional ultrasonic
anemometer. The 1 m deep boundary-layer was artificially generated using five 1.2 m tall
Irwin spires located at 1 m from the tunnel inlet (Irwin, 1981), a 190 mm tall barrier placed
0.77 m from the inlet and a staggered distribution of roughness elements (20 mm tall and 80
mm wide) covering all the floor of the tunnel with a lateral spacing of 240 mm. Boundary-
layer properties were measured with a two-component LDA system, based on a DANTEC
BSA 57N11 burst analyzer after seeding the flow with a solution of sugar and water using a
commercial haze generator placed just outside the tunnel inlet. The LDA was operated at an
average acquisition frequency of 100 Hz. Velocity and turbulence measurements were taken
on the tunnel centreline at three positions, X, downwind of the sources (i.e. X=0, X=0.5 m
and X=1.1 m). In Figure 1a the flow velocity profile is shown, this being the spatial average
of the three positions. Similar averages were performed for
u
,
w
and <uw> and the
resulting profiles are plotted in Figure 1b. The mean velocity profile was characterised by a
logarithmic form, with u*/U
ref
=0.056 and z
o
=1.6 mm, up to 0.5 m, which is also the height
range in which the Reynolds stress <uw> was approximately constant. This logarithmic
profile is shown in Figure 1a together with a power-law profile with an exponent of 0.23,
using the boundary layer height (1m) as reference height. A photograph of the wind tunnel
set-up is reported in Figure 2a and other details were reported in Contini et al. (2006).
The sources (a single stack or two identical stacks) were placed 14 m downwind of
the tunnel inlet on a PC-controlled turntable. The emission was a three gas mixture: helium,
air and propane mixed in appropriate proportions. The propane trace gas concentration was
changed from case to case as required to optimize instrument response; air and helium were
mixed in order to provide a buoyant plume with a density difference relative to the
environmental air density
a
equal to /
a
=0.79. The two-stack model, both 300 mm tall
with internal diameter D=5 mm and separation d=70 mm, is shown schematically in Figure
2b together with the definition of the reference system used to present and discuss the results.
The flow-rate, Q, from the stacks was kept constant at 10 l/min through each stack. The flow
125
through the two stacks was balanced using mass-flow regulators and two rotameters. Point
concentration measurements were carried out by using two fast FID (HRF 400 Cambustion
Ltd) placed on the traversing system of the wind tunnel. These were equipped with a tube 250
mm long and 1.2 mm in internal diameter for calibration purposes; a second smaller was the
actual sampling tube (internal diameter 0.305 mm, estimated flow-rate 1.24 mg/s). The FFIDs
sampled at the same position Z (above the wind tunnel floor) and at the same distance x
(downwind of the source) and were 50 mm apart in the Y direction (cross-wind). The FFIDs
response time measured during laboratory tests as response to a large step concentration
change was about 5 ms (between 5% and 95% of a step change in concentration).

10
100
1000
0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
U / Uref
z

(
m
m
)
Measurements
Log profile
Exponential law profile
(a)
0
200
400
600
800
1000
1200
-0.02 0 0.02 0.04 0.06 0.08 0.1 0.12
U
/ Uref, W
/ Uref, -<uw>/U
2
ref
z

(
m
m
)
Horizontal
Vertical
Stress
(b)

Figure 1: Characteristics of the neutral boundary-layer in the wind tunne. (a) Vertical profile of average wind
velocity and (b) vertical profiles of
u
,
w
and <uw>.

Figure 2: (a) Photograph of the wind tunnel set-up and (b) scheme of the two-stack configurations with
explanation of symbols.
Concentration maps at different distances downwind of the sources were measured on a
regular grid with a number of points variable between 130 and 340. At each position the
FFIDs outputs were recorded at a sampling frequency of 500 Hz for about 3.5 minutes. The
two FFIDs were frequently calibrated (every 1-2 hours of continuous operation) and during
calibration the background concentration in the wind tunnel was also measured and
subtracted from successive measurements. Calibrations were made with two or three points:
clean air and one or two span gases. The stability and repeatability of the FFID calibrations
were good throughout the measurements, with output voltage changes between one

y
D=5 mm
(a)
Origin of the
reference system
Up-wind stack
Down-wind stack
Wind

x
Separation d=70 mm
(b)
(a)
126
calibration and the successive usually less than 0.5% and with changes through the day
usually less than 1%.
To evaluate the stationarity of the results during the averaging times, the measured
time-series have been divided in 4 segments of equal length and the different statistics (mean,
standard deviation, skewness and kurtosis) were evaluated in each segment. The relative
difference between the calculations in each segment and those of the entire data series is an
indicator of the stationarity of results. In Figure 3 an example of variability in the different
segments is reported for single plume and for double plume. In general terms the stationarity
of results is lower at the edge of the plume and, in some cases, relative differences up to 80-
100% have been observed.

-0.08
-0.06
-0.04
-0.02
0
0.02
0.04
0.06
0.08
1 2 3 4
Segment number
R
e
l
a
t
i
v
e

d
i
f
f
e
r
e
n
c
e
Average
Variance
Skewness
Kurtosis
-0.2
-0.15
-0.1
-0.05
0
0.05
0.1
0.15
0.2
1 2 3 4
Segment number
R
e
l
a
t
i
v
e

d
i
f
f
e
r
e
n
c
e
Average
Variance
Skewness
Kurtosis

Figure 3: Evaluation of stationarity of statistical parameters. Relative variability for the first four concentration
moments for a single plume in the centreline at X=60D (left) and for two-stack configuration with =45
measured at X= 120D, Y=0 at the lower side of the plume.
A systematic check of the quality of the point concentration measurements has been
performed through repeatability analysis (Contini et al., 2006). Generally the repeatability is
good in all the cases analysed and variability is usually within 10% for single point average
concentration measurements and for measurements of concentration variances.
3 INTERMITTENCY, PROBABILITY DENSITY FUNCTIONS AND POWER
SPECTRA
The intermittency, I, of concentration was evaluated as I=1-, where is the fraction of
measurement time non-zero concentration. This actually means the fraction of time in
which the measured concentration is above a certain (small) threshold. In this work the
threshold for calculation of has been chosen as max(1 ppb; <c>) with =0.05, being c and
<c>, respectively, the instantaneous and the average concentrations. This because our
analysis shows that a 1 ppb concentration is generally not distinguishable from background.
The absolute value of I depends on source size, and the effect of size is larger for elevated
sources with respect to ground level sources, but the overall shape of the curves at different
positions is similar for different source sizes (Fackrell and Robins, 1981). Because
concentration is zero outside the plume, knowing I, it is possible to derive statistical
conditioned parameters (in-plume). However this has not been done in this work because the
goal is to work with complete statistics. There is not much to gain in using conditioned
127
statistics because what it could be really useful is to measure concentrations at a fixed
position with respect to the plume centre rather than conditioned statistics.
In Figure 4a it is reported an evaluation of I as function of and in Figure 4b it is
reported the intermittency evaluated for single plume as function of X for three different
positions: the centerline, the upper side of the plume and the lower side of the plume (in the
figure, upper and lower side of the plumes are points in the middle of the plume Y=0 with
concentration roughly equal to 15% of the centerline concentrations). Error bars have been
obtained as one standard deviation of several repeated experiments in nominally identical
conditions.
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1
Fraction
I
n
t
e
r
m
i
t
t
e
n
c
y
,

I
Single plume
Two-plume 45
0
0.2
0.4
0.6
0.8
1
0 50 100 150 200 250
X / D
I
n
t
e
r
m
i
t
t
e
n
c
y
,

I
Centreline
Upper side
Lower side

Figure 4: (left) Intermittency as function of for the same cases of Fig. 3. (right) Intermittency, as function of
X, for single plume calculated in the middle of the plume (Y=0) and different Z: centreline, upper side and
lower side of the plume.
In Figure 5 the vertical profiles of intermittency (at Y=0) are reported for the different
configurations evaluated at X=20D and X=120D. Results show that the vertical asymmetry is
present in both single and two-stack configurations. Further, at low values of X the effect of
the mixing is to reduce the intermittency for cases in which the two stacks are almost aligned
with the crossflow (low values of ) and to increase the intermittency for 45. The
dependence from the angle happens because for small values of the lower plume that is
smaller and more concentrated is guided in the middle of the largest upper plume by the
counter-rotating vortex pair (CVP) instead, for large values of , the mixing is less efficient
because the two CVPs can enter in contact having opposite vorticity. At large values of X the
intermittency of two-stack configurations is smaller than the corresponding single stack cases
for all values of . Further, the differences between intermittency at the centreline of the
plume and at the edge decreases downwind of the source for both single and two-stack
plumes.
The probability density function of instantaneous concentrations c has been evaluated,
as in Fackrell and Robins (1982), as function of
c
c c
> <
= . Results indicate a significant
difference in the shape of the pdf with higher being more frequent at the lower side of the
plume. The power spectra Sc of concentration fluctuations are reported in the normalised
form:
2
c
Sn
f Sc
= , where Sn is the spectrum in standard units and f is the normalised

frequency,
U
nZ
f = (being n the natural frequency and U the wind velocity). In Figure 6 the
128
pdf and the power spectra, evaluated at X=20D and Y=0, for different Z: plume centreline,
upper and lower part of the plume are reported. In the Figure it is also included the
3 / 2
f

line
indicating the presence of an inertial subrange with a limited extension with respect to typical
wind velocity power spectra (Fackrell and Robins, 1982; Hanna, 1986). The strong peaks at
c=0 in the probability density functions are not an artefact due to instrumental noise rather
they are a genuine behaviour that reflects the intermittency of the concentration field.
In the upper part of the plume the high concentration fluctuations are less frequent
with respect to the lower part of the plume. Power spectra clearly show that high frequency
contributions is reduced in the upper part of the plume with respect to the centreline and the
lower part of the plume. Instead, the low frequency contribution is larger. This means that the
characteristic time scales of concentration fluctuations are larger in the lower part of the
plume (where it is located the stable boundary of the plume) and are larger in the upper part
of the plume (where there is the unstable boundary of the plume).

0
5
10
15
20
25
30
0 0.2 0.4 0.6 0.8 1
Intermittency, I
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=20 D
-20
-10
0
10
20
30
40
50
60
0 0.2 0.4 0.6 0.8 1
Intermittency, I
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=120 D

Figure 5: Vertical profiles of intermittency at two different X in the middle of the plume (Y=0). H is the heigth
of the stacks.
In Figure 7 the evolution with X of the pdf and power spectra along the centreline has been
reported for single plume configuration. It appears that there is an evolution of the pdf along
the centreline that, at large X, becomes almost exponential-like (excluding the step at c=0)
similarly to the simple model in Csanady (1967) and to the results in Fackrell and Robins
(1982). This is accompanied by an evolution of the power spectra (Fig. 7) that present a
decrease of high frequency contribution with X and an increase of the intermediate frequency
(around f=1). This evolution can be interpreted considering that at short distances from the
source the plume, internal turbulence and meandering are dominant over the external
crossflow turbulence. However, when the distance increases and the plume is diluted with
environmental air the importance of internal turbulence is reduced and concentration
fluctuations are driven by external turbulence in the crossflow.
The evolution of the pdf (not shown) and of the power spectra in the two-stack
configurations is similar to that of single plume (Fig. 8) but slower because the effect of the
mixing process are still evident a 120 diameters downwind of the sources. Further, there is a
clear dependency on the angle . At low values of it is possible to have two distinct peaks
in the power spectra at very different frequencies one at low frequency and another at high
frequency likely associated to the different dynamics of the two mixing plumes having
significantly different dilutions.
129
-2 0 2 4 6 8 10
10
-4
10
-3
10
-2
10
-1
10
0
10
1
Normalised concentration
P
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n
X=20 D
Centreline
Upper-plume
Lower-plume
10
-2
10
-1
10
0
10
1
10
2
10
-3
10
-2
10
-1
Normalised frequency
N
o
r
m
a
l
i
s
e
d

s
p
e
c
t
r
u
m
X=20 D
Centreline
Upper-plume
Lower-plume
f
-2/3

Figure 6: Pdf and power spectra for single plume at X=20D evaluated on the centreline, on the upper side and
on the lower side of the plume at Y=0.
-2 0 2 4 6 8 10
10
-4
10
-3
10
-2
10
-1
10
0
10
1
Normalised concentration
P
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n
Single plume X=20D
X=60D
X=120D
X=240D
10
-2
10
-1
10
0
10
1
10
2
10
-4
10
-3
10
-2
10
-1
N
o
r
m
a
l
i
s
e
d

s
p
e
c
t
r
u
m
Single plume
X=20D
X=60D
X=120D
X=240D
f
-2/3

Figure 7: (left) Pdf for single plume at the centreline as function of X. (right) spectra of concentration
fluctuations for single plume along the centreline.
10
-2
10
-1
10
0
10
1
10
2
10
-3
10
-2
10
-1
N
o
r
m
a
l
i
s
e
d

s
p
e
c
t
r
u
m
Two-stack 0
X=20D
X=30D
X=60D
X=120D
f
-2/3
10
-2
10
-1
10
0
10
1
10
2
10
-3
10
-2
10
-1
N
o
r
m
a
l
i
s
e
d

s
p
e
c
t
r
u
m
Two-stack 90
X=20D
X=30D
X=60D
X=120D
f
-2/3

Figure 8: (left) Concentration spectra along the centreline for the two-stack configuration with =0 and (right)
the same analysis for the configuration with =90.
130
4 VARIANCE, SKEWNESS AND KURTOSIS OF THE CONCENTRATION FIELDS
The variance is studied in the normalised form
2
2
c
c
V
> <
= , the skewness is calculated as

( )
3
c
3
c c
S
> > < <

= and the kurtosis as
( )
4
c
4
c c
K
> > < <

= . The vertical profiles of variance, for
the different geometrical configurations investigated, is reported in Figure 9 at Y=0 for two
distances from the sources: 20D and 120D. Results clearly shows that the effect of the mixing
process is still visible at 120D downwind of the sources with a decrease of the relative
fluctuations and the magnitude of the decrease is depending on the angle . At short distances
from the source there is a decrease of relative variance for low values of the angle and an
increase for 45 with respect to single plume. At large distances from the stack the general
decrease of V for two-stack configurations is particularly evident in the lower part of the
plume. This increases the vertical asymmetry of the plume and the recovery towards a
symmetrical condition is slower than for single plume. The vertical profiles of skewness, for
the same cases of Figure 9, are reported in Figure 10 and the vertical profiles of kurtsosis are
reported in Figure 11. At short distances from the sources also the normalised skewness and
the normalised kurtosis present a decrease for low values of and an increase for large values
of with respect to the single stack case. Instead, at large X both S and K are smaller for
two-stack configurations with respect to single plume. However the reduction is not uniform
at different heights. For angles lower than 15 the values of S and K are similar to single
stack case at the lower part of the plume and the reduction is present on the plume centreline
and at the upper part of the plume. Instead, for 30 the reduction of S and K is particularly
evident in the lower part of the plume. This means that the mixing process influence the
profiles of high order moment of concentrations in a way that is dependent on the geometry
of the sources and the memory of this geometry is still evident at relatively high distances
downwind (i.e. at X=120D).

0
5
10
15
20
25
30
0.1 1 10 100 1000
Normalised variance, V
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=20 D
-20
-10
0
10
20
30
40
50
60
1 10 100
Normalised Variance, V
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=120 D

Figure 9: Vertical profiles of normalised variances, V, at Y=0 for the different configurations of sources at
X=20D (left) and X=120D (right).

Results also show that the normalised variance, skewness and kurtosis, on the plume
centreline, increase with the distance X from the stack for the single plume over the range
studied. However, for the two-plume configurations, a minimum was found between X=30D
and X=60D, depending on the angle .

131
0
5
10
15
20
25
30
-2 0 2 4 6 8 10 12 14 16 18 20
Normalised Skewness, S
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
x=20 D
-10
0
10
20
30
40
50
60
0 2 4 6 8 10 12 14 16
Normalised Skewness, S
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=120 D

Figure 10: Vertical profiles of normalised skewness, S, at Y=0 for the different configurations of sources at

0
5
10
15
20
25
30
0 5 10 15 20 25 30 35 40 45 50
Normalised Kurtosis, K
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=20 D
-10
0
10
20
30
40
50
60
0 5 10 15 20 25 30 35 40 45 50
Normalised Kurtosis, K
(
Z

-

H
)

/

D
Single Two 0
Two 15 Two 30
Two 45 Two 60
Two 90
X=120 D

Figure 11: Vertical profiles of normalised kurtosis, K, at Y=0 for the different configurations of sources at
5 CONCLUSIONS
Wind tunnel experiments have been conducted to analyse concentration fluctuations during
the mixing of two plumes released into a crossflow by two identical stacks. An asymmetry
between the upper part of the plume was observed, for both single stack and two-stack
plumes, with the upper part, having larger intermittency and increased occurrence of small
concentration fluctuations. The power spectra show that the lower part of the plume presents
larger contributions at high frequencies (i.e. f= n Z / U > 10) and smaller contributions at low
frequencies (f < 1) with respect to the upper part of the plume. The asymmetry tends to
disappear at large distances downwind of the stacks, however, traces are still visible in the
probability density functions and in the intermittencies at x=240 D. For two-plume
configurations the asymmetry is dependent on the angle and, for geometrical configuration
at low values of ( < 45), it is visible at larger X with respect to the single plume meaning
that the evolution towards a symmetric condition is slower. The probability density functions
of concentration fluctuations on the plume centreline evolve towards an exponential-like
shape for both single stack and two-stack plumes. Power spectra of concentration fluctuations
have two distinct peaks at different frequencies when evaluated near the sources and for small
values of the alignment angle with a shape depending on . At large distances from the
sources the power spectra for two-stack plumes evolve, with X, in a similar way to those of
132
single stack presenting a decrease of high frequency contributions and an increase of the
intermediate frequency (around f=1). Results show that the normalised variance, skewness
and kurtosis, on the plume centreline, increase with the distance X from the stack for the
single plume over the range studied. However, for the two-plume configurations, a minimum
was found between X=30D and X=60 D, depending on the angle ; further, skewness and
kurtosis in the mixing plumes were generally smaller than those observed in the single plume.
6 AKNOWLEDGEMENTS
The authors wish to thank Dr Paul Hayden and Mr. Tom Lawton at EnFlo, University of
Surrey, for their help in setting up the experiments and in developing software.
7 REFERENCES
Bara B.M., Wilson, D.J., Zelt, B.W., 1992. Concentration fluctuation profiles from a water channel simulation
of a ground level release. Atmospheric Environment, 26A, 1053-1062.
Csanady, G.T., 1967. Concentration fluctuations in turbulent diffusion. Journal of the Atmospheric Sciences, 11,
21-28.
Contini, D., Robins, A.G., 2004. Experiments on the rise and mixing in neutral crossflow of plumes from two
identical sources for different wind directions. Atmospheric Environment, 38, 3573-3583.
Contini, D., Hayden, P., Robins, A.G., 2006. Concentration field and turbulent fluxes during the mixing of two
buoyant plumes. Atmospheric Environment, 40, 7842-7857.
Fackrell J.E., Robins A.G., 1981. The effects of source size on concentration fluctuations in plumes. Boundary-
Layer Meteorology, 22, 335-350.
Fackrell J.E., Robins A.G., 1982. Concentration fluctuations and fluxes in plumes from point sources in a
turbulent boundary layer. Journal of Fluid Mechanics, 117, 1-26.
Hanna, S.R., 1984. Concentration fluctuations in a smoke plume. Atmospheric Environment 18, 1091-1106.
Hanna, S.R., 1986. Spectra of concentration fluctuations: the two time scales of a meandering plume.
Atmospheric Environment, 20, 1131-1137.
Huq, P., Stewart, E.J., 1997. Measurements of density fluctuations in steady, buoyant plumes in crossflow.
Klein, P.M., Young, D.T., 2011. Concentration fluctuations in a downtown urban area. Part I: analysis of Joint
Urban 2003 full-scale fast-response measurements. Environmental Fluid Mechanics, 11, 23-42.
Klein, P., Leitl, B., Chatzmann, M., 2011. Concentration fluctuations in a downtown urban area. Part II: analysis
of Joint Urban 2003 wind-tunnel measurements. Environmental Fluid Mechanics, 11, 43-60
Jorgensen, H.E., Mikkelsen, T., Pecseli, H.L., 2010. Concentration fluctuations in smoke plumes released near
the ground. Boundary-Layer meteorology, 137, 345-372.
Levellen, W.S., Sykes, R.I., 1986. Analysis of concentration fluctuations from Lidar observations of
atmospheric plumes. Journal of Climate and Applied Meteorology, 25, 1145-1154.
Mavroidis, I., Andronopoulos, S., Bartzis, J.G., Griffiths, R.H., 2007. Atmospheric dispersion in the presence of
a three-dimensional cubical obstacle: modelling of mean concentration and concentration fluctuations.
Mortarini, L., Franzese, P., Ferrero, E., 2009. A fluctuating plume model for concentration fluctuations in a
plant canopy. Atmospheric Environment, 43, 921-927.
Yee, E., Chan, R., Kosteniuk, P.R., Chandler, G.M., Biltoft, C.A., Bowers, J.F., 1994. Experimental
measurements of concentration fluctuations and scales in a dispersing plume in the atmospheric Surface
Layer obtained using a very fast response concentration detector. Journal of Applied Meteorology, 33, 996-
1016.
Yee, E., Chan, R., 1997. A simple model for the probability density function of concentration fluctuations in
atmospheric plumes. Atmospheric Environment, 31, 991-1002.
Wilson, D.J., Fackerell, J.E., Robins, A.G., 1982a. Concentration fluctuations in an elevated plume: a diffusion-
dissipation approximation. Atmospheric Environment, 16, 2581-2589.
Wilson, D.J., Fackerell, J.E., Robins, A.G., 1982b. Predicting the spatial distribution of concentration
fluctuations from a ground level source. Atmospheric Environment, 16, 497-504.
133
Influence of urban roughness on mean and turbulent
wind fields in the city of Hamburg
C. Peeck
a
, B. Leitl
a
, M. Schatzmann
a

a
Meteorological Institute, KlimaCampus, Hamburg University, Germany
christine.peeck@zmaw.de, bernd.leitl@zmaw.de
ABSTRACT: The analysis of flow characteristics in the near-ground region of the atmos-
pheric boundary layer is one of the focal points in micro-meteorological research. Particularly
the highly unsteady flow fields within densely built-up cities are of great interest with regard
to wind comfort assessment or the derivation of pollutant concentration fields. The charac-
terization of mean and time-dependent flow fields within and above the city centre of Ham-
burg, Germany, is the focus of the present study. Flow measurements of high spatio-temporal
resolution were conducted in a 1:350 scale model of the Hamburg harbor area and parts of the
inner city centre that was built up in the boundary-layer wind tunnel WOTAN of the Uni-
versity of Hamburg. The neutrally stratified surface-layer approach flow was modeled on the
basis of long-term statistics derived from a meteorological measurement mast which is sited
in proximity to the city.
1 INTRODUCTION
The number of inhabitants in cities has increased continuously over the last decades. As a re-
sult, dealing with pollution and environmental problems is getting more complicated. Fur-
thermore, the increased number of inhabitants is a challenge for a possible evacuation of an
urban area. A reliable tool for the practical management in cases of incidents is required. The
direction of propagation and the travel time of pollutants is influenced by the local wind
fields, which are highly influenced by the near by buildings. Understanding the mean and
turbulent wind patterns is substantial for the successful reaction of the helping forces. Meas-
urements of flow conditions and dispersion processes in the wind-tunnel model of Hamburg
are used for the validation of an emergency response tool (see also Harms et al., 2011,
Hertwig et al., 2011 and Boris et al., 2002).
The measurement plan was designed to document the development of the boundary layer
across the city, to characterize the influence of heterogeneous roughness pattern on local flow
features, as well as to study unsteady turbulent wind fields. Parts of the results will be dis-
cussed in the following paragraph. The measured vertical wind profiles, for example, show
that the influence of the local conditions can last up to 3 times the mean building height.
Fields of spatially high resolved measurements document bimodal distributions of fluctua-
tions in the stream-wise and span wise velocity component together with characteristic devia-
tions in wind speed and turbulence intensity at roof level. The visualization of wind fields
around characteristic urban building structures show increasing wind speeds due to channel-
ing effects as well as circulating regimes. The documentation of the locally influenced wind
fields will be the focus of the analysis.

134
2 EXPERIMENTAL SETUP AND BOUNDARY LAYER MODELING
For the model experiments in the boundary-layer wind tunnel a 4 m wide and 10.5 m long
model area of the inner city centre and the port of Hamburg was installed. Figure 1 shows the
section that was chosen for the model area. The input conditions for the model area were de-
rived out of velocity time series from 2007 to 2009 captured at the Hamburg weather mast
that is located southeast of the city centre. The laboratory flow measurements were made
with Laser-Doppler Velocimetry; a Prandtl tube was used for recording a reference wind
speed at the inlet of the tunnel.

Figure 1: (a) Partial model of the city of Hamburg in the wind tunnel, (b) aerial view of the model area and (c)
picture of the Hamburg weather mast with its measuring heights.

The building structure in the proximity of the weather mast is in the southwest in its density
and height similar to that of the harbor area, that is located upstream of the city centre for the
inflow direction modeled in the wind tunnel (see Fig. 1a). This is of fundamental importance
since no data from field measurements located within the model area are available. On the
basis of long-term statistics from the weather mast, a southwesterly approach flow direction
was chosen that mirrors a frequent meteorological conditions for the Hamburg area. The
wind-tunnel model thus also includes parts of the harbor of Hamburg as a potential area of
pollutant releases. In order to define the boundary-layer parameters the field data were ana-
lyzed for a 60 wide sector of 235 + / - 30. To ensure that changes in the inflow direction
within the 60 broad sector do not result in significant changes of the boundary-layer parame-
ters, the field data were further divided into 10 wide sectors and the resulting variability of
the wind profile exponents and roughness lengths were examined (see Fig. 2). The study
showed that the distribution of the parameters and Z
0
does not significantly change by vary-
ing the assumed height of the Prandtl layer. Within the 10 wide bins falling within the cru-
cial sector of 235 +/- 30 no jumps occur in the distributions of the two parameters. The sec-
tor 0 + / - 30 is affected by the orientation of the measuring equipment, so that these results
should not be used. For winds from the southeast or southwest the results are quite homoge-
neous. According to the VDI Guideline, the roughness lengths and the wind profile exponents
are all falling in the very rough (urban) roughness regime.
Since the modeling of the atmospheric boundary layer in WOTAN could only be done un-
der neutral stratification conditions, the records of field measurements were filtered for this
stability class. To this end, the time series were examined with regard to the stability parame-
ter (z / L
M
), where L
M
is the Monin-Obukhov length scale. Only those velocity records were
considered for which (z / L
M
) <0.001. Since the height of the Prandtl layer influences the
(a) (b) (c)
135
boundary layer parameters, Z
0
and the wind profile exponent were derived as functions of the
Prandtl layer height.

Figure 2: Wind profile exponent and roughness length Z
0
as a function of flow direction for different Prandtl
layer heights.
3 APPROACH FLOW CONDITIONS
The approach flow conditions were determined by the derived boundary-layer parameters of
the measuring mast. By the use of roughness elements and turbulence generators at the inlet
and within the approach section of the tunnel it was possible to model a very rough classified
surface-layer flow. The similarity of the modeled boundary layer with the specified natural
boundary layer is crucial for the reliability and transferability of the laboratory results. In
order to provide a quality-controlled boundary layer, the characteristics of the inflow
conditions were tested against the VDI Guideline 3783/12. With respect to turbulence
intensities, characteristic turbulent length scales as well as the roughness length and profile
exponent the modeled boundary layer lies within the specified range of a very rough
boundary layer. An independency of the characteristics to changes in the inlet velocity was
successfully tested. Despite the size of the model, it generates no significant obstruction of
the measurement cross-section. At the beginning of the model area, the influence of the
turbulence generators could be seen in the measured velocities. This led to variations in the
turbulence intensities, which were still in the given very rough magnitude. As further quality
checks longitudinal pressure gradients along the test section were eliminated by adjusting the
tunnel ceiling. The approaching flow is horizontally homogeneous. The length of the time
series was determined by convergence analysis to be 170s model scale.

The measurements of the u-and v-component, and the u-and w-component were carried out
on three lateral positions in the tunnel center and at a distance of 0.5 m to the left and right.
Results are shown in Figure 3c. The distance to the vortex generators was relatively short, so
that it was expected that the immediate influence of the generators was recordable. By aver-
aging the total of six individual profiles, a mean wind profile was generated, which matches
qualitatively very well with the natural wind profile, see Figure 3a. In Figure 3b, the func-
tional relationship between and Z
0
is shown. The results are slightly above the empirically
defined curves from Counihan (1975).

136

Figure 3: Mean vertical wind profiles derived from field measurements and from the modelled wind-tunnel
approach flow (a) and wind profile exponents & roughness lengths of the laboratory approach flow for different
assumed heights of the Prandtl layer (b), vertical velocity profiles measured for different lateral positions (c).
4 RESULTS
In Figure 4 the measured positions of the vertical profiles for the boundary layer development
are depicted inside a map of the model area. The measuring positions were chosen to be non-
equidistant in stream-wise direction. In the downtown area more positions were located than
in the harbor area. Areas where densely spaced measurements were conducted were chosen
with regard to their microclimatic features. They differ in the geometry of the adjacent build-
ings and in the orientation to the mean flow direction. The aim of the measurements is to
make out characteristic flow conditions that are directly influenced by the geometry of the
surrounding buildings.
4.1 Boundary Layer Development
The dimensionless height z/h
mean
was separately defined for measurement positions in the
harbor area (profile 1 to 6) and for the inner city centre (profile 7 to 15). The harbor area is
mainly characterized by low and less dense building structures with a mean building height of
21 m. The mean building height for the city centre was defined as 34.3 m

Figure 4: Overview of the measurement positions for measurements of boundary layer development. Image
from Googel Earth.
Figures 5a and 5b show the vertical velocity and Reynolds stress profiles for all measurement
locations indicted in Figure 4. The mean normalized velocity begins to converge from a
height of 2*z/h
mean
. The influence of local roughness structures is no longer recognizable at a
137
height of 3*z/h
mean
. According to Cheng and Castro (2002) the height of the convergence cor-
responds to the upper edge of the roughness layer. In this case, the height of the roughness
layer ranges between 60 and 65 m. The influence of the river Elbe and the downstream inner
city that marks a significant change in the roughness structure is obvious. The velocity pro-
files for measuring positions in the harbor area are clearly separated for heights <2*z/h
mean

from those profiles measured in and above the city part. The integral length scale (not shown)
is limited by the structure of the buildings for heights 1*z/h
mean
. Above the roof level the
length scale increases and is independent of the local roughness structure above a height of
3*z/h
mean
. The vertical momentum transport is done by sweeps and ejections, which intensi-
ties depend on the density and height of the surrounding buildings. Figure 5c shows the de-
viation of the dominance of sweeps over ejections. It is again obvious that the development of
the urban boundary layer along the city is significantly influenced by the river Elbe. The de-
viation dS is split up into two separated height profiles, in the same way as observed before.

Figure 5: (a) Vertical profiles of the mean normalized velocity, (b) normalized Reynoldsstress, and (c)
dominance of sweeps over ejections.
4.2 Spatially high resolved flow measurements
The first measurement field with high spatial resolution is represented by a street canyon with
approximately +30 deviation from the prescribed inflow direction, see Figure 6a. The hori-
zontal measurement plane consists of 87 measuring positions and is located in a measuring
height of 10 mm above ground. This corresponds to a height of 3.5 m full-scale. The second
field is located at an intersection. Figure 4b shows the distribution of measurement positions
inside the measuring fields. The crossing street canyons are rotated by 45 to the mean flow
direction. On the crossing is a source installed for propagation modeling. The intersection is
located in the inner city close to Hamburg Town Hall.

(a) (c) (b)
138

Figure 6: Positions of measurement points and field areas for streetcanyon (a) and intersection of roads (b).
4.2.1 Street canyon flow
In Figure 7a, the mean normalized horizontal wind vectors are shown. The length and color
of the vectors characterize the strength of the velocity. Figures 5b and 5c show the variance
of the u- and v-component of the velocity vector, respectively. Fields with lower normalized
wind speeds, such as field A in which positions 1 to 4 are located (see Fig. 6a); show a large
variance of the u-component, while the variance of the v-component is smaller. Low average
wind speeds were also measured at positions 9 to 11 (box B). Here the variance of the v-
component is greater than that of the u-component. The same appears even further down-
stream at position 12 and 21. At positions 5 to 8 that are located windward, the ratio is re-
versed for the variances. Here the average variance of the u-component is greater than for the
v-component.

Figure 7: (a) Mean normalized horizontal wind vectors; variance of u (b) and v (c) components inside the street
canyon.
The observed accelerated inflow perpendicular to the mean flow direction within the canyon
in field D, leads to a large variance of the v-component in positions 50 (box D), 68 and 69
(box F), and to a large variance of the u- component at position 52. Large values of the vari-
ance indicate a high temporal variability of local wind intensity and gusts at the measuring
point.
4.2.2 Intersection flows
Analogous to the averaged normalized wind speeds shown in section 4.2.1, the wind vectors
from the measurements of the second highly resolved measurement field are presented in
Figure 8a.
(a) (b)
139

Figure 8: (a) Mean normalized wind speed; variance of u (b) and v (c) components, crossroads.
Since the second intensive measurement field is located further inside the city as the street
canyon, the mean normalized wind speeds at the same measuring height are lower on aver-
age. In box A at positions 3 and 4 (see Fig. 6b), compared to the positions 5 to 8, downstream
just behind the building, very low wind speeds were found. At positions that are facing the
wind in the street canyon, fields C and D, a significant acceleration is evident particularly in
the area of the intersection. In Figure 9a-d the distributions of the measured instantaneous
wind directions are plotted. In fields A and B positions 3 and 16 show bimodal distributions.
At position 3 very low wind speeds were measured that show only 20% of the magnitude of
the wind speeds observed at position 16. In field B, however, the positions 13 to 15 show
equal distributions of wind directions, these three positions are located downstream in front
of the buildings. The fields C and D are similar in their distributions of the horizontal wind
direction. The maximum of the distribution in both cases is about -20, while measurements
in field D show higher maxima.

Figure 9:(a) - (d) Frequency distributions of the horizontal wind directions of all 4 fields within the intersection.
Looking at Figure 8b-c, it is striking that at the positions at which bimodal distributions of
wind direction occur large variances of wind speed relative to the mean wind are found. At
position 3, the average fluctuations in the values of the u- and v-component are large. Posi-
tion 16 shows the v-component with a large variance. At position 17 (field C) the average
deviation of the u-component is remarkably large. The distributions of the components of
horizontal wind speed show a clear left-tailed distribution of the u-component, while the dis-
(a) (b)
(c) (d)
140
tribution of the v-component is quite steep. This shows that not only positions with bimodal
distributions of the wind direction show large variances in the velocity components, but also
positions with monomodal distributions and significantly skewed distributions of velocity
components.
5 CONCLUSIONS & SUMMARY
The main purpose of the extensive flow measurements was to document the development of
the boundary layer within and above the city. Here, an independence of the flow
characteristics of the inlet velocity was detected. The different building structures in the two
different sections of the model considerably differ in density and height; they cause a
constant adjustment of the flow to the local roughness. Taking into account the average
building height of both model sections the height of the roughness layer is approximately
3*z/h
mean
. Another focus was to investigate the local influence of urban roughness to the
near-surface wind field. In different intensive field measurements with high spatial resolution
characteristic flow patterns are found which are generated by the local geometry. The com-
plex local flow patterns were documented by mean normalized wind vectors. In particular,
the observation of bimodal distribution functions of the horizontal components of wind speed
and wind direction show that the description of wind conditions by mean wind vectors are not
a reliable representation of the expected wind conditions in the natural case. The information
about the temporal evolution of the flow pattern is lost in this kind of representation.
6 ACKNOWLIDGEMENTS
Financial support by the German Federal Office of Civil Protection and Disaster Assistance
as well as by the Ministry of the Interior of the City of Hamburg is gratefully acknowledged.
Parts of the wind-tunnel model construction were financially supported by the KlimaCampus
at the University of Hamburg.
7 REFERENCES
Boris, J. P., Obenschain, K., Patnaik, G., Young T. R. Jr. CT-ANALYST: Fast and accurate cbr emergency as-
sessment. Laboratory for Computational Physics and Fluid Dynamics, U.S. Naval Research laboratory.
Cheng, H., Castro, I. P., 2002. Near wall flow over urban-like roughness. Boundary-Layer
Meteorology, 104, 229-259.
Counihan, J., 1975. Adiabatic atmospheric boundary layers: A review and analysis of data from the period
1880-1972. Atmospheric Environment, 9, 871-905.
Harms, F., 2011. Characterization of transient dispersion processes in an urban environment, in: Proceedings of
PHYSMOD2011 International Workshop on Physical Modeling of flow and Dispersion Phenomena,
Hamburg, Germany.
Hertwig, D., Harms, F., Patnaik, G., Obenschain, M.Y., Leitl, B. and Schatzmann, M., 2011. On aspects of LES
validation for urban flow fields, in: Proceedings of PHYSMOD2011 International Workshop on Physical
Modeling of flow and Dispersion Phenomena, Hamburg, Germany.
VDI Guideline 3783 Part 12, 2000. Environmental meteorology. Physical modeling of flow and dispersion
processes in the atmospheric boundary layer. Application of wind tunnels.
141
Comparisons of turbulence structures over various types
of surface geometry
Hiroshi Takimoto
a
, Ayumu Sato
b
, Takenobu Michioka
c
,
Atsushi Inagaki
d
, Manabu Kanda
e

a
Tokyo Institute of Technology, Tokyo, Japan, takimoto.h.aa@m.titech.ac.jp
b
Central Research Institute of Electric Power Industry, Abiko, Japan,
ayumu@criepi.denken.or.jp
c
Central Research Institute of Electric Power Industry, Abiko, Japan,
michioka@criepi.denken.or.jp
d
Tokyo Institute of Technology, Tokyo, Japan, inagaki.a.ab@m.titech.ac.jp
e
Tokyo Institute of Technology, Tokyo, Japan, kanda.m.aa@m.titech.ac.jp
ABSTRACT: The aim of this study is to reveal the dependency of turbulence structures on
the types of surface geometry. The characteristics of the turbulence structures in the
horizontal sections at the several measurement heights over the four types of surfaces, (1)
Square Array, (2) Height Variation, (3) 2D Canyon, (4) Flat surface, were investigated using
Particle Image Velocimetry technique. Our attention mainly goes to the aspect ratio of
turbulence structures, and they are quantitatively measured from the distributions of two-
point correlation coefficients. The aspect ratios (streamwise length / spanwise width) of the
structures, which were extracted using the threshold of Ruu=0.4, were found to be accurately
parameterized by the shear strength at that height. The interesting result is that there is a
critical value in the shear strength. The aspect ratio of turbulence structures is linearly
increased with increasing shear strength until the critical value, and then the aspect ratio
converges a constant value 3.5 regardless of the magnitude of the velocity gradient.
1 INTRODUCTION
Turbulent flow contains wide-ranging scales of coherent structures. The smallest scale of
coherent structure is just a several fold of Kolmogorov scale (Tanahashi et al., 2004),
whereas the largest scale, e.g. in the atmospheric boundary layer, is up to a few kilo meters
(Kanda et al., 2004). Among them, hairpin-like vortex is one of the most important structures
in wall turbulences. Willmarth and Lu (1971) and Blackwelder and Kaplan (1976) revealed
the dominant influences of the associated ejection flow and sweep flow to the transport
processes of momentum. Pantons (2001) review paper gives us a clear view of self-
sustaining mechanism of turbulences over smooth surfaces, and packet structure model well
explains the chain-reaction generation of hairpin vortices (Adrian et al., 2000). Christensen
and Adrian (2001) suggested that the morphology of the packet structures is not largely
influenced by Reynolds number, and Hommema and Adrian (2003) revealed the similar
packet structures in the high-Reynolds number atmospheric turbulence.
A packet structure can be recognized as a streaky structure in the horizontal cross section
near the surface, and this type of structures has been found in wall turbulences with many
different types of surface geometry, including very rough surfaces such as urban-like and
plant-like canopies. Within the surface layer over urban-like canopys, Kanda et al. (2004)
indicated the presence of low speed streaks which have the similar shapes to that in flat
142
boundary layers, and Coceal et al. (2007) successfully visualized the hairpin vortices
developed over cubic arrays. For plant canopies, Shaw et al. (1995) revealed the distributions
of two-point correlation coefficients of wind velocity to quantitatively express the shape and
the size of turbulence structures. The results demonstrated the presence of longitudinally
elongated structures, and it was reinforced by Watanabe (2004) from the snapshots of the
flow fields over a plant-like canopy layer.
However, many of the studies on such coherent flow structures focused on just one type of
surface geometry, and only a few investigations have been done with multi-type surface
geometries (Inagaki et al., 2009; Lee et al., 2011). Thus little is known of the influences of
surface geometry on the characteristics of turbulence structures.
In this study, we used Particle Image Velocimetry (PIV) technique to visualize and
compare the turbulence structures over a flat surface and three types of simplified urban-like
surfaces, and examined the influence of surface geometry to the turbulence structures. Since
it is difficult to discriminate streaky structures from the wakes of buildings in the roughness
sublayer over urban canopy layers, our focus mainly goes to the turbulence structures in
logarithmic layer.
2 EXPERIMENTAL DESCRIPTIONS
Experiments were conducted in the second test section of Twinned Wind Tunnel
(TWINNEL) at Central Research Institute of Electric Power Industry, Japan (Sato et al.,
2009). The test section is 10 m-long with a cross-sectional area of 1 m x 1 m. All
measurements were done under the neutral atmospheric stability conditions to exclude the
thermal buoyancy effects. Figure 1 shows the schematics of the four types of surface
geometries used in this study; (a) Square array (SQ), (b) Height variation (HV), (c) 2D street
canyon (2D), and (d) Flat surface (FL). In SQ case, wooden cubes were aligned regularly
with a plan area density of 0.25. The height of each cube was 35 mm and described as H,
hereinafter. HV case has the same volume of building models per unit area as SQ case, but
two different heights of blocks were used. 2D model consists from wooden bars whose length
is 900 mm, that is equivalent to 26H. Here, the streamwise, the spanwise, and the vertical
directions are described as x, y, and z-axis, where x = 0 lies at the entrance of the wind tunnel,
y = 0 at the lateral centre of the test section, and z = 0 at the floor level. In the case of SQ, HV,
and 2D, 106 rows of building models were aligned with the regular interval starting from x =
0.3 m. Free stream velocity was set to 2.0 m s
-1
at z = 525 mm (15H). Turbulence structures
in x-y plane were visualized and measured using PIV, and the measurement sections were
centred at x = 6.5 m in every case (figure 2). The fetch length from the leading edge of the
internal boundary layer is 177H for SQ, HV, and 2D case. Measurement heights of each case
are shown in table 1, and they correspond to the layers of log region and some portions of the
surrounding layers, including the upper part of roughness sublayer and the bottom part of
wake region. Three types of normalization, z
+
= z/
, = z/, z/H, were used in this study,

where
expresses viscous length scale calculated as
=/u
*
, and represents the
boundary layer height.
The size of the measurement section is 669 mm x 502 mm (19.1H x 14.3H) for the
measurement height of z = 70 mm (2H), but it differs slightly depending on the height
(maximum 10% in length). Double pulse images were taken by a CCD camera (1600 x 1200
pixel, Dantec Dynamics) with a sampling frequency of 6 Hz. Measurement time was set to 5
minutes, and a total of 1800 vector maps were obtained for each measurement case. Tracer
particles (average diameter of 1 m) were generated by a fog generator (SAFEX), and were
143
released from the entrance of the wind tunnel. The size of interrogation window, which
represents the spatial resolution of PIV measurement, is 13.4 mm. Boundary layer depth was
estimated as the height where the mean velocity take 99% of the free-stream velocity.
Friction velocity was calculated by non-linear least-square method together with roughness
length
0
z and displacement height d. Reynolds number
/ Re
*
u of each case is also given
in table 1.

(a)
(b) (c) (d)
1H
1H
1H
0.5H
1H
1.5H 1H
1H
1H
1H

Figure 1: Schematic view of the surface geometries; (a) SQ, (b) HV, (c) 2D, (d) FL.

Figure 2: Schematics of (a) wind tunnel and (b) building array (SQ).

Case
Building
height
Plan area
density
Boundary
layer depth
[m]
Measurement height Displacement
height
d [m]
Friction
velocity
u
*
[m/s]
Reynolds
number
Re

z
+
z/H
SQ 1H 0.25 0.35
642, 888, 1382,
1876, 2863, 3850
0.13, 0.18, 0.29,
0.39, 0.59, 0.80
1.5, 2, 3,
4, 6, 8
0.007 0.22 4935
HV
0.5H,
1.5H
0.25 0.32
1024, 1630,
2235, 3447, 4658
0.19, 0.31, 0.42,
0.66, 0.89
2, 3, 4,
6, 8
0.011 0.27 5451
2D 1H 0.5 0.28
350, 518,
855, 1191
0.14, 0.20,
0.34, 0.47
1.5, 2,
3, 4
0.016 0.15 2692
FL 0 0 0.18 164, 245, 327
0.20, 0.30,
0.40
- 0 0.072 808

Table 1: Conditions of the wind tunnel experiments.

Flow
(a)
(b)
144
3 ESTIMATING THE SIZE OF TURBULENCE STRUCTURES
In this study, streaky structures are observed in every case. Figure 3 shows a sample
velocity field obtained in HV case (=0.31). Coloured regions represent the area of the
positive second invariant of velocity gradient tensor, Q, multiplying the sign of vertical
vorticity,
z z
/ . The positive second invariant indicates the intensity of rotating motions,
and it can be a good marker of low-speed streaks as is shown in figure 3. It is possible to
estimate the size of turbulence structures from this index, or we can also utilize the value of
vorticity or swirling strength to extract each turbulence structures (e.g. Ganapathisubramani
et al., 2005; Hirooka et al., 2007). However in this study, the measurement area is not large
enough to look the entire picture of those coherent structures which has severalfold scale of
the boundary layer thickness. It is also difficult to distinguish individual structures from
coalescent structures. Thus for the estimation of the scale of turbulence structures we adopted
a method that uses the distributions of two-point correlation coefficients. Two-point
correlation coefficient, R
uu
, was calculated from the fluctuation of streamwise velocity u at
the reference point, (x, y) = (x
ref
, y
ref
), and each grid point. Figure 4 is an example of the
distributions obtained from SQ case (=0.18). It reveals that the correlation soon decays in
spanwise direction, whereas the correlation keeps the significant value in streamwise
direction. This indicates the dominance of the longitudinally elongated structures in velocity
fluctuations. In this study, the isoline of R
uu
= 0.4 was used as the boundary of the averaged
turbulence structures. The value 0.4 corresponds to relatively large scale structures. The
length of the averaged turbulence structure, L
x
, was estimated on the line of y = y
ref
as the
distance between two intersecting points of R
uu
= f(x, y
ref
) and R
uu
= 0.4. The width L
y
was
calculated in the same way on the line of x = x
ref
. Ensemble averages of L
x
and L
y
were taken
with multiple reference points to improve the statistical stability.

-8 -6 -4 -2 0 2 4 6 8
-6
-4
-2
0
2
4
6
Ref. Vector (m/s)
1
-0.4
-0.3
-0.2
-0.1
0
0.1
0.2
0.3
y
/
H
x'/H
Wind

Figure 3: Snapshots of the flow fields in HV case. Velocity vectors express the fluctuation from the horizontal
average.
-8 -6 -4 -2 0 2 4 6 8
-6
-4
-2
0
2
4
6
x'/H
y
/
H

Figure 4: Example of the distributions of two-point correlation coefficients for SQ case (z/H = 2).
145
4 RESULTS
4.1 Length scale
Figure 5 shows the vertical profiles of L
x
normalized with boundary layer depth for each
surface geometry. For three urban cases, profiles have large curvature at certain height, and
the trends are completely different between above and below these levels. In the lower layer,
L
x
is getting longer with the distance from the surface. In general, eddy size is linearly
increased with the distance from the wall, and the merging of packet structures can also be a
cause of this elongation (Tomkins et al., 2003). The gradient of L
x
is very similar for all cases,
including FL case, with the growth rate of L
x
1.86z. In contrast to this, the upper layer has
the shorter L
x
with increasing height. The heights of maximum L
x
are not the same for each
case.
However, the variations of L
x
with velocity gradient, '
*
kz u z u , shown in figure 6 reveal
a good matching of L
x
regardless of the types of surface geometry. This indicates the
importance of the shear strength for the determination of the size of turbulence structures.
The upper limit of the lower layer can be predicted by the shear, and the number of streaky
structures or the intensity of streaky structures are drastically decreased above this layer. FL
case has similar variations to the urban types at least in the logarithmic layer. Note that the
boundary layer depth is a good scaling parameter for L
x
.
To confirm this agreement, snapshots of the flow fields are presented in figure 7. As noted
earlier, measurement area of the current system is not enough to capture the coherent
structures. However, owing to the frequent sampling of the PIV imaging, consecutive
snapshots were overlapped with each other, and they can be combined by considering the
matching of the wind velocity distributions between the two snapshots. Matching can be
quantified by the cross correlation coefficients of two images. Although this method is
assuming Taylors frozen hypothesis, it is usable for qualitative analysis, and it enables us to
look the distributions of coherent structures.
Figure 7 is obtained by combining several tens of sequential vector maps, and shows the
distributions of velocity fluctuation divided by its standard deviation. They are typical scenes
of each surface geometry, and the dimensions of the figures are normalized by boundary
layer thickness to compare the normalized turbulence structures. The measurement heights of
each case correspond to the almost same shear level (u
*
/kz 6). The length, width, and the
meandering feature of coherent structures look similar with each other, and these results also
support the findings from figure 6.

Figure 5: Vertical profiles of L
x
. Figure 6 Variations of L
x
with velocity gradient.

146

-3 3

Figure 7: Snapshots of the combined flow fields; (a) SQ, (b) HV, (c) 2D, (d) FL.

4.2 Aspect ratio
Figure 8 shows the variations of L
x
/ L
y
, the aspect ratio of turbulence structures, with
velocity gradient. The plots surprisingly agree well for all surface geometries, and this fact
indicates that the aspect ratio of turbulence structures in the logarithmic layer are irrelevant to
the surface geometry, the magnitude of drag, or even the origin of eddies. Not only the
statistical similarity found in the framework of Monin-Obukhov similarity theory, but also
the structural similarity of coherent turbulence is expected to exist.
For small velocity gradient, aspect ratios are linearly increased with velocity gradient.
These heights are corresponding to the layer that has decreasing L
x
with height shown in
figure 5. On the other hand, the aspect ratio becomes saturated above a certain velocity
gradient level around u
*
/kz = 6. Although it is unclear from the present dataset whether FL
case shows the saturation area in the same way as urban cases or not, Lee et al. (2011)
revealed that the aspect ratio can still become longer as the surface gets closer even in the
buffer layer. This trend is different from the urban cases, and thus the disturbances from the
wake of blocks seem to be the main cause of this saturation. In addition, Hutchins et al.
(2007) proved the underestimation of L
x
due to the meandering property of streaky structures
in a method using two-point correlation coefficients, and this can lead to the limit of the
aspect ratio.
Here, the constant aspect ratio has the value of about 3.5 with the threshold of R
uu
= 0.4.
This value is different from the other thresholds of R
uu
, but the trends are almost same for R
uu

= 0.3, and R
uu
= 0.5.

Figure 8: Variations of the aspect ratio L
x
/ L
y
with velocity gradient.
y /
=0.20
=0.34 y /
0
1
-1
0
1
-1
0
.5
-.5
0
.5
-.5
y /
y /
=0.42
=0.39
(a)
(b)
(c)
(d)
147
5 CONCLUSIONS
The differences and the similarity of the turbulent structures over several types of surface
geometry were investigated, and following results are obtained:
1. The length of turbulence structure is dependent on shear strength and boundary layer
depth.
2. Over urban-like surface, aspect ratio of the mean turbulence structures are independent of
surface geometry.
3. Flat surface also has the similar aspect ratio to urban-type in logarithmic region.
6 REFERENCES
Adrian, R.J., Meinhart, C.D., Tomkins, C.D.: 2000, 'Vortex organization in the outer region of the turbulent
boundary layer', J. Fluid Mech., 422, 1-54.
Blackwelder, R.F., Kaplan, R.E.: 1976, On the wall structure of the turbulent boundary layer, J. Fluid Mech,
76, 89-120.
Christensen, K.T., Adrian, R.J.: 2001, Statistical evidence of hairpin vortex packets in wall turbulence, J. Fluid
Mech., 431, 433-443.
Ganapathisubramani, B., Hutchins, N., Hambleton, W.T., Longmire, E.K., Marusic, I.: 2005, Investigation of
large-scale coherence in a turbulent boundary layer using two-point correlations, J. Fluid Mech., 524, 57-
80.
Hirooka, S., Inagaki, A., Kanda, M.: 2007, A method of identification of the turbulent organized structure,
Hydraulic Eng., 51, 241-246 (in Japanese).
Hommema, S.E., Adrian, R.J.: 2003, Packet structure of surface eddies in the atmospheric boundary layer,
Boundary-Layer Meteorol., 106, 147-170.
Hutchins, N., Marusic, I.: 2007, Evidence of very long meandering features in the logarithmic region of
turbulent boundary layers, J. Fluid Mech., 579, 1-28.
Inagaki, A., Maruyama, A., Kanda, M.: 2009, Spatial and temporal scales of coherent turbulence over outdoor
reduced urban scale model, The 7
th
Int. Conf. on Urban Climate, P2-12.
Kanda, M., Moriwaki, R., Kasamatsu, F.: 2004, Large-eddy simulation of turbulent organized structures within
and above explicitly resolved cube arrays, Boundary-layer Meteorol., 112, 343-368.
Lee, J.H., Sung, H.J., Krogstad, P.-.: 2011, Direct numerical simulation of the turbulent boundary layer over a
cube-roughened wall, J. Fluid Mech., 669, 397-431.
Panton, R.L.: 2001, Overview of the self-sustaining mechanisms of wall turbulence, Prog. Aerosp. Sci., 37,
341-383.
Sato, A., Takimoto, H., Michioka, T.: 2009, Impact of wall heating on air flow in urban street canyon, Proc. of
International Workshop on Physical Modelling of Flow and Dispersion Phenomena.
Shaw, R. H., Brunet, Y., Finnigan, J. J., Raupach, M. R.: 1995,Awind tunnel study of air flow in waving
wheat: Two-point velocitystatistics , Boundary-Layer Meteorol., 76, 349-376.
Tomkins, C.D., Adrian, R.J.: 2003, Spanwise structure and scale growth in turbulent boundary layers, J. Fluid
Mech., 490, 33-74.
Watanabe, T.: 2004, Large-eddy simulation of coherent turbulence structures associated with scalar ramps over
plant canopies, Boundary-Layer Meteorol., 112, 307-341.
Willmarth, W.W., Lu, S.S.: 1971, Structure of Reynolds stress near the wall, J. Fluid Mech, 55, 65-92.

148
Session 2 Papers
S-II. Validation
Tuesday, August 23
Validation part 1
10.4011.00, S-II.1
11.0011.20, S-II.2
11.2011.40, S-II.3
11.4012.00, S-II.4
Validation part 2
13.2013.40, S-II.5
13.4014.00, S-II.6
14.0014.20, S-II.7
14.2014.40, S-II.8
14.4015.00, S-II.9
Validation part 3
15.2015.40, S-II.10
15.4016.00, S-II.11
16.0016.20, S-II.12
16.2016.40, S-II.13
149
NOTES AND COMMENTS:
150
NOTES AND COMMENTS:
151
NOTES AND COMMENTS:
152
Buoyant Flow in a Street Canyon: Comparison of CFD
Simulations and Wind Tunnel Measurements
Jonas Allegrini
a
, Viktor Dorer
b
, Jan Carmeliet
c,d

a
Empa, Swiss Federal Laboratories for Materials Science and Technology,
Laboratory for Building Science and Technology, Dbendorf, Switzerland,
jonas.allegrini@empa.ch
b
viktor.dorer@empa.ch
c
ETHZ, Swiss Federal Institute of Technology Zrich, Chair of Building
Physics, Zrich, Switzerland, carmeliet@arch.ethz.ch
d
jan.carmeliet@empa.ch
ABSTRACT: PIV measurements of the flow inside a street canyon with bottom heating were
carried out in the new boundary layer wind tunnel of ETH/Empa in Dbendorf. The
measured velocities and turbulent kinetic energies were compared with 2D RANS CFD
(Computational Fluid Dynamics) simulations with a realizable k- model. The measured
results show that for low free stream velocities, the velocities and turbulent kinetic energies
inside the street canyon are increased for the heated cases. For high free stream velocities the
flow field does not change significantly with bottom heating. The CFD simulations show the
same trend and large scale flow structures, but the profiles on the vertical centerline are
significantly different. The CFD simulations overestimate the velocities and turbulent kinetic
energies close to the bottom and underestimate the effect of the buoyancy. This study
demonstrates the difficulties to predict buoyant flow in street canyons with CFD.
1 INTRODUCTION
A significant part of the worlds energy consumption is used for heating and cooling of
buildings. Today about 50% of the population lives in urban areas (UN Population Division
2007). This proportion will rise to 70% in the next 40 years. Minimizing the energy demand
of buildings in urban areas has a great energy-saving potential (Santamouris et al. 2001). The
energy demand is influenced by the convective heat flux at the building faades and therefore
by the flow field around the building. In street canyons buoyancy plays a very important role.
During the day the building faades are heated by the solar radiation. The heated walls
significantly affect the flow field in the street canyon (Xie et al. 2007 and Kovar-Panskus et
al. 2002). For a hot summer day with calm winds and high faade temperatures buoyancy
influences the convective heat fluxes significantly. Very few experimental studies have been
conducted to analyse buoyant flow fields inside street canyons. Kovar-Panskus et al. (2002)
measured the flow inside a street canyon with heated walls in a wind tunnel and found some
influence of the heating on the flow field. Garbero et al. (2011) compared wind tunnel
measurement with numerical simulations and showed a qualitative agreement comparing
153
contour and vector plots but no velocity profiles. Water tank experiments with bottom heating
of a non-symmetrical street canyon were conducted by Huizhi et al. (2003). They found that
the flow was completely driven by buoyancy for low ambient wind speeds. Louka et al.
(2001) measured the flow field in a full-scale street canyon in Nantes and found only a thin
thermal boundary layer that was influenced by the buoyancy.
For this study wind tunnel experiments are conducted for a street canyon with bottom
heating. PIV is used to measure the flow field in the street canyon. The measured wind
velocities and turbulent kinetic energies are used for a comparison with CFD (Computational
Fluid Dynamics). Results of these CFD simulations in terms of convective heat transfer
coefficients are to be used in building energy simulations to assess the building energy
demand for cooling and heating by means of (Allegrini et al. 2011).
Similarity criteria have to be respected for the wind tunnel measurements with the scaled
model. The Richardson number Ri was used to evaluate the importance of the buoyancy
effects:
2
)
0
(
U
H T
W
T g
Ri
(1)
where g is the gravitational acceleration, T
w
the wall temperature, H the building height, U
the free stream velocity and T
0
the temperature at the inlet boundary. For buildings high
scaling factors are needed for wind tunnel measurements, thus high surface temperatures
have to be used to achieve the same Ri as in full-scale. However, the surface temperatures
should not be too high to avoid high changes of the density of the air. Therefore it is difficult
to have the same Ri for the model and the full-scale street canyon. The largest Ri number in
this study is 0.61. This still is a rather low value, but buoyancy effects can already be seen in
the measurements.
This study was conducted in the closed loop ETHZ/Empa atmospheric boundary layer wind
tunnel in Dbendorf. To model the flow in a street canyon a 150x20x20cm cavity was used
(figure 1a). The flow direction was normal to the axis of the cavity. The windward and the
leeward walls were made out of wood and the ground of the cavity was made out of
aluminium. The ground was heated by heating mats to surface temperatures of 40C and
60C. The approach flow and unheated wall temperatures were 25C. Because the heated
surfaces could not be perfectly isolated from the unheated surfaces, at the edges temperatures
above 25C could not be avoided. Free stream velocities were 0.55, 1.4 and 2.75m/s and the
largest Richardson number was 0.61. The Reynolds number was between 7000 and 35000
using the free stream velocity and the street canyon height as reference speed and height. The
floor of the wind tunnel was smooth, no roughness was used. The upstream fetch was 7.5m
long and therefore it is assumed that a stable boundary layer is formed in front of the cavity.
The velocity and the turbulent kinetic energy of the boundary layer were measured and used
as inlet boundary conditions for the CFD simulations. The cross-section of the wind tunnel
was 98x190cm.
Particle image velocimetry (PIV) was used to measure the flow field (figure 1b). The
advantage of this method is that the instantaneous flow field can be measured in a 2D plane.
A camera with a resolution of 2016 x 2016 pixels was used. As the strong reflections of the
laser sheet on the surfaces of the cavity are prone to damage the camera detectors, the field of
view had approximately the same dimensions as the cavity (20x20cm) but did not include the
region very close to the surfaces. 1576 images were recorded at a frequency of 200Hz the
154
minimal frequency of the laser. The 16GB memory of the camera allows to measure for
7.88s, what is a rather short time period for such a flow. A time average of these
measurements was used for the comparison with the CFD simulations described in section 3.

Figure 1: Dimensions of the wind tunnel measurement (a) and sketch of the measurement equipment (b).
3 NUMERICAL SIMULATION
Steady 2D RANS CFD simulations using ANSY Fluent 12.0 were conducted with a
realizable k- turbulence model and low-Reynolds number modeling (LRNM) at the walls.
The dimensions of the cavity were the same as for experimental measurements. The
dimension of the grid was chosen according to Franke et al. (2007). A 2D structured grid was
built based on grid sensitivity analysis and Franke et al. (2007) and consisted of 3500 cells.
The grid was refined towards the walls with maximum y+ values of 5.
At the inlet of the domain the measured boundary layer profiles were used. A gradient of the
boundary layer cannot be avoided in the CFD simulations (Blocken et al. 2007). Therefore
CFD simulations with different boundary layer profiles at the inlet for the same free stream
velocities were used and it was found that the flow inside the street canyon is not very
sensitive to the boundary layer profiles. The boundary layer of the turbulent dissipation rate
was computed as a function of the turbulent kinetic energy. At all surfaces a no-slip boundary
condition with zero roughness was imposed, because no surface roughness can be specified
for LRNM in ANSYS Fluent. For the surfaces in front and behind the cavity an adiabatic
boundary condition was set. The temperatures inside the cavity were set according the
temperatures of the measurements. To model the buoyancy the Boussinesq approximation
was applied. Pressure-velocity coupling was taken care of by the SIMPLEC algorithm. The
PRESTO! spatial discretization scheme was used for the pressure interpolation and a second
order spatial discretization scheme for the convection of the governing equations.

Figure 2: Computational domain.
(a) (b)
155
4 RESULTS
4.1 Measurements
The boundary layer was measured at the same position as the street canyon before the street
canyon was installed. In figure 3 the boundary layer profiles of the horizontal velocity and the
turbulent kinetic energy are given for the three free stream velocities specified (0.55, 1.4,
2.75m/s).

Figure 3: Boundary layer profiles of the horizontal time averaged velocity (a) and turbulent kinetic energy (b).

Figure 4: Vector field of the wind tunnel measurement with free stream velocity of 2.75m/s and ground
temperature of 60C (a) and contour plots of the velocitiy magnitude of wind tunnel measurements with free
stream velocities of 0.55m/s (b), 1.4m/s (c) and 2.75m/s (d) and a ground temperature of 60C.
(a) (b)
u=0.55m/s
u=1.4m/s
u=2.75m/s
u=0.55m/s
u=1.4m/s
u=2.75m/s
(a) (b)
(c) (d)
156
The measurements were conducted without roughness elements, because in the CFD
simulations no roughness can be specified with LRNM. Therefore the boundary layer is
rather thin.
Figure 4a shows the vector field for the measurement with a free stream velocity of 2.75m/s
and a temperature at the ground of 60C. It can be seen that one main vortex is formed in the
center of the street canyon. This flow structure can be found for all the measurements of this
study, except for the case with a free stream velocity of 0.55m/s and a ground temperature of
60C. The contour plot of the averaged velocity for this case is given in figure 4b. The center
of the vortex is shifted to a more upstream and higher position. A small second vortex can be
found in the bottom right corner. Due to the buoyancy the velocities inside the street canyon
are increased (see also the velocity profiles in section 4.2). The velocities close to the
windward wall have a similar magnitude as the ones of the measurement with free stream
velocity of 1.4m/s and a ground temperature of 60C (figure 4c). For the two higher free
stream velocities the center of the vortex remains in the center of the street canyon for a
ground temperature of 60C (fig 4c and 4d).
4.2 Numerical Simulations and Comparison
The CFD simulations show a similar main flow structure as the measurements for most of the
cases. The contour plots of the velocity magnitude for the three free stream velocities and a
ground temperature of 60C are given in figure 5. For all studied cases the vortex remains in
the center of the street canyon. Therefore the CFD results for the case with a free stream
velocity of 0.55m/s and a ground temperature of 60C (figure 5a) differs from the measured
results (figure 5b). The CFD simulations underestimate the buoyancy effects. Comparing
figures 4b-d with figures 5a-c shows that the CFD simulations overestimate the velocities for
all the cases inside the street canyon.

Figure 5: Contour plots of the velocitiy magnitude of the CFD simulations with free stream velocities of
0.55m/s (a), 1.4m/s (b) and 2.75m/s (c) and a ground temperature of 60C.
The measured results for the velocities and the turbulent kinetic energies on the vertical
centerline inside the street canyon are compared with the CFD results. In figure 6 the profiles
of the horizontal velocities are given for the nine studied cases. In the center of the street
canyon the results are comparable. For the case with a free stream velocity of 0.55m/s and a
ground temperature of 60C the measured results on the vertical centerline significantly differ
from CFD results. This is due to that fact that in the measurement the main vortex shifted in
direction of the leeward wall and therefore the horizontal centerline does not go through the
center of the vortex. Measurements and CFD simulations show increasing velocities close to
the ground with increasing ground temperatures due to buoyancy. This effect is less
(a) (b) (c)
157
pronounced for higher free stream velocities and is negligible for the highest free stream
velocity analyzed and is more pronounced in the measurements than it is predicted by CFD.
A cause for this could be that the Boussinesq approximation underestimates the buoyancy
effect. For the cases without bottom heating the CFD simulations predict higher velocities
close to the ground than measured in the wind tunnel, this probably due to the surface
roughness that cannot be modeled with LRNM in the CFD simulation. However, the
aluminum plate is smooth, and therefore other reasons might cause these differences. The
wooden walls could also slow the vortex down, but they are rather smooth as well. The
velocities at the top of the street canyon are also higher in the CFD results, probably due to
the same raison as for the higher velocities close to the ground.

Figure 6: Vertical profiles of the time averaged horizontal velocities on the centerline of the street canyon for
flows with free stream velocities of 0.55m/s (a), 1.4m/s (b) and 2.75m/s (c).
In figure 7 the turbulent kinetic energies on the vertical centerline of the street canyon are
given for all the studied cases. The differences for the turbulent kinetic energies between the
measurements and the CFD simulations are higher than for the velocities. Note that the
measurement time of 7.88s actually is too short to fully satisfy the requirements for averaging
the turbulent kinetic energy values, but it is assumed that they are good enough to do a
qualitative comparison. For the lowest free stream velocity the CFD simulations fail to
predict the turbulent kinetic energies. For the unheated case the results of the CFD
simulations are much higher than the wind tunnel results. For CFD and wind tunnel the
turbulent kinetic energies are increasing with higher ground temperatures, but the increase is
much stronger for the wind tunnel than for the CFD results. As already mentioned above,
CFD is underestimating the buoyancy effect. For the two higher free stream velocities the
results of the comparison are slightly better. The profiles have a similar shape and the same
order of magnitude. As for the velocity profiles the predicted turbulent kinetic energies by
CFD are higher close to the ground and the top of the street canyon than for the measured
results. Further it can be noticed that also for the turbulent kinetic energies buoyancy
becomes less important for higher free stream velocities for both the simulated and measured
results. For the highest free stream velocity the shapes of the turbulent kinetic energy profiles
are very similar for the different ground temperatures. This is an indication that the quality of
these profiles is not too low even though the measuring time is rather short.
This comparison shows the importance of validation the CFD results for street canyon with
mixed convective flows with e.g. wind tunnel measurements. The CFD simulations are able
to predict the main flow structures but fail to predict the flow close to the surfaces. Further
the CFD simulation does not predict the buoyancy effects correctly. Therefore a new wind
tunnel model was constructed for the ETH/Empa wind tunnel that will allow more detailed
(a) (b) (c)
158
validation experiments. The two walls and the bottom surface can be heated individually and
therefore allow for tests of different combinations of windward wall, leeward wall and
bottom heating. The new model will also allow to measure flow fields in street canyons with
different aspect ratios.

Figure 7: Vertical profiles of the turbulent kinetic energies on the centerline of the street canyon for flows with
free stream velocities of 0.55m/s (a), 1.4m/s (b) and 2.75m/s (c).
5 DISCUSSION
Wind tunnel measurements were conducted for a street canyon with bottom heating and the
results were compared with the results of 2D RANS CFD simulations with a realizable k-
model. These numerical models are not the best choice to simulate a buoyant flow field inside
a street canyon, but were chosen as reasonable approach in a coupled CFD- building energy
simulation (BES) environment where the CFD results in terms of convective heat transfer
coefficients are transferred to the BES. The wind tunnel experiments were built up in such a
way that the flow should be 2D. This is to a certain extend confirmed by the measured flow
field. The time averaged results show a closed vortex, what indicates that there are no strong
3D flow structures. The formation of 3D helix was avoided by closing both sides of the street
canyon. This was verified by smoke visualization.
Due to the limited camera memory the flow field was measured for a rather short time.
Therefore the comparison in this paper was more qualitative than quantitative. Because the
shapes of the profiles for the different measurements are very similar, it is assumed that the
measured flow field can be used for a qualitative comparison.
6 CONCLUSION
The flow field in a street canyon with bottom heating was measured and compared with CFD
simulations. The comparison was done for three different free stream velocities and three
different ground temperatures. It was found that for both the wind tunnel measurements and
the CFD simulations there exists only one main vortex inside the street canyon. In the wind
tunnel this vortex shifts slightly to the leeward wall and higher up for the lowest free stream
velocity and the highest ground temperature. This effect could not be seen in the CFD results.
But the acceleration of the flow inside the street canyon due to the buoyancy can be seen in
the wind tunnel and CFD results. This effect is more pronounced for higher free stream
(a) (b) (c)
159
velocities. The CFD simulations underestimate the buoyancy effects for all the studied cases.
This underestimation can be noticed in the profiles of the turbulent kinetic energies as well.
For the turbulent kinetic energies the differences between the measurements and the
simulations are in general higher. The shapes of the profiles are similar but the CFD
simulations predict much higher turbulent kinetic energies close to the ground and the top of
the street canyon. For the lowest free stream velocity the CFD simulations fail to predict the
turbulent kinetic energies inside the street canyon.
This study shows the importance of experimental validation of CFD simulations of buoyant
flows in street canyons.
7 REFERENCES
Allegrini, J., Dorer, V., Carmeliet, J., 2011. Analysis of convective heat transfer at building faades in cities and
its influence on the energy demand in buildings, in: Proceedings of the 13th International Conference on
Wind Engineering, Amsterdam, NL.
Blocken, B., Stathopoulos, T., Carmeliet, J., 2007. CFD simulation of the atmospheric boundary layer: wall
function problems. Atmospheric Environment , 41, 238-252.
Franke, J., Hellsten, A., Schlnzen, H., Carissimo, B., 2007. Best practice guideline for the CFD simulation of
flows in the urban environment, COST Action 732: Quality assurance and improvement of microscale
meteorological models. Hamburg, Germany.
Huizhi, L., Bin, L., Fengrong, Z., Boyin, Z., Jianguo, S., 2003. A laboratory model for the flow in urban street
canyons induced by bottom heating. Advances In Atmospheric Science, 20, 554564.
Kovar-Panskus, A., Moulinneuf, L., Savory, E., Abdelqari, A., Sini, J.F., Rosant, J.M., Robins, A., Toy, N.,
2002. A wind tunnel investigation of the influence of solar-induced wall-heating on the flow regime within a
simulated urban street canyon. Water, Air and Soil Pollution, Focus 2, 555-571.
Louka, P., Vachon, G., Sini, J.-F., Mestayer, P.G., Rosant, J.-M., 2001. Thermal effects on the airflow in a street
canyon - Nantes 99 experimental results and model simulations. Water, Air and Soil Pollution, Focus 2,
351364.
Santamouris, M., Papanikolaou, N., Livada, I., Koronakis, I., Georgakis, C., Argiriou, A., Assimakopoulos,
D.N., 2001. On the impact of urban climate on the energy consumption of buildings. Solar Energy, 70, 201
216.
Garbero, V., Salizzoni, P., Marro, M., Berrone, S., Soulhac, L., 2011. Influence of heat fluxes on the flow within
a two-dimensional street canyon: a comparison between wind tunnel measurements and CFD simulations, in:
Proceedings of the 13th International Conference on Wind Engineering, Amsterdam, NL.
UN Population Division, World Urbanization Prospects: The 2007 Revision Population Database,
http://esa.un.org/unup (retrieved January 2009)
Xie, X., Lui, C.-H., Leung, D., 2007. Impact of building faades and ground heating on wind flow and pollutant
transport in street canyons. Atmospheric Environment, 41, 9030-9049.

160
Validation study of flow and concentration fields in a
semi-idealized city
G. C. Efthimiou
a
, D. Hertwig
b
, F. Harms
b
, J. G. Bartzis
a
, B. Leitl
b

a
Department of Mechanical Engineering, University of West Macedonia,
Kozani, Greece
b
Meteorological Institute, KlimaCampus, University of Hamburg, Hamburg,
Germany
ABSTRACT: The study of flow and dispersion processes in urban areas is of great practical
interest in various fields ranging from wind comfort assessment and urban planning to
emergency response activities. One of the key issues in research on dispersion in complex
urban areas is the ability to predict the individual exposure. Within the bilateral research
project MODEX Modeling individual exposure from airborne hazardous releases this
topic is addressed by evaluating experimental and numerical data of flow and dispersion in
urban areas. The reliable prediction of peak concentrations and dosages requires an adequate
prediction of the wind flow and the concentration field. On the basis of state-of-the-art
validation metrics established in the COST Action 732 (e.g. Schatzmann et al., 2010) urban
flow and concentration fields computed by the Computational Fluid Dynamics (CFD)
Reynolds-Averaged Navier Stokes (RANS) code ADREA are compared to validation data
from boundary layer wind tunnel measurements. In a systematic study the quality of the
numerical prediction of wind and concentration fields is evaluated with a focus on the
identification of model strengths and limitations.
1 INTRODUCTION
CFD-RANS models allow for fast predictions of mean wind and concentration fields in
complex environments and are thus wide-spread tools for the investigation of pollutant
transport, for engineering wind comfort studies or in micro-scale meteorological applications.
With regard to these fields of applications the models should undergo systematic evaluations
to assess the quality of their predictions and establish a basis of confidence for users as well
as decision-makers. Dedicated measurements in boundary-layer wind tunnels can establish a
basis for such systematic validation studies given the high time and space resolution of state-
of-the-art measuring techniques. Since inflow and boundary conditions are well-defined and
documentable, laboratory data can provide high statistical confidence levels of measured
quantities, whereas the potential of field measurements in this regard is limited by the
natural variability of the atmosphere. Within the bilateral research project MODEX
Modeling individual exposure from airborne hazardous releases the topic of validation of
numerical models is addressed by evaluating experimental and numerical data of flow and
dispersion in urban areas. In this study urban flow and concentration fields computed by a
CFD-RANS code are compared to validation data measured in the boundary-layer wind
tunnel facility at the University of Hamburg. On the basis of state-of-the-art validation
metrics established in the COST Action 732 (Schatzmann et al., 2010) urban flow and
concentration fields computed by the Computational Fluid Dynamics (CFD) Reynolds-
Averaged Navier Stokes (RANS) ADREA code is compared to validation data from
161
boundary layer wind tunnel measurements. In a systematic study the quality of the numerical
predictions of mean wind and concentration fields is evaluated with regard to the
identification of model strengths and limitations. Based on the local identification of areas
of good or bad comparison the study showed that unsteady flow effects deep within street
canyons are a major cause for discrepancies between numerical and experimental results.
Another emphasis was put on identifying the influence of incorporated model
parameterizations, boundary conditions or grid resolution aspects as well as on a discussion
of possible implications for an improved fitness-for-purpose of experimental data.
2 EXPERIMENTAL AND NUMERICAL METHODS
2.1 Wind tunnel measurements
Flow and dispersion experiments were conducted in the boundary layer wind tunnel
WOTAN at the Environmental Wind Tunnel Laboratory of the University of Hamburg.
The 18m long and 4m wide test section of the tunnel is equipped with an adjustable ceiling
that allows the modeling of zero-pressure gradient atmospheric boundary layers and flows
within and above urban geometries.
The test case is the wind and concentration field within and above a 1:225-scale wind-tunnel
model of a semi-idealized urban complexity (Michel-Stadt) that is part of the online
validation data base CEDVAL-LES (www.mi.uni-hamburg.de/CEDVAL-LES-
V.6332.0.html). This building structure comprehends distinct characteristics of typical central
European cities (Fig. 1). With sharp building corners, characteristic courtyards, and complex
intersection structures the model was designed to pose a challenge to numerical models while
still being an approximation of a genuine urban roughness. Whereas the street canyon width
was kept constant, the height of the flat-roof buildings varied between 15, 18, and 24m full-
scale.

Figure 1: (a) Wind tunnel model of semi-idealized urban complexity (wind tunnel scale 1:225). (b)
Computational domain (full scale).
2.2 Numerical simulations
The numerical simulations have been performed using the research code ADREA (Bartzis et
al., 1991). For the present study the code solves the RANS equations for mass and
momentum of a fully turbulent and isothermal flow as well as the transport equation for the
mean concentration. The turbulence closure is obtained using the eddy-viscosity concept and
the two equation k- model (Bartzis, 2005) is used, where is the wave number scale. Finite
volume methods are used to discretize the conservation equations.
(a)
(b)
162
The computational domain is presented in Fig. 1. It should be noted that the simulation was
performed in full scale. Hexahedral cells have been used with a discretization of 191 x 118 x
41 in X, Y, and Z directions respectively. The minimum/maximum sizes of the discretization
cells along X-, Y- and Z- directions were taken 7.8/27.0m, 8.3/16.69m and 3.2/3.5m,
respectively. The computational grid was uniform and dense in the area between the
buildings and had a logarithmically increasing distance in the lateral areas.
The simulations were performed in two steps. First, the experimental approach flow was
simulated in order to obtain the vertical profiles of the velocity, the turbulent kinetic energy
and the wave number expected to be in agreement with the measurements. This was achieved
by solving the 1-D equations in the vertical direction with a very rough wall (experimental z
0
)
and prescribing constant experimental values at the top of the domain for U, k and . The
computational grid was the same as for the 3-D simulations in the vertical direction i.e.
consisting of 41 grid points and extending up to 144m.
The second computational step consisted of performing the full 3-D flow calculations of the
urban flow and dispersion using the computational results of the first step as inlet boundary
conditions. At the outlet of the domain outflow boundary conditions were imposed, i.e. zero
horizontal gradient, while at the lateral planes symmetry boundary conditions were
implemented. Similar to the first step, constant experimental values for U, k and were
specified on the top of the domain. In order to be consistent with the experiment the ground
floor was divided in three regions. The ground area upwind of the buildings was treated as
fully rough wall prescribing the experimental z
0
. The ground area between and after the
buildings was treated as partly rough wall using a very small z
0
equal to 0.0625m
corresponding to z
0
=0.3mm in wind tunnel scale. Concerning the pollutant source, it has been
modeled by an irregular surface that has been placed inside the computational domain at the
same location and height as in the experiment. Here, the source was placed on the roof top of
a building well upstream of the city center. The area of the surface, the release rate and the
physical properties of the actual pollutant (ethane) were modelled according to the
experimental reference. The second computational step was treated as a true transient state
problem. The total calculation time was selected as large as possible in order for the
pollutants to cross the entire computational domain.

163
3.1 Mean velocities
Figure 2 shows the scatter plots of modelled versus measured values for the along wind
velocity component U/Uref, and the lateral component V/Uref. In these plots all data points
(2,158) are included, and thus there are measurement positions both below and above roof
heights.

Figure 2: Scatter plots for the velocity components U/Uref and V/Uref for the ADREA simulations. The points
are colour-coded with respect to the measurement positions.
Concerning the velocity U/Uref the model appears to predict successfully the observed values
especially at high elevations (i.e. above 18m). However, a high scatter in the data is obvious
and this is mainly due to the coarse grid used (nearly 1million cells). It is worth to notice also
that the model appears much less successful in performance of U/Uref for the measurements
at the street canyon 2m than for the other locations. At this height (close to the ground) the
model presents a similar systematic underprediction of the observations. This trend will be
analyzed further in section 3.4. From Figure 2 it is also apparent that at the street canyon
levels (i.e. at 2m, 9m and 18m) the range of magnitudes of the velocity V/Uref is high due to
the arrangement of the buildings that drives the wind flow at these heights. From the Figure it
seems that the model has the ability to predict the V/Uref quite well. Again, the high scatter
of the results is most likely due to the coarse grid used. A linear trend is obvious around the
1:1 line. Additionally to the scatter plots and in order to quantify the model performance
validation metrics have been calculated and have been presented in Efthimiou et al. (2011)
based on the COST 732 guidelines (Schatzmann et al., 2010). It is shown that at street canyon
levels, the metrics for the individual layers (2m, 9m and 18m) showed a clear dependency on
the measurement height, with a general tendency of worse predictions close to the ground due
to the presence of the buildings.
In addition, Figure 3 shows locations within the 2m measurement plane at which the
validation metric factor of two of observations (FAC2) (Schatzmann et al., 2010) was
either within the bounds of acceptance (i.e. 0.5) or not (i.e. < 0.5) for the streamwise (left)
and the spanwise (right) velocity component calculated with ADREA (for a detailed
discussion of the validation metrics see also Efthimiou et al., 2011). Many of the locations at
which the model does not reach the acceptance criterion are representative of flow situations
in which RANS-based models are known to struggle due to the presence of unsteady flow
effects.
164

Figure 3: FAC2 of the ADREA predictions for locations at the 2m street canyon level for U/Uref (left) and for
V/Uref (right). Black circles indicate FAC2<0.5 and white circles FAC20.5. The arrow indicates the approach
flow and the numbers correspond to the heights of individual buildings.
Figure 4 shows the velocity vectors at the 2m measurement plane (left) and at 18m (right) for
the ADREA calculations in comparison to the experimentally determined flow field.

The code is able to capture the dominant features of the mean horizontal flow field at both
heights. However, concerning the local magnitude and direction of the wind vectors some
tendencies can be identified. ADREA shows underpredictions of the wind speeds especially
in streamwise oriented street canyons and the open space for the 2m measurement height.
Here, the influence of the prescribed boundary conditions is dominant and likely caused the
discrepancies. At 18m height, which roughly roof level height for most buildings, the
agreement is found to be very good.

Figure 4: Vector plots showing horizontal wind speed at 2m (left) and 18m (right). The large arrows indicate
the approach flow and the numbers correspond to the height of individual buildings. Green vectors indicate the
observations and black vectors the predictions.
165
3.2 Mean concentrations
Figure 5 presents the scatter plot for the mean normalized concentration as obtained by the
model. The 1:1 line as well as the FAC2 (factor of two) lines are presented also in addition. It
is obvious that a large majority of the results lies in the area delimited by the FAC2 lines.
Also it is interesting to note that the higher concentration values follow the 1:1 line quite
well, an ideal fact for individual exposure studies. The underestimation of the experimental
measurements, especially for small concentration values, is most likely due to the
inefficiency of the RANS methodology to predict unsteady effects at intersections and at
crosswind street canyons as will be described below.

Figure 5: Scatter plot of the mean normalized concentration between wind tunnel measurements and numerical
results (all sensors included).
In order to evaluate the total performance of the mean concentration model, validation
metrics have been used. These are the fractional bias (FB), the normalized mean square error
(NMSE) and the factor of two of observations (FAC2).
It should be noted that all the observed measurements were non-zero. The results for the
mean concentration validation metrics of the ADREA code are presented in Table 1. The
sensor results have been grouped into two sets. The first set consists of 40 sensors that were
placed far away from the ground on a horizontal plane at a height of 30m. The second set
consists of 106 measurement locations and contains sensors that were placed near the ground
(z=2m) where the human exposure is more direct.
The second column of Table 1 shows the models overall performance (all measurement
positions included). Concerning the model performance with regards to metrics, the COST
732 guidelines require the following quality acceptance criteria: FAC2 > 50%, |FB| < 0.3 and
NMSE < 4 for mean concentrations. According to Table 1, the FAC2 is relatively high for all
measurements (72%), indicating that the model predicts quite well the observed mean
concentrations. Also the FB and the NMSE values (0.21 and 0.13, respectively) indicate that
the model gives a small underprediction and scatter of the observations, respectively.
Overall the metrics computed from the entire set of measurements fulfill the proposed quality
acceptance criteria. Concerning the measurements at the 30m plane well-above roof top it is
obvious that the FAC2 with 73% is well above the limit (i.e. >50%). Acceptable values for
the NMSE (0.17) and the FB (0.36) have been calculated, too.
166
Table 1: Validation metrics for the normalized mean concentration. The results have been grouped in all
measurements and street canyon measurements at 2m and 30m.
Metrics All measurements Measurements at 2m 30m
FAC2 0.72 0.62 0.73
NMSE 0.13 0.09 0.17
FB 0.21 0.14 0.36

In case of the street canyon plane at 2m, the results are rather good. It is worth to notice that
for the measurements at the 2m plane the FB is small indicating a very small underprediction
of the observations. The scatter between the two planes is similar. Overall all the metrics at
the street canyon level fulfill the quality acceptance criteria.

One important factor for the discrepancies of the mean concentration prediction at the 2m
measurement height seems to be the predicted wind flow, which is the driving mechanism of
the dispersion process. In Efthimiou et al. (2011) it was shown that the unsteady flow effects
deep within the street canyons are a major source of discrepancies between the numerical and
experimental results. In order to understand the cause of these discrepancies a systematic
point-by-point comparison of the predicted flow and concentration with the experimental data
needs to be conducted.

Figure 6 shows the sensor locations at the 2m measurement plane where both flow and
concentration measurements have been performed in the wind tunnel.

Figure 6: Sensor locations at the 2m measurement plane where both flow and concentration measurements have
been performed.
In Figure 7, the 2D field of the normalized predicted and experimental mean concentration is
presented at the horizontal plane of 2m. Measurements and simulations were done at the
locations displayed in Figure 6. The source is located at approximately 400m full-scale
distance upstream of the first measurement point. As expected, the evolution of the mean
concentration away from the source is similar between the simulation and the experiment, i.e.
higher values near the source and lower ones farther downstream are found. It is interesting to
note that at the two upper corners of the central building there is a distinct difference in the
spread of the plume. As has been stated earlier this is mainly due to the unsteady effects of
the wind flow, where the model failed also to predict exactly the experimental velocity
measurements in the same area.
167

Figure 7: Horizontal plane of the normalized mean concentration of the experimental and numerical results at
the height of 2m, showing the evolution of the mean concentration around a building.
4. DISCUSSION AND CONCLUSIONS
The ability of a CFD-RANS model to adequately predict the wind flow and dispersion inside
a semi-idealized urban structure was studied. Based on wind tunnel validation data a
systematic study of the model performance revealed that unsteady flow effects deep within
the street canyons are a major source of discrepancies between the numerical and
experimental results. This pointwise evaluation enables to distinguish these kinds of
uncertainties from numerical or modeling errors. Overall the presented results are very good,
a fact that clearly strengthens the models robustness to be used for individual exposure
studies. The next steps of the analysis will concentrate on further investigation of the
experimental velocity in flow time series in terms of frequency distribution in order to
determine possible dependencies of the models agreement or disagreement on the shape of
the distributions.
5. ACKNOWLEDGEMENTS
We thank IKY and the DAAD IKYDA program for financial support in the MODEX project.
6. REFERENCES
Bartzis, J. G., Venetsanos, A., Varvayani M., Catsaros, N., Megaritou, A., 1991. ADREA-I: A three dimen-
sional transient transport code for complex terrain and other applications. Nuclear Technology 94, 135148.
Bartzis, J. G., 2005. New approaches in two-equation turbulence modelling for atmospheric applications.
Boundary-Layer Meteorology 116 (3), 445-459.
Bastigkeit, I., Fischer, R., Leitl, B, Schatzmann, M., 2010. Fundamental quality requirements for the gen-eration
of LES-specific validation data sets from systematic wind tunnel model experiments, in: Proceedings of
CWE2010. Chapel-Hill, NC, USA.
Efthimiou, G C., Hertwig, D., Fischer, R., Harms, F., Bastigkeit, I.,Koutsourakis, N., Theodoridis, A., Bartzis, J.
G., Leitl, B., 2011. Wind flow validation for individual exposure studies, in: Proceedings of ICWE13,
Amsterdam, The Netherlands.
Fischer, R., Bastigkeit, I., Leitl, B., Schatzmann, M., 2010. Generation of spatio-temporally high resolved
datasets for the validation of LES-models simulating flow and dispersion phenomena within the lower
atmospheric boundary layer, in: Proceedings of CWE2010. Chapel-Hill, NC, USA.
Schatzmann, M., Olesen, H., Franke, J., (Eds.), 2010. COST 732 model evaluation case studies: Approach-es
and results. University of Hamburg.
VDI Guideline 3783/12, 2000. Environmental Meteorology, Physical Modelling of Flow and Dispersion
Processes in the Atmospheric Boundary Layer Applications of Wind Tunnels. Beuth Verlag, Berlin.
VDI, 2005. Environmental meteorology Prognostic microscale windfield models Evaluation for flow around
buildings and obstacles. VDI guideline 3783, Part 9. Beuth Verlag, Berlin.
168
LES of flow and plume dispersion within and over
various obstacle arrays
Hiromasa Nakayama
a
, Klara Jurcakova
b
and Haruyasu Nagai
c

a
Japan Atomic Energy Agency, Ibaraki, Japan, nakayama.hiromasa@jaea.go.jp
b
Institute of Thermomechanics, Academy of Sciences of the Czech Republic, Prague, Czech
Republic, klara.jurcakova@it.cas.cz
c
Japan Atomic Energy Agency, Ibaraki, Japan, nagai.haruyasu@jaea.go.jp
ABSTRACT: We generated a spatially-developing boundary layer flow by using the existing
turbulent inflow technique and performed LESs of turbulent flow and plume dispersion in
obstacle arrays with various obstacle densities. In this paper, we examine the basic
performance of the LES model by comparing to wind tunnel experiments.
1 INTRODUCTION
Studies of atmospheric dispersion of contaminated materials in urban areas are one of the
most challenging tasks. In urban areas where the ground surface is covered with many
buildings and obstacles, distribution patterns of concentration should be investigated
accounting for the geometrical arrangement of urban buildings. For the assessment of human
health hazard from harmful substances, not only mean but also high concentration peaks in a
plume should be considered. For understanding plume dispersion behaviors within urban
areas, it is helpful to investigate the dispersion characteristics within regular arrays of
obstacles as simplified surface geometries for actual urban areas as a first step.
There are typically two approaches to predict plume dispersion in urban areas; one is wind
tunnel experimental technique and the other is numerical simulation technique by
computational fluid dynamics. The wind tunnel experiment is used as a reliable tool that can
accurately provide the data of flow and dispersion considering effects of urban-like obstacles.
For example, Davidson et al. (1996) investigated the effects of different obstacle array
configurations on streamwise variations of mean concentrations, and vertical and spanwise
spreads of a plume. Pascheke et al. (2009) investigated the effects of regularly arrayed
obstacles with uniform and variable heights on ventilation of scalars. Bezpalcova and Ohba
(2008) investigated the effects of obstacle arrangements and densities on the characteristics of
mean and root mean square (r.m.s.) concentrations.
The numerical simulation technique using Large-Eddy Simulation (LES) also has come to
be regarded as a useful tool, with the rapid development of computer technology. Dejoan et al.
(2010) conducted LESs of flow and plume dispersion in an obstacle array and investigated
the effects of incident wind angle deviation on the mean velocity and mean concentration
fields in comparison with the Mock Urban Setting Test field experiment. Boppana et al.
(2010) conducted LESs of flow and dispersion in obstacle arrays with uniform and variable
heights and investigated distribution patterns of mean concentrations. Branford et al. (2011)
also conducted LESs of flow and plume dispersion in an obstacle array and investigated the
effects of different wind directions on mean concentration. These numerical studies have
focused on plume dispersion in obstacle arrays with one obstacle density. As Grimmond and
Oke (1999) mentioned, actual urban surface geometries are characterized by a wide range of
169
building densities. In order to examine plume dispersion within actual urban areas, urban-
type surface geometries with a wide range of obstacle densities need to be considered.
In this study, we perform LESs of flow and plume dispersion in regular arrays of cubic
buildings with various obstacle densities and examine the basic performance of the LES
model by comparing to wind tunnel experiments.
2 NUMERICAL MODEL
The basic equations for our LES model are the spatially filtered continuity equation,
Navier-Stokes equation and the scalar transport equation. The subgrid-scale Reynolds stress
is parameterized by the standard Smagorinsky model (Smagorinsky, 1963) with the
Smagorinsky constant of 0.1. The subgrid-scale scalar flux is also parameterized by an eddy
viscosity model and the turbulent Schmidt number is set to 0.5 (Nakayama et al., 2011).
The coupling algorithm of the velocity and pressure fields is based on the Marker and Cell
method (Harlow and Welch, 1965) with the second-order Adams-Bashforth scheme for time
integration. The Poisson equation for pressure is solved by the Successive Over-Relaxation
method which is an iterative method. For the spatial discretization in the basic equation, a
second-order accurate central difference is used. Cubic Interpolated Pseudo-particle (CIP)
method proposed by Takewaki et al. (1985) is used only for the advection term of the scalar
transport equation. The time step interval tU
/H is 0.005 (t: time step). The maximum

CFL (Courant-Friedrich-Levy) number is about 0.15.
3 TEST SIMULATIONS
3.1 EXPERIMENTS FOR EVALUATING THE MODEL PERFOMANCE
The experiments were carried out by Bezpalcova and Ohba (2008) in the Boundary Layer
Wind Tunnel at Wind Engineering Center of Tokyo Polytechnic University, Japan. The
experimental set-up consists of buildings with dimensions: 70 mm (width), 70 mm (length),
and 70 mm (height). In this paper, obstacle density
f
is defined as the ratio of the total floor
projection area of buildings to the plan area of the study site. Buildings are arranged in the
regularly square array with
f
=0.16, 0.25, and 0.33. There are 157, 189, and 209 building
arrays with
f
=0.16, 0.25, and 0.33, respectively. The ground-level point source is located at
the center just behind the building of the 7th row and the 4th column, the 8th row and the 5th
column, and the 9th row and the 5th column of the arrays in cases of
f
=0.16, 0.25, and 0.33,
respectively. Here, the rows are numbered in increasing order in the streamwise direction
from the leading edge of the array and the columns are numbered in increasing order in the
spanwise direction.
In their experiment, the lower part of the neutral atmospheric boundary layer is simulated
by vortex generators set up at the wind tunnel section and roughness blocks. The scale of the
modeled boundary layer is 1:400, i.e. the lowest 120 m of the boundary layer in the full scale
is modeled. The mean wind velocity vertical profile of approach flow can be approximated
by a power law exponent of 0.25. Wind velocity was measured by Thermoanemometry using
a split-fibre probe. The uncertainties of flow measurement were 5% for both mean and r.m.s.
quantities. Concentration is measured using a fast-response flame ionization detector. The
uncertainties of concentration measurement were 9% and 17% for mean and r.m.s. quantities,
170
respectively. In their wind tunnel experiment, the building Reynolds number based on the
cubical building height and wind speed at the building height is about 14,000.
3.2 COMPUTATIONAL SETTINGS
Figure 2 shows a schematic illustration of the numerical model. Two computational
domains are set up: The main region for a simulation of turbulent flow and plume dispersion
in a building array and the driver region for generation of a spatially-developing boundary
layer flow. First, a thick boundary layer flow is generated by using the inflow turbulence
generation method of Kataoka and Mizuno (2002) into an upstream part of the driver region
and, then, a wind flow with strong turbulent fluctuations is produced by a tripping fence and
roughness blocks placed at the downstream of the recycle station. The fluctuating part of the
velocity at the recycle station is recycled and added to the inflow as shown in Fig. 2(a). This
unsteady wind flow is imposed at the inlet of the main region and calculations of turbulent
flow and plume dispersion in a building array are performed as shown in Fig. 2(b).
In the driver region, a conventional convective boundary condition is applied at the exit, a
free-slip condition for streamwise and spanwise velocity components is imposed and vertical
velocity component is 0 at the top. A periodic condition is imposed at the side and a non-slip
condition for each velocity component is imposed at the ground surface. Assuming that the
scale of the simulated boundary layer by LES is 120m in the full scale condition, the size of
the driver region is 1700m460m600m in streamwise, spanwise and vertical directions,
respectively. The number of grid points is 460250100. The Van Driest damping function
(Van Driest, 1956) is incorporated to account for near-wall effects. Building effects are
represented by immersed boundary method proposed by Goldstein et al. (1993).
In the main region, there are 156, 258 and 289 obstacle arrays with
f
=0.16, 0.25 and
0.33, respectively. A ground-level point source is located just behind an obstacle of the 7th
row and the 3rd column, the 8th row and the 4th column, and the 9th row and the 5th column
of the arrays in cases of
f
=0.16, 0.25 and 0.33, respectively. Each obstacle of the array is
resolved by 161624 grids in the streamwise, spanwise and vertical directions, respectively.
At the inlet of the main region, the turbulent inflow data obtained near the exit of the driver
region is imposed. The other boundary conditions in a flow field are the same as those in the
driver region but the damping function to account for near-wall effects is not incorporated. In
a concentration field, zero gradient is imposed at all the boundaries. The size and the number
of grid points for the main region are 2200m460m600m and 780250100 in streamwise,
spanwise and vertical directions, respectively. The lengths of the domain in front of the first
row and behind the last row are both 360m. The grid resolution above the ground surface is
the same as the one in the driver region. The origin of the coordinate is the location of a
plume source point, thus, x/H=0.0, y/H=0.0 and z/H=0.0 (H: an obstacle height). The plume
source is represented by two grids in order to be placed at just center behind a building. The
grid resolution for the point source is twice the real diameters of the point source.

Figure 1: Schematic diagram of the numerical model. (a) Driver region for generating boundary layer flow. (b)
Main region for plume dispersion within a building array.
(a) (b)
171
The length of the simulation run to calculate the time averaged values of velocity and
concentration TU
/H (T: averaging time) is 500. The length of the simulation run before
releasing the scalar is TU
/H is 250. In the present LES model, the building Reynolds

number is almost 5,000.
4 RESULTS
4.1 APPLOACH FLOW
Figure 2 compares the LES results of turbulence characteristics of approach flow with the
wind tunnel experimental data of Bezpalcova and Ohba (2008) and the recommended data of
Engineering Science Data Unit 85020 (ESDU 85020, 1985). ESDU 85020 provides
comprehensive turbulence characteristics of neutrally stratified atmospheric boundary layer
based on independent experimental measurements ranging from the ground surface to 300m.
ESDU 85020 recommends vertical profiles of turbulence intensities for each wind component
and their relationship in dependence on surface roughness. The experimental data are shown
with the error bars described in section 3.1.The profile of the mean wind velocity of LES is
found to fit the experimental profile of 0.25 power law. LES approach flow turbulence
intensities are copying shape of vertical profiles recommended by ESDU 85020 very well
between the recommended data for moderate rough and rough surfaces up to 100m. The
experimental data agree with LES and ESDU data only for vertical component, experimental
streamwise and spanwise turbulence intensities are slightly overpredicted and underpredicted,
respectively. The vertical profile of Reynolds stress of LES shows a constant profile in the
range 10m-60m. According to the review paper of Counihan (1975), it is shown that the
average height of the constant shear stress layer is 100m. The LES data lies within this range
shown by Counihan (1975).
The LES approach flow corresponds to a neutral atmospheric boundary layer based on
comparison with the ESDU 85020 recommended data. Although some of the turbulence
characteristics by LES are quantitatively different from those by the experiment, they both
reasonable well model the neutral boundary layer above rough surface and can be compared
taking in account their differences.

Figure 2: Schematic diagram of the numerical model. (a)Mean wind velocity. (b)Streamwise turbulence
intensity. (c)Spanwise turbulence intensity. (d)Vertical turbulence intensity. (e)Reynolds stress
4.2 DISPERSION CHARACTERISTICS
4.2.1 SPANWISE AND VERTICAL PROFILES OF CONCENTRATIONS
(a) (b) (c) (d) (e)
172
Figures 3 and 4 compare the LES results with the wind tunnel experimental data of
Bezpalcova and Ohba, (2008) of the spanwise profiles of mean (C
ave
) and r.m.s. (C
r.m.s.
)
concentrations at a height of 0.29H at the 3rd, 4th, and 5th row behind the source in
f
=0.16,
0.25, and 0.33, respectively. The positions are located at streamwise distance x/H=8.17H,
8.43H, and 8.86H behind the source, respectively. The mean and r.m.s. concentrations are
normalized by wind velocity at the obstacle height (U
H
), the obstacle height and the source
emission (Q). The experimental data are shown with the error bars. In the case of
f
=0.16,
mean concentrations of the LES are generally smaller than those of the experiment, while in
both cases of
f
=0.25 and 0.33 mean concentrations show good agreement with those of the
experiment. However, tendencies such as the formation of high concentration region in the
range -1.0<y/H<1.0 due to the enhanced spanwise spreads of the plume by obstacle arrays
and the decrease of mean concentrations towards the plume edge are similar to the
experiment in each case. The r.m.s. concentrations of LES are generally smaller than those of
the experiment in the case of
f
=0.16, while the LES data are similar in magnitude to the
experimental data in both cases of
f
=0.25 and 0.33. However, the tendencies to show the
local minimum at y/H=0.0 and the local maximum around y/H=-1.0 and 1.0 is the same as
the experiments in each case.
Figures 5 and 6 compare the LES results with the wind tunnel experimental data of the
vertical profiles of mean and r.m.s. concentrations at the central street canyon and crossing
Figure 3: Spanwise profiles of mean concentrations at height of 0.29H. (a) at the 3rd row behind source
location in the case of
f
=0.16. (b) at the 4th row behind source location in the case of
f
=0.25. (b) at the 5th row
behind source location in the case of
f
=0.25.
Figure 4: Spanwise profiles of r.m.s. concentrations at height of 0.29H. (a) at the 3rd row behind source
location in the case of
f
=0.16. (b) at the 4th row behind source location in the case of
f
=0.25. (b) at the 5th row
behind source location in the case of
f
=0.25.

Figure 5: Vertical profiles of mean and r.m.s. concentrations at central street canyon position. (a)-(c) indicate
mean concetrations in cases of
f
=0.16, 0.25, and 0.33, respectively. (d)-(f) indicate r.m.s. concetrations in cases
of
f
=0.16, 0.25, and 0.33, respectively.
(a) (b) (c)
(a) (b) (c)
(a) (b) (c) (d) (e) (f)
173
Figure 6: Vertical profiles of mean and r.m.s. concentrations at crossing position. (a)-(c) indicate mean
concetrations in cases of
f
=0.16, 0.25, and 0.33, respectively. (d)-(f) indicate r.m.s. concetrations in cases of
f
=0.16, 0.25, and 0.33, respectively.
positions at the 3rd, 4th, and 5th row behind the source in cases of
f
=0.16, 0.25, and 0.33,
respectively. At the central street canyon position, mean concentrations of the LES especially
within the obstacle height are smaller than those of the experiments in each case. However,
the tendencies to show nearly constant values within the obstacle height and decrease higher
are the same as the experiments. R.m.s. concentrations of the LES show constant within the
obstacle height and a peak just above the obstacle and gradually decrease with height.
Although r.m.s. concentrations of LES are underestimated within the obstacle height in each
case, these tendencies are the same as the experiments.
4.2.2 FREQUENCY DISTRIBUTIONS OF CONCENTRATION
Figures 7 and 8 compare the LES results with the wind tunnel experimental data of time
series of instantaneous concentrations at a height of 0.29H at central street canyon and
crossing position at the 3rd, 4th and 5th row behind the source in
f
=0.16, 0.25 and 0.33,
respectively. The instantaneous concentrations are normalized by the mean concentrations. At
the central street canyon, instantaneous concentrations of the experiments are found to
fluctuate around the average level smoothly and continuously in each case. Although
instantaneous concentrations of the LES also fluctuate around the average level, the
variability is seemed to be a little smaller than that of the experiment in each case. At the
crossing position, instantaneous high concentrations of the experiments which significantly
exceed the average level frequently occur in each case. Also in the LES, the peak
concentrations which exceed the average level also frequently occur and the variability of
concentration fluctuations is similar to that of the experiments in each case.
Figures 9 and 10 show normalized frequencies of instantaneous concentrations at the
central street canyon and crossing positions. Instantaneous concentrations are normalized by
the mean concentrations. At the central street canyon position, the normalized frequencies of
the experiments show sharp peak around the average levels and rapidly decrease towards
larger and smaller values of instantaneous concentrations in each case. Although the LES
results also show such a tendency in each case, the instantaneous concentration distribution is
sharper especially for the cases of denser arrays,
f
=0.25 and 0.33. At the crossing position,
the normalized frequencies of the experiments show wide flat peak around the average levels
and then gradually decrease with increase of instantaneous concentrations in each case.
Although the LES results also show such a tendency, especially for the cases of denser arrays,
f
=0.25 and 0.33, the normalized frequency of the mean concentrations are a little
overestimated. However, the shapes of the normalized frequency distributions are the same as
those of the experiments in each case.
According to numerical experiments of Xie et al. (2006) and Santiago et al. (2008), it is
reported that each building should be resolved by at least 15-20 grid points in each dimension
in order to accurately simulate turbulent behaviors around a building. In our LES model, the
(a) (b) (c) (d) (e) (f)
174

Figure 7: Time series of instantanoues concentration at central street canyon position in wind tunnel
experiments and LESs.

Figure 8: Time series of instantanoues concentration at crossing position in wind tunnel experiments and LESs.

Figure 9: Normalized frequencies of instantaneous concentrations at central street canyon position.

Figure 10: Normalized frequency of instantaneous concentrations at crossing position.
grid points for each obstacle are enough to capture turbulent behaviors (as mentioned in
Section 3.2). However, for denser arrays of obstacles, the differences of r.m.s. concentrations
and variability of concentration fluctuations between the wind tunnel experiments and the
LES model are observed. Important issues still remain in determining appropriate model
constant and the number of grid points for individual obstacles to accurately predict not only
turbulent flows but also unsteady behaviors of a plume within obstacles arrays.
(a) at central street
canyon in
f
=0.16
(b) at central street
canyon in
f
=0.25
(c) at central street
canyon in
f
=0.33
(d) at crossing position
in
f
=0.16
(e) at crossing position
in
f
=0.25
(f) at crossing position
in
f
=0.33
(a)experiment,
f
=0.16
(d)LES,
f
=0.16
(b)experiment,
f
=0.25
(e)LES,
f
=0.25
(c)experiment,
f
=0.33
(f)LES,
f
=0.33
(a)experiment,
f
=0.16
(d)LES,
f
=0.16
(b)experiment,
f
=0.25
(e)LES,
f
=0.25
(c)experiment,
f
=0.33
(f)LES,
f
=0.33
175
5 CONCLUSION
We generated a spatially-developing boundary layer flow by using the existing turbulent
inflow technique and performed LESs of turbulent flow and plume dispersion in obstacle
arrays with various obstacle densities. The approach flow of LES corresponds to a neutral
atmospheric boundary layer based on comparison with the ESDU 85020 recommended data.
Although some of the turbulence characteristics by LES are quantitatively different from
those by the experiment, they both reasonable well model the neutral boundary layer above
rough surface and can be compared taking in account their differences. The distribution
patterns of mean and r.m.s. concentrations and the variability of concentration fluctuations of
LES depending on obstacle densities are similar to those of the experiments. From these facts,
it is considered that our LES model gives satisfactorily results. However, important issues
still remain in determining appropriate model constant and the number of grid points for
individual obstacles to accurately predict not only turbulent flows but also unsteady behaviors
of a plume especially for denser arrays of obstacles.
6 REFERENCES
Bezpalcova,K.,and Ohba,M., 2008. Advective and turbulent vertical fluxes of the passive contaminant inside an
urban canopy, in: Proceeding of 20th National Symposium on Wind Engineering, Tokyo, Japan, 20, 19-24.
Boppana, V.B.L., Xie, Z.T., and Castro, I.P.: Large-eddy simulation of dispersion from surface source in arrays
of obstacles, Boundary-Layer Meteorology, 135, 433-454, 2010.
Branford,S.,Coceal,O.,Thomas,T.G.,and Belcher,S.E.,2011.Dispersion of a point-source release of a passive
scalar through an urban-like array for different wind directions,Boundary-Layer Meteorology,139,3,367-394.
Counihan, J., 1975. Adiabatic atmospheric boundary layers: a review and analysis of data from the period 1880-
1972, Atmospheric Environment, 9, 871-905.
Dejoan, A., Santiago, J.L., Martilli, A., Martin, F., and Pinelli, A., 2010. Comparison between large-eddy
simulation and Reynolds-averaged Navier-Stokes computations for the MUST field experiment. Part2:
Effects of incident wind angle deviation on the mean flow and plume dispersion, Boundary-Layer
Davidson, M.J., Snyder, W.H., Lawson. R.E., and Hunt, J.C.R., 1996. Wind tunnel simulations of plume
dispersion through groups of obstacles, Atmospheric Environment, 30, 3715-3731.
Engineering Science Data Unit., 1985. Characteristics of atmospheric turbulence near the ground Part2 Single
point data for strong winds (neutral atmosphere), ESDU Item, 85020.
Goldstein, D., Handler, R., and Sirovich, L., 1993. Modeling a no-slip flow boundary with an external force
field, Journal of Computer Physics, 105, 354-366.
Grimmond, C. S. B., and Oke, T.R., 1999. Aerodynamic properties of urban areas derived from analysis of
surface forms. Journal of Applied Meteorology, 38, 1262-1292.
Harlow, F., and Welch, J,E., 1965. Numerical calculation of time-dependent viscous incompressible flow of
fluid with a free surface, Physics of Fluids, 8, 2182-2189.
Nakayama, H., and Nagai, H., 2011. Development of local-scale high-resolution atmospheric dispersion model
using large-eddy simulation Part2: Turbulent flow and plume dispersion around a cubical building, Journal
of Nuclear Science and Technology, 48, 3, 374-383.
Kataoka, H., and Mizuno, M., 2002. Numerical flow computation around aeroelastic 3D square cylinder using
inflow turbulence, Wind and Structures, 5, 379-392.
Pascheke, F., Barlow, J.F., and Robins, A., 2008. Wind-tunnel modeling of dispersion from a scalar area source
in urban-like roughness, Boundary-Layer Meteorology, 126, 103-124.
Santiago, J.L., Coceal, O., Martilli, A., and Belcher, S, E., 2008: Variation of the sectioanl drag coefficient of a
group of buildings with packing density, Boundary Layer Meteorology, 128, 445-457.
Smagorinsky, J., 1963. General circulation experiments with the primitive equations, Monthly Weather Review,
91, 3, 99164.
Takewaki, H., Nishiguchi, A., and Yabe, T., 1985. Cubic Interpolated Pseudo-particle method (CIP) for solving
hyperbolic-type equations," Journal of Computer Physics, 61, 261-268.
Van Driest, E.R., 1956. On turbulent flow near a wall, Journal of Aerospace Science, 23, 1007-1011.
Xie, Z.T., and Castro, I.P., 2006. LES and RANS for turbulent flow over arrays of wall-mounted obstacles,
Flow Turbulence Combustion, 76, 291-312.
176
Turbulent flow around a surface-mounted cube: time-
averaged and time-resolved measurements
Enrico Paterna
a
, Peter Moonen
b
, Viktor Dorer
c
, Jan Carmeliet
d,e

a
Chair of Building Physics, ETH Zurich, Zurich, Switzerland,
paterna@arch.ethz.ch
b
Laboratory for Building Science and Technology, Empa, Dubendorf,
Switzerland, Peter.Moonen@empa.ch
c
Switzerland, Viktor.Dorer@empa.ch
d
Chair of Building Physics, ETH Zurich, Zurich, Switzerland,
carmeliet@arch.ethz.ch
e
Switzerland, Jan.Carmeliet@empa.ch
ABSTRACT: Time-averaged and time-resolved PIV measurements of the flow around a
surface mounted cube have been carried out in the new ETH/Empa wind tunnel in Zurich.
The first measurements have characterized the average flow around a cube at several vertical
planes and under several Reynolds numbers. The second measurements have shown the time-
resolved development of multiscale vortical structures downstream of the cube at the vertical
plane crossing the center of the cube.
1 INTRODUCTION
Bluff bodies like cubes or prisms of different aspect ratios are generally considered for wind
engineering studies in urban contexts as simplified model of buildings. Measurements of the
flow around bluff bodies have been extensively studied in the past, by mainly employing
point-wise measurements or flow visualizations initially (Ogawa et al., 1382, Martinuzzi et
al., 1993, Becker et al., 2002, Vardoulakis et al. 2011), while more recently also by applying
time-resolved PIV (Vlachos et al., 2002). Moreover other fundamental studies have stressed
the importance of Reynolds number independency of flows around bluff bodies (Uehara et
al., 2002, Lim et al., 2007).
The aim of the paper is a general characterization of the average flow around a surface-
mounted cube and the understanding of the basic dependency of the flow interaction with the
cube on the Reynolds number of the flow. The measurements presented in this article are part
of a wider measurement campaign, which is in progress at the new ETH/Empa wind tunnel,
applying time-resolved and time-averaged PIV measurements to the study of the flow around
a cube immersed in a turbulent boundary layer. The aim of the campaign is to compare the
results to those obtained by Lattice Boltzmann simulations, which, being computationally
expensive, require low Reynolds numbers of the measurements to be set.

177
2 MEASUREMENT SETUP
The measurements were carried out in the ETH/Empa wind tunnel having a test section 1.9 m
wide, 1.3 to 1.6 m high (by means of an adjustable ceiling for longitudinal pressure control)
and 10.4 m long and allowing both open- and closed loop operations. Flow uniformity is
achieved by several screens and a honeycomb panel while homogeneity is ensured by a row
of turning vanes at each of the four corners of the facility. The blowing fan has a diameter of
1.8 m and a nominal power of 110 kW and it allows free stream velocities from 0.5 to 25 m/s.
The camera used is part of a stereoscopic PIV system of two cameras equipped with a 12 bits
CMOS sensor having maximum resolution of 2016 x 2016 pixels up to 639 Hz, and it was
used with a 135mm F2.0 Canon objective. The camera was placed on a translation stage
installed around the perimeter of the cross section test section of the wind tunnel. The images
were acquired and post-processed by means of the LaVision GmbH software DaVis 7.2. The
flow is seeded by aerosol particles produced from DEHS liquid by a particle generator.
The two laser system used for the measurements are a Nd:YAG laser having 532 nm
wavelength and 100 mJ per pulse, a pulse duration of 5-8 ns, and a Nd:YLF laser with a
wavelength of 527 nm, a maximum energy per pulse of 30 mJ at 1kHz repetition rate and a
pulse duration of 150 ns. The laser beam for both of them is driven through a flexible guiding
arm having at the end of its path a cylindrical lens of -50 mm, producing a laser sheet of a
width of 4 mm.
2.1 Time-averaged flow measurements
The first set of measurements was conducted on the flow around a cube having dimension
H=4 cm mounted on a wooden plate in the wind tunnel. The strong reflections generated by
the laser sheet directed downwards from the wind tunnel top on horizontal surfaces were
hidden by having the model on a flat plate above the wind tunnel floor (Fig. 1) and by placing
the camera slightly below the height of the plate and by turning it in order to avoid the
camera to be exposed to the strong light intensities. The laser system employed was the
Nd:YAG laser, the acquisition frequency of the PIV system was set to 50 Hz, the minimum
allowed by the system, and 1000 images were acquired. The laser sheet width was 4 mm and
the field of view was 196 mm (4.9H) in both dimensions. If considering the Reynolds number
Re
H
based on the cubes height H and free-stream velocity
, two subset of measurements

were conducted, the first subset only on the vertical symmetry plane of the cube and with
Re
H
=1460 11400 in order to check the Reynolds number independency of the flow, the
second one at Re
H
=3290 and on 7 vertical planes spanning from the symmetry plane at z=0 to
z=1.5H (Fig. 2). The Reynolds number chosen for the latter subset of measurements
represented a good compromise between Reynolds number independency and the needs for a
low Re
H
for future comparison with Lattice Boltzmann simulations (Chikatamarla, 2008).

Figure 1: Setup for time-averaged measurement Figure 2: Measurement planes
178
2.2 Time-resolved flow measurements
The second set of measurements was conducted on the flow around a cube having dimension
H=15 cm and placed on the wind tunnel floor. In this case the laser was directed by a mirror
which directed it horizontally from downstream of the cube. If compared to the latter
described, this setup allowed to better resolve the flow attached to horizontal surfaces since
no reflections were obtained on them, but, as a drawback, the flow in the upstream region of
the cube could not be measured since it remained in a shadow. The laser sheet width was 4
mm and the field of view 214 mm in both dimensions, and measurements were taken at
Re
H
=3860 at the vertical symmetry plane at different distances downstream of the cube up to
13 H. The laser system employed was the Nd:YLF, the acquisition frequency of the PIV
system was set to 200 Hz, the minimum allowed by the system, and 1500 images were
acquired. Such, the requirements for, ensemble averaging of turbulence statistics were not
fully met yet, as the intention for the measurements mainly was to verify the capability of
time-resolved PIV to resolve different length- and timescales of the flow around the cube.
2.3 PIV post-processing
For both the measurements the vector field was computed by means of a standard cross-
correlation function via FFT employing multigrid analysis with two refinement steps with
refinement ratio 2 and final interrogation window size of 32x32 pixels with 50% overlapping.
The average dimension of seeded particles in recorded images is around 2-3 pixels thus
avoiding peak locking. The software used applies a Gaussian peak fit as a three-point
estimator for the correlation peak with an order of accuracy of 0.1-0.05 pixels (Tropea et al.,
2007). Validation process included a median filter which compares the median of each vector
with the median of the 8 neighboring vectors to replace bad vectors.
2.4 Space and time scales
A few considerations can be done about the capability of the current PIV system (with the
Nd:YLF laser) installed at the ETH/Empa wind tunnel. Its beyond the scope of this work to
measure the Kolmogorov length- and time scale due to the not sufficient spatial resolution of
the measured data of the first experiment to compute the dissipation rate of the turbulent
kinetic energy . By introducing the strong assumption that the turbulence is isotropic and
that the integral length and time scales of the flow are similar to the characteristic length and
time of the problem, it is possible to draw a few qualitative considerations about the
measurements length and time scales that was possible to resolve, with respect to the smallest
in the flow, the Kolmogorov scales, by considering the following relations (Pope, 2000):

-
(1)

The ratio between the vector grid cell dimension and the Kolmogorov length scale is
only a function of the Reynolds number and the field of view (FOV) of the camera (Fig. 3),
therefore to resolve the length scale as close as possible to the lowest scale in the flow a small
FOV and a low Reynolds number should be considered. The ratio between the inverse of the
PIV acquisition frequency and the Kolmogorov time scale
is depending on the Reynolds

number and on the characteristic length of the problem (Fig. 4), therefore to resolve the time
scales of the flow as close as possible to the Kolmogorov scale a large characteristic length
and a low Reynolds number should be chosen.
179

Figure 3: Length scale ratios Figure 4: Time scales ratios
As a result of that to better resolve the small-scale eddies in the flow it was decided, for the
second set of measurements, to keep a Reynolds number as low as possible and it was chosen
a cube dimension as large and a FOV as small as possible both considering the aim of the
measurements.
3 RESULTS
3.1 Time-averaged flow measurements
3.1.1 Re
H
independency study
From the data measured at different Re
H
it was possible to estimate the critical Re
H
of the
flow around the cube, and it was set for the measurements on several vertical planes around
the cube. If considering the wake downstream of the cube the vertical profiles of
show
high similarity in the range Re
H
=4710-11400 (Fig. 5a) while show differences for profiles in
the range Re
H
=1460-2050 (Fig. 5b) especially in the range y=0.5-1.5H. This phenomenon is
more evident in the wake within H from the leeward surface of the cube and becomes less
evident further downstream.

Figure 5: Vertical profiles of
downstream of the cube

(a)
(b)
180
If considering the vertical profiles of
in the cubes wake (Fig. 6) it is even more evident

how in the range Re
H
=1460-2050 the profiles are separated from the rest of the profiles taken
at higher Re
H
especially from 0.5H from the leeward wall of the cube and further downstream
although with a decreasing magnitude.


At the top of the cube the vertical profiles of
show consistent differences in the range

but its not possible to distinguish a clear Reynolds dependency from them
(Fig. 7).

on the top of the cube

If considering horizontal profiles of vorticity averaged in time and in space in the range
mm, in order to capture the vorticity associated to the horseshoe vortex, and non-
dimensionalized with the cube dimension and the free stream velocity, it was found that when
increasing Re
H
, for Re
H
< 3290, the vortex approaches the windward wall of the cube and
decreases in magnitude (Fig. 8). While the first aspect might be due to the decrease of the
length of the upstream separation region upstream of the cube for an increase of the free
stream velocity, the latter aspect might be mainly due to the higher content of non-
dimensional inlet vorticity at lower free stream velocities (Fig. 9) and might indicate that the
flow in the separation region is Reynolds number independent for Re
H
> 3290.

Figure 8: Horizontal profiles of
Figure 9: Vertical profile of
x=0
at
181
3.1.2 Flow around a cube at Re
H
=3290
The upstream flow while approaching the cube separates from the plate surface showing a
stagnation point on the windward side of the cube at y=0.7 H and generating a recirculation
region below the latter point (Fig. 12a). The vortex thus created, known as horseshoe vortex,
is located on the symmetry plane at 0.3 H upstream of the block. The vorticity peak
associated with the horseshoe vortex is approaching
the windward surface of the cube if moving from
plane at z=0 to the plane at z=0.5H (Fig. 10), and if
considering the mean turbulent kinetic energy the
peak associated to the latter vortex develops from a
circular shape, from plane z=0 (Fig. 12d) to z=0.75
H into a more elongated structure at z=H (Fig. 12e).
This latter observation can be explained by the
horseshoe shape of the vortex which, at z=H is
turning around the cube forming laterally and
downstream of the cube a highly turbulent regions
attached to the bottom surface which is then entrained in the low pressure region downstream
of the cube. At the top edge of the cube on its windward side the flow sees an acceleration
due to the blocking effect of the high momentum flow by the windward surface, and a
recirculation region above the top of the cube. This recirculation region reaches its maximum
extent at the center plane (Fig. 12a) and becomes smaller if moving towards the side of the
cube (Fig. 12b, Fig. 11). If moving laterally further away from the cube at z=0.75H the
average horizontal velocity shows an horizontal acceleration of the flow at 0.3H downstream
of the windward surface due to the lateral corner stream (Fig. 12c). Downstream of the cube
the average flow shows a
recirculation region and a
reattachment length equal to 1.6 H
at z=0 and 1.4 H at z=0.25H. If
moving from z=0 towards
z=1.5H, as the interaction of the
flow gets less influential, the
vertical profiles of
(Fig. 13)
and of
(Fig. 14) tend to the

typical boundary layer profiles.

Figure 12: Contours of
at z=0 (a),
at z=0.5H (b),
at z=H (c),
at z=0 (d),
at z=H (e),
at z=0 (f).
on the top of the cube

Figure 10: Horseshoe vortex displacement
(a) (b)
(e)
(c)
(d)
(f)
182



3.1.3 Uncertainty estimation
Following the central limit theorem the statistical uncertainty of the measured data can be
estimated by the standard deviation of measured data divided by , where N is the
number of independent samples acquired. In computing the standard deviation a confidence
level of 95% was considered corresponding to . If considering a median value of
uncertainty computed for the area of the flow under investigation (width=4.9H, height=2H),
the relative uncertainty of
and
of was found to be within 3%, 20% and 16%

respectively for the whole set of measurements.
3.2 Time-resolved flow measurements
The second set of measurements is a preliminary
demonstration of the time-resolved development of
turbulent eddies on the top and in the wake of a cube.
An higher cube dimension (15 cm) together with an
higher acquisition frequency (200 Hz) allows to better
resolve the turbulent time- and length-scales. In figure
15 an instantaneous flow field of the recirculation
region downstream of the cube is shown, while in
figures 16a-c instantaneous plots of
show the
development of vortexes at the top edge of the
windward wall of the cube.

Figure 16: Instantaneous vorticity fields at t=0s (a), t=0.1s (b), t=0.2s (c)
Figure 15: Instantaneous velocity
magnitude at t=0s
(a) (c) (b)
183
4 DISCUSSION
The results obtained from the first set of measurements show the developments of the main
flow patterns of the flow around a cube, and the dependency on the incoming boundary layer.
The critical Re
H
for the case being studied was found to be in the range
in the
flow downstream of the cube, and, on the top, it was found to be higher, but due the low
number of cases above Re
H
=5x10
3
it was not possible to determine it more precisely.
Moreover, from the upstream vertical and horizontal profiles of non-dimensional vorticity , it
was found that that the critical Re
H
number upstream of the cube is in the range
.
Due to the fact that the cube is not immersed in an equilibrium boundary layer in the wind
tunnel but in a boundary layer generated by the plate on which it is placed, some of the
deductions from this paper are more qualitative than quantitative and require further
measurements with more appropriate settings to be validated and further investigated.
5 CONCLUSION
The flow around a surface mounted cube has been measured in a wind tunnel by using PIV.
Two sets of measurements have been conducted, the first one with the aim of characterizing
the average flow and turbulence statistics around the model and of understanding the
influence on the average flow fields of changing the Reynolds number of the flow. The
critical Re
H
has been found to be in the range 2-3 x 10
3
upstream and downstream of the cube
and higher, but not determined, on the top. Finally the second set of measurements has
qualitatively shown the time-resolved development of multiscale vortical structures due to the
cube-flow interaction at the vertical symmetry plane of the cube.
6 REFERENCES
Becker, S., Lienhart, H., Durst F., 2002. Flow around three-dimensional obstacles in boundary layers. Journal of
Wind Engineering and Industrial Aerodynamics, 90, 265-279.
Chikatamarla S. S., 2008, Hierarchy of Lattice Boltzmann models for Fluid Mechanics, ETH Dissertation No.
17893, Zurich.
Lim, H.C., Castro, I. P., Hoxey, R. P, 2007. Bluff bodies in deep turbulent boundary layers: Reynolds-number
issues. Journal of Fluids Mechanics, 571, 97-118.
Martinuzzi, R., Tropea, C., 1993. The flow around surface-mounted, prismatic obstacles in a fully developed
channel flow. Journal of Fluids Engineering, 115, 85-92.
Ogawa, Y., Oikawa, S., Uehara, K., 1983. Field and winds tunnel study of the flow and diffusion around a
model cube I. Flow measurements. Atmospheric Environment, 17, 1145-1159.
Pope, S. B., 2000, Turbulent Flows. Cambridge University Press, New York.
Tropea C. et al., 2007, Handbook of Experimental Fluid Mechanics, Springer, Berlin
Uehara, K., Wakamatsu, S., Ooka, R., 2003. Studies on critical Reynolds number indices for wind-tunnel
experiments on flow within urban areas. Boundary Layer Meteorology, 107, 353-370.
Vardoulakis, S., et al., 2011, Numerical Model Inter-comparison for Wind Flow and Turbulence Around Single-
Block Buildings. Environmental Modeling and Assessment, 16, 169181
Vlachos, P. P., Hajj, M. R, 2002. A time-resolved DPIV study of the unsteady character of the flow over a
surface-mounted prism. Journal of Wind Engineering and Industrial Aerodynamics, 90, 543-553.

184
Characterization of transient dispersion processes in an
urban environment
F. Harms
a
, D. Hertwig
a
, B. Leitl
a
, M. Schatzmann
a
, G. Patnaik
b

a
Meteorological Institute, KlimaCampus, University of Hamburg,
Bundesstrasse 55, D-20146 Hamburg, Germany, frank.harms@zmaw.de
b
Naval Research Laboratory, Washington DC 20375, USA
ABSTRACT: Validating LES-based flow and dispersion models for the purpose of predicting
transient flow and dispersion phenomena is more demanding than validating RANS-based
codes. Since the model output is no longer related to stationary or quasi-stationary boundary
conditions, and since the model results are not restricted to mean flow and average dispersion
patterns, an evaluation of the model based on mean results is no longer adequate. A more
sophisticated but also more complex validation approach based on statistically representative
ensembles is required. Reference data is needed which reliably identifies mean as well as
extreme values of concentration, dosage and cloud travel times for a given dispersion
scenario. An example of how systematic wind tunnel measurements can characterize transient
dispersion processes of puffs in a complex urban environment is given.
1 INTRODUCTION
Dispersion processes in an urban environment are highly complex. Numerical simulations
that predict these processes require substantial computational effort. In the past simpler
models such as empirical (Gaussian) models, diagnostic models (which use only the mass
conservation equation) or CFD models with full parameterization of turbulence, i.e.,
Reynolds-averaged NavierStokes (RANS) codes were used for these complex tasks.
Nowadays increasing computer power enabled the possibility to use Large Eddy Simulation
(LES) models for urban flow and dispersion simulations. Sagaut (2005) states that Large
Eddy Simulations are an effective intermediate approach between Direct Numerical
Simulations (DNS) and the RANS methods. A basic requirement of any numerical model is
the validation. Validation data for numerical models are not just any experimental data; they
must fulfill certain requirements with respect to completeness, spatial and temporal
resolution, accuracy, representativeness and documentation of the measured results
(Schatzmann and Leitl 2002). If these requirements are not met, too many degrees of freedom
remain in the set up of numerical model runs. A wide variety of numerical results can be
generated with reasonable assumptions for the input data, with the consequence that a solid
conclusion concerning the model quality cannot be reached. Hence validation datasets that
match the complexity of specific groups of models are needed. In order to validate an urban
LES simulation validation data is required that contains data of flow and turbulence fields in
combination with concentration fields measured with high resolution in space and time. Field
measurements do not fulfill these high validation requirements, unless they have been carried
out over long periods of time with many repetitions of individual situations. In the laboratory,
however, and under certain limiting conditions, such datasets can be generated under
carefully controlled conditions in well- equipped boundary layer wind tunnels.
185
The results presented here were obtained as part of a pilot study initiated by the German
Federal Office of Civil Protection and Disaster Assistance to test the LES-based model
FAST3D-CT and the emergency response tool CT-Analyst using Hamburg, Germany as a
pilot city. The models, developed by the Naval Research Laboratory (NRL), were
successfully verified based on several field and laboratory data sets compiled for complex
urban geometries which are typical for cities in the US. Dense built-up areas with tall
buildings and a substantial local variability in building height are characteristic features of
US urban areas. The question however was how the tool would perform in other urban
environments such as typical European cities. Hamburg offers a variety of typical European
urban features as well as some very specific threats resulting from a large harbor area
included in the bounds of the city. The focus of the study is on validating the size of
estimated danger zones and cloud travel time in a typical European city.
2 THE FAST3D-CT AND CT-ANALYST MODEL
FAST3D-CT, a detailed physics-based numerical LES-model (Boris 2002), was developed at
the Naval Research Laboratory to accurately predict plume evolution and the contamination
footprints resulting from these releases. It is a general-purpose fully 3D computational fluid
dynamics (CFD) model for the simulation of transport processes in complex urban
geometries. It is based on the high-resolution, time accurate Flux-Corrected Transport
algorithms developed at NRL.
The model output from the FAST3D-CT CFD calculations are summarized and distilled into
memory efficient, time-independent Dispersion Nomograph
TM
data sets. These are
interpreted and evaluated by CT-Analyst. There are no time-dependent integrations
performed explicitly by CT-Analyst. Instead, all predictions produced by CT-Analyst are
simply the result of applying an interpolation procedure utilizing the appropriate nomograph
data set based upon the high-resolution CFD results. This same interpolation procedure can
be used in both upwind and downwind directions with equal effectiveness. These simple
geometric operations are used to determine the probable source zone upwind of each sensor.
CT-Analyst

provides the capability to immediately backtrack and simultaneously determine
the location of multiple unknown sources simply based on sensor readings and
meteorological parameters. The plume predictions from CT-Analyst, based on a
quantitative Figure of Merit, agree, within 80 to 90%, with the FAST3D-CT CFD simulations
on which they are based and yet are available much faster than corresponding Gaussian
plume estimates (Boris et al., 2002, 2004). Furthermore, the underlying CFD technology is
uniformly convergent so answers automatically get better with increasing computer power
because higher resolution CFD simulations can be used to build the Dispersion Nomograph
data sets (Patnaik et al., 2003).
3 BRIEF OVERVIEW OF THE WIND TUNNEL EXPERIMENTS
In order to create a high quality reference dataset, which is adequate to fulfill major model-
and application-specific validation data requirements for an LES-based, urban flow and
dispersion model, numerous flow and concentration measurements were carried out in a
extended seven month wind tunnel campaign. A 1:350 scale model of the city center of
Hamburg including the harbor area was built for this study. The model covers a total area of
1.4 x 3.7 kilometer. Hamburg was selected as model city in order to analyze dispersion
186
processes in a typical European city. During the first part of the wind tunnel campaign
experimental work focused on the characterization of the modeled approach flow condition as
well as on flow measurements within the model area (Peeck 2011). An extensive set of flow
measurements with high temporal and spatial resolution was carried out. Using 2D Laser-
Doppler- Anemometry (LDA) delivered component-resolved flow data at sampling rates up
to several hundred Hertz under favorable conditions, resolving even small-scale turbulence in
time.
During the second part of the wind tunnel campaign the dispersion of clouds of pollutants
(puffs) was captured and analyzed by corresponding measurements. In order to characterize
the dispersion of puffs properly, large and statistically representative ensembles of puffs were
released and the concentration time traces subsequently analyzed. In order to estimate the
number of repetitive releases required for achieving a reasonable confidence interval of the
ensemble-averaged results a series of pre-tests were carried out. During these pre-tests the
scalability of the puff dispersion results was analyzed regarding the released amount of tracer,
with respect to the wind speed and in relation to the release duration. During these pre-tests
more than 10.000 individual puffs were released. An additional 32.000 puff releases for
different source locations and measurement locations within the modeled area of Hamburg
were carried out in order to create a comprehensive validation dataset.
3. Wind tunnel facility and the Hamburg model
The wind tunnel measurements were carried out in the large boundary layer wind tunnel
WOTAN. A general drawing of the facility is shown in Figure 1. The 25 m long wind
tunnel provides an 18 m long test section equipped with two turn tables and an adjustable
ceiling. The cross section of the tunnel measures 4 m in width and 2.753.25 m in height
depending on the position of the adjustable ceiling.
While in the test section free stream wind speeds of more than 20 m/s can be reached, the
typical wind velocities chosen for atmospheric flow and dispersion modeling are in the range
of 515 m/s. The model boundary layer flow is generated by a carefully optimized
combination of turbulence generators (spires) at the inlet of the test section and a floor
roughness (Peeck 2011).

Figure 1: Drawing of the large boundary layer wind tunnel facility WOTAN of the University of Hamburg
Figure 2 shows the Hamburg model mounted in the wind tunnel. The model is 4 meter wide
and 10.5 meter long, corresponding to an area of 1.4 x 3.7 kilometer at full scale. For
dispersion modeling and measurements several point emission sources were flush-mounted in
the model floor. The circular release area had a diameter of 7 mm (model scale),
corresponding to 2.1 m at full scale. In order to avoid the formation of a significant vertical
jet at higher emission rates, the source area was covered by a lid, 3.5 mm (1.05 m full scale)
above the ground level. In order to simulate instantaneous puff releases, a continuous by-pass
187
flow of tracer gas was temporarily switched to the source by means of a fast solenoid micro-
valve. With this setup, the release rate could be kept absolutely constant for repetitive
releases lasting much less than a second at model scale. The precise repeatability of releases
and the consistency of puff modeling were verified by extensive systematic tests prior to the
experiments. The puff dispersion was measured by a Fast Flame Ionization Detector mounted
to a traverse system. In order to avoid flow disturbances the instrument was located well
above the urban structures.

Figure 2: Wind tunnel model of the city center of Hamburg.
5 BOUNDARY LAYER MODELING
Basic requirement of the wind tunnel measurements was to model a boundary layer flow
similar to the conditions found upwind of the city center of Hamburg. In an iterative process
the shape and arrangement of the spires and the floor roughness elements were varied until
the modeled boundary layer was in reasonable agreement with the full scale conditions
measured at a 300 m tall mast located upstream of the city center of Hamburg (Peeck 2011).
The proper scale of the modeled boundary layer was verified by comparing integral length
scales and spectral distributions of the turbulent kinetic energy with those of the real
atmosphere. A careful adjustment of the modeled boundary layer enabled even large scale
turbulent wind fluctuations up to a time scale of approximately 45 min to be replicated at
scale in the wind tunnel.
6 PUFF DISPERSION MEASUREMENTS
A focus of the project was put on the puff dispersion measurements. One of the specific
features of the generated benchmark database is the provision of systematic and statistically
representative test data for puff dispersion in urban areas. The large ensembles of individual
releases carried out under identical mean wind and release conditions enable a probabilistic
approach to be used for the comparison of wind tunnel data with the corresponding CFD
results or field data. Comparing individual transient puff signals is not an adequate approach
because of the large variation in the shape of time traces. Figure 3 illustrates the variability of
individual puff concentration vs. time traces recorded at a measurement location in the wind
tunnel for seven identical releases. The instants of releases are indicated by the black bars and
the red line states the measured concentration in ppm
v
at the measurement location. The
figure shows that two of the seven released clouds completely miss the measurement location
188
while the measured concentrations for the other five releases differ significantly. Though the
mean boundary conditions were identical for all releases, the observed differences are caused
by the turbulent flow field.

Figure 3: Section of a typical measured concentration signal for seven consecutively released puffs. The release
of tracer for each puff is indicated by the black bar.
Although a mean puff can be defined from a sufficiently large ensemble of releases, the
mean puff is not adequate for comparison with a single release from the field test. A pure
pattern-based comparison is far to elaborate for both the wind tunnel data and the CFD
results. To simplify the comparison and to enable a quantitative comparison, a puff can be
characterized by a number of parameters such as arrival time (at), peak time (pt), leaving time
(lt) and dosage (dos) or peak concentration (pc). Each of the parameters illustrated in Figure
4a can be calculated for each of the individual puff signals recorded at a given measurement
location. For detecting the arrival time and leaving time from a recorded time series a dosage
based method was used. As indicated in Figure 4a the arrival time and leaving time define the
time interval after the release when 5% and, respectively, 95% of the total dosage of a puff
reached the measurement location.
It was found that the dosage based detection method provides, in contrast to other threshold
criteria, a uniform arrival time and leaving time identification for puffs with significantly
different concentrations. Furthermore, a puff can be characterized by the duration (lt-at), the
ascent time (pt-at), and the descent time (lt-pt).
Plotting a sufficiently large ensemble of derived puff parameters, a well-defined and
sufficiently smooth frequency distribution can be achieved. Figure 5b shows a frequency
distribution plot generated from wind tunnel measurements for the arrival time parameter. In
order to estimate the number of repetitive releases required for achieving a reasonable
confidence interval of the ensemble-averaged results, a series of pre-tests were carried out.
For a variety of possible measurement locations, several hundreds of individual releases were
carried out and ensemble-averaged values of puff dispersion parameters were calculated for
gradually increasing ensemble sizes. Additionally for each ensemble size the mean values
were calculated by selecting the results of different puff releases. It was found that a
minimum of about 200 releases were required in order to reach a confidence interval
qualified for model validation while still maintaining a reasonable experimental effort.

189
6
(a) (b)
Figure 4: (a) Puff parameters for statistical study. The arrival time is defined as the time after release when the
dosage exceeds the threshold of 5%. The leaving time is defined as the time when 95% of the total dosage is
reached. (b) Frequency distribution plot for the 'arrival time'-parameter generated by 400 puff releases.

Figure 5 shows a typical result of a convergence analysis for the measured puff travel time. In
this case 300 identical puffs were released from a source located at a crossing in the city
center of Hamburg, modeling a mean wind direction of 235 and a mean wind speed of 2.5
m/s at the height of 80 m above the ground just upstream the city center. The measurement
location was located about 400 meters further downstream. The figure illustrates the
uncertainty in defining the mean arrival time if the ensemble size is limited.

Figure 5: Mean arrival time calculated for different ensemble sizes
As expected the uncertainty in defining the mean value decreases with increase in ensemble
size. In this particular case 200 releases allow defining the mean arrival time with an
uncertainty of 5%. This uncertainty increases to at least 16% if the mean value is
calculated from 50 releases only. For field measurements it has to be considered that due to
changing weather the number of puff measurements that can be carried out under similar
boundary conditions is typically much less than 50 releases.
As illustrated by the black line in Figure 5 the reduction in the uncertainty with increase in
ensemble size can be described by

(1)

190
where y describes for each ensemble size n the maximum difference to the mean arrival time,
which was calculated from all 300 puffs. A and B are the curve parameters whose values may
vary depending on the considered dispersion scenario.
The finding is in line with statistical theory according to which the reduction in the
uncertainty with increase in ensemble size n is proportional to . The wind tunnel
measurements document that the uncertainty in defining mean values depends also on other
parameters. For measurements close to the source location the observed concentration
gradients are stronger and an increased number of releases are required to define the desired
mean values with the same statistical confidence. In addition, it was found that at the same
measurement location the uncertainty range differs for various puff parameters. The dosage
and peak concentration seem to be the parameters with the largest uncertainty levels. This
information is essential when mean values from wind tunnel tests, field experiments, or
numerical results are compared with each other.
7 COMPARISON OF MEASURED AND PREDICTED DANGER ZONES
One objective of the presented study was to validate the predicted danger zones of CT-
Analyst. A danger zone marks the area which can be reached by a released tracer for a
selected source location and a selected mean wind speed. For this analysis two different
source locations and a wind direction of 235 were selected. The tracer was released
continuously during the measurements. Figure 6 shows the result of this comparison for the
two dispersion scenarios.

Figure 6: Comparison of the predicted and measured danger zones for two different dispersion szenarios within
the city center of Hamburg.
The red area indicates in each case the predicted danger zone by CT-Analyst. The triangles
and squares in Figure 6 represent the results of the wind tunnel measurements. A triangle
states that no concentration was detected during a 4 minute wind tunnel measurement and a
square indicates that within the 4 minute measurement a concentration of 5 ppm
V
was
exceeded at least once. Hence the area between a triangle and a square marks the edge of the
wind tunnel plume. It has to be considered that due to the model scale of 1:350 a 4 minute
wind tunnel measurement corresponds to a 24 hour measurement at full scale under identical
weather conditions. In order to analyze the effect of the release rate the measurements were
repeated for different release rates. It was found that increasing the release rate by a factor of
191
ten has no effect to the detected danger zone. Figure 6 shows that the edge of the plume
detected by wind tunnel measurements is in good agreement with the danger zone predicted
by CT-Analyst.
8 CONCLUSIONS
Time-dependent Large Eddy Simulation is a cost effective approach, which has a complexity
in between DNS and the RANS methods. Increases in computing power have enabled LES-
based modeling to be applied on a routine basis for urban flow and dispersion problems.
However, the validation of time-dependent, eddy-resolving LES codes is not as
straightforward as it is for models based on RANS methods. A qualitative and quantitative
evaluation of an LES code requires statistically valid model- and application-specific test
data, and a commonly accepted and scientifically justified validation strategy. Validation
procedures become more complex because comparisons must not just be based on mean
quantities but on frequency distributions of statistically representative ensembles of results as
well. It was found that due to the enormous variability of puff signals, locally measured in an
urban environment, a huge number of repetitions of individual releases under identical mean
boundary conditions are necessary to estimate the bandwidth of possible results for a
particular release configuration. Therefore, validating LES-based numerical models with
results from individual puff releases is not meaningful. As shown, carefully controlled wind
tunnel measurements provide the possibility to estimate the bandwidth of possible dispersion
results even for complex transient dispersion situations. However, sufficiently high
experimental standards have to be met in order to ensure credibility of wind tunnel tests and
to achieve data qualified for a rigorous validation of eddy-resolving CFD models.
9 REFERENCES
Boris, J.P. Obenschain, K., Patnaik, G., and Young, T.R., 2002. CT-ANALYST
TM
, Fast And Accurate CBR
Emergency Assessment, in: Proceedings of the 2
nd
International Conference on Battle Management,
Williamsburg, (VA), USA.
Boris, J.P., 2002. The Threat of Chemical and Biological Terrorism: Preparing a Response, Computing in
Science and Engineering 4(2), 22-32.
Boris, J.P., Fulton, J., Obenschain, K., Patnaik, G., and Young, T.R., 2004. CT-ANALYST, Fast and Accurate
CBR Emergency Assessment, in: Proceedings of the SPIE Defense and Simulation Symposium, SPIE Paper
5416-01, Orlando (FL), USA.
Patnaik, G., Boris, J.P., Grinstein, F.F., and Iselin, J.P., 2003. Large Scale Urban Simulations with the MILES
Approach. AIAA Paper 2003, 4104.
Peek, C., 2011. Influence of urban roughness on mean and turbulent wind fields in the city of Hamburg, in:
Proceedings of the Physmod conference 2011, Hamburg, Germany.
Sagaut, P., 2005. Large Eddy Simulation for Incompressible Flows: An Introduction, third ed. Springer.
Schatzmann, M., and Leitl, B., 2011. Issues with validation of urban flow and dispersion CFD models, Journal
of Wind Engineering and Industrial Aerodynamics, Volume: 99, Issue: 4, 169-186.

192
LES of flow and plume dispersion in actual urban area
in a spatially-developing boundary layer flow
Hiromasa Nakayama
a
, Bernd Leitl
b
, Haruyasu Nagai
c
and Frank Harms
d

a
Japan Atomic Energy Agency, Ibaraki, Japan, nakayama.hiromasa@jaea.go.jp
b
University of Hamburg, Hamburg, Germany, bernd.leitl@zmaw.de
c
Japan Atomic Energy Agency, Ibaraki, Japan, nagai.haruyasu@jaea.go.jp
d
University of Hamburg, Hamburg, Germany, frank.harms@zmaw.de
ABSTRACT: We performed an LES of turbulent flow and plume dispersion in an actual
urban area immersed in a spatially-developing boundary layer flow and examined the basic
performance of the LES model by comparing to wind tunnel experiments. Although some of
the turbulence and dispersion characteristics are quantitatively different from those by the
experiment due to the insufficient grid points for urban buildings and street canyon, the
distribution patterns of turbulent flow and dispersion are generally similar to those of the
experiment. It is considered that our LES model gives satisfactory results.
1 INTRODUCTION
An accurate analysis of atmospheric dispersion is important for emergency responses
against accidental and intentional releases of hazardous and radioactive materials within
populated urban areas. It is a challenging task for higher-level consequence assessments to
quantitatively predict the spatial extent of contaminated regions and the characteristics of
mean and fluctuating concentrations, including the occurrence of peak concentration.
So far, wind tunnel experimental studies of plume dispersion in idealized urban canopy
have been conducted. For example, Davidson et al. (1996) investigated the effects of different
obstacle array configurations on streamwise variations of mean concentrations, and vertical
and spanwise spreads of a plume. Pascheke et al. (2008) investigated the effects of regularly
arrayed obstacles with uniform and variable heights on ventilation of scalars. Bezpalcova and
Ohba (2008) investigated the effects of obstacle arrangements and densities on the
characteristics of mean and root mean square (r.m.s.) concentrations.
Recently, computational fluid dynamics using Large-Eddy Simulation (LES) also has
come to be regarded as a useful tool, with the rapid development of computer technology. For
example, Dejoan et al. (2010) conducted LESs of flow and plume dispersion in an obstacle
array and investigated the effects of incident wind angle deviation on the mean velocity and
mean concentration fields in comparison with the Mock Urban Setting Test field experiment.
Bopana et al. (2010) conducted LES of flow and dispersion in obstacle arrays with uniform
and variable heights and investigated distribution patterns of mean concentrations. Branford
et al. (2011) also conducted LES of flow and plume dispersion in an obstacle array and
investigated the effects of different wind directions on mean concentration. These numerical
studies have implied that LES technique is enough potential for prediction of plume
dispersion in urban-like obstacles.
However, actual urban surface geometries are highly complex because buildings and
obstacles with various shape and size are randomly arranged. The geometries of urban
obstacles employed in these studies are too simplified to directly apply the results of the
193
experimental and numerical studies to real urban settings. In this study, we extend to an LES
of flow and plume dispersion in an actual urban area and examine the basic performance of
the LES model for emergency responses against accidental and intentional releases of
hazardous and radioactive materials by comparing to wind tunnel experiments.
2 NUMERICAL MODEL
The basic equations for the LES model are the spatially filtered continuity equation,
Navier-Stokes equation and the transport equation for concentration. The subgrid-scale
Reynolds stress is parameterized by using the standard Smagorinsky model (Smagorinsky,
1963), where the Smagorinsky constant is set to 0.1 for estimating the eddy viscosity
(Murakami et al, 1987). The subgrid-scale scalar flux is also parameterized by an eddy
viscosity model and the turbulent Schmidt number is set to 0.5 (Nakayama et al., 2011).
The coupling algorithm of the velocity and pressure fields is based on the Marker and Cell
method (Harlow and Welch, 1965) with the second-order Adams-Bashforth scheme for time
integration. The Poisson equation is solved by the Successive Over-Relaxation method which
is an iterative method for solving a Poisson equation for pressure. For the spatial
discretization in the governing equation of the flow field, a second-order accurate central
difference is used. For the dispersion field, Cubic Interpolated Pseudo-particle (CIP) method
proposed by Takewaki et al. (1985) is used for the advection term. CIP is a very stable
scheme that can solve generalized hyperbolic equations in space. For diffusion term, a
second-order accurate central difference method is used.
3 TEST SIMULATIONS
3.1 WIND TUNNEL EXPERIMENTS FOR EVALUATING THE BASIC
PERFORMANCE OF THE LES MODEL
In this study, we evaluate the basic performance of our LES model in comparison with the
experimental datasets from the wind tunnel simulations of turbulent flow and plume
dispersion in an actual urban area, thus, COST Action 732 (http://www.mi.uni-
hamburg.de/Official-Documents.5849.0.html). The experiments were carried out in the new
Large Boundary Layer Wind Tunnel WOTAN at Hamburg University. The study site is
Oklahoma City. In the experiment, the lower part of the neutral atmospheric boundary layer
is simulated by spires set up at the wind tunnel section and floor roughness elements. The
scale of the modeled boundary layer is 1:300, i.e. the lowest 300 m of the boundary layer in
the full scale is modeled. The mean wind velocity vertical profile of approach flow can be
approximated by a power law exponent of 0.18. Wind velocity was measured by 2D Laser
Doppler Anemometry. Concentration is measured using a fast flame ionization detector. In
the wind tunnel experiment, the Reynolds number based on scaling length (L
scale
) of 1m and
wind speed (U
ref
) at the reference height of 0.27m at model scale is about 400,000. U
ref

indicates the undisturbed approach flow at the reference height. The scaling length and the
reference height at model scale correspond to 300m and 81m in the full scale condition,
respectively.
In this study, to evaluate the model performance, we compare our LES results with these
wind tunnel experimental data.

194
3.2 COMPUTATIONAL SETTINGS
Figure 1 shows a schematic illustration of the numerical model. In our LES model, the
driver regions for generation of a spatially-developing turbulent boundary layer flow and the
main analysis region for a simulation of turbulent flow and plume dispersion in an actual
urban area. First, a boundary layer flow is generated by the turbulent inflow technique of
Kataoka and Mizuno (2002) in the driver region1 and then, the wind velocity near the exit of
the driver region1 is imposed at the inlet of the driver region2 at each time step. In the driver
region2, the wind flow with active turbulent fluctuations is reproduced by a tripping fence
and roughness blocks as shown in Fig.1 (a).
In the driver regions, a conventional convective boundary condition is applied at the exit, a
free-slip condition for streamwise and spanwise velocity components is imposed and vertical
velocity component is 0 at the top. A periodic condition is imposed at the side and a non-slip
condition for each velocity component is imposed at the ground surface. Assuming that the
scale of the simulated boundary layer by LES is 300m in the full scale condition, the total
size for the driver regions are 3060m1200m750m in streamwise, spanwise and vertical
directions, respectively. The total number of grid points for the driver regions 51030090.
The Van Driest damping function (Van Driest, 1956) is incorporated to account for near-wall
effects. Building effects are represented by the external force proposed by Goldstein et al.
(1993).
In the main analysis region, the study site of Oklahoma City is set up. At the inlet of the
main analysis region, the turbulent inflow data obtained near the exit of the driver region2 is
imposed. The other boundary conditions in a flow field are the same as those in the driver
regions but the damping function to account for near-wall effects is not incorporated. In a
concentration field, zero gradient is imposed at all the boundaries. The origin is at the
intersection of Park Avenue and Robinson Avenue and a plume source is located near the
Botanical Garden as shown in Fig.2. x, y and z indicate streamwise, spanwise and vertical
directions, respectively. The size and the number of grid points for the main analysis region
are 2000m1200m750m and 45030090 in streamwise, spanwise and vertical directions,
respectively. The mesh size is 4m in the horizontal directions and 1m-36m stretched in the
vertical direction. According to this mesh arrangement, Park Avenue in the range y=-200m-
0m is resolved by 8 grid points in the streamwise direction. Buildings at the upstream and
downstream of Park Avenue in the range y=-200m-0m are resolved by about 12 and 8 grid
points, respectively.
The time step interval tU
ref
/ L
scale
(t: time step) is 0.001. The maximum CFL (Courant-
Friedrich-Levy) number is about 0.15.The length of the simulation run to calculate the time
averaged values of velocity and concentration TU
ref
/L
scale
(T: averaging time) is about 150.
The length of the simulation run before releasing the scalar is TU
ref
/L
scale
is about 75. In the
present LES, the Reynolds number based on scaling length and wind speed at the reference
height is about 10,000.

(a)
195

Figure 1: Schematic diagram of the numerical model. (a) Driver region for generation of a spatially-developing
boundary layer flow. (b) Main analysis region for turbulent flow and plume dispersion in an actual urban area.

Figure 2: Coordinate system. The solid and dot lines indicate Robinson Avenue and Park Avenue. The black
circle indicates the origin located at the intersection of Robinson Avenue and Park Avenue. The star indicates
the location of a point source. The dot-dashed line indicates the area of Botanical Garden. The white circle,
triangle and square indicate the measurement locations of the vertical profiles of turbulent statistics in Figs. 4-6,
respectively. x and y indicate streamwise and spanwise directions, respectively.
4 RESULTS
4.1 APPLOACH FLOW
Figure 3 compares the LES results with the wind tunnel experimental data of turbulence
characteristics of approach flow (COST Action 732) in the full scale condition. The
experimental data are shown with the error bars. The turbulent statistics are normalized by
the reference wind speed. The vertical profile of mean wind velocity of the LES is consistent
with the experimental profile of 0.18 power law. Each component of the turbulence intensity
profiles of the LES is found to be in good agreement with the experimental data although the
LES data are underestimated at heights greater than about 150m. The Reynolds stress profile
of the LES shows a constant profile in the range from 40m to 120m although the LES data
are a little overestimated in that range and more rapidly decrease with height than the
experimental data. According to the review paper of Counihan (1975), it is shown that the
average height of the constant shear stress layer is 100m. The LES data nearly corresponds to
the range shown by Counihan (1975).
Although some of the turbulence characteristics by LES are quantitatively different from
those by the experiment, it is considered that the LES approach flow that corresponds to a
neutral atmospheric boundary layer is obtained.
(b)
196

Figure 3: Turbulence characteristics of approach flow. (a)Mean wind velocity. (b)Streamwise turbulence
intensity. (c)Spanwise turbulence intensity. (d)Vertical turbulence intensity. (e)Reynolds stress.
4.2 TURBULENCE STRUCTURES IN AN ACTUAL URABAN AREA
Figures 4-6 compare the LES results with the wind tunnel experimental data (COST Action
732) of vertical profiles of mean wind velocity (U) and each component for turbulence
intensity (
u
,
v
,
w
) at y=-15m, -90m and -135m, respectively. Each turbulent statistics is
normalized by the reference wind velocity. At y=-15m, mean wind velocity of the LES
slightly shows a negative peak at 45m height, while that of the experiment shows a negative
sharp peak at a height of 40m. Streamwise and vertical turbulence intensities of the LES are
underestimated by more than 50% in the range from 10m to 70m. At y=-90m, mean wind
velocity of the LES is consistent with that of the experiment. Streamwise turbulence intensity
of the LES is underestimated below 40m height although that of the LES is consistent with
that of the experiment above 40m height. Spanwise and vertical turbulence intensities of the
LES are also underestimated below 70m, although the LES data are consistent with the
experiment data above 70m. At y=-135m, mean wind velocity of the LES is consistent with
that of the experiment. Streamwise, spanwise and vertical turbulence intensities of the LES
are generally underestimated.
In our LES model, it is shown that the vertical profiles of mean wind velocity show good
agreement with those of the experiment. However, turbulence intensity for each component
by the LES is underestimated especially within urban canopy, although that of the LES is in
similar to the experimental data above urban canopy. According to numerical experiments of
Xie et al. (2006) and Santiago et al. (2008), each building should be resolved by at least 15-
20 grid points in each dimension in order to accurately complex behaviors of turbulent flows
around obstacles. Kanda (2006) mentioned that the values of turbulent statistics are
underestimated within urban canopy in case the resolution for each building is less than 10
grid points in each dimension. Franke et al. (2007) also described that the resolution for street
canyon should be at least 6-8 grid points. As mentioned in Section 3.2, the grid points for
buildings around Park Avenue in streamwise direction is 8-12 grid points and insufficient to
accurately simulate turbulent behaviors. Although the resolution for street canyon seems to
be marginal, the underestimation of the values of the turbulent statistics is clearly due to the
insufficient grid points for buildings.
On the other hand, Bou-Zeid et al. (2009) and Tseng et al. (2006) mentioned that the basic
turbulent flow patterns around a bluff body can be simulated even with 4 grid points across a
building, although this grid points seems to be marginal. In our LES model, each urban
building and obstacle is clearly resolved by at least 4 grid points in each dimension.
Therefore, the shape of the turbulence intensity profiles is considered to become similar to
that of the experiment.
(a) (b) (c) (d) (e)
197

Figure 4: Turbulence structures at the position of y=-15m. (a)Mean wind velocity. (b)Streamwise turbulence
intensity. (c)Spanwise turbulence intensity. (d)Vertical turbulence intensity.


4.3 DISPERSION CHARACTERISTICS
Figures 7 and 8 compare the LES results with the wind tunnel experimental data of the
spanwise profiles of mean (C
ave
) and r.m.s. (C
r.m.s.
) concentrations at the position of y=-200m-
0m at a ground level and a height of 48m, respectively. The mean and r.m.s. concentrations
are normalized by U
ref
, L
scale
and the source strength (Q). The experimental data are shown
with the error bars. Mean and r.m.s. concentrations of the LES at a ground level gradually
decrease towards y=-200m (r.m.s. concentration shows a slight peak at y=-50m), while the
experimental data generally decrease with small fluctuations. At a height of 48m, mean and
r.m.s. concentrations of the LES also generally decrease towards y=-200m, while those of the
(a) (b) (c) (d)
(a) (b) (c) (d)
(a) (b) (c) (d)
198
experiment rapidly decrease between y=-25m and 0m, and generally decrease with small
fluctuations towards y=-200m. It is found that the LES model cannot capture such local
distributions of mean and r.m.s. concentrations of a plume. This is possibly due to the
insufficient grid points for not only urban buildings but also street canyon. Important issues
remain in determining the appropriate grid points and mesh arrangement for urban buildings
and obstacles. However, the LES data are generally in good agreement with the experimental
data especially at a ground level and the tendency to generally decrease towards y=-200m is
the same as the experiment. From these results, it is considered that our LES model for
turbulent flow and plume dispersion in actual urban area gives satisfactory results.

Figure 7: Mean and r.ms. concentrations at the position of y=-200m-0m at a ground level. (a)mean
concentration. (b)r.m.s. concentration.

Figure 8: Mean and r.ms. concentrations at the position of y=-200m-0m at a height of 48m. (a)mean
concentration. (b)r.m.s. concentration.
5 CONCLUSION
We performed LESs of turbulent flow and plume dispersion in an actual urban area
immersed in a spatially-developing boundary layer flow and examined the basic performance
of the LES model by comparing to wind tunnel experiments. Due to the insufficient grid
points for urban buildings and obstacles, the values of the turbulent statistics of the LES
within urban canopy are underestimated. Furthermore, our LES model cannot capture local
distributions of mean and r.m.s. concentrations of a plume within street canyon. Although
important issues still remain in determining appropriate grid points and mesh arrangement for
urban buildings, the shape of turbulent statistics and mean and r.m.s. concentrations profiles
is similar to that of the experiment. It is considered that our LES model gives satisfactory
results.
(a) (b)
(a) (b)
199

6 REFERENCES
Bezpalcova,K., and Ohba,M., 2008. Advective and turbulent vertical fluxes of the passive contaminant inside an
urban canopy, in: Proceeding of 20th National Symposium on Wind Engineering, Tokyo, Japan, 20, 19-24.
Boppana, V.B.L., Xie, Z.T., and Castro, I.P., 2010, Large-eddy simulation of dispersion from surface source in
arrays of obstacles, Boundary-Layer Meteorology, 135, 433-454, 2010.
Bou-Zeid,E., Overney,J., Rogers, B,D., and Parlange, M,B., 2009, The Effects of Building Representation and
Clustering in LESs of Flows in Urban Canopies, Boundary Layer Meteorology,132,415436.
Branford,S., Coceal, O.,Thomas,T.G., and Belcher,S.E.,2011. Dispersion of a point-source release of a passive
scalar through an urban-like array for different wind directions,Boundary-Layer Meteorology,139,3,367-394.
COST Action 732, 2005-2009, Quality assurance and improvement of micro-scale meteorological models.
http://www.mi.uni-hamburg.de/Official-Documents.5849.0.html.
Counihan, J., 1975. Adiabatic atmospheric boundary layers: a review and analysis of data from the period 1880-
1972, Atmospheric Environment, 9, 871-905.
Dejoan, A., Santiago, J.L., Martilli, A., Martin, F., and Pinelli, A., 2010. Comparison between large-eddy
simulation and Reynolds-averaged Navier-Stokes computations for the MUST field experiment. Part2:
Effects of incident wind angle deviation on the mean flow and plume dispersion, Boundary-Layer
Davidson, M.J., Snyder, W.H., Lawson. R.E., and Hunt, J.C.R., 1996. Wind tunnel simulations of plume
dispersion through groups of obstacles, Atmospheric Environment, 30, 3715-3731.
Engineering Science Data Unit., 1985. Characteristics of atmospheric turbulence near the ground Part2 Single
point data for strong winds (neutral atmosphere), ESDU Item, 85020.
Franke, J., Hellsten, A., Schlnzen, H., and Carissimo, B., 2007: Best Practice Guideline for the CFD simulation
of flows in the urban environment. COST Office Brussels.
Goldstein, D., Handler, R., and Sirovich, L., 1993. Modeling a no-slip flow boundary with an external force
field, Journal of Computer Physics, 105, 354-366.
Harlow, F., and Welch, J,E., 1965. Numerical calculation of time-dependent viscous incompressible flow of
fluid with a free surface, Physics of Fluids, 8, 2182-2189.
Kanda, M., 2006, Large-eddy simulations on the effects of surface geometry of building arrays on turbulent
organized structures, Boundary Layer Meteorology, 118, 151-168.
Nakayama, H., and Nagai, H., 2011. Development of local-scale high-resolution atmospheric dispersion model
using large-eddy simulation Part2: Turbulent flow and plume dispersion around a cubical building, Journal
of Nuclear Science and Technology, 48, 3, 374-383.
Kataoka, H., and Mizuno, M., 2002. Numerical flow computation around aeroelastic 3D square cylinder using
inflow turbulence, Wind and Structures, 5, 379-392.
Pascheke, F., Barlow, J.F., and Robins, A., 2008. Wind-tunnel modeling of dispersion from a scalar area source
in urban-like roughness, Boundary-Layer Meteorology, 126, 103-124.
Santiago, J.L., Coceal, O., Martilli, A., and Belcher, S, E., 2008: Variation of the sectional drag coefficient of a
group of buildings with packing density, Boundary Layer Meteorology, 128, 445-457.
Smagorinsky, J., 1963. General circulation experiments with the primitive equations, Monthly Weather Review,
91, 3, 99164.
Takewaki, H., Nishiguchi, A., and Yabe, T., 1985. Cubic Interpolated Pseudo-particle method (CIP) for solving
hyperbolic-type equations," Journal of Computer Physics, 61, 261-268.
Tseng, Y. H., Meneveau, C., and Parlange, M, B., 2006, Modeling flow around bluff bodies and predicting
urban dispersion using large eddy simulation, Environmental Science and Technology, 40, 26532662.
Van Driest, E.R., 1956. On turbulent flow near a wall, Journal of Aerospace Science, 23, 1007-1011.
Xie, Z.T., and Castro, I.P., 2006. LES and RANS for turbulent flow over arrays of wall-mounted obstacles,
Flow Turbulence Combustion, 76, 291-312.
200
On aspects of LES validation for urban flow fields
D. Hertwig
a
, F. Harms
a
, G. Patnaik
b
, M.-Y. Obenschain
b
, B. Leitl
a
, M. Schatzmann
a

a
Meteorological Institute, KlimaCampus, Hamburg University, Germany
denise.hertwig@zmaw.de
b
Laboratory for Computational Physics and Fluid Dynamics, U.S. Naval
Research Laboratory, Washington D.C., USA
ABSTRACT: In a systematic study wind-tunnel measurements and large-eddy simulation
(LES) results of turbulent flow in the inner city of Hamburg, Germany, are compared. The
focus of the validation exercise is the comparison of time-series information and the characte-
rization of turbulent flow structures within and above the urban canopy. On the basis of
densely spaced measurements in vertical profiles and horizontal flow layers the developments
of the turbulent boundary-layer within the city as well as typical street-canyon flow scenarios
provide the framework for this analysis. Particular challenges with respect to the validation of
time-resolved simulations in contrast to standard approaches are discussed with an emphasis
on specific demands in the case of urban flow fields.
1 INTRODUCTION
The continuing increase of computational capacities rapidly augmented the use of time-
resolved numerical approaches like LES for the prediction of flow and dispersion fields in
complex urban geometries. In contrast to numerical codes based on Reynolds-averaged con-
servation equations (RANS) these eddy-resolving approaches have the capability to adequate-
ly reproduce spatially complex turbulent flow regimes together with their time evolution.
Within the urban canopy layer unsteady flow effects are strongly enhanced by the presence of
buildings, leading to flow situations that could formerly not be described by numerical ap-
proaches (cp. recent urban LES studies by Xie and Castro, 2009; Letzel et al., 2008, or Pat-
naik et al., 2007).
Verification of the numerical result against suitable reference data is a crucial step in order to
establish credibility of the prediction and assess its reliability for cases in which the truth is
not known a priori. In this context, the accuracy of the simulation in terms of expectable
bounds of uncertainty should be determined primarily by quantitative means. The thorough
review by Oberkampf and Trucano (2002) addresses these points in detail.
The physical character of LES adds new aspects to the validation problem. Together with the
gain of information about the flow, in particular with respect to its turbulent eddy structures,
there is an increasing demand on the quality and quantity of reference data. Aspects of valida-
tion data requirements for LES in contrast to RANS are for example discussed by Adrian et
al. (2000) and Kempf (2008). The strategies pursued in the model validation itself have to go
beyond a pure comparison of statistical moments but should additionally provide an assess-
ment whether the simulation reproduces the spatio-temporal behavior of turbulent eddies rea-
listically. However, so far there is no consensus about standards for such an elementary LES
validation that would really give consideration to this issue. It could be demonstrated, how-
ever, that mathematical tools from the field of advanced signal analysis and pattern recogni-
tion might have the potential to establish a basis for comparisons between experiment and
LES simulation (cp. Hertwig et al., 2011).
201
The aim of the present study is the validation of an urban LES code based on information
drawn from numerical and experimental time series that were obtained in wind-tunnel mea-
surements. Specifically it is focused on comparing time characteristics of the flow associated
with the presence of energy-dominating eddy structures that should be well reproduced in an
LES simulation.
2 EXPERIMENTAL & NUMERICAL METHODS
Laboratory measurements in specialized boundary-layer wind tunnels can provide an ideal
validation data basis supplementary to information from field sites. Well definable and con-
trollable boundary conditions together with the potential to repeat experimental runs under
the same constraints as often as required result in high statistical confidence levels of the
measured quantities.
2.1 Wind-tunnel measurements
The reference measurements for this study were performed in the boundary-layer wind tunnel
WOTAN at the University of Hamburg.
The wind-tunnel model comprises the city center of Hamburg together with industrial harbor
sites that are separated from the downtown area by the river Elbe. In total, the model domain
encompasses an area of 3.7km x 1.4km in full-scale dimension (compare extent of the inner
frame in Fig. 1a). The physical model was built on a scale of 1:350, including terrain and a
3.5m high water front. Effects of urban greenery are not accounted for. Figure 1b shows the
buildings incorporated in the wind tunnel on a model area of 42m.
The flow is approaching from the southwest (235), mirroring a quite frequent meteorological
condition for that area, and was physically modeled to feature urban (i.e. very rough) turbu-
lence characteristics (~0.29; z
0
~1.5m) under neutral atmospheric stratification.
All flow measurements were conducted by non-intrusive laser Doppler velocimetry. The in-
flow was constantly monitored and documented through Pitot tube measurements in the first
section of the tunnel. For further details about the measurement campaign and specifics about
the urban flow characteristics it is referred to Peeck et al. (2011).
2.2 Numerical simulation
Numerical results are obtained from simulations with the urban aerodynamics LES model
FAST3D-CT that handles the dynamical effects of sub-grid scales implicitly through numeri-
cal diffusion. The model is developed and operated by the U.S. Naval Research Laboratory
and is based on a monotone integrated large-eddy simulation (MILES) methodology that of-
fers high computational efficiency. Details on physics and numerics within FAST3D-CT are
given in Patnaik and Boris (2010).
The 3D CFD simulation for Hamburg was performed on a 4.0km x 4.0km region of the inner
city with a 2.5m grid resolution (cp. outer frame in Fig. 1a). The calculation was run on 62 or
64 CPUs of a SGI Altix computer, took over three weeks for 350,000 time steps at
0.05sec/time step generating over 4 hours of real time data. The average wind direction is
235 rotated clockwise from due south. The wind speed was approximately 7.0m/s at a height
of 190m. To match the FAST3D-CT conditions with the wind-tunnel experiments as closely
as possible, all temperature related effects such as buoyancy and surface heating as well as
drag effects of trees have been turned off. Time-dependent wind data were collected every
0.5 seconds for over 4 hours at various heights up to 130 meters.
202

Figure 1: (a) Google Earth image showing the wind-tunnel model domain (inner frame) and the simulation
region for FAST3D-CT (outer frame) of the inner city part of Hamburg. (b) Wind-tunnel model with an
indication of the reference location (framed dot) above the river.
2.3 Data selection & preparation
For the validation exercise 22 measurement locations within the model domain were chosen
for which highly resolved time series of the horizontal wind components (and partly also for
the vertical component) are available in densely spaced profiles and horizontal flow layers.
The selection was made to include areas of the city that feature characteristic urban flow situ-
ations that also pose challenges to numerical models. Thus, the locations also include narrow
street canyons, complex intersections, and measurement points close to the ground.
Wind detectors in the numerical calculations were deployed to match the specified locations
in the wind-tunnel experiment as closely as possible. The nearest neighbor extraction was
chosen in order to avoid contamination of the results by interpolating data in order to have an
exact spatial match. This procedure led to slight offsets of the x, y, and z positions of the
comparison points that were in the range of a few centimeters up to a maximum of 1.5m.
Experimental and numerical data were homogenized by referencing all velocities and their
derivatives to a reference wind speed at a fixed location. This monitoring point was defined at
a height of 49m above the river Elbe at approximately 1km upstream distance from the city
center (see indication of that location in Fig. 1b).
3 VALIDATION RESULTS
First results of the validation study are presented in the next sections. Although the emphasis
of the analyses is put on the comparison of time-series characteristics, the starting point of the
study was set by the validation of the mean flow. Examples of these results will be presented
in the following paragraph.
3.1 Mean flow comparison
Figure 2 shows comparisons of vertical profiles of the streamwise velocity component from
wind-tunnel measurements and FAST3D-CT simulations. The profile locations differ in the
arrangement of the surrounding buildings. Figure 2a shows velocity profiles above the river
Elbe (the location is identical with the reference point indicated in Fig. 1b). Being situated
well upstream of the densely built-up city center the good agreement between experimental
and numerical profiles mirrors a good match of the mean inflow conditions.
(a) (b)
reference location
above river
203

Figure 2: (a)-(f) Comparison of mean streamwise velocity profiles from wind-tunnel measurements (symbols)
and numerical simulations with FAST3D-CT (lines) at different locations within the city. Scatterbars attached to
the experimental values represent the reproducibility of the data based on repetition measurements. The
incoming flow is from left to right. Area images are extracted from Google Earth.
A good agreement is also found for positions at which the flow is strongly influenced by the
building structure. Figure 2b shows a profile measured in a very narrow street canyon. In
Figure 2c the measurement position is located in an open plaza exhibiting a strong recircula-
tion regime that is captured quite well by the code. Measurements shown in Figures 2d-f were
conducted at intersections that trigger complex flow behavior. At elevations below the mean
building height (approx. H
mean
~35m by averaging over the city center) there is a slight trend
towards an underprediction of velocities, whereas higher wind speeds than in the reference
are observed at heights larger than 2.5H
mean
. The slight offsets observed within the street can-
yon might be explained by the close proximity of building walls and the effect of their physi-
cal treatment inside the simulation. The stronger acceleration well above the canopy has to be
investigated further and might reflect an excess of TKE in the numerical inflow prescription.
3.2 Time-series analysis
Next, experimental and numerical time series were analyzed in terms of frequency distribu-
tions, energy spectra, and joint time-frequency characteristics of the signals. It has to be noted
that both signals differ in their length and their time resolution under full-scale conditions.
While the 170s measurement time in the wind tunnel results in a full-scale length of 16.5h,
the length of the numerical time series is 4.5h. Especially at low elevations within street can-
yons the full-scale temporal resolution of 2Hz of the FAST3D-CT signals is better than the
scaled wind-tunnel data rate that is strongly affected by the local seeding conditions.
3.2.1 Wind roses
Figure 3 shows wind-rose diagrams of horizontal wind speeds and directions observed (Fig.
3a) and simulated (Fig. 3b) at four different heights within and above a street canyon.
(a) (b) (c)
(d) (e) (f)
204
Wind tunnel

FAST3D-CT

Figure 3: Wind-rose diagrams showing frequency distributions of horizontal wind speeds and wind directions
for wind-tunnel measurements (a) and FAST3D-CT simulations (b) at four different heights within and above a
street canyon (same location as in Fig. 2f). Arrows on the left indicate the inflow direction.
An indication of the measurement location is given in Figure 2f together with the mean
streamwise velocity profiles. The wind-rose bars display the fractional frequency at which
certain wind speeds (color-coded) were observed from the respective class of wind directions.
At first view the graphs show that the model predicts the deflection of wind directions inside
the canopy quite well, together with the adjustment to the wind direction of the inflow at
roof-top level and well above at 57.75m (i.e. 1.65H
mean
). The spread about the central direc-
tion is largest at roof-top height and smallest at the highest elevation in both the experiment
and the simulation. However, discrepancies in velocity magnitudes are observed inside the
canopy, especially for the lowermost point at 2.5m and 2.75m, respectively. As discussed ear-
lier in connection with the mean flow validation, the lower magnitudes are most likely due to
the influence of wall boundary conditions prescribed at the ground and at upright building
surfaces. Despite these differences the analysis indicates that the LES code is able to repro-
duce the directional fluctuation levels caused by unsteady flow effects quite reliably.
3.2.2 Turbulence spectra
Auto-spectral energy densities of the turbulent streamwise velocity component are studied in
order to analyze the spectral content associated with different eddy structures found in the
flow. The spectra were obtained using an FFT algorithm. In order to make the spectra inter-
pretable in terms of characteristic energetic ranges, two averaging techniques are used. First,
the time series is separated into fragments of equal lengths and it is averaged over the spectra
obtained from these sub-samples. Next, this averaged spectrum is smoothed by taking the
mean over equal intervals with respect to the logarithm of frequency. Original values are only
kept for the lowest frequencies that are connected to the largest structures in the flow.
Figures 4a-d show scaled frequency spectra obtained from numerical and experimental veloc-
ities at various locations at heights of 17.5m (~0.5H
mean
) and 45.5m (~1.3H
mean
), respectively.

235
(a)
(b)
205

Figure 4: (a)-(d) Autospectral energy densities of the fluctuating streamwise velocity component from wind-
tunnel measurements and simulations with FAST3D-CT at various locations within the city at heights of 17.5m
(~0.5H
mean
) and 45.5m (~1.3H
mean
). The dashed lines separate the low frequency parts of the spectra that can be
directly resolved by the numerical model given the grid resolution of =2.5m and the respective mean wind
speeds from the subgrid-scales affected by numerical diffusion. Area images are extracted from Google Earth.
A very good agreement of the production and energy-containing range of the spectra is found
at all positions. The energetic peaks associated with integral length scale eddies coincide very
well for the measurements shown in Figures 4b and 4d, whereas at the other positions the
peaks are shifted for more than a decade towards higher frequencies. In order to investigate
this further, next analyses will concentrate on comparisons of integral length scales that can
be determined from autocorrelation time scales invoking Taylors hypothesis.
Common to all of the numerical spectra is their fast roll-off in the high frequency range that
marks the onset of the influence from the dissipation scheme. At most of the investigated lo-
cations this influence becomes noticeable approximately one decade after the spectral peak
was reached resulting in a shortened extent of the inertial range.
In consideration of the fact that FAST3D-CT was particularly designed to simulate dispersion
processes in urban areas, the very good match of the energy-containing ranges associated
with eddies that play a dominant role for scalar transport confirms the models fitness for that
purpose. However, it should be studied whether an extension of the inertial range is possible
in order to add to the physical character of the LES.
3.2.3 Wavelet transform
The continuous wavelet transform (CWT) is a representative of joint time-frequency analysis
methods whose capabilities in the field of turbulence research and coherent structure detec-
tion were thoroughly investigated by Farge (1992).
The CWT of a time-dependent, square integrable 1D function u(t) is given by the convolution
(a)
(c)
(b)
(d)
206
of the signal and the family of so-called wavelet functions
s,n
:

(1)

where the asterisk denotes the complex conjugate, n refers to the translation parameter and
s>0 to the scale parameter (Addison, 2002). Through n, the wavelet function is translated in
time, covering different parts of the signal. By adjusting the scale, the wavelet function can
be compressed or stretched, acting as a mathematical microscope that zooms in and out of
signal features to resolve components of high and low frequencies. The scale is thus inversely
proportional to the frequency. The wavelet coefficients W
n
(s) simultaneously comprehend
time and frequency information of u(t).
The CWT is applied to numerical and experimental time series of the streamwise velocity
component using the Mexican-hat function as the mother wavelet. The numerical imple-
mentation follows Torrence and Compo (1998) for a computation in Fourier space.
In order to make both signals comparable in the time-frequency domain, dimensionless times
and sampling frequencies were adjusted. Figure 5 shows wavelet coefficients of the stream-
wise velocity component determined through Equation (1) from a wind-tunnel time series
(Fig. 5a) and from a FAST3D-CT simulation (Fig. 5b). The signals were sampled at a height
of 45.5m. The location is the same as in Figure 2b and the corresponding energy spectrum is
shown in Figure 4d. The coefficients are presented in a non-dimensional time/scale frame-
work. Clearly noticeable are large scale (i.e. low frequency) undulation pattern found for both
signals that are associated with large eddy structures passing the sensors and that could be
successfully separated from high frequency noise. Future analyses could now concentrate
on determining and comparing the frequency of occurrence of these large eddies and studying
their energetic contributions in terms of wavelet variances.
4 DISCUSSION & OUTLOOK
This study identified possible strategies concerning an in-depth LES validation. Wind-tunnel
measurements of flow fields within a genuine physical model of the downtown area of Ham-
burg, Germany, provided the reference basis for the validation of the implicit LES model
FAST3D-CT.

Wind tunnel FAST3D-CT

Figure 5: Contour plots of wavelet coefficients (lower panels) from the fluctuating streamwise velocity
component (upper plots) measured in the wind tunnel (a) and simulated by FAST3D-CT (b). The measurement
position is at a height of 45.5m. The line marks the cone of influence above which coefficients are not reliable.
(a) (b)
207
The focus of the analysis was put on the extraction of information from numerical and expe-
rimental time series in terms of histograms, turbulent energy spectra and joint time-frequency
information. Performed in terms of a blind test, the study documented that the code is able
to capture the effects of dominant eddy structures in terms of wind fluctuation levels and as-
sociated energetic properties of the turbulent flow. First results of a wavelet analysis of the
signals showed that the time evolution of turbulent structures can be tracked down, offering
great potential to validate the model in terms of its time-dependent characteristics. The next
steps of the study will address this topic in more detail. Other statistical measures that provide
insight into turbulent eddy characteristics (e.g. autocorrelation time scales or Reynolds stress
components) are subjects for further studies.
In addition to the definition of new validation standards there is also a need to agree upon
quantitative measures of the model performance in order to assess whether the results are ac-
ceptable for the respective purpose. Similarly to the approaches taken in the case of micro-
scale atmospheric RANS models (see Schatzmann et al., 2010) a compilation of new best
practice directives for LES validation should be considered in the future.
5 ACKNOWLEDGEMENTS
Financial support by the German Federal Office of Civil Protection and Disaster Assistance
as well as by the Ministry of the Interior of the City of Hamburg is gratefully acknowledged.
Parts of the wind-tunnel model construction were financially supported by the KlimaCampus
at the University of Hamburg.
6 REFERENCES
Addison, P.S., 2002. The illustrated wavelet transform handbook. Institute of Physics Publishing, London.
Adrian, R.J., Meneveau, C., Moser, R.D., and Riley, J.J., 2000. Final report on Turbulence Measurements for
LES workshop. TAM report No. 937, University of Illinois at Urbana-Champaign.
Farge, M., 1992. Wavelet transforms and their application to turbulence. Annual Review of Fluid Mechanics,
24, 395-457.
Hertwig, D., Leitl, B., and Schatzmann, M., 2011. Organized turbulent structures Link between experimental
data and LES. Journal of Wind Engineering and Industrial Aerodynamics, 99, 296-307.
Kempf, A.M., 2008. LES validation from experiments. Flow, Turbulence and Combustion, 80, 351-373.
Letzel, M.O., Krane, M., and Raasch, S., 2008. High resolution urban large-eddy simulation studies from street
canyon to neighbourhood scale. Atmospheric Environment, 42, 8770-8784.
Oberkampf, W. L., and Trucano, T. G., 2002, Verification and validation in computational fluid dynamics.
Progress in Aerospace Sciences, 38, 209-272.
Patnaik, G., Grinstein, F.F., Boris, J.P., Young, T.R., and Parmhed, O., 2007. Large-scale urban simulations, in:
Grinstein, F. F., Margolin, L. G., and Rider, W. J., (Eds.), Implicit Large Eddy Simulation: Computing Tur-
bulent Fluid Dynamics, Cambridge University Press, pp. 502-530.
Patnaik, G., and Boris, J.P., 2010. FAST3D-CT: An LES model for urban aerodynamics, in: Proceedings of the
5th International Symposium on Computational Wind Engineering, Chapel Hill (NC), USA.
Peeck, C., Leitl, B., and Schatzmann, M., 2011. Influence of urban roughness on mean and turbulent wind fields
in the city of Hamburg, in: Proceedings of PHYSMOD2011 International Workshop on Physical Model-
ling of Flow and Dispersion Phenomena, Hamburg, Germany.
Schatzmann, M., Olesen, H., and Franke, J., (Eds.), 2010. COST 732 model evaluation case studies: Approaches
and results, University of Hamburg.
Torrence, C., Compo, G.P., 1998. A practical guide to wavelet analysis. Bulletin of the American Meteorologi-
cal Society, 79, 61-78.
Xie, Z.-T., and Castro, I.P., 2009. Large-eddy simulation for flow and dispersion in urban streets. Atmospheric
Environment, 43, 2174-2185.
208


Compiling an LES-specific validation data base from
systematic wind tunnel modeling of flow and dispersion
phenomena within the lower atmospheric boundary layer
Rasmus Fischer, Ilona Bastigkeit

, Bernd Leitl, Michael Schatzmann

Meteorological Institute, Klima Campus, University of Hamburg, Germany,
rasmus.fischer@zmaw.de
ABSTRACT: In order to develop and improve new concepts for grid-adaptive eddy-
resolving numerical simulations of natural flow phenomena, the German Research Foun-
dation (DFG) established the priority program MetStrm. The program is intended to
bundle the specific expertise currently available in the fields of mathematics, engineering
and natural sciences, fostering the further development of the Large Eddy Simulation
(LES) approach. As part of MetStrm, the Environmental Wind Tunnel Laboratory
(EWTL) at Hamburg University is tasked to generate data qualified for the validation of
LES models. Based on the conceptual work provided by Bastigkeit et al [1], an extensive
set of experiments has been designed and partially already realized. Measurements of suf-
ficiently high temporal and spatial resolution were carried out in two boundary layer wind
tunnel facilities (WOTAN and BLASIUS) using mainly Laser-Doppler-Anemometry
(LDA) and Particle-Image-Velocimetry (PIV) techniques. Adopting the style of the al-
ready existing validation data base for RANS-type models (CEDVAL), systematic test
data sets were generated at different levels of complexity. The paper illustrates the de-
manding experimental efforts needed to provide reference data qualified for the validation
of LES-based codes. Special attention is given to data quality assurance procedures and
an automated post-processing of the experimental results.
1 INTRODUCTION
Numerical simulations of flow and dispersion phenomena in urban environments are very
complex and require comprehensive computational efforts. In the past it was common
practice to use simple models such as analytical (Gaussian) models, diagnostic models,
based on the mass conservation equation only or computational fluid dynamics (CFD)-
models with full parameterization of turbulence (RANS-models) for these complex tasks.
While such models are able to compute mean urban flow and dispersion within a reason-
able time, they are unable to replicate the inherently unsteady eddy and plume dynamics
driven by the urban geometry. The highest accuracy of numerical simulation in computa-
tional fluid dynamics is provided by Direct Numerical Simulations (DNS). Unfortunately,
they are prohibitively expensive for most practical flows at moderate-to-high Reynolds
numbers, and especially so for urban flow and dispersion studies. An effective intermedi-
ate between DNS and RANS is provided by large eddy simulation approaches (LES).
LES-models only simulate the large eddies directly, whereas the effect of the smaller tur-
bulent structures is still parameterized. LES provides higher accuracy than the usual
RANS methods at much lower costs than a direct numerical simulation. The increasing
computer power enables the use of LES models for urban flow and dispersion simula-
tions. However, particularly for the validation of numerical models simulating transient
atmospheric flow and dispersion phenomena, the provision of qualified experimental data
is still required.
209


Within `MetStrm` the EWTL at Hamburg University is tasked to generate data with
high spatial and temporal resolution, qualified for the validation of LES-models. For this
purpose, a comprehensive concept of requirements for LES-specific validation data has
been developed (Bastigkeit [1]) and an extensive set of experiments was designed. Thus
for, measurements were carried out at different scales using mainly LDA and PIV meas-
urement techniques. Adopting an established validation data base concept for RANS-type
models (CEDVAL), the post processed data are published in a new validation data base
specifically serving the needs of LES-models (CEDVAL-LES). Similar to the existing
CEDVAL, CEDVAL-LES also provides test data sets at different levels of complexity.
For this reason, measurements in MetStrm are carried out at four different levels of test
complexity. The scope of experiments is ranging from undisturbed boundary layer flows
without any obstacles to only slightly idealized models of typical European urban geome-
tries (Michel-Stadt). The much higher demands on validation data for LES codes com-
pared to RANS-models are expressed in a significantly higher quality as well as a large
quantity of required test data. Qualified test data sets must contain sufficiently long time
series of test data, providing an adequate temporal resolution as well as a high enough
spatial resolution. In order to detect coherent structures within turbulent flows, correlated
LDA measurements are desired as well as a mapping of turbulent structures by PIV. To
handle the resulting huge amount of data and to allow for data quality assurance, the post
processing of experimental results had to be standardized and automated. The software
tools developed within the scope of the project allow data verification and quality testing
to be carried out immediately after individual test runs.
In the following paragraphs the conception of the different levels of complexity will be
explained in detail and selected experimental setups and test series as well as the proce-
dure of data quality assurance and data post processing will be illustrated.
2 COMPLEXITY OF TEST CASES
According to Bastigkeit et al [1], validation data bases specifically designed for LES-
models are supposed to include test data sets at different levels of physical complexity. A
structured set of test cases at different levels of complexity enables codes to be validated
during the process of model development. The increase in test case complexity towards a
real urban geometry should be realized in well-defined and small enough steps.
Within the DFG-priority program MetStrm, Bastigkeit [1] developed a comprehen-
sive scheme of requirements for LES-validation. The methodology of increasing com-
plexity is taken on and several levels of test case complexities are defined. The actual ex-
periments have been designed to form four different levels of complexity:

- Complexity 0
As requested by cooperating LES-modelers the lowest level of complexity was defined as
a neutrally stratified, undisturbed boundary layer flow similar to different full-scale
roughness conditions. In these cases, the wind tunnel floor is only covered by roughness
elements adequate for boundary-layer modeling. In contrast to higher levels of complexity
there are no individual obstacles mounted in the test section.
- Complexity 1
At the second level of complexity flow patterns around individual obstacles are analyzed.
Corresponding obstacles are supposed to have a very simple geometry and the tests will
be carried out within the model boundary layer flows generated at complexity level 0.
Common structures to be tested are cubes and cube-like houses with slanted or flat roofs.
- Complexity 2
The third level of complexity is characterized by more or less regular arrays of structures
then as tested at level 1. This test case will not require all individual obstacles to have ex-
210


actly the same geometry. For example, systematic variations in the roof-shape are al-
lowed. With regard to real urban structures, this level of complexity already represents a
complex idealized urban geometry.
- Complexity 3
The fourth level of complexity is defined to be formed by urban structures less idealized
and more realistic than at level 2. For this higher level of complexity Bastigkeit [1] de-
signed a virtual urban geometry based on building shapes and aspect ratios as they are
typical for central-European cities. The synthesized city quarter will enable systematic
variations of the urban geometry with respect to building heights, roof shapes and other
parameters dominating flow and dispersion patterns within urban geometries.

Figure 1: examples for the different levels of complexity realized within the MetStrm-project (top left:
level 0; top right: level 1; bottom left: level 2; bottom right: level 3)
To date (May 2010) four known experimental campaigns have been carried out in two
different boundary layer wind tunnel facilities. While the test section of the larger facility
WOTAN measures 18 m in length, 4 m in width and 2.75 to 3.25 m in height, the test
section of the smaller wind tunnel BLASIUS is 11.5 m long, 1.5 m wide and approxi-
mately 1 m in height. Both facilities are equipped with one (BLASIUS) and two (WO-
TAN) turn tables in order to model different wind directions. In order to minimize the
longitudinal pressure gradient along the model area, the ceiling is adjustable in height in
both facilities. Computer-controlled 3d-traverse systems allow the measuring instruments
to be positioned to a spatial accuracy better than 0.1 mm.
Point wise high resolution flow measurements are carried out by means of 2D LDA at
minimum sampling rates of 500 Hz (model scale). By operating two synchronized LDA-
systems simultaneously, spatially correlated measurements of wind components can be
achieved. In order to obtain simultaneous flow data not only from one or two points in
space simultaneously, a PIV was used. At 15 Hz the PIV sampling rate is lower than the
one provided by LDA-systems but PIV provides a far better experimental insight into tur-
bulent flow structures. For dispersion measurements slow and a fast Flame Ionization De-
tectors (FIDs) are used. While the slow FID usually monitors the background concentra-
tion of tracer in the approach flow, the fast FID is recording turbulent concentration
fluctuations within the model area at sufficiently high frequencies.
211


In order to follow the data quality standard for flow and dispersion data defined in
Bastigkeit et al [1], the EWTL posses a set of certified calibration references for all test
equipment utilized.
4 DATA POST-PROCESSING
Due to the demanding requirements for LES-validation data, huge amounts of data accu-
mulate during each measurement campaign. In order to allow for an automated data proc-
essing, analysis, and quality assurance, a strategy for efficient data post-processing was
developed and employed. Ideally, data are completed by essential information on the test
conditions within the scope of post-processing and then can be published in the corre-
sponding data base shortly after a test series.
The software-tools developed were used first time for developing the requested
boundary layer flows. In this, usually time consuming, process, efficiency could be im-
proved significantly. The ability to analyze data immediately after a test run enables the
experimentalist to respond to results instantly. Test setups and measurement strategies
can be adapted if necessary with almost no loss of time. Additionally, the applied post-
processing is transforming individual test results into a consistent and uniform data for-
mat of time series and statistical results.
4.1 Data base CEDVAL-LES
The structure of the CEDVAL-LES database is adopting the style of the already existing
CEDVAL database. The data folder structure of CEDVAL-LES is shown in Figure 2,
whereas the structuring is maintained throughout the entire database (marked by ).
The highest level of folders is indicating the four different levels of complexity. As sec-
ond level in the data archive, the three different types of boundary layer flows were cho-
sen. The corresponding folders are labeled by the model scale (scale 1:500, scale
1:300, scale 1:225). The next level provides a detailed description of the model setup
(MS) for the particular combination of model boundary layer flow and test configura-
tion or complexity. The description includes sketches and photos of the wind tunnel facil-
ity, turbulence generators, roughness elements and the model setup itself. The measured
flow data (FL) are sorted by simultaneously measured wind components (UV; UW),
while the concentration date (CC) are sorted by the kind of release (puff release, con-
tinuous release). Following the structure of the output files provided by the post-
processing software tools, these data folders are divided into sub-folders containing statis-
tics files (ST), post-processed time series (TS) and analysis results such as fluctuations
of wind direction (FLUC) or spectra of turbulent kinetic energy (SPEC) for all meas-
ured wind components. Currently, the data structure is extended by specific folders for
dispersion and pressure data.
212


CEDVAL-LES
complexity 0 complexity 1 complexity 2 complexity 3
... ... ...
scale 1:500 scale 1:300 scale 1:225
... ...
MS
model setup
FL
flow data
CC
concentration data
DSC
description
DRAW
drawings
IMG
images
DOC
documentation
ST
statistics
TS
timeseries
FLUC
fluctuarions of winddirection
and windspeed
SPEC
spectra
DOC
documentation
ST
statistics
TS
timeseries
DOC
documentation
ST
statistics
TS
timeseries
PUFF
puff releases
PLUME
continuous releases

Figure 2: temporary structure of the LES-validation data base CEDVAL-LES
5 SELECTED RESULTS
To illustrate the extend and quality of data, selected results compiled for complexity level
0 (boundary layer flow) are presented. Two model boundary layer flows, of which one is
classified as 'very rough' and modeled at a scale of 1:225 and one is classified as 'rough'
and modeled at a scale of 1:300 were tested in the large wind tunnel facility. A 'moder-
ately rough' boundary layer flow was modeled at a scale of 1:500 in the small wind tunnel
facility. Customized turbulence generators and floor roughness elements were used to
generate the desired neutrally stratified model boundary layer flows. All three flows were
found to be consistent with corresponding full-scale conditions.
5.1 Documentation of modeled boundary layers
From the processed time series measured, a complete documentation of the model bound-
ary layer flows was compiled. Representative mean flow profiles are shown in the left
part of Figure 3. The data is non-dimensionalized with a reference wind speed measured
at 111 m above ground in full scale.
Fitting the measured profiles to a power law as well as to the logarithmic wind profile
assumed to be present within the Prandtl-layer, the corresponding power law exponent
and roughness length can be derived and the flow conditions can be classified according
to VDI 3783/12 [2]. Table 1 summarizes the power law exponents and roughness lengths
derived.

Table 1: characteristic parameters of the modeled environmental boundary layers

Class of roughness (VDI)
0
z
[m] (full scale)
[-] Scale
Moderately rough 0.1 0.16 1:500
Rough 0.2 0.18 1:300
Very rough 1.53 0.27 1:225
In order to roughly check the physical consistency of the modeled boundary layer
flows, the power law exponent can be plotted as function of the corresponding roughness
length (Counihan [3]). The right graph in Figure 3 illustrates the consistency with Couni-
hans analysis.
213


U
mean
/U
ref
[-]
Z
f
s
[
m
]
0 0.2 0.4 0.6 0.8 1 1.2 1.4
0
30
60
90
120
150
moderately rough
rough
very rough
z
0
[m]
[
-
]
10
-3
10
-2
10
-1
10
0
10
1 0
0.1
0.2
0.3
0.4
field data
theory
moderately rough
rough
very rough

Figure 3: the mean vertical wind profiles of the U-component for the three modeled boundary layers (left);
the ratio of the profile exponent to the roughness length
0
z for the three modeled boundary layers af-
ter Counihan [3] (right)

In a further analysis, the vertical distribution of mean turbulence intensity profiles is
evaluated. The corresponding graph in Figure 4 documents the expected decay of turbu-
lence intensity corresponding with increasing height above ground, whereas the shape of
the profiles and the magnitude of the values observed drop good with corresponding ref-
erence values from full scale.

I
U
[%]
Z
f
s
[
m
]
0 10 20 30 40 50
0
30
60
90
moderately rough
rough
very rough
-U'W'
mean
/U
ref
2
[-]
Z
f
s
[
m
]
0 0.002 0.004 0.006 0.008
0
50
100
150
200
250
moderately rough
rough
very rough
constant shear layer

Figure 4: the vertical distributions of the turbulence intensity of the u-wind component for the three mod-
eled boundary layers (left); the vertical distribution of temporally averaged turbulent momentum fluxes for
the three modeled boundary layers (right)

Another parameter to be evaluated carefully is the forming of a constant shear layer
within the lowest 10% of the modeled boundary layer flow. In both wind tunnels, the ceil-
ing was adjusted carefully in order to minimize the longitudinal pressure gradient along
the test section, which is a prerequisite for obtaining a constant shear layer in the wind
tunnel. As documented in Figure 4, the modeled boundary layer flows contain a constant-
flux-layer for heights up to approximately 100 m above ground in full scale.
Additionally, a complete boundary layer documentation requires the integral length
scales of the turbulent structures to be evaluated. For all measurement heights above
ground, the corresponding parameters such as L
UX
are calculated and plotted against em-
pirical relationships. In Figure 5, the left graph shows the corresponding data for the 3
model boundary layer flows developed. The symbols consistently follow the correspond-
214


ing empirical curves whereas the best agreement was achieved for the moderately rough
and the very rough type boundary layer flow.
In Figure 5 (right) the spectral distribution of turbulent kinetic energy was calculated at
different heights above ground for just one of the three model boundary layer flows. In
the modeled boundary layer flows, the spectral distribution of turbulent kinetic energy
follows the expected full-scale distributions over a sufficiently wide range of relevant
turbulence frequencies as exemplarily illustrated by the graph.

L
UX
[m]
Z
f
s
[
m
]
10
1
10
2
10
3 10
1
10
2
10
3
z
0
10 m(theory)
z
0
1.0 m(theory)
z
0
0.1 m(theory)
z
0
0.01 m(theory)
field data (low z
0
)
field data (highz
0
)
moderately rough
rough
very rough
f z / U [-]
f
S
U
U
(
f
,
z
)
/
U 2
[
-
]
10
-4
10
-3
10
-2
10
-1
10
0 10
-4
10
-3
10
-2
10
-1
10
0
z=25m
z=38m
z=85m
Kaimal z=85m
Simiu z=85m
Karman z=85m
very rough boundary layer

Figure 5: vertical distributions of the integral length scales of turbulence for the three modeled boundary
layers (left); spectra of turbulence of the very rough boundary layer at different heights (right)
5.2 Detection of coherent flow structures
The new data sets generated are intended to provide a better insight into the turbulent
phenomena dominating flow and dispersion within the lower atmospheric boundary layer.
LES-based models are expected to replicate most of the dominating turbulent structures
properly. Also a method of comparing the results of an LES-code with corresponding
wind tunnel test data has to be established. In order to 'test' the adequacy of data in this
regard, Hertwig [4] used the new data set for their applicability to POD-, LSE- and Joint-
Time-Frequency-Analysis techniques. It was found that generally the data is qualified for
structural analysis.
5.3 Concentration measurements
One of the specific features LES-based models would deal with are transient dispersion
phenomena such as puff dispersion (Harms et al [5]). Mei [6] carried out systematic puff
dispersion experiments within a semi-idealized urban roughness of complexity level 2.
The test was carried out in the 'very rough' boundary layer flow modeled in the large wind
tunnel facility. Transient puff concentration data were collected in a statistically represen-
tative way by measuring large ensembles of repetitive puff releases. Long concentration
time series containing at least 200 individual puff releases were recorded under system-
atically varied boundary conditions. Parameters such as release duration, source strength,
and mean approach flow velocity were systematically varied and analyzed. Both the puff
concentration time series as well as the results of a comprehensive systematic analysis of
the data will be published in the CEDVAL-LES data base.
215


6 CONCLUSIONS
Compiling validation data qualified for the testing LES-based numerical flow and disper-
sion models is found to be a non-trivial task. The model-specific requirements in data
quality and data quantity require substantially higher experimental efforts with respect to
data collection, data handling and quality assurance of experimental. However, as shown,
carefully designed tests carried out under controlled boundary conditions in a wind tunnel
can deliver data qualified for testing eddy-resolving models systematically.
As the MetStrm project continues, further systematic test series will be carried out.
For example, in the large wind tunnel facility the model of a complex synthesized urban
structure has been investigated. Tests in the small wind tunnel facility BLASIUS will
concentrate on flow measurements and correlated flow-pressure measurements in model
structures representing complexity level 2 will be carried out. For this purpose a set of
rectangular rings of buildings will be mounted on a turn table to simulate different ap-
proach flow directions.
7 ACKNOWLEDGEMENTS
Support by the German Science Foundation through the Cluster of Excellence "Inte-
grated Climate System Analysis and Prediction (CliSAP)" and through the DFG Priority
Program 1276 "MetStroem" is gratefully acknowledged.
8 REFERENCES
[1] Bastigkeit, I., Fischer, R., Leitl, B., Schatzmann, M., Fundamental quality requirements for the genera-
tion of LES-specific validation data sets from systematic wind tunnel model experiments, International
Symposium on Computational Wind Engineering (CWE 2010), 23-27 May 2010, Chapel Hill, USA
[2] VDI-Richtlinien (2000), Environmental meteorology Physical modeling of flow and dispersion proc-
esses in the atmospheric boundary layer, Application of wind tunnels, VDI 3783/12, December 2000,
VDI/DIN-Handbuch Reinhaltung der Luft, Band 1b.
[3] Counihan, J., Adiabatic atmospheric boundary layers: A review and analysis of data from the period
1880-1972, Atmospheric Environment Vol. 9. pp. 871-905. Pergamon Press 1975.
[4] Hertwig, D., Bastigkeit, I., Leitl, B., Schatzmann, M., Organized turbulent structures The link be-
tween experimental data and LES modeling, International Symposium on Computational Wind Engi-
neering (CWE 2010), 23-27 May 2010, Chapel Hill, USA
[5] Harms, F., Leitl, B., Schatzmann, M., Patnaik, G., Validating LES-based flow and dispersion models ,
International Symposium on Computational Wind Engineering (CWE 2010), 23-27 May 2010, Chapel
Hill, USA
[6] Mei, M., Leitl, B., Fischer, R., Schatzmann, M., Systematic analysis of puff dispersion in a semi-
idealized urban roughness, International Workshop on Physical Modeling of Flow and Dispersion Phe-
nomena (PHYSMOD 2009), 24-26 August 2009, Brussels, Belgium
216
Comparison between WRF calculations and
observations of vertical profiles of wind velocity
Masanori Mochizuki
a
, Ryuchiro Yoshie
b

a
Tokyo Polytechnic University, Atsugi, Kanagawa, Japan, m1064010@st.t-kougei.ac.jp
b
Tokyo Polytechnic University, Atsugi, Kanagawa, Japan, yoshie@arch.t-kougei.ac.jp

ABSTRACT: We intend to use the WRF (The Weather Research and Forecasting Model to
prepare standard wind data at high altitude for the assessment of pedestrian wind
environment, and also to investigate urban heat island phenomena. Before doing these
investigations, it is necessary to confirm the prediction accuracy of WRF. In this study, the
measured occurrence frequencies and vertical wind velocity profiles obtained by Doppler
Soda at Minami-senju (inland area in Tokyo) were compared with those calculated by the
WRF model. In order to well regenerate the vertical profile of wind velocity by WRF,
appropriate surface roughness should be given to the WRF calculation. Thus, we used GIS
data to appropriately classify land-use categories and to give a roughness length to the WRF
calculation. The WRF calculation results using GIS data were compared with those of WRF
using default values. The occurrence frequency and the vertical profiles of wind velocity
calculated by WRF using GIS data agreed very well with those of the observation data. The
calculated results of WRF using GIS were much better than those of WRF using default
values.
1 INTRODUCTION
To assess the pedestrian wind environment around tall buildings based on occurrence
frequencies of wind velocities we need reliable statistical wind observation data from near
their construction sites. However, wind observatories are not always located near
construction sites. Even if they do, the observation height is sometimes not high enough and
the wind data are affected by surrounding buildings. Meso-scale simulation can be an
alternative to direct observation. The authors intended to use the WRF (The Weather
Research and Forecasting Model), a meso-scale simulation model, in order to prepare
standard wind data at high-altitude for the assessment of the pedestrian wind environment.
We also plan to use WRF for research on urban heat island phenomena, which is becoming
serious in large cities in Japan. One of the effective countermeasures against heat island
phenomena is to lead cool air of sea breeze into urban canopies. This strategy strongly
depends on the vertical profile of wind velocity and the temperature of the sea breeze. Before
doing these investigations using WRF, it is necessary to confirm how WRF can correctly
regenerate the occurrence frequencies and vertical profiles of wind velocities. For this
validation, observation data measured by Doppler Sodar in the Minami Senju district in
Tokyo (Miyashita et al. 2002) were used. In order to well regenerate the vertical profile of
wind velocity by WRF, it is considered that an appropriate surface roughness should be given
to the WRF calculation. However, the default setting of WRF based on USGS (United States
Geological Survey) expresses urban areas as only by one category and gives a uniform
roughness length regardless of building densities and heights. Thus, we used GIS
(Geographic Information System) data to appropriately classify urban land-use categories and
to give roughness lengths to the WRF calculation. We conducted two cases of WRF
217
calculations using a default setting and using GIS data, and the results were compared with
the observation data.
2 OUTLINE OF CALCULATIONS
2.1 Calculation Period, Initial and Boundary Conditions, Physics Scheme
Calculations were carried out for a period of one month from May 1 to May 31, 2000. To set
initial and boundary conditions we used FNL (Final) Operational Global Analysis data from
the NCEP (National Center for Environmental Prediction). Table 1 gives an overview of the
Physics Schemes used for these calculations.

Table 1: Physics Schemes
microphysics WRF 6 graupel class
Longwave Rapid Radiative Transfer Model
Shortwave Gudhia showtwave
Surface layer Monin-Obukhov (Janjic Eta) scheme
Land surface Noah land surface model
Planetary boundary layer
Mellor-Yamada-Janjic scheme
(2.5 level closure model)
Cumulus no
Urban canopy no
Physics scheme

Table 2 22 2: :: : Computational Domain and Grid Resolution
Domain (km)
X Y Z
Grid Size
X Y Z
Horizontal grid intervals
(km)
Domain 1 450 450 20 50 50 60 9
Domain 2 180 180 20 60 60 60 3
Domain 3 60 60 20 60 60 60 1

We used unequal intervals for the vertical grid and small intervals for areas near the ground surface.

MINAMISENJU
Domain 2
Domain1
Domain 3
MINAMISENJU
Domain 2
Domain1
Domain 3

Figure 1: Computational Domain
218
2.2 Calculation Domains and Grid Intervals
The calculations were conducted using three-stage, two-way nesting grids. The calculation
domains and grid resolution are shown in Table 2 and Figure 1. The Minami Senju district,
used for the comparison between observation and WRF calculations, is included in Domain
3.
2.3 Setting Land-Use Categories and Ground-Surface Parameters
Default setting of WRF employs the USGSs (United States Geological Survey) 24 categories
of land use and their corresponding ground-surface parameters such as roughness length,
albedo, and so forth. However, in the USGS, urban area is expressed by only one category
and its roughness length is uniformly 0.8m regardless of building density and building height.
Moreover, some areas that would be considered urban in reality are classified as pasture,
grassland, and other categories in the USGS. Thus, the USGS does not sufficiently express
the actual circumstances of Tokyo. For these reasons, we used the Digital Map 5000 and
National Land Numerical Information Report (both referred to as GIS hereafter) published
by the Ministry of Land, Infrastructure, Transport, and Tourism to appropriately classify
land-use categories. Furthermore, we classified the urban area into three categories, Low,
High, and Commercial. We mainly used the Digital Map 5000, adding supplementary
data from the National Land Numerical Information Report for places not covered by the
former. The procedures for classifying land-use and determining ground-surface parameters
are as follows.
Firstly, we netted the GIS in a 1km 1km mesh, which is the same size as the grid in
Domain 3 of the WRF calculation. Then, the area for each land-use category in each
1km1km grid was calculated, and the land-use category occupying the largest area within
each grid was determined to be a representative land-use category in the grid.
Secondly, we classified urban area into three categories to reflect the diverse urban
configurations of Tokyo. (As described above, in the default setting of WRF with USGS, the
urban area is expressed by only one category regardless of building density and building
height.). We used land-use categories from the Digital Map 5000 and assigned them to three
different urban categories: Low, High, and Commercial. General Low-Rise
Residential area, High-Density Residential area, and Others (including Land for Public
Facilities) in the Digital Map 5000 were assigned to the Low urban category; Medium-
and High-Rise Residential area to the High urban category; and Area for Commercial
and Business Use to the Commercial urban category. We totaled the areas occupied,
respectively, by the Low, High, and Commercial category in each grid square, and the urban
category with the largest area was determined to be a representative urban category for that
grid square.
Finally, GIS land-use categories were corresponded to the USGS land-use categories, and
surface parameters were determined as shown in Table 3. The USGS uses a uniform
roughness length of 80cm for the urban area, but for this study we used 70cm, 100cm, and
150cm for Low urban area, High urban area, and Commercial urban area, respectively. We
adopted these values by referring to Grimmond and Oke (1999). Figure 2 shows land-use
categories based on the USGS (Figure 2a) and the GIS (Figure 2b), which shows large
differences between them.

219

Table 3: Land-use Categories and Surface Parameters
No.
USGS/SIB
Land-Use Category
GIS
Land-Use Category
Urban
Category
Surface Z0
[cm]
Albedo
[%]
Soil
Moisture
[%]
Surface
Emmisivity
[%]
Low 70
High 100
Commercial 150
2
Dryland Cropland
and Pasture
Farm
-
15 17 30 98.5
3
Irrigated Cropland
and Pasture
Paddy Field
-
10 18 50 98.5
7 Grassland Park and Grass - 12 19 15 96
15 Mixed Forest Forest - 50 13 30 97
16 Water Lake and Sea - 0.01 8 100 98
19
Barren or Sparsely
Vegetated
Soil Surface
-
1 25 2 90
10 88 1 Urban Land
House and
Pablic Building
15

2a) USGS
Urban
Grass, Forest, etc
Water
Low-rise Urban
High-rise Urban
Commercial Urban
Low-rise Urban
High-rise Urban
Commercial Urban
2b) GIS
Grass, Forest, etc
Water
2a) USGS
Urban
Grass, Forest, etc
Water
Low-rise Urban
High-rise Urban
Commercial Urban
Low-rise Urban
High-rise Urban
Commercial Urban
2b) GIS
Grass, Forest, etc
Water

Figure 2: Land-use Categories in Domain 3

2.4 Calculation Cases
220
We carried out two kinds of calculations: one using the default setting of WRF with USGS
(Case 1), and the other using the land-use categories and surface parameters based on GIS
data as described in 2.3 (Case 2).
3 Comparison of WRF-Calculated Results with Observation Data
3.1 Occurrence Frequencies of Wind Velocities at High Altitude
Figure 2 compares the probabilities of exceedance of mean wind velocity (10 minutes
averaging 300m high) between the results of WRF and observation data. Case 1 and case 2
indicate WRF calculations using the default value and GIS data, respectively. The
probability of exceedance of wind velocity calculated by Case 2 agreed very well with the
observation data. The probability of exceedance calculated by Case 1 was higher than that of
Case 2. However, the difference was not so large at this high altitude.
0
0.2
0.4
0.6
0.8
1
0 5 10 15 20
Wind Speed (m/ s)
F
r
e
q
u
e
n
c
y
OBS
case1
case2

Figure 3: Probability of Exceedance for Mean Wind Speed

3.2 Vertical Profiles for Mean Wind Velocities
For this comparison, we firstly extracted southerly wind (wind direction=18033at 200m
high) for both observation and calculation. Then the vertical profiles of wind velocity of the
observation data were classified into eight clusters by cluster analysis. Corresponding clusters
of calculated results were made so that each cluster at the same time of the observation data
became the same cluster.
Figure 4 shows vertical profiles of wind velocities in the eight clusters. Averaged values 1
(standard deviation) at each height in each cluster are plotted in the figures. The value shown
in the caption for each cluster is the occurrence frequency for that cluster. Except for Figure
4f (Cluster 6), there is a pretty good match between observation and calculated results. In
particular, Case 2 using GIS is remarkably close to the observation values. Case 1 using the
default setting of USGS generally show higher values than the observation data. The
calculated wind velocities for Case 6 do not correspond with observation data because
Cluster 6 includes many data obtained during rainy weather. In both Case 1 and Case 2,
221
during rainy weather the WRF calculation tends to produce a higher wind velocity than the
observations.

0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
4a) Cluster 1 (Occurrence Frequency 17.4%) 4e) Cluster 5 (Occurrence Frequency 19.1%)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
4b) Cluster 2 (Occurrence Frequency 9.8%) 4f) Cluster 6 (Occurrence Frequency 10.5%)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
4c) Cluster 3 (Occurrence Frequency 13.5%) 4g) Cluster 7 (Occurrence Frequency 11.4%)
222
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case1 (USGS)
0
50
100
150
200
250
300
350
400
0 5 10 15 20
Wind Speed(m/ s)
H
e
i
g
h
t
(
m
)
OBS
WRF
Case2 (GIS)
4d) Cluster 4 (Occurrence Frequency 10.6%) 4h) Cluster 8 (Occurrence Frequency 7.7%)
Figure 4: Vertical Profile of Mean Wind Speed
4 Conclusion
Two cases of WRF calculations were conducted using the default setting (Case1) and based
on GIS-derived land-use categories and surface roughness lengths (Case2). The calculated
results were compared with the observation data measured by Doppler Soda at Minami Senju.
The calculated probabilities of exceedance of wind velocity at high altitude (especially
Case2) corresponded very well with that of the observation. In addition, it was apparent that
calculated vertical profiles of mean wind velocity of Case 2 were very close to those of the
observation.

Reference
Chung and Yoshie, 2010, Classification of vertical profiles of wind velocity and temperature in sea breeze using
cluster analysis, Summaries of technical papers of Annual Meeting Architectural Institute of Japan ,D-1,975-
976

Grimmond, C. S. B. and T. R. Oke , 1999,Aerodynamic Properties of Urban Areas Derived from Analysis of
surface Form, J. Applied Met., 38, 1262-1292

Miyashita, K.,Suda, K., Iwatani, Y., Hibi, K., Ishibashi. S., and Tamura, Y., 2002,Observations of wind speed
profiles over various surface roughness sites using Doppler sodars : Part 19 Characteristics of natural winds
in Tokyo city area, Summaries of technical papers of Annual Meeting Architectural Institute of Japan ,B-
1,103-104

223
Models in Wind Energy Assessment
Graciana Petersen
a
, Ulrich Ghde
b
, Martin Hoffmann
b
, Bernd Leitl
a
, Michael
Schatzmann
a

a
Meteorological Institute, KlimaCampus, University of Hamburg, Hamburg,
Germany, graciana.petersen@zmaw.de
b
Department of Philosophy, Theoretical Philosophy, University of Hamburg,
Hamburg, Germany
ABSTRACT: In this paper, model comparison with regard to wind energy assessment is
examined. Motivation for this analysis is the strong political and economical demand for
improvement of wind energy assessment tools. Additionally, the current scientific
methodology lacks a generally accepted methodological framework to determine the process
of model construction. Based on the so-called semantic metatheoretical approach, a five-step
model comparison is suggested and related to the process-oriented model evaluation
procedure of COST 732. The COST 732 action concerning quality assurance and
improvement of microscale meteorological models is used as a successful example for model
comparison in practice. Then, model improvement is analyzed structurally. It is found that
model improvement is an empirical and recursive process and demands high quality
validation data, which is not as trivial as it sounds. Fundamental principles, case restrictions
and strengths of windtunnel modeling are reviewed shortly. It turns out that windtunnel data
in wind energy assessment serve as high quality validation and high quality input data for
generation of artificial experience, a term that is going to be defined with regard to the
process of model improvement.
1 INTRODUCTION
The current status of wind energy assessment is the following: An uncertainty of 30-40% in
todays prediction of wind energy output for the next ten years at an average wind energy site
is assumed, Rodrigo, 2010. In order to improve quality of wind assessment models, models
are to be compared. Models in this case are numerical models and physical models.
Numerous fundamental questions arise: How can models be compared? What is the relation
between wind tunnel, numerical model and reality? What do we consider as reality in this
context? Exactly these questions Models and their role towards theory and reality are
ongoing research areas in philosophy of science, Magnani et al., 1999, Morgan and Morrison,
1999. In order to find qualified answers for above questions with regard to comparison and
quality assurance of wind energy assessment tools, concepts and tools developed within the
semantic view of empirical theories are utilized.

First, we will differentiate between two concepts of models used in the philosophy of science:
the mathematical concept of models and the concept of models. In a second step, the
concept of models as mediators is applied to a special case: the relation between models
understood in this sense as, theory, reality and data in wind energy assessment is examined.
Finally, a comparison of models is considered in some detail. With this, we deduce a
seemingly trivial lemma 1: Data of model 1 and model 2 are only comparable if model 1 and
model 2 are comparable. Surprisingly, this statement is not self-evident. With regard to
224
practice, a five-step comparison is suggested and related to the six-step model evaluation
procedure of the COST 732 action, Schatzmann et al., 2010. After this, the process of model
improvement is examined and the term artificial experience is introduced. In most cases, the
process of model improvement is neither machine learning, nor artificial intelligence, but
generation and application of artificial experience. It is emphasized that reference data-sets
that are used as validation data have to be of good quality. Literature review shows that this is
not trivial, Petersen et al., 2011, Petersen et al., 2011b. After analyzing the relation between
models, theory and reality in wind energy assessment the role of windtunnel modeling is
examined. The fundamental principles of accurately modeling atmospheric boundary layer
flow are briefly reviewed. Model simplifications and shortcomings, model case restrictions
and model strengths are pointed out.
In conclusion this leads to lemma 2 on the role of wind tunnel modeling in wind energy
assessment: Windtunnel data in wind energy assessment is only useful, if it is of high quality.
Then it
1. serves as high quality validation and high quality input data for generation of artificial
experience in compatible numerical models.
2. predicts atmospheric flow for certain sites and in a certain range of meteorological
boundary conditions.
2 SHORT INTRODUCTION TO MODELS IN PHILOSOPHY OF SCIENCE
The various versions of the so-called semantic view of empirical theories as proposed by
Suppes, van Fraassen, Sneed, and others aim to characterize empirical theories by their sets
of models. Initially, for that purpose, the mathematical concept of models was used.
According to this concept, models of an axiomatized mathematical theory are tuples,
consisting of sets and relations defined on these sets, that fulfill the axioms of that theory.
This mathematical concept of models was then applied to empirical theories: in a first step,
the theories were reconstructed in an axiomatized form; in a second step, the corresponding
sets of models were defined as the sets of all structures (tuples) fulfilling these axioms.

In a seminal publication, however, Morgan and Morrison (1999) pointed out that in scientific
practice, the concept of models is used in an entirely different way which they tried to
characterize by means of the slogan models as mediators. The introduction of this idea is to
be understood against the background of the empirical sciences, wherein an increasing
number of systems are investigated that are highly complex and cannot be described
theoretically in any direct way. In order to nonetheless provide a theoretical description of
these systems, highly idealized and simplified models of the systems in question are
employed. These models serve as mediators between theory and reality hence the slogan.
With their help it becomes possible to analyze some though by no means all aspects of
these systems. Additionally, the complexity of these models is considerably reduced as
compared to the complexity of the original phenomenon.

This massive reduction of complexity allows the models to fulfill their role as useful
instruments to mediate between theory and empirical phenomenon. They can be treated as
specialized representations of parts of reality that are far easier to understand than the original
application which is sometimes too complex to be understood. Furthermore, complexity
reduction is one reason for another pragmatically fruitful feature of models: their high degree
of flexibility. On the one side their small number of free parameters can flexibly be fitted to
predictions of more general theories. On the second hand, they can be fitted to measured data
structures. Exactly these properties of models make them manageable tools. They allow
models to be used as mediators to explore, to correct, and to develop the theory in question.
225
However, one should not overlook the fact that complexity reduction also has some
drawbacks, i.e., it is burdened with some theoretical costs. The feature of complexity
reduction involves at least two aspects: (i) idealization: the properties of the original
phenomenon are roughly described and in many aspects a simplified way, e.g. a complex
correlation between two variables in reality may be approximately represented as a linear
function; and (ii) incomprehensiveness: the model represents just a few (sometimes only two
or three) properties of the many properties that describe the original phenomenon. These two
aspects help us to indicate two problematic consequences of complexity reduction: First,
complexity reduction can be achieved in numerous ways. This is because there is a large
variety of possible answers to the questions (i) which idealizations are carried out and to what
degree, and (ii) which properties of the empirical application are chosen as parameters of the
model. At present we do not have scientifically justified, or even consensually accepted
methodological rules, which could give guidance to answer these questions. Second, since the
models are offering an idealized and incomprehensive picture of reality, it is probable that the
predictions of these models behave in a radically different way than the real world
phenomena. This is especially a problem for the present application of models in engineering
sciences, where the production of reliable predictions for real world settings is often judged
as the most important virtue of models.

In summary, a model normally is reducing the complexity of its empirical application. This
leads to a high degree of variability and hence to a new type of complexity. This new type
of complexity concerns methodological decisions in model construction. At present the
scientific methodology lacks a generally accepted methodological framework to determine
the process of model construction. The following will show that this may have consequences
for the comparability of different models. Before it is possible to work out criteria for helping
to determine which models can be compared in which respects, it is important to conduct a
careful examination of existing procedures to combine models with data structures. The
following case study is intended to do a first step to develop methodological criteria for the
comparison of models. Of course, a case study can only have an explorative character. It is
our aim to work out and conceptualize the questions one has to face in the present context.

As it becomes clear from these considerations, the empirical application of models should not
be confused with the models in mathematical model theory. In fact, the concept of models as
mediators differs fundamentally from the concept of models that is used in mathematics.
Where mathematical models refer to one specific theory, in the construction of models as
mediators there are often several different empirical theories working together in a
complicated and obscure way. Furthermore, in the construction of these models, numerous
additional auxiliary hypotheses are generally involved some of which are only legitimized
by the explanatory and predictive success they provide. In what follows, we shall refer to
'models as mediators', referring to the concept of Morgan and Morrison as described above.
3 MODELS IN WIND ENERGY ASSESSMENT
3.1 Models, theory, reality and data in wind energy assessment
Regarding the understanding of models as mediators, how does field-data and model-output
data modify the relation between model and reality? Assume a model is constructed with the
help of one or more empirical theories plus several auxiliary hypotheses. Its purpose is to
provide a simplified image of reality and to enable both predictions and explanations. Based
226
on Cartwrights framework, the relation between theories, models and reality can be
visualized as follows:

Note, that this diagram is only a rough visualization of the relation between reality, theories
and models. Actually, every arrow and arrow head in the above diagram symbolizes a
difficult and different relation. This should be kept in mind for further discussion.

In scientific practice, how is a model used? We shall discuss this problem with respect to a
special case: the use of models for the prediction of wind speeds at a certain site that is
interesting for the placement of wind turbines. The corresponding models make use of data
obtained by physical or numerical experiments. This means, measurements or computations
are done and analogously to field measurements every step induces uncertainties and
errors.
The process chain is following:

Thus, data from field studies is used to represent reality and, equivalently, data from
experiments is used to represent the physical or numerical model. In conclusion, in order to
consider the relation between model and reality, model-data output and field-data output is
compared. In fact, the time-line is as follows: First, the relation between model and theories is
examined (I). Then, in order to understand the relation between model and reality (II), data
are compared (III), in diagram:

How can it be known whether a model represents the real world properly? With regard to
wind energy assessment, it is currently done as follows: If data from the numerical or
physical experiment fits well with field data, the model is considered to be good. More
explicit, if averaged model wind speed data match field data in a certain range for certain
points, the model is considered to be a good representation of reality concerning wind
resource assessment.
3.2 Comparison of models
In the diagram above, the term model refers to any element out of the set of scientific
models used for wind energy assessment, namely physical or numerical models. Praxis shows
that competing models exist at the same time, say model 1 and model 2. Assume model 1
performs well in specific cases A, and model 2 in specific cases B. An obvious idea is to
REALITY FIELD DATA
MODEL
uncertainty
THEORIES
(I)
MODEL OUTPUT DATA
(II)
(III)
THEORY
REALITY
MODELS
as mediators
MODEL
experiment
Measurements/
Computations
data processing
MODEL OUTPUT DATA
227
merge both models. In practice this is often not reasonable. Combining models does not
necessarily lead to improvement. Another idea is to achieve partial synergy effects, such as:
model 1 learns from model 2 for cases B, and model 2 learns from model 1 for cases A.
However, the problem of choosing between competing models is as following: We want to
know how the wind blows at a certain site x, but we only have meteorological data from an
airport which is 50 kilometer away. As before, let us assume that model 1 performs well for
cases of type A, model 2 for cases of type B. If we want to make predictions for atmospheric
flow at x and run model 1 and model 2, two sets of data are generated. How do we know,
which model is better if we do not know if x belongs to A or to B or maybe to cases of type
C that are neither A nor B?

Typically, the answer is: Do a blind comparison of the models. Blind indicates that physical
and numerical modelers get a certain restricted set of field data and are asked to reproduce the
remaining data-set. For example the Bolund hill is a well received field study and blind
comparison, Bechmann et al., 2007, where inflow conditions serve as a restricted starting
data set and measurements from the masts are to be reproduced. Another example for a
successful study in model comparison is the COST-Action 732, 2005-2007, see Schatzmann
et al., 2010. COST is an intergovernmental European framework for international cooperation
between nationally funded research activities. The objective of COST 732 was to improve
model evaluation and provide a methodology to ensure the quality of microscale
meteorological models used to predict flow and transport processes in urban or industrial
environments. Although urban and industrial environments are not wind energy sites, the
wind energy community can benefit from the methods outlined in the COST action. Within
COST 732, a structured quality assurance procedure was developed based on data for model
validation. We will come back to the COST-Action later.

Blind comparison competitions are being done in the scientific community to determine how
well the models perform. However, it has to be noted that physical and numerical models
provide a finer spatial resolution than field measurements, due to financial limitations for
field instruments. In consequence, only few data-points can be compared between model
output-data and field-data and the large amount of additional model-data can only be
compared among other models. The question is how can models be compared?
What is mostly done in practice is the following:

In words: modelers let their numerical model or the wind tunnel run and obtain data. The
datasets obtained thereby are compared, indicated by arrow (I) in the diagram above. Similar
to the diagrams before, the diagram above is only a rough visualization of the relation
between model and reality, theories and data. Can the relation between model and reality be
further specified?

REALITY
MODEL 1
uncertainty
THEORIES
MODEL OUTPUT DATA
MODEL 2 MODEL OUTPUT DATA
(I)
REALITY
MODEL 1
THEORIES
MODEL OUTPUT DATA
FIELD DATA
(II)
(III)
(IV)
(I)
228
As illustrated in the diagram, the relation between model and reality is most often obtained
from analyzing field data. Then, the relation between model and reality is approached by the
relation between model output data and field data. Moreover, the relation (I) is deduced from
the relation between model data output and field data (II), and from the representational
function from model to model data-output (III). Further, there is some known relation
between model and theories (IV).

In consequence, if competing models 1 and 2 exist they are linked together by the relation of
their data towards the data that represent reality. It is emphasized once again that we do not
know what reality is, all we can do is interpret data obtained by measurements. Additionally,
model 1 and model 2 are linked together by their relation to theory. The relation to theory is
not exactly the same for both models. At least the implementation of theory, or parts of the
theories such as auxiliary hypothesis must be different, otherwise models are the same. In
practice, relation to theory differs extremely from model to model. This is not only the case
for comparison of physical with numerical models. In contrary, within the set of numerical
models, the relations to theory are already very broad and sometimes based on contradicting
theories. For example, contradicting theories can be found concerning the handling of
turbulent closure in the Navier-Stokes equations for turbulent atmospheric boundary layer
flow.

In summary, for the case of two competing models, the situation between models, theories,
reality and data can be illustrated as follows:

In words: Reality is interpreted by means of field data. Aspects of the reality are described by
models that depend on theories. Models produce output data which is to be interpreted with
regard to reality. Mostly, interpretation is done by quantified comparison of data, that is to
say model output data is compared with field data or output data of other models.

Examples for quantified comparison are the Bolund blind comparison, Petersen et al., 2011b,
and COST 732, Schatzmann et al., 2010. In the Bolund study, quantification of data
comparison is done with common metrics such as differences of mean wind speed values and
differences of wind speed standard deviations. Thus, comparison of models is quantified by
comparison of the data. The problem is that we do not learn anything from quantified
comparison of data if we do not know about the models from which data is generated. Thus:

Lemma
1
1: Output data of model 1 and model 2 is only comparable if model 1 and
model 2 are comparable.

It is very tempting to just compare data and forget about the models. But how can the
difference in numbers be compared reasonably if it is not taking into account where the
numbers come from? In most cases, the relation between models and reality can only be
approximated by means of comparing the model output-data with field-data. In practice, the

1
Here, the term Lemma is lent from mathematics and shall emphasize the importance of the statement within
the article. The lemma is not proven in a strict sense but followed by a paragraph with an argumentative
derivation, ended by .
MODEL 2 DATA
REALITY
MODEL 1 DATA
DATA
THEORIES
229
focus often remains on the data: data of the models is compared instead of stepping back
and first considering the structure. At the end, the scientist is even forced to interpret the
results of the comparison. We claim that to compare output data of models the relation
between the models has to be clear. The structure is following:

In other words: Comparison between out-put data of model 1 and model 2 (I) is only possible
if model 1 and model 2 are comparable (II). This leads to the Lemma.

What does comparability of models mean in practice? We suggest a five step comparison
which starts in the following diagram on the left hand side and moves toward the right. At the
end it takes into account the whole picture. The five steps of model comparison are practical
for two models and for more than two:

First, the relation between models and theory must be made clear: What theories are involved
in the models? What are the assumptions of the theories? This relation is indicated by arrow
(I) in the following diagram.
Second, assumptions for the different theories involved in different models are to be
compared, arrow (II).

Third, the relation between model and output data, (III), is to be compared. This includes
experimental conditions and means, the experimental set-up, conduction of measurements
or computations in case of computational models the technical facilities and the
uncertainties have to be identified and compared. A detailed documentation of the
experimental set-up is obviously the precondition for such a comparison. In practice, it is not
always the case; see Schatzmann et al., 2010 and Petersen et al., 2011.

Fourth, out-put data are to be compared. Quantification of comparison has to be done
carefully and with regard to underlying theories and assumptions. For example the
dependence of comparison results on the choice of the comparison metric has to be
considered. (IV)
Fifth, comparison of data is to be interpreted with regard to reality. If field-data is available it
is used for comparison with model output-data. The aim of the interpretation is to obtain a
quantitative diagnosis of how the model data-output relates, how the models qualitatively
correspond and how the results of the comparison can be interpreted with regard to theory
and reality. (V)

Is this feasible in reality? The answer is: yes. For example the model evaluation procedure
outlined in COST 732 consists of the following six steps: model description database
description scientific evaluation code verification model validation and user-oriented
assessment. COST 732s model evaluation procedure and the five step model comparison
suggested above are strongly connected. The former is the process-oriented formulating of
the latter which is based on a semantic (model-theoretic) approach:

MODEL 2 DATA
REALITY
MODEL 1 DATA
DATA
(I)
(II)
230
The step database description refers to the fact that participants of COST 732 agreed on the
use of a database as a reference data set. It was employed for the evaluation of the model and
was windtunnel data-output. Thus, database description in terms of COST 732 means that
the reference data set and its model was worked through step (I) to (V).
The term model validation is defined in various papers, for example in Oberkampf et al.,
2002. In COST 732, the term model validation means a structured comparison of model
predictions with experimental data and is based on statistical analysis of selected variables.
It seeks to identify and quantify the differences between the model predictions and the
evaluation datasets; it provides evidence as to how well the model approximates to reality.
Schatzmann et al., 2010, p. 9. It has to be mentioned critically that the last statement is
misleading. Comparing model-data with reference-data from another model is the
comparison of models. In order to obtain evidence as to how well the model approximates
reality, the reference data-set and comparison results would have to be interpreted with regard
to reality (V).

In summary, if different models exist and are to be compared it is not sufficient to compare
model data-output. It is concluded that data from models is only comparable if models are
comparable. A five-step comparison is suggested: First, the relation of the models towards
theory is clarified. Second, the theoretical assumptions are compared this is not trivial since
models are assumed to be different and thus they do have a different relation towards theory.
Third, experimental set-up and conduction is compared. Fourth, model data-output is
compared and quantified in a reasonable way. Fifth, comparison of model-data is interpreted
with regard to theory and reality.

Recall the motivation for model-comparison in the case of wind energy assessment: The aim
is the improvement of models. As mentioned in the introduction chapter, there is a strong
political and economical demand for reduction of wind energy assessment errors. How model
improvement is achieved in practice of wind energy assessment is examined in the following
section.
3.3 Artificial experience
In wind energy assessment, a model is considered to be good if it fits reality in two ways: The
model reproduces and predicts correctly what is observed in reality. Reproduction mostly
means that a restricted set of field-data is used for the simulation (model experiment) as
starting condition. Then, the model output is compared with the full set of field-data.
As mentioned before, appropriate field data for this evaluation are difficult to find and are
limited in measurements. For example, one single mast is mounted within the site of interest
and wind speed measurements are taken for 3 months. This meteorological data delivers
neither sufficient, nor unique data as a starting condition for numerical models. First, the
THEORIES
(I)
MODEL 2 DATA
REALITY
MODEL 1 DATA
(II)
(III)
(IV) (V)
scientific
evaluation
(code)
verification
model
validation
database
description
user-oriented
assessment
model
description
(I) - (V)
231
spatial resolution is extremely low since data is just from one point or with luck from 3
points arranged vertically. However, most numerical models need more grid points as
boundary and initial conditions. Simply stated, these models are solving modified Navier-
Stokes equations - that is to say differential equations that require initial and boundary
conditions.
Second, with regard to the amount and variability of meteorological parameters the
measurement time period is not statistically representative due to the variability in
meteorological boundary conditions. In order to obtain reasonable information on wind
speed, field data are divided into equivalence classes
2
depending on the following
parameters: wind direction and stratification which contains vertical temperature gradient
and mean wind speed and surrounding meteorological conditions such as pressure and
humidity. Thus, these 4 parameters can appear in nearly infinitely many states where all four
parameters exist in hundreds of thousands of combinations.. In consequence, classification of
data is a tough question and if the classification scheme is refined, few data remain within
each equivalence class. In other words, actually a small amount of data is practically usable.
Equivalence classes can be extended by softening the condition of nearly constant
meteorological parameters. For example segmentation into wind direction can be coarsened.
In consequence, boundary conditions within each equivalence class are less stationary. The
ambiguity in the right choice of input data for model simulations concerns every field data
set, not only those in which the model input can be selected from several meteorological
towers. Schatzmann et al., 2010, p. 12.

In conclusion, validation of numerical model output data with field data in order to obtain
knowledge about the relation between model and reality is only reasonable if a) ambiguity of
field data can be minimized and b) purpose of the model matches with conditions of available
field data, especially location of field study, meteorological boundary conditions as well as
the spatial and temporal resolution.
An alternative to field-data is windtunnel data where sufficient spatial and temporal
resolutions exist and steady state boundary conditions can be obtained. For numerical models
that run in steady state mode this is a crucial advantage of using windtunnel data versus field
data. This was the case in the COST 732 action.
Improvement of models proceeds in a three-stepped loop:

This means, a model is used with input data and produces data-output. On the other hand,
model data-output is compared with reference data. This leads to adjustments of the model. A
new simulation is conducted, following the same steps again and again. Also, with help of

2
The term equivalence class is chosen because of its meaning in mathematics. In this case it emphasizes the
fact that within an equivalence class the meteorological parameters are assumed to be nearly constant.
simulation
data-output
adjustments
input
prognosis
reference
data
reality
232
model data-output a prognosis is deduced and is checked to reality and reference data. In
summary, the model is adjusted due to validation results with a reference data set. Most often,
parameters are added or adjusted in order to approximate the validation data set by the model
output. In this way, model improvement is an empirical and recursive process without regard
to theoretical justification. In other words, model improvement is a learning process by
models that are fed with evaluation data which, in practice, are sometimes of high and
sometimes of low quality. In terms of philosophy of science, fundamental theories which
can also consist of empirical theories are mixed with data of empirical phenomena and
adjusted recursively. This is a sort of learning empirically and means to gain experience.
With regard to wind energy assessment, experience comes out of parameterization and
adjustment of equations and comparison of model data-output. In most cases, the process of
model improvement is neither directly machine learning, nor artificial intelligence, but a
sort of artificial experience. There exist exotic prognosis tools that work with neural
networks, for example. However, in most cases the process is not automated, but a mixture of
machine and human learning based on model data-output comparison thus the wording
artificial experience is chosen here.

In conclusion, reference data-sets that are used as validation data have to be of good quality,
which is not trivial, as mentioned before. Model improvement is strongly based on the
recursive process using input data, producing output data, comparing the output data with a
reference data set, adjusting the model and starting from the beginning. Further on, we
mentioned that reference data sets have to match the requirements of the model, for example
concerning a complete set of boundary conditions. The model data-output depends on the
model-input-data. If boundary conditions have to be guessed, model adjustment does not
make sense. In order to gain useful artificial experience and improve models in wind energy
assessment, coherent and solid validation data is necessary.
3.4 Physical models in wind energy assessment
After analyzing the relation between models, theory and reality in wind energy assessment,
the role of windtunnel modeling is examined explicitly. First of all, fundamental principles
for modeling atmospheric boundary layer flow are shortly reviewed.
3.4.1 Fundamental principles
The purpose of physical flow modeling is to accurately simulate the dynamics of the flow
that exists in reality. Thus, a similarity criterion has to be formulated: If x then the dynamics
of the flow in the fluid model can be called similar to those in reality. A generally accepted
similarity analysis for fluid modeling of atmospheric phenomena is described in detail by
Snyder, 1981, Cermak, 1984. The analysis is based on the equations of motion, explicitly
conservation of momentum, continuity and energy.

The main idea is to convert the equations into a dimensionless form by inserting reference
quantities that determine the flow dynamics. Reasonable reference quantities assumed to be
supplied through the boundary conditions are a specific length, velocity, density,
temperature and angular velocity. Using these reference quantities the equations of motions
can be nondimensionalized, and flow-characterizing coefficients can be isolated, for example:
the Rossby-, Froude-, Reynolds- and Peclet-Number. From this it follows that if and only if
the characteristic numbers are identical, as well as the nondimensionalized boundary
conditions, the solutions of the modified set of equations are identical. Thus, the dynamics of
the fluid with the same characteristic numbers and nondimensionalized boundary conditions
are similar. In conclusion, every atmospheric flow that can be described by the modified
233
equations of motion can be modeled by another flow provided the characteristic coefficients
are equal and dimensionless boundary conditions correspond.

For practical use, the requirement is softened. For example, length-scale ratios larger than
10:1 lead to the problem that Reynolds- and Froude-number cannot be matched
simultaneously. Thus, deviations of characteristic coefficients occur but, in some cases, are
empirically negligible. For example, air density is assumed to be variable within a certain
range without affecting flow dynamics. Additionally, fluid dynamics are assumed to be
similar for different Reynolds-numbers beyond a critical Reynolds-number. The critical
Reynolds number determines the transition point of laminar or turbulent flow to so called
fully turbulent flow. The theory of Reynolds-number independency for fully turbulent flow
is helpful for wind tunnel modeling since the Reynolds-number from atmospheric flow can
never be matched in the wind tunnel. Instead, fully turbulent flow can be achieved by
turbulence generators at the windtunnel inlet and roughness elements on the floor. Thus, for
length and time scaling above the similarity criterion this is the basis of windtunnel modeling.

Here a critical remark: deduced from the similarity analysis, windtunnel data is interpreted
for different length- and time-scales. Further, if the assumption on self-similarity of fully
turbulent is correct, windtunnel time-series from fully turbulent flow must be self-similar and
show a fractal structure as well as field time-series at almost steady-state boundary
conditions. Still, the interpretation of one and the same time series for continuously varying
pairs of time- and length-scales leads to an enormous simplification of the flow model. So
far, mean value and standard deviation are the only tools to examine flow properties with
regard to wind energy assessment. For mean value and standard deviation, distortion of time
resolution in the time series is not important. But, for wind turbines, wind gusts and small-
scale turbulence get more and more important. Thus, it is interesting to examine whether up-
and down-scaling of wind-tunnel data leads to a distortion of small-scale turbulence statistics.
3.4.2 Model simplifications
Regarding state-of-the-art tasks for wind energy assessment, simplifications of windtunnel
modeling in relation to reality are:
The Coriolis force in the model does not resemble the Coriolis force in reality. It is assumed
that effects of Coriolis force for atmospheric length scales smaller than 5 kilometers are
negligible. Thus, the larger the area modeled in the windtunnel, the larger the error due to
mismatch of Coriolis force. The model area within the windtunnel is physically restricted by
walls, therefore the atmospheric turbulence larger than the windtunnel dimension can
physically not be reproduced. This is why length scaling is important with regard to
windtunnel size, for example in a 1m wide 1m high windtunnel at length ratio 1:100, the
largest reproducible eddy is about 100 m width and length in full scale. Assuming turbulence
is isotropic and homogenous, and that the mean advection time of the eddy is 10 m/s,
turbulence scales of more than 10 seconds length does not occur in the time series.
Additionally, the Kolmogorov length scale in the windtunnel translated to full scale is too
large. In other words, the smallest eddies are not resolved in the windtunnel model since
dissipation of the turbulent flow is not down scaled. Explicitly, for scaling ratios between
1:100 and 1:500, eddies of about 10 - 50 cm in full scale are neglected in the windtunnel
model. However, fluid dynamics are mainly driven by large eddies. Thus, compared to large
eddies which carry the main part of flow energy, small eddies are neglectable. Summarizing
the latter two points, the turbulence spectrum in the windtunnel compared to that of the field
is cut at low and at high frequencies and has to be considered with regard to windtunnel size
and the choice of length scale ratio.

234
3.4.3 Model case restrictions
The restrictions of windtunnel modeling are: most wind tunnels are not built for the
simulation of thermal effects, thus in the majority of cases windtunnel modeling is restricted
to neutral stratification. Alteration of wind direction is possible, where the model area can be
turned on a turntable for every change in wind direction. However, inflow conditions have to
be adjusted carefully and in practice this technique is labor-intensive and, in most cases
simulations are restricted to 1-3 mean wind directions.
3.4.4 Model strengths
Windtunnel simulation is based on similarity criteria derived from the equations of motion
(Navier-Stokes-equations), beyond this the Navier-Stokes equations are not needed for
windtunnel simulation. This is the main advantage since the NavierStokes existence and
smoothness problem is an open problem in mathematics. Instead of solving simplified
equations, the physical model uses real flow to replicate flow physics.
Compared to field data, windtunnel data is very cost-effective and rapidly available. In
addition, windtunnel data can be provided in a statistically representative sample size. Also,
inflow conditions in windtunnel modeling are well defined in the sense of completeness of
information since they can be measured with high spatial and temporal resolution. Further,
boundary conditions are steady state, which is both a weakness and a strength. We come back
to that point in Ad 1. However, the boundary conditions are controllable which is very
important with regard to model comparison.

Applying previous considerations to the role of windtunnel simulation in wind energy, leads
to:

Lemma 2: Windtunnel data in wind energy assessment has to be of high quality and
then
1. serve as high quality validation and high quality input data for generation of artificial
experience in compatible numerical models.
2. predict atmospheric flow for certain sites and in a certain range of meteorological
boundary conditions.

Ad 1: As stated before, to improve wind energy assessment models are compared with
evaluation data and adjusted recursively. Therefore, reference data sets have to match the
requirements of the model, for example concerning a complete set of boundary conditions.
Since windtunnel modeling delivers data with high spatial and time-resolution and well-
known boundary conditions, it serves as coherent, solid and reasonable validation data. In
some cases, for example concerning mesoscale numerical models, windtunnel simplifications
can be incompatible, such as neglecting Coriolis-force and assuming steady-state boundary
conditions. However, for compatible numerical models, wind tunnel data serves as high
quality validation and high quality input data for generation of artificial experience.
Ad 2: As model of atmospheric flow, windtunnel measurements predict atmospheric flow for
certain cases. It is specifically useful in cases where flow behavior can not be described
physically and be solved adequately by numerical models. For example around steep hills, no
generally accepted turbulence closure model exists. Thus, with respect to model restrictions
and simplifications, wind tunnel modeling is a useful tool for prediction of atmospheric flow
at certain sites and in a certain range of meteorological boundary conditions.

235
4 CONCLUSIONS
In this article, the concept of models as mediators between reality and theory as a modern
perception of models in science was introduced into wind energy assessment. Models are
specialized representations of parts of reality and reduce the complexity of reality. This kind
of reduction of complexity causes a new type of complexity concerning methodological
decisions in model construction. Examining the relation between models, theories and reality
is necessary if we want to improve models and the interpretation of model output data with
regard to reality.
Another important issue in this article was the comparison of models. It turned out that data
of different models is only comparable if the models are comparable, (Lemma 1). This is
important and in practice non-trivial. However, the quantification of data comparison is not
sufficient if models are to be improved and the results are interpreted with regard to reality. It
is necessary that the underlying aspects are clarified, such as assumptions of theories
involved in the models. Further on, model improvement was analyzed. It was found that
model improvement is an empirical and recursive process and thus depends on the quality of
the validation data. It turned out that windtunnel data in wind energy assessment can serve as
high quality validation and high quality input data for generation of artificial experience in
compatible numerical models, (Lemma 2).
To end, the authors look forward to fruitful discussion and obtaining feedback on that topic.
It is a difficult task to understand the role of models and their applications. Models are being
used for predictions towards reality and thus as basis for economical, political and social
decisions. Only with strong interdisciplinary work the benefit of models can be increased, not
only for wind energy assessment but in general.
5 REFERENCES
Rodrigo, J.S.; State-of-the-Art of Wind Resource Assessment; CENER, 2010

Magnani, L.; Nersessian, N. J. & Thagard, P. (ed.); Model-based reasoning in scientific discovery;
Kluwer Academic/ Plenum Publishers, 1999

Morgan, M. S. & Morrison, M. (ed.); Models as Mediators; The Press Syndicate of The University of
Cambridge, 1999

Schatzmann et al.: COST 732: Quality assurance and improvement of microscale meteorological models -
Model evaluation case studies: approach and results, 2010

Petersen, G., Leitl, B & Schatzmann, M; On Proper Physical Simulation of Turbulent Atmospheric Flow over
Hills; Brussels EWEA proceedings, 2011

Petersen, G., Leitl, B & Schatzmann, M; ABL flow over hills: A review on theory and critics of recent wind
tunnel studies; ICWE 13 proceedings, Amsterdam, 2011b

Oberkampf, W. L. & Trucano, T. G.; Verification and validation in computational fluid dynamics
Progress in Aerospace Sciences, 2002, 38, 209 - 272

Bechmann, A.; Johansen, J. & Sorensen, N.; The Bolund Experiment -Design of Measurement Campaign using
CFD; 2007

Snyder, W. H.; Guideline for Fluid Modeling of Atmospheric Diffusion; United States Environmental
Protection Agency, 1981

Cermak, J. E.; Physical modelling of flow and dispersion over complex terrain; Boundary Layer Meteorology,
1984, 30, 261-292
236
Model evaluation methodology for wind energy
applications
H. A. Holmes
a
, M. Schatzmann, B. Leitl
Meteorological Institute, University of Hamburg, Hamburg, Germany,

a
heather.holmes@zmaw.de
ABSTRACT: Numerical models used for wind energy prediction undergo verification and
validation by the developers to determine the reliability of the results. However, each
modeler utilizes a different procedure for this evaluation process, and has access to different
datasets for the evaluation. To improve the results obtained from wind energy models a
database of quality checked experimental data, from physical models and field data, must be
accessible to numerical modelers. The objective is to develop methods that provide guidance
and outline quality check procedures for experimentalists and numerical modelers to ensure
consistency and improve the modeled results. Due to the different scales being modeled,
varying model applications and computational advancements several types of models are
implemented to evaluate wind energy potential. It is expected that each type of model will
require a separate evaluation procedure. However, this paper will present a general
evaluation procedure with purposed validation metrics, and provide an example model
evaluation process for flow in complex terrain.
1 INTRODUCTION
Literature shows a large range of wind turbine power output predictability, with under and
over prediction, from blind comparisons of numerical simulations with both field and wind
tunnel test data. Blind comparisons from the National Renewable Energy Laboratory (NREL)
in the United States indicate a range of power output predictability from 60% underprediciton
to 150% overprediction, for a wind turbine with simple unyawed, unstalled operating
conditions (Leishman, 2002). While it is unlikely to be achieved, The European Wind Energy
Technology Platform (TPWind) in their Strategic Research Agenda published in 2008
declared that the goal should be to reduce model uncertainties to within 3% in wind energy
assessment by the year 2030. Standardized quality assurance procedures need to be
developed to improve the reliability of power output calculations, especially because the
economic feasibility assessment for wind energy projects relies on these models. Model
developers must verify and validate the numerical model output, but currently a standardized
procedure for this model evaluation does not exist. To improve the results obtained from
wind energy models a common evaluation methodology and database of quality checked
experimental data must be accessible to numerical modelers.

The objective is to develop methods that provide guidance and outline quality check
procedures for experimentalists and numerical modelers to ensure consistency and improve
the modeled results. The intent is not to evaluate and rank individual models, but to work
simultaneously with modelers and experimentalists in developing a consensus to establish
quality assurance methods. Due to the different scales being modeled, varying model
applications and advancements in computational fluid dynamics (CFD) several types of
237
models are implemented in wind energy assessment. Each type of model will require a
different evaluation procedure, especially for the surface mean wind speed models versus the
fine spatial resolution models that investigate flow in complex terrain or around a wind
turbine. This work will focus on the fine resolution models, typically CFD models, to
investigate wind turbine wake behavior and flow phenomena in complex terrain.
Specifically, the focus will be model evaluation for CFD models used to investigate flow in
complex terrain for wind energy applications.

The work to be presented will focus on the methodology for model evaluation and the
importance of establishing a common procedure for validation. Additionally, proposed
quantities of interest and validation metrics will be given, including an example of using
these metrics for flow in complex terrain. The purpose is to present an initial framework of
the evaluation methodology specific to wind energy, to initiate discussion so input from the
research community can be obtained.
2 BACKGROUND
The procedure for this work will benefit form the work done in the COST Action 732, which
determined model evaluation guidance and protocols for urban dispersion modeling (Franke
et al., 2007; Britter and Schatzmann, 2007b). While the objectives in wind energy modeling
differ from those of urban dispersion, similar methodology is used to develop the physical
and mathematical models. Therefore, the procedures outlined in COST 732 can be utilized as
a starting point, particularly the emphasis on high quality experimental data and the fitness
for purpose criteria. The background review and investigation into previous model
evaluation activities reported in the COST 732 documentation identified six steps to properly
evaluate a numerical model (Britter and Schatzmann, 2007b). Below the six steps are listed
and briefly summarized, where the reference to model implies a numerical simulation.

1. Model Description
Brief description of the model and the purpose for which the model was developed. The
theoretical background of the model should be presented, including the limitations,
assumptions and applicable range. If the model is derived from experimental data this
should be mentioned, explicitly stating the dataset used for development so it will not be
used for evaluating the model performance.

2. Database Description
Description and justification of the experimental datasets chosen to be compiled into an
accessible database for numerical modelers. The datasets selected to make up the
database should be of high quality, and must follow quality assurance methods for data
collection and processing steps. The uncertainty and variability in the data should be
estimated and included in the database.

3. Scientific Review
Ensuring the equations used to describe the real world physical process are acceptable.
This includes a description of the equations, explanation and justification of the
numerical model and its limitations with respect to applications of use.

238
4. (Code) Verification
Evaluation of the implementation of the equations into a computer model. This step is
done to ensure there are no mistakes in the computer code or inaccuracies in the
mathematical equations.

5. Model Validation
Comparing the model outputs with experimental data in a quantitative statistical manner.
This step is intended to determine how well the numerical model predicts the real world
process. Quantities of interest should be chosen, along with the validation metrics to
quantitatively determine the performance of the numerical simulation.

6. User-oriented Assessment
Providing documentation and information on the numerical model so an individual user
is able to read about the model development and applicability. This includes material
that allows the user to install, setup and operate the numerical code or software.
Additionally, training or best-practice guidelines for using the model are included here.

All six of these steps are necessary for the proper evaluation of a numerical model. Of the six
steps listed the most difficult and ambiguous is step 5, Model Validation. This step does not
have explicit requirements, and can be approached with several questions regarding the
criteria and methodology to validate a numerical model. Determining the quantities of
interest are the first priority, then deciding if these should be evaluated as time or space
dependent quantities follows. Along with the space and time dependence, the question of
averaging times and locations for the comparison arise. Once the quantity of interest is
identified and the appropriate spatial, temporal and averaging method used for the
comparison are selected the steps for statistical comparison are implemented.

Again, this may seem straight forward, but there are many ways to statistically compare the
different quantities of interest. These are the validation metrics, and are necessary to create a
uniform method for comparing numerical results with experimental data. As part of a
validation procedure these metrics must be clearly defined, and be accompanied with
quantitative criteria for determining the success of the model output. These validation metrics
and criteria should not be viewed as ranking or excluding numerical models, but rather as a
way for the modeling community to compare the model outputs in a uniform manner.
It is important to mention that there are several sources for error and uncertainty in the
evaluation methodology. In every step of the evaluation procedure uncertainties influence the
accuracy of the model validation. The numerical model itself has uncertainty in the ability of
the model to accurately describe a real world process. The data used to develop the models,
as inputs to the model or for the model comparison, also have uncertainty due to limitations
in the measurement equipment or experimental setup. Finally, uncertainties can come from
the variability inherent in the physical process being modeled, for wind energy this is due to
atmospheric processes and turbulence.
3 MODEL EVALUATION
In this section an outline of the suggested evaluation methodology for numerical simulations
used in wind energy is given, with an emphasis on blind comparison. The objective of model
validation is not to use experimental data for model improvement, but to blindly test the
239
outputs from numerical models with experimental data. Included in this outline are protocols
to ensure proper data collection methods so high quality datasets will be obtained. Also,
proper documentation for experimental studies (i.e., field and wind tunnel) is required for the
data to be used in numerical model validation. Guidelines for validation experiments for CFD
can be found in Oberkampf and Trucano (2002) and should be considered in the development
of experimental guidelines. The following are critical tasks that will be addressed in the
development of the model evaluation procedure.
1: Develop guidelines for proper documentation and post-processing of field data
2: Develop procedures for physical modeling to ensure quality data collection
3: Create database with high quality experimental data, quantifying uncertainty
4: Using the above database, design evaluation procedures for numerical models
5: Implement evaluation tool, ensures accurate comparison and provides documentation
The items listed above will influence the quality of data used in each step of the model
evaluation. It is recommended that databases of models and experimental data be created to
simplify the record keeping and documentation required for validation. For the numerical
models, this can be done in the format of a modeling inventory with information collected
using a standardized questionnaire with all relevant model information documented. Also, a
data repository is suggested to keep the validation datasets in one location and prevent
manipulation of the datasets.
3.1 Validation Procedure
To apply the methods presented above to wind energy applications the first step is to define
the quantities of interest and the proper metrics for validation so they are the most useful.
These steps can be referred to as the validation objective, and should be clearly defined prior
to conducting the evaluation study. The list presented here are proposed quantities of interest
and validation metrics, given to prompt discussion. Input from numerical modelers in the
wind energy community will be instrumental in defining the validation objective and guiding
the selection of validation metrics and their criteria. The following is a brief summary of
each step required to develop a model validation methodology. More detail for this
procedure and additional background information can be found in Holmes et al. (2011).
Step 1. Determine Quantity of Interest
Mean Wind, Wind Direction, Turbulent Kinetic Energy, Turbulence Intensity

Step 2. Important Considerations Required for Validation
What time average should be used for comparison?
Should extreme events be considered in addition to means?
How many spatial locations are required for each case, where should they be?
Vertical profiles, what should experimentalists measure (difficult in field studies)?

Note: These considerations will greatly influence experimental data collection. While
limitations exist in wind tunnel modeling, the data for comparison allows for a better
spatial and temporal resolution of experimental data. Therefore, these considerations
should be evaluated in the context of both field and wind tunnel experiments.

Step 3. Generate Plots to Visualize Comparison (Qualitative)
Scatter Plot: correlation between experimental data and model outputs for a quantity
of interest
Quantile-quantile Plot: correlation between the probability distributions of a quantity
of interest
240
Residual Plot: model performance (ratio of predicted to observed) as a function of
independent variables (i.e., stability, time of day, wind speed) to investigate
the influence of these variables on the results

Step 4. Use Validation Metrics for Statistical Comparison (Quantitative)
The following are commonly used validation metrics, correlation coefficient (R),
Fractional Bias (FB), Normalized Mean Square Error (NMSE), Geometric Mean
(MG), Geometric Variance (VG), Fraction of predictions within a factor of two of the
observations (FAC2) and Hit Rate (Holmes et al., 2011). Each of these metrics
represents a unique way to relate the model data with observations, therefore the
metric chosen for validation should be the most relevant for the application of interest.
Determining validation metrics is not a trivial task and several criteria must be established
and met to ensure that the chosen metric is useful. Several key components of a validation
metric can be found in Oberkampf and Barone (2006). Mainly, the metric should serve as a
key variable in quantifying the agreement between the numerical simulation and
experimental data for the particular real world process of interest. In addition, the metrics
should take into account different sources of error and uncertainty that arise during the
evaluation procedure.
4 VALIDATION EXAMPLE
The following is an example of how to apply the above procedure to a test case using data
from the Bolund blind comparison study (Bechmann et al., 2009). The Bolund study was
selected to improve flow modeling in complex terrain, therefore flow over an isolated hill
was investigated. Figure 1 shows the elevation profile of the Bolund hill, with the numbers
along the top denoting the locations of meteorological towers for measurements. For this
example only westerly winds are investigated, or flow moving from left to right in Figure 1.
Step 1, Identify quantities of interest; in this case mean wind speed and the turbulent kinetic
energy (TKE) are considered. The mean wind speed in this example is represented as the
total mean wind speed obtained from vector scaling (S = (U
2
+V
2
+W
2
)
1/2
, where a capital
letter denotes an average) and accounts for the wind speed in all three directions.
Step 2, Important considerations, for this study these were already selected by the researchers
conducting the blind comparison: 10-min averaged field data and only time periods with a
neutral atmospheric boundary layer (ABL). The 10-min periods that met the criteria for
wind direction and stability were ensemble averaged for the comparison.

Figure 1: Profile of Bolund hill showing elevation and monitoring locations for meteorological data.
241
Figure 2: Plot showing the fractional speed-up ratio (FSR = (S-S
0
)/S
0
) for the Bolund field data and two
numerical simulations, top: z= 5m and bottom: z = 2m.
Step 3, Plot quantities of interest, shown in Figure 2 for mean wind speed represented as the
fractional speed-up ratio (FSR = (S-S
0
)/S
0
). The red star in front of the hill (x = -180m) is the
reference mast and the wind speeds at all locations are normalized by the wind speed at this
location. Two heights are shown in the figure, five meters in the top figure and two meters in
the bottom, with data from two different numerical simulations and green circles representing
the field data.
Step 4, Plot the visual metrics to compare the numerical output with experimental data, as
shown in Figure 3 for the total mean wind speed (S) for Bolund. Note that using field data
limits the number of data points that can be compared, for this example there are nine data
points that can be directly compared for the numerical output and experimental data.
Step 5, Compute the validation metrics for a quantitative comparison between the models and
experimental data. This provides a uniform way of comparing the models so they can be
compared side by side. Five validation metrics are computed for two numerical models,
using two different quantities of interest, shown in Table 1. These metrics show the
importance of selecting the proper quantity of interest and metric for a specific application.

Figure 3: Scatter plot comparing mean wind speed (S) for Bolund field data with two numerical simuatlions.
R
2
= 0.8649 R
2
= 0.8146
R
2
= 0.9332
R
2
= 0.7608
0
0.2
0.4
0.6
0.8
1
1.2
1.4
0 0.2 0.4 0.6 0.8 1 1.2 1.4
S/S0 [Field Data]
S
/
S
0

[
N
u
m
e
r
i
c
a
l
]
CFDWind1 z=2m CFDWind2 z=2m
CFDWind1 z=5m CFDWind2 z=5m
Linear (CFDWind1 z=2m) Linear (CFDWind2 z=2m)
Linear (CFDWind1 z=5m) Linear (CFDWind2 z=5m)
242
Table 1: Validation metrics to compare the performance of two numerical models with field data.
CFDWind1 CFD Wind2

S TKE S TKE Perfect
Fractional Bias (FB)
0.149 0.140 0.133 0.239 0
Normalized Mean Square Error (NMSE)
0.051 0.544 0.057 0.668 0
Geometric Mean (MG)
1.108 0.994 1.066 1.094 1
Geometric Variance (GV)
1.046 1.199 1.082 1.188 1
Factor of Two Observations (FAC2)
1 0.778 1 0.889 1
For example, both models behave similarly based on the metrics for S, but for TKE the
CFDWind1 model performs better according to the FB and NMSE, while according to the
FAC2 the CFDWind2 model is better. Therefore, the selection of these quantities of interests
and validation metrics is not a trivial task, and it is critical to reach a consensus for the
selection of both in the wind energy community.
5 DISCUSSION
The Bolund example provides a starting point to develop a validation methodology for wind
energy applications. However, due to the limited amount of data and limitations in the spatial
and temporal resolution of the field study, the use of wind tunnel data for these investigations
arises. While the atmospheric variability is difficult to capture in wind tunnel investigations,
modeling of neutral ABL processes can be done to obtain experimental datasets with high
spatial and temporal resolution.
A scatter plot (Fig. 4) taken from the work done in COST 732 shows a comparison of mean
wind speeds from a numerical simulation with wind tunnel data, using 566 data points for the
comparison. While there appears to be a significant amount of scatter and the validation
metrics may perform worse with more data points, the better spatial resolution ensures a
better estimation of the small scale flow physics. Therefore, considerations for metrics that
include the use of more data points should be included to weight the incorporation of flow
complexity in the model evaluation.

Figure 4: Left: Locations for comparison wind tunnel data (blue closed circle) and numerical output (red open
circle) each point has up to 29 vertical locations. Right: Scatter plot comparing mean wind speed from wind
tunnel data with a numerical simulation, using 566 data points.
R
2
= 0.9023
-0.5
0
0.5
1
1.5
-0.5 0 0.5 1 1.5
U/U0 [Wind tunnel]
U
/
U
0

[
M
o
d
e
l
]
243
6 SUMMARY
This paper gives a brief background for a numerical model evaluation methodology and
provides an example validation procedure. Considerations specific to validating numerical
models used in wind energy assessment are given to generate discussion among the research
community. While wind tunnel data was not available for this analysis, the benefit of
incorporating wind tunnel data into the model evaluation procedure for wind energy
applications is discussed. Establishing a consensus among wind energy modelers for the
development of these methods is desired. Input from modelers and experimentalists is
critical to ensure the use and applicability of the quality assurance protocols to be developed.
The final objective is to establish threshold criteria for model validation and apply the
methodology to a test case model evaluation specific to wind energy assessment.
7 ACKNOWLEDGEMENTS
The authors are grateful for financial support from the EU FP7-PEOPLE program, under
WAUDIT Marie-Curie Initial Training Network.
8 REFERENCES
AIAA, 1998. Guide for the Verification and validation of computation fluid dynamics simulations. American
Institute of Aeronautics and Astronautics, Reston, VA, AIAA-G-077-1998.
Bechmann, A., Berg, J., Courtney, M., Jrgensen, H., Mann, J., Srensen, N., 2009. The Bolund Experiment:
overview and background. Ris DTU report Ris-R1658(EN).
Britter, R., Schatzmann, M. (Eds.), 2007a. Background and justification document to support the model
evaluation guidance and protocol document. COST Office, Brussels.
Britter, R., Schatzmann, M. (Eds.), 2007b. Model evaluation guidance and protocol document. COST
Office, Brussels.
Chang, J. C., Hanna, S. R., 2004. Air quality model performance evaluation. Meteorology and Atmospheric
Physics, 87, 167-196.
Franke, J., Hellsten, A., Schlnzen, H., Carissimo, B. (Eds.), 2007. Best practice guideline for the CFD
simulation of flows in the urban environment. COST Office, Brussels.
Holmes, H., Schatzmann, M., Leitl, B., Sanz Rodrigo, J., 2011: Quality assurance of wind energy assessment
models, in: Proceedings of the 13
th
International Conference on Wind Engineering, Amsterdam, the
Netherlands.
Leishman, J. G., 2002. Challenges in modelling the unsteady aerodynamics of wind turbines. Wind
Energy, 5, 85-132.
Oberkampf, W. L., Barone, M. F., 2006. Measures of agreement between computation and experiment:
Validation metrics. Journal of Computational Physics, 217, 5-36.
Oberkampf, W. L., Trucano, T. G., 2002. Verification and validation in computational fluid dynamics.
Progress in Aerospace Sciences, 38, 209-272.
Sanderse, B., van der Pilj, S. P., Koren, B., 2011. Review of computational fluid dynamics for wind
turbine wake aerodynamics. Wind Energy, doi: 10.1002/we.458.
Schatzmann, M., Leitl, B., 2002. Validation and application of obstacle-resolving urban dispersion models.
Schatzmann, M., Olesen, H., Franke, J. (Eds.), 2010. COST 732 model evaluation case studies: approach
and results. COST Office, Brussels.
Schlesinger, S., 1979. Terminology for model credibility. Simulation, 32, 103-104.
Troen, I., Petersen, E. L., 1989. European wind atlas. Ris National Laboratory, Roskilde.
Vermeer, L. J., Srensen, J. N., Crespo, A., 2003. Wind turbine wake aerodynamics. Progress in
Aerospace Sciences, 39, 467-510.
244
Session 3 Papers
S-III. Applied Physical Modelling
Wednesday, August 24
9.009.20, S-III.1
9.209.40, S-III.2
9.4010.00, S-III.3
10.0010.20, S-III.4
10.4011.00, S-III.5
11.0011.20, S-III.6
11.2011.40, S-III.7
11.4012.00, S-III.8
13.2013.40, S-III.9
13.4014.00, S-III.10
14.0014.20, S-III.11
14.2014.40, S-III.12
14.4015.00, S-III.13
15.2015.40, S-III.14
15.4016.00, S-III.15
16.0016.20, S-III.16
16.2016.40, S-III.17
245
NOTES AND COMMENTS:
246
NOTES AND COMMENTS:
247
NOTES AND COMMENTS:
248
Atmospheric dispersion simulation:
The Dutch round-robin wind tunnel test
drs. S.W. van Ratingen
a
, ir. K. Artois
b
, ir. N.L.S. Moonen
c
a
TNO, Utrecht, Utrecht, The Netherlands, sjoerd.vanratingen@tno.nl
b
German-Dutch Wind Tunnels (DNW), Emmeloord, Flevoland,
The Netherlands, koen.artois@dnw.aero
c
Peutz bv, Mook, Limburg, The Netherlands, n.moonen@mook.peutz.nl
ABSTRACT: TNO, German-Dutch Wind Tunnels (DNW) and Peutz participated as
independent wind tunnel laboratories in the development of a measurement protocol for
atmospheric dispersion simulation in wind tunnels and a consecutive wind tunnel round-robin
test. This round-robin test was organized to compare results of different laboratories all
working according to the measurement protocol as well as to obtain a comparison with long
term measurements along the Dutch highway A28. This paper describes the different
approaches in conducting the experiments and compares the results of the three laboratories.
1 INTRODUCTION
DNW, Peutz and TNO participated in the development of a measuring protocol for
atmospheric dispersion simulation: Measuring protocol wind tunnels Air quality (draft
version). Development of the protocol was initiated by the Air Quality Innovation
Programme (IPL) for the Ministries of Transport and Environment (VenW and VROM). The
protocol was developed in order to set a uniform quality standard for dispersion experiments
by means of wind tunnel simulation. To evaluate the draft protocol a round-robin test for
wind tunnel laboratories was organized to compare results of different laboratories all
working according to the measurement protocol. Another goal of the round-robin test is to
obtain a comparison with long term field measurements at a specific test location along the
Dutch highways. The round-robin experiment was specifically aimed at determining the
effect of acoustic screens on air quality. This paper will describe the test setup and
atmospheric boundary layer simulation for the three wind tunnels and present the measured
screen effects.
2 THE PROTOCOL
The protocol was developed to give a uniform set of boundary conditions for dispersion
experiments aimed at determining the effects of acoustic screens or tunnel exhausts on air
quality by means of wind tunnel simulation. The objective of the protocol is to ensure that
wind tunnel simulation provides a reliable prediction of air quality that can be reproduced and
verified also by other wind tunnel laboratories. Key elements of the protocol are requirements
on the velocity profile and turbulence properties of the boundary layer as specified by
ESDU(2001) and VDI(2000), tracer gas injection, gas analysis, independence of Reynolds
number, independence of source strength and repeatability in time.
249
3 FIELD EXPERIMENT CHARACTERISTICS OF THE ENVIRONMENT
The location of interest is an experimental setup of noise barriers situated along highway A28
near Putten close to Nulde beach. The A28 is oriented in north south direction. In each
direction the highway has two lanes and a hard shoulder. The barriers are positioned on the
western side of the road at 8m from the roadside. The surrounding area is mainly grassland
on the eastern side, grassland, some trees and water on the western side. The field
measurements are conducted on the western side at 5m, 10m and 28.5m behind the screens.
An overview of the location is shown in figures 1 and 2.
The setup consists of 3 barriers, each of which is approximately 100m long and an equally
long stretch of open area to the north of the barriers to serve as a reference. Two of the
barriers are 4m high (north), one is 7m high (south). Between the 7m barrier and the 4m
barrier there is a gap of approximately 11m. Between the 4m barriers there is a gap of
approximately 68m, this gap is filled with several rows of trees of approximately 10m high.
To the south of the 7m barrier there are several rows of trees approximately 15m high. The
field experiment consists of several long-term measurements to determine the effect of noise
barriers on the dispersion of road traffic emissions.
4 TEST SETUP DNW
4.1 The wind tunnel
The DNW-LST is an atmospheric, closed-circuit wind tunnel. The wind speed can be varied
from 1.5m/s up to 80m/s. The DNW-LST has a test section with a cross section of 3.0m wide
and 2.25m high. The total length of the test section is 8.75m. The forward part with a length
of 5.75m is used for aeronautical testing, whereas the aft part is used for non-aeronautical
(industrial aerodynamics) testing. So for atmospheric boundary layer simulation the available
length for the roughness area in front of the model is about 6m. The non-aeronautical part of
the DNW-LST test section has a turntable of 2.4m in diameter.
4.2 The wind tunnel model
To model the area and the noise barriers situated along highway A28 near Putten a scale of
1:400 is applied. This scale is chosen mainly because a correct scaling of turbulence
parameters could only be guaranteed at a 1:400 scale. With a 1:400 scale also the entire test
field including a reasonable area around the barriers on the LST turntable can be captured
with a single model. The modeled area is shown in figure 2.
With a 1:400 scale a single barrier becomes about 10mm (4m full scale) high and 250mm
(100m full scale) long in the tunnel. This means the barrier become smaller than the 20mm
minimum height requirement in the measuring protocol. However DNW considered dynamic
similarity not to be an issue for a sharp-edged model like the Putten model.
The highway was modeled by 2 x 3 line source elements of each 600mm. So two lines of
1,8m (720m full scale) are used to simulate 2 x 2 traffic lanes. Traffic induced turbulence was
not simulated in the wind tunnel.
To compensate for the lack of surface roughening elements in the open area around the test
field in Putten, the surface of the model was covered with very rough sand paper.
The concentrations are measured using small suction tubes with an outer diameter of about
5mm. These tubes are located at 5m, 10m, 28.5m and 100m behind the center of the barriers.
A picture of the model in the DNW-LST is shown in figure 3.
250
5 TEST SETUP PEUTZ
5.1 The wind tunnel
The Peutz wind tunnel is a closed return type tunnel with a test section of 3.5m wide and
1.8m high, the model is placed on a rotating table of 2.3m diameter and can be rotated for
simulation of different wind directions. The length of the roughness area in front of the model
is approximately 9m.
In the wind tunnel model the highway is simulated with line source elements. Each lane is
simulated with one row of line source elements. Altogether four lanes of line source elements
were used to simulate the highway. For practical reasons the measurements were performed
for two lanes at a time. So all measurements were taken for lanes 1 and 3 at the same time
and for lanes 2 and 4 later on. The wind tunnel measurements are conducted at 5m, 10m,
28.5m and 100m behind the screens. In addition to these measurement stations Peutz took
measurements at 57m and 200m from the barriers.
Boundary layer properties matched for a model scale of 1:400. For wind directions with a
small angle to the road axis the concentrations at the measurement stations will be influenced
by a long stretch of the road upstream, limited by the size of the wind tunnel. Model scale
was also chosen as 1:400 to capture a large part of the road for these conditions. At this
model scale the 4m barriers become smaller than the 20mm minimum height requirement in
the protocol. However Peutz considers dynamic similarity not an issue for this model.
Surface roughness was applied to the model area to prevent large changes in the boundary
layer profile over the model area. The dispersion measurements were taken at 1.5m over the
surface (full-scale). An overview of the model in the Peutz wind tunnel is shown in figure 4.
6 TEST SETUP TNO
6.1 The wind tunnel
The wind tunnel experiments of TNO have been carried out in open circuit wind tunnel with
a cross-section of 2x3 m
2
. The model is placed on a turntable with a diameter of 2.3m. The
total length of roughness area in front of the model is about 15m.
The measurements have been carried out according to the Dutch Measuring Protocol Wind
Tunnels Air Quality. In this protocol a minimum height of the screens of 20mm is given.
Therefore, given the lowest screens of 4m, the model scale must be 1:200 or lower according
to the protocol. On this scale it is not possible to model the whole area in one time in the
wind tunnel on a model scale 1:200. The total length of the measuring area, including groves,
is more than 500m. Therefore two models with a substantial overlap have been used: a model
of the northern area with the zero measuring line and the two screens with the height of 4m
and a model of the southern area with the 4m screen, the 7m screen and groves. No other
obstacles are present in the Nulde area. Figure 2 gives a lay-out of the situation.
251
The highway has been modeled by two line sources in the middle of each side of the road.
The line sources consist of a line of small holes with a plate above it to eliminate the vertical
momentum of the speed through the holes.
The measurements have been carried out statically, without traffic-induced turbulence.
Traffic-induced turbulence decreases the concentration levels along a highway. Therefore the
effect of the traffic-induced turbulence on the concentrations along the highway has been
introduced afterwards by means of a correction factor in the static measurements. This
correction factor has been derived from dynamic measurements (with traffic-induced
turbulence included in the wind tunnel experiment).
The model of the southern part of the Nulde area is given in figure 5.
7 RESULTS
7.1 Boundary layer simulation
The results of the three wind tunnels are referred to as W1, W2 and W3.
Figure 6 and 7 show the vertical profiles for the mean velocity and the longitudinal
turbulence intensity as measured by the participating wind tunnels. The roughness height,
derived from the velocity profiles, as well as the turbulence intensities lie, for the largest part,
within the range for grassland, prescribed by ESDU (2001).
7.2 Comparison between wind tunnels
All tunnels showed independence of the C* on wind tunnel velocity and source strength.
Figure 8 shows the concentration coefficients (C*) as measured by the three participants. C*
are plotted for 3 wind directions (60, 90 and 120 ) and 3 measurement lines corresponding tot
the 0-situation, the 4 meter high screen and the 7 meter high screen. Each measurement line
contains 4 measurement locations at distances 5, 10, 28.5 and 100 from the acoustic screen
position.
The results show no dependence of the C* on the choice of model scale. Therefore the
demand in the protocol regarding the minimum barrier height, which implies that the size of
the scale model should be scale 1:200 or larger, seems too strict.
Comparison between 1:200 and 1:400 scales for some measurement points at wind directions
30 and 150 (wind direction 90 is perpendicular to the highway) was not possible. For these
combinations of wind directions and distance to road, the differences in modeled full-scale
length of the line source, can affect the measured concentrations. In wind tunnel studies, the
contribution to the concentration from outside the modeled area can be computed using
dispersion modeling.
The screen effect, based on concentration coefficients, is defined as 1 C*
screen
/ C*
0-situation.
Figure 9 shows the screen effect for the 4 high and the 7 meter high screen, averaged over the
measured wind directions. For distances from the screen up to 28.5 meters, the absolute
difference in measured screen effect between the three wind tunnels varies between 2% and
17%. The differences between the screen effects, measured at large distance from the screen
(100 meter) can become large because of the small values of the measured concentrations
contributions. Air quality studies in wind tunnels, however, mainly focus on the high
concentrations contributions which are measured close to the highway.
The post processing of the measured C* into yearly average pollutant concentrations PM
10

and NO
2
is performed by each wind tunnel in a different manner. The correction for vehicle
252
induced turbulence is performed either by multiplication of the C* by a constant factor, or by
using a correction depending on hourly wind speed.
Yearly average pollutant concentrations (including NO/NO
2
conversion) are calculated either
by using an hour-by-hour method or by using a finite number of meteorological classes (12
wind directions, 3 wind speeds).
The average screen effect behind the 4 meter screen ranges from 29% to 36% for PM
10
and
24% to 30% for NO
2
. Behind the 7m screen this is 33% to 46% for PM
10
and 31% to 39% for
NO
2
. Figure 10 shows the NO
2
screen effect.
8 CONCLUSIONS
Despite the use of a commonly established measuring protocol there are still interesting
differences in the approaches of the Dutch wind tunnel laboratories in conducting the
atmospheric dispersion experiments. Evaluation of the protocol and the ring test resulted in
some adjustments of the protocol which included a less strict demand on the choice of model
scale, less frequent measurements of the boundary layer profiles and a more elaborate
definition of the concentration coefficient.
For distances from the screen up to 28.5 meters, the absolute difference in measured screen
effect between the three wind tunnels varies between 2% and 17%.
Differences between the concentrations coefficients as measured by the wind tunnels in the
present study may be caused by different line source behavior. Stricter demands on line
source behavior could improve the agreement between the wind tunnel results even more.
The post processing of C* to pollutant concentrations requires further harmonization.
9 ACKNOWLEDGEMENT
The authors would like the thank ir. E. Willemsen, ing. J. Takens, ir. S.P.M. van den Akker,
ir. J.F.W. Koopmans, ing. G.Th.Visser, ir. T.A.J. Cornelissen, ing. H. Spoelstra and Dr. B.
Leitl for their contribution to the experiments and the preparation of this paper.
10 REFERENCES
ESDU, 2001. ESDU 85020 Characteristics of atmospheric turbulence near the ground. Part II: single point data
for strong winds (neutral atmosphere).
VDI, 2000. VDI 3783 Environmental meteorology. Physical modeling of flow and dispersion processes in the
atmospheric boundary layer. Application of wind tunnels. Beuth Verlag, Berlin.

253
FIGURES

Figure 1: Overview of the test field along highway A28 near Putten (picture: Maquette Studio Stens).

Figure 2: Overview of the location and the model area.
254
4 m screen 4 m screen 7 m screen
Measurement probes
4 m screen 4 m screen 7 m screen
Measurement probes

Figure 3: Close-up of the modeled noise barriers in the DNW-LST.

Figure 4: Overview of the model for the 7m barrier in the Peutz wind tunnel.

Figure 5: Model of the southern area in the TNO wind tunnel.
255

1
10
100
1000
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
U/U
wt
[-]
H
fu
ll s
c
a
le

[
m
]

W1Uwt=11.46m/s
W2Uwt=5m/s
W3Uwt=12m/s
W1fit logz0=0.08m
W2fit logz0=0.04m
W3fit logz0=0.052m
1
10
100
1000
5 10 15 20 25 30
Turb intensity [%]
H
fu
ll s
c
a
le

[
m
]
W1
W2
W3
ESDUz0=0.01m, v10=10m/s
ESDUz0=0.1m, v10=10m/s
ESDUz0=0.01m, v10=4,5m/s

Figure 6: Longitudinal velocity component Figure 7: Longitudinal turbulence intensity

Figure 8: Concentration coefficients (C*) for the three participating wind tunnels as a function of configuration
and measurement line.

Measurement line without screen
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
Measurement line with 4 meter heigh screen
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
Measurement line with 7 meter heigh screen
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
0
0.1
0.2
0.3
0.4
0.5
0.6
0 20 40 60 80 100 120
distance to screen
C
*
W1
W2
W3
w
i
n
d

d
i
r
e
c
t
i
o
n

=

6
0

w
i
n
d

d
i
r
e
c
t
i
o
n

=

9
0
w
i
n
d

d
i
r
e
c
t
i
o
n

=

1
2
0
256
NO2 screen effect: screen height 4 meter, angle average
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0 20 40 60 80 100 120
distance to screen
S
c
r
e
e
n

e
f
f
e
c
t
W1
W2
W3
average
NO2 screen effect: screen height 7 meter, angle average
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0 20 40 60 80 100 120
distance to screen
S
c
r
e
e
n

e
f
f
e
c
t
W1
W2
W3
average
C* screen effect: screen height 4 meter, angle average
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0 20 40 60 80 100 120
distance to screen
S
c
r
e
e
n

e
f
f
e
c
t
W1
W2
W3
C* screen effect: screen height 7 meter, angle average
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0 20 40 60 80 100 120
distance to screen
S
c
r
e
e
n

e
f
f
e
c
t
W1
W2
W3

Figure 9: C* screen effect averaged over wind directions for measurement line behind 4 meter screen and
measurement line behind 7 meter screen.

Figure 10: NO
2
screen effect, averaged over wind directions for measurement line behind 4 meter screen and
measurement line behind 7 meter screen.
257
Atmospheric Dispersion Wind Tunnel Simulation
The first Dutch Round robin-test at Putten
Lessons learned in wind tunnel simulation and field
data analysis
ir. N.L.S. Moonen
a
, ir. S.P.M. van den Akker
b
, ir. J.F.W. Koopmans
c
a
Peutz bv., Mook, The Netherlands, n.moonen@mook.peutz.nl
b
Peutz bv., Mook, The Netherlands, s.vandenakker@mook.peutz.nl
c
Peutz bv., Mook, The Netherlands, f.koopmans@mook.peutz.nl
ABSTRACT: The Dutch Directorate-General for Public Works and Water Management
(Rijkswaterstaat, RWS) initiated the development of a measurement protocol for atmospheric
dispersion simulation by means of wind tunnel testing. A round robin test for wind tunnel
laboratories was organized to compare results of different laboratories all working according
to the measurement protocol as well as to obtain a comparison with long term field
measurements. This paper aims to share some of the practical lessons learned in exploring
new fields of wind tunnel research.
1 INTRODUCTION
The Dutch Directorate-General for Public Works and Water Management (Rijkswaterstaat,
RWS) initiated the development of a measurement protocol for atmospheric dispersion
simulation by means of wind tunnel testing [1]. A round robin test for wind tunnel
laboratories was organized to compare results of different laboratories (Peutz, TNO, DNW)
all working according to the measurement protocol, as well as to obtain a comparison with
long term measurements along the Dutch highway E232/A28. The measurement program was
specifically aimed at determining the effects of noise barriers on air quality.
The test site was characterised by a very low roughness (fields and water) which provided
challenges in the boundary layer modelling and for the line source simulation. From
comparison of the results of the windtunnels it was learned that the test site caused several
unexpected results. Lessons were learned for further improvement of the protocol and further
harmonisation of procedures as well as in the comparison of field data to wind tunnel data.
The usual field of use of wind tunnel dispersion studies is the urban environment. The low
roughness number and the lack of other obstacles means exploring new areas of research.
The test location consisted of a setup of three sound barriers of approximately 100m long.
Two barriers with a height of 4m, one with a height of 7m. Next to the setup of sound barriers
a 0-situation without obstacles is situated. The barriers are situated at 8m from the roadside.
Field measurements were taken at 5m, 10m and 28.5m from barriers. Wind tunnel
measurements were taken at 5m, 10m, 28.5m and 100m. All measurements are taken at 1.5m
above ground level (full scale). A more comprehensive outline of the test setup is described
in [2] and [3].
258
2 Lessons learned
The goal of the round robin test was to test the measurement protocol and to compare wind
tunnel data to field data. From analysis of the wind tunnel results it turned out that the test
location showed unexpected behavior. As a result of working according to the protocol a high
data quality was achieved. Comparison of results of all wind tunnel laboratories and analysis
of differences showed that some important improvements still could be made.
2.1 Reynolds number effects
A key element in the measurement protocol is the requirement that states that the
measurement results should be independent of Reynolds nuber and source strength. Therefore
several series of wind tunnel test velocity (U
10
) and source strength (Q) variations are
measured to show independence of these parameters. Initial tests on the 0-situation (no
barrier) showed consistent results for several combinations of test velocity and source
strength. In figure 1 an example of the results for several ratios of source strength to test
velocity (Q/U
10
) are shown, independence of the results is clearly shown.
Surprisingly the wind tunnel test setup with noise barriers did not behave independently of
test velocity and source strength, see figure 2 (dashed lines) for the extreme conditions (7m
barrier at lowest and highest windspeed). All wind tunnels encountered this behavior in some
form.
This might have led to the conclusion that the flow and the dispersion around the (sharp
edged) noise barriers did not behave Reynolds number independent. In the round robin
experiment two laboratories chose a model scale of 1:400, one laboratory of 1:200. In case of
a Reynolds no. similarity problem one would expect the results between the two model scales
to be different, but also the 1:200 model suffered from the problem.
Further data analysis showed a very strong correlation to test velocity and a weaker
dependency on source strength. This revealed that there was no Reynolds similarity problem
of the flow around the noise barriers but severe pressure gradients disturbing the lateral
Figure 1: Wind tunnel test velocity (U10) and source strength (Q)
variations, situation no barrier (1.5m above ground level)
0 20 40 60 80 100 120
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Wind tunnel data
Situation no barrier, wind perpendicular to road axis, lanes 1+3
Q/U10=1.4
Q/U10=0.9
Q/U10=2.0
Distance to roadside [m]
C
*

[
-
]
259
homogenity of the tracer gas injection. In the test setup the noise barriers are placed very
close to the road and therefore close to the line sources. Due to the very low roughness and
the lack of any other obstacles upwind of the barriers, they are exposed to the maximum
dynamic pressure in the flow causing a high stagnation pressure region upwind of the barrier.
All wind tunnels used different types of line sources but all systems are pressure driven
which means that the actual local source strength depends on the local differential pressure
over the inside and outside of the source.
Aknowledging this effect the simulations can be executed at favorable conditions i.e. low
wind tunnel speed (reduction of stagnaton pressure) and high source strength (increase of line
source pressure drop). Also a changed geometry of the line source could be implemented in
order to increase the pressure drop.
In this simple geometrical situation a theoretical source strength correction could be worked
out and applied to the affected results. Knowing the pressure drop over the line source at a
given source strength and the pressure difference caused by the model the reduced source
strength can be calculated. For a simple geometrical model this source strength correction can
be applied at least for wind directions perpendicular to the road. Incorporating the effect of
the changing angle of incidence of the wind to the barriers also a correction for deviations
around the perpendicular direction can be applied. The correction is shown in equation 1a-1d.
Q
cor
= f
Q
Q
1a)
f
Q
=
QQ
Q
1b)
Q=
P
model
dP
dQ
source
1c)
P
model
=
1
2
U
top barrier
2
sin
2
f
geom
1d)
In which:
Q Source strength
Qcor Corrected source strength
fQ Source strength correction factor
Q Source strength inhomogenity due to pressure change in model
Pmodel Pressure rise in model at source location due to obstacles
(dP/dQ)source Source design parameter
Utop barrier Wind velocity at top of the barrier
Angle of incidence
fgeom Geometrical parameter depending on geometry of barrier and source in the model (here 0,22)
The result of the pressure correction is also shown in figure 2 (continuous lines) and provides
good results for all wind speeds and source strength conditions. The correction varies from
almost zero at the low wind speed and high source strength to almost 25% at the highest wind
speed and lowest source strength.
260
The usual field of use of wind tunnel dispersion studies is in the urban environment. Due to
the low roughness number and the lack of other obstacles other than the noise barriers, which
are placed very close to the line source, the pressure gradient effects were much larger than
generally experienced and expected. In this case the protocol proved very useful in exploring
new areas of research.
2.2 Tracer gas injection
Another fenomena that occurred in the results of one windtunnel was a strange behavior at
the first and second measurement station (1.5m above ground level). Generally it would be
expected that the concentrations are highest close to the road and decrease with distance in
the undisturbed situation. Some measurement series of two of the wind tunnels however
showed a peak value at the second measurement station, see figure 3.
Figure 2: Pressure correction on wind tunnel data (1.5m above
ground level)
0 20 40 60 80 100 120
0.00
0.05
0.10
0.15
0.20
0.25
Stagnation pressure correction
Situation 7m barrier, wind perpendicular to road axis, lanes 1+3
U10=3 m/s; Q=3.3
l/min; p_cor
U10=8 m/s; Q=3.3
l/min; p_cor
U10=3 m/s; Q=3.3
l/min
U10=8 m/s; Q=3.3
l/min
Distance from road side [m]
C
*

[
-
]
Figure 3: Underprediction of concentration at first measurement
station
0 20 40 60 80 100 120
0.0
0.1
0.2
0.3
0.4
0.5
0.6
Wind tunnel data
Situation no barrier, wind perpendicular to road axis, lanes 1+3
Q/U10=1.4
Q/U10=0.9
Q/U10=2.0
C
*

[
-
]
261
By comparison to other windtunnel results it was shown that the measurements at the second
station and further from the road showed a good match and therefore the first station was the
outlier. To further analyse this effect height profiles of tracer distribution close to the road
were measured at both windtunnels. A simplified setup with on row of line sources was used
for this experiment. A 4 m barrier was placed along the full length of the line source at 10 m
from the road axis. In figure 4 the height profiles are shown at a distance of 10 m from the
road axis for W1 and at a distance of 9 m to the road axis for W2. It is obvious that a
difference in the initial height distribution of the tracer gas can cause large differences in
measured concentrations close to the road. At W1 several modifications to the source were
studied to influence the initial height distribution is it was regarded to be too shallow, causing
large gradients in measured concentrations close to the road and close to ground level.
Between wind tunnels it was agreed to aim at an initial height distribution of the tracer which
should be homogeneous for a height range from ground level up to 1-2 m above ground level.
The motivation of this choice is given by an estimate of the average wake size behind
vehicles in which the exhaust gasses are assumed to be dispersed homogeneously.
Figure 4 also shows the result of the modified version of the W1 source which clearly shows
a more homogeneous concentration between 0-1.5 m than the non-modified version. Still
some differences between line sources used in different laboratories have to be accepted.
Measurements with and without a barrier at 10 m from the road axis were taken to show the
effect on the barrier performance. Figure 5 shows vertical concentration profiles at several
stations from the road axis for the non-modified source W1, with and without barrier. Figure
6 shows the vertical concentration profiles for a second modification to the source:
W1_mod2, with and without barrier. It becomes clear that the largest influence of the initial
height distribution is seen in the configuration without barrier, but also at 10 m behind the
barrier (at 20 m from the road axis) a clear difference in the vertical concentration profile is
seen, especially in the region just above the barrier between heights of 4-6 m. It also becomes
clear that the effect of a 4m heigh barrier is still visible up to 160 m form the road axis. In
figure 7 the barrier effect is compared for configurations W1 and W1_mod 2 at h=1.5 m. It is
clearly shown that small changes in initial tracer gas height distribution can cause significant
changes in the measured barrier effect, which also depends on the height. It can be concluded
that a clear guidance in the protocol on the desired initial height distribution of the tracer gas
is needed to obtain comparable results from different experiments.
Figure 4: Comparison of height profile of W1 and W2 at 9-10m
from road axis
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
2
4
6
8
10
12
Comparison of tracer gas heigth profiles W1 and W2
at 10m and 9m from road axis
W1 10m 1:400
W1 modified 10m
1:400
W2 9m 1:200
C* [-]
H

[
m
]
262
Figure 5: Height profiles with and without barrier source
configuration W1
0 0.2 0.4 0.6 0.8 1 1.2 1.4
0
5
10
15
20
25
30
Height profile of concentrations (at distance to road axis [m])
Configuration source W1 vs source W1 with 4m barrier at 10m from road axis (1:400)
W1_barrier 20m
W1_barrier 30m
W1_barrier 40m
W1_barrier 80m
W1_barrier 160m
W1 20m
W1 30m
W1 40m
W1 80m
W1 160m
C* [-]
H
e
i
g
h
t

[
m
]
Figure 6: Height profiles with and without barrier, source
configuration W1_mod2
0 0.2 0.4 0.6 0.8 1 1.2 1.4
0
5
10
15
20
25
30
Height profile of concentrations (at distance to road axis)
Configuration source W1_mod2 vs source W1 mod 2
with 4m barrier at 10 m from road axisl (1:400)
W1_mod2_barrier
20m
W1_mod2_barrier
30m
W1_mod2_barrier
40m
W1_mod2_barrier
80m
W1_mod2_barrier
160m
B515_20
B515_30
B515_40
B515_80
B615_160
C* [-]
H
e
i
g
h
t

[
m
]
Figure 7: Comparison of barrier effect for source configuration
W1 vs W1_mod2
0 20 40 60 80 100 120 140 160 180
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
Barrier effect = 1 - (C*_barrier/C*_no barrier) at h=1.5m
Configuration W1 vs W1_mod2
W1_mod2 h=1.5m
W1 h=1.5m
Distance to road axis [m]
B
a
r
r
i
e
r

e
f
f
e
c
t

[
%
]
263
3 Analysis of field data and comparison to WT data
Initially the analysed field data (2007-2008) as provided to the wind tunnel laboratories after
the wind tunnel experiments, were presented by calculating the yearly averaged noise barrier
effect as: =1
C
barrier
C
no barrier
2)
The same was done for the windtunnel results. These data are also shown in [3]. The road
traffic contribution however is relatively small. This means that the measurement uncertainty
in the absolute concentration level causes a much larger uncertainty on the noise barrier effect
obtained from the field data. Therefore each configuration of noise barriers was re-analysed
separately to compare the absolute road traffic contribution per configuration between wind
tunnel and field data.
For the field data the NO
x
measurements were used because of the good signal to noise ratio.
Unfortunately no usable traffic intensity data were available from the test location at the time
of the field measurements. A representative traffic intensity profile was obtained from a
permanent monitoring station Strand Horst Ermelo nearby, based on hourly values of 2004
and 2005. A threshold value on traffic intensity was set to eliminate noise at the low traffic
hours resulting in an analysis of the hourly concentration values between 8.00h and 20.00h
daily. Although this is the best data available it should be kept in mind that this approach
contributes to an increase in uncertainty of the field data traffic source strength. All
measurements were then grouped by wind sector and reprocessed to concentration
coefficients.
The data sets acquired in the field test and wind tunnel tests were obtained from [4]. Field
experiment data are based on the hourly NOx concentration averages for the indicated wind
direction 15, for hours between 08:00 and 20:00 (UTC). Hourly estimated traffic emissions
on the E 232 are based on a traffic intensity of 65,000 vehicles per day, 6.9% buses/vans,
9.1% HGV, emission data for 2007 as supplied by V+W. Hourly wind direction and potential
wind speed U
10
data from meteorological station Lelystad supplied by the Royal Dutch
Meteorological Institute (http://www.knmi.nl/samenw/hydra).
From the comparison of wind tunnel and field data in figure 8 it is seen that wind tunnel
measurements provide a good estimation of the concentrations behind the barriers for the
wind directions 60 and 90. In the situation without the barrier a clear over-prediction of the
concentration is seen. An over-prediction of the concentrations without the barriers will cause
an over-prediction of the effectivity of the barrier in reducing the concentration level. It is
expected that modification of the sources as shown in chapter 2 will provide a more accurate
prediciton of the concentrations especially in the situation without the barrier and therefore a
more accurate prediciton of the barrier effect will result.
The performance of the wind tunnels seems to be worse for wind direction 120. However it
is seen that the wind tunnel results are similar to the results for 60, wich is expected for the
measurements behind the barriers as a high degree of symmetry is seen in the model. The
field data results are remarkable lower than the results at 60, no explanation for this behavior
has been found yet.
264
4 Conclusions
Genarally it is concluded that working according to the protocol provides high quality data
that show good repeatability and are consistent between different laboratories working
according to the same protocol. However based on the analysis of the results of the first
round robin test improvements were made to the protocol to avoid unfavorable pressure
gradients in the model by chosing favorable test settings. Also additional requirements on the
initial height distribution of the tracer gas were incorporated in the protocol [6] to harmonise
the line source behavior. Additional round robin tests will show the effect of the enhanced
protocol. Despite the improvements that still can be made it is concluded that wind tunnel
simulations already can provide accurate information on concentration levels of pollutants
along highways and the effect of barriers on the concentration level.
5 REFERENCES
[1] Measuring protocol wind tunnels, Air quality, Semi final version, 6 december 2007.
[2] Artois, K., Moonen, N., Ratingen van, S, 2009. Atmospheric dispersion simulation, The Dutch round robin
wind tunnel test, in: Prodeedings Physmod 2009 International workshop on physical modelling of flow and
dispersion phenomena, Rhode-St-Genese, Belgium.
[3] Ratingen van, S, Artois, K., Moonen, N., 2011. Atmospheric dispersion simulation, The Dutch round robin
wind tunnel test, in: Prodeedings Physmod 2011 International workshop on physical modelling of flow and
dispersion phenomena, Hamburg, Germany.
[4] Dutch Ministry of Transport, Public Works and Water Management (V+W), 2009. IPL-database Archivering
van alle binnen het IPL verzamelde meetgegevens, rapportnummer IPL-7. The Dutch Directorate-General
for Public Works and Water Management, Delft.
[5] Dutch Ministry of Transport, Public Works and Water Management (V+W), 2009. Lessons Learned: meten
van en rekenen aan luchtkwaliteit, rapportnummer IPL-9. The Dutch Directorate-General for Public Works
and Water Management, Delft.
[6] Measuring protocol wind tunnels, Air quality, Draft version, 18 june 2010.
Figure 8: Comparison of wind tunnel to field test data for wind directions 60, 90, and 120
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
Comparison of wind tunnel and field data
Situation no barrier, wind 60 degrees relative to road axis
W1 no barrier
W2 no barrier
W3 no barrier
Field test no barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
Situation 4mbarrier, wind 60 degrees relative to road axis
W1 4m barrier
W2 4m barrier
W3 4m barrier
Field test 4m barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
W1 7m barrier
W2 7m barrier
W3 7m barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
Comparison of wind tunnel data and field data
Situation no barrier, wind direction perpendicular to road axis
W1 no barrier
W2 no barrier
W3 no barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
Comparison of wind tunnel data and field data
Situation 4mbarrier, wind perpendicular to road axis
W1 4m barrier
W2 4m barrier
W3 4m barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
Situation 7mbarrier, wind perpendicular to road axis
W1 7m barrier
W2 7m barrier
W3 7m barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
Situation no barrier, wind 120 degrees relative to road axis
W1 no barrier
W2 no barrier
W3 no barrier
Distance to roadside[m]
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
W1 4m barrier
W2 4m barrier
W3 4m barrier
C
*
[-
]
0 20 40 60 80 100 120
0.00
0.10
0.20
0.30
0.40
0.50
0.60
W1 7m barrier
W2 7m barrier
W3 7m barrier
C
*
[-
]
265
Atmospheric Dispersion of traffic exhaust emissions.
A proposal for a theoretical model and parameter
estimation through field data analysis.
Ir. S.P.M. van den Akker
a
, Ir. J.F.W. Koopmans
b
a
Peutz bv, Mook (Nijmegen), the Netherlands, s.vandenakker@mook.peutz.nl
b
Peutz bv, Mook (Nijmegen), the Netherlands, f.koopmans@mook.peutz.nl
ABSTRACT:
A theoretical model is presented to account for the effect traffic induced dilution on
concentration of air pollution near roads. The model is validated through an analysis of field
data from urban and rural areas, for different distances and wind directions. This analysis also
results in a quantitative estimate of the model parameter describing traffic induced dilution.
It is concluded that the theoretical model provides an adequate description of air pollution
levels as a function of wind speed and traffic intensities, although the model parameter
describing traffic induced dilution may vary considerably depending on the distance to the
road.
1 INTRODUCTION
The dispersion of traffic exhaust emissions is governed to a large extent by atmospheric
conditions like wind speed, wind direction and the amount of atmospheric turbulence.
Numerous models exist to predict the results of these atmospheric dispersion processes.
In addition to these large scale atmospheric processes, traffic itself produces turbulence.
Downwind from a road the amount of turbulence can be significantly higher from the upwind
conditions. The additional turbulence is produced by moving vehicles, and increases the
dilution of the exhaust emissions. Dispersion models that fail to take this traffic induced
dilution into account invariably overestimate the concentrations of air pollution near roads.
2 A THEORETICAL MODEL FOR TRAFFIC INDUCED DILUTION
Behind moving vehicles the air flow separates from the vehicle body, creating turbulence.
The exhaust gasses are emitted into this area. The locally increased turbulence causes mixing
with clean ambient air. As a hypothesis it is proposed that this process may be characterised
by a dilution volume flow (
v
, in m per second per meter road length). Given the total
emission of exhaust gasses (Q, in kg per second per meter road length) the resulting
concentration of exhaust gasses C
TIT
(in kg/m) is defined by:
(1)
266
Several studies suggest that the effect of traffic induced turbulence is primarily dependent on
the vehicle speed, total traffic intensity (N) and traffic composition (fraction heavy goods
vehicles and buses). The studied field data at present does not include sufficient variation in
vehicle speed and traffic composition. These dependencies are therefore beyond the scope of
this study.
.
The dependency on the total traffic intensity (N) is incorporated in the proposed model by
assuming that the dilution volume flow is proportional to the traffic intensity:
v
= pN (2)
Roadside concentrations are the result of a combination of traffic induced dilution and
atmospheric dilution and dispersion. Assuming that the atmospheric dilution is proportional
to the wind speed (U, in m/s) the roadside concentration of exhaust gasses may be defined by:
(3)
In this equation the concentration coefficient K (= 1/A, in 1/m) is introduced, which depends
on the wind direction. Equation (3) implies that traffic induced dilution is localised in the area
behind the vehicle body. In reality, significant amounts of traffic induced turbulence may be
present downwind from the road, where it can cause additional dilution. The robustness of the
proposed model in the case of traffic induced turbulence downwind from the road will be
investigated.
3 FIELD DATA
Field data from several sources was processed and analysed to quantify the effect of traffic
induced dilution in both rural and urban situations.
3.1 Processing
As a first step in processing field data the background concentration where eliminated from
the concentration measurements, isolating the concentration contribution (C) of the traffic.
The first order influence of the wind speed (U) variations and the traffic emission (Q)
variations was eliminated from C by calculating concentration coefficients C*, defined as:
(4)
If the dispersion is influenced exclusively by atmospheric conditions, then C* will be largely
independent of the wind speed U. If the effect of traffic induced dispersion is significant,
substitution of C in definition (4) with model equation (3) indicates that C* is not
independent of the wind speed U:
(5)
Hourly average C* values where calculated from the available field data. In order to limit the
variation of the concentration coefficient K within a data set, C* values where classified
within wind sectors of 20 or 30. The resulting data sets contain 320 to 2267 individual C*
267
values. A nonlinear least-squares (NLLS) Marquardt-Levenberg algorithm was used to
simultaneously estimate the value of K and
v
.
3.2 Available field data
C was determined from field data sets that provided simultaneous measurements of upwind
and downwind concentrations of nitrogen oxides (NO
x
). Traffic emissions of NO
x
where
based on measured and estimated traffic intensities, combined with suitable emission factors.
The wind direction and potential wind speed U
10
data was obtained from the Royal
Netherlands Meteorological Institute (see http://www.knmi.nl/samenw/hydra), and was based
on measurements at nearby airports.
The trajectory of the exhaust gasses between the road to the measurement position is
significantly longer when a noise barrier is present. Measurements behind a noise barrier may
therefore be indicative of the dispersion at much greater distances from the road. Traffic
induced turbulence may be completely dissipated before the exhaust plume reaches a
measurement position behind a noise barrier. Measurements behind a noise barrier from two
locations where incorporated in the study.
3.2.1 E 232 (A28) near Putten, the Netherlands
In 2008 and 2009 the Dutch Directorate-General for Public Works and Water Management
(Rijkswaterstaat, RWS) performed an extensive measurement campaign of air pollutants
along the E 232 (A28) near Putten. Concentrations of air pollutants where measured upwind
from the road as well as at distances of 10, 15 and 33 meters downwind from road. Situations
with noise barriers and without noise barriers where measured simultaneously. Traffic data
and vehicle emission factors where also provided by RWS.
Figure 1 gives a typical example of individual hourly values of C* along the E 232, plotted as
+ against the potential wind speed U
10
.
Figure 1: Hourly average values of C* at a distance of 33 meters from the side of the E 232 near Putten (no
noise barrier). Wind direction 60 15 relative to the road axis.
268
The upper curve in figure 1 shows the predicted effect of
v
(traffic induced dilution per
vehicle p = 5.7 m/vehicle) on C* for the observed lower peak hour traffic intensity (N =
3.100 vehicles/h). The lower curve in figure 1 represents the predicted effect of
v
on C*(U)
for the highest peak traffic intensity (N = 5.500 veh/h).
3.2.2. E35 (A2) near Breukelen, the Netherlands
The national air pollution monitoring network of the Dutch National Institute for Public
Health and the Environment (RIVM) has a measurement station at 15 meters east of the road
side of the E35 (A2) near Breukelen. This measurement station provides long term
concentration data. Measurements where used from 2007 and 2008. Simultaneous upwind
NO
x
concentrations where obtained from the Zegveld regional measurement station of the
same monitoring network.
3.2.3. E19 (A13) in Rotterdam, the Netherlands
The Overschie area surrounding the E19 in Rotterdam gained national notoriety in the 1990's
for being one of the most polluted residential area's of the Netherlands. To monitor the air
quality in this area the DCMR Environmental Protection Agency for the Rotterdam
metropolitan area has a measuring station at 15 meters east of the E19. For north-westerly
wind directions the Schipluiden measuring station of the national air pollution monitoring
network was used as background station. From both stations the available measurements for
the year 2008 where used.
Figure 2 gives a typical example of individual hourly values of C* along the E 19, plotted as
+ against the potential wind speed U
10
.
Figure 2: Hourly average values of C* from NOx concentration measurements at a distance of 15 from the E 19
in Rotterdam-Overschie, behind a 4.5 meters high noise barrier. Wind direction 65 5 relative to the road axis.
The upper curve in figure 2 shows the predicted effect of
v
(traffic induced dilution per
vehicle p = 5.1 m/vehicle) on C* for the observed lower peak hour traffic intensity (N =
269
9.000 vehicles/h). The lower curve in figure 2 represents the predicted effect of
v
on C*(U)
for the highest peak traffic intensity (N = 12.600 veh/h).
3.3 Analysis and discussion of the results
The results of the nonlinear least-squares estimates of the amount of traffic induced dilution
per vehicle (p =
v
/N, see equation 2) are presented in table 1. Estimated values that are
based on a relatively small number of C* values (< 450 values) generally have a large
uncertainty. These values where placed in parentheses in table 1, and where excluded from
the determination of average values.
Table 1: Estimates of traffic induced dilution per vehicle (p = v/N) in different situations, for several distances
to the road side.
No sound barrier Behind sound barrier
R Road 10 m 15 m 33 m 10 m 15 m 33 m
30 E19 (A13) in Rotterdam - - - - (4.5) -
30 E35 (A2) near Breukelen - 3.9 - - - -
30 E 232 (A28) near Putten (5.1) (5.4) (6.8) (6.9) (7.2) (8.1)
60 E19 (A13) in Rotterdam - - - - 4.7 -
60 E 232 (A28) near Putten 3.7 4.1 5.7 5.2 5.5 6.8
90 E 232 (A28) near Putten 3.2 3.6 5.1 5.0 5.3 6.4
120 E 232 (A28) near Putten 3.7 3.9 5.7 5.5 5.2 (6.9)
150 E 232 (A28) near Putten 3.5 3.9 (5.5) 4.4 4.6 (5.9)
Average (all roads) 3.5 3.9 5.5 5.0 5.1 6.6
Figures 1 and 2 provide examples that show that equation 3 provides an adequate description
of C* as a function of wind speed, although the estimated values for traffic induced dilution
per vehicle vary considerably depending on the distance to the road side. Traffic induced
dilution per vehicle increases for larger distances to the road side from 3.5 m/veh to
5.5 m/veh. This increase indicates that traffic induced dilution is not limited to the area
behind and between vehicles. Traffic induced turbulence is causing additional dilution
downwind from the road.
Estimates for positions behind noise barriers may be indicative of traffic induced dilution at
much greater distances from the road. Directly behind a noise barrier the amount of traffic
induced dilution is fairly constant at 5.0 to 5.1 m/veh, roughly comparable to the value of
5.5 m/veh in a position at 33 meters from the road side without noise barrier. These higher
values may represent the final amount of traffic induced dilution for situations where traffic
induced turbulence has dissipated. However, the value of 6.6 m/veh at a distance of 33 m
behind a noise barrier may indicate that this final value may be even higher.
Comparing the results for the rural Putten and Breukelen situations to the urban Rotterdam
situation, it is concluded that traffic induced dilution per vehicle does not depend
significantly on the type of the surrounding area.
270
Comparing the results for Putten (65.000 vehicles per day) to Rotterdam and Breukelen
(> 100.000 vehicles per day), it is concluded that traffic induced dilution per vehicle does not
depend significantly on the traffic intensity. From this it may be concluded that equation 2
provides an adequate description of the relationship between traffic induced dilution and
traffic intensities.
The amount of traffic induced dilution may also depend on vehicle speed and traffic
composition. However, the studied field data at present does not include sufficient variation
in vehicle speed and traffic composition. These dependencies are therefore beyond the scope
of this study.
271
Experimental parametric study of the influence of an idealized
upstream ridge on the flow characteristics over Alaiz mountain
Boris Conan
a, b
, Fernando Carbajo
a
, Jeroen van Beeck
a
, Sandrine Aubrun
b

a
von Karman Institute, Rhodes-St-Gense, Belgium, conan@vki.ac.be,
carbajo@vki.ac.be, vanbeeck@vki.ac.be
b
PRISME laboratory, Orlans, France, sandrine.aubrun@univ-orleans.fr
ABSTRACT: This paper presents a preliminary quantification of the influence, at the top of a
main topography, of an upstream obstacle compared to the influence of inlet conditions. A
simplified 2D model is built and several inlet conditions are coupled with different upstream
relief distances. The velocity deficit and the turbulence increase are discussed. The main
result is the higher influence of the inlet conditions compared to the obstacle position.

This work is performed within the WAUDIT Marie-Curie Initial Training Network (Wind
Resource Assessment Audit and Standardization) funded by the European Commission.
1 INTRODUCTION
Physical modeling in a wind tunnel is a common tool for numerous applications like
atmospheric dispersion investigations, wind comfort assessment or wind loads on buildings
studies. In the booming wind energy sector, wind tunnel tests can be a suitable tool for the
assessment of the wind resource, especially for wind turbine micro-siting in complex terrain.
The simulation of atmospheric flows in the wind tunnel requires the verification of a number
of assumptions. Taking as implicit the similarity criteria (dimensionless numbers) described
by Cermak (1971), a certain number of parameters have to be taken into account: the
Reynolds number dependence, the modelling of the local roughness on the mock-up, the
reproduction of inflow conditions and the choice of the modelled area.
The latter is a major parameter that drives the choice of the scaling factor and influences all
the other parameters. The choice of the area to model is the result of a compromise between:
having a large area in order to reproduce closely the effects of the surroundings topographies,
the limitation of the wind tunnel dimensions (blockage) and the difficulties to reproduce
realistic flows at very high scaling factors (model roughness, reproduction of inflow
conditions, measurements limitations).

The Alaiz site (Fig. 1.a) tested in the wind tunnel (Conan, 2011) illustrates this conflict. The
terrain is a 1130m high mountain situated next to Pamplona (Spain); it is a very complex
terrain stretching over 10km in the W-E direction and 8km in the N-S direction (Cabezon,
2006). When the wind comes from the North, one of the dominant directions, it faces a ridge
before reaching the mountain (position x = 0.75m in Fig. 1.a). The ridge is around 1/3 of the
main mountain's height and 7km upstream. An influence is expected, it is then chosen to
include it in the wind tunnel mock-up. Giving the test section size (2m x 3m), that leads to a
very large scaling factor of 1/5300.
Experimental tests are realized at the von Karman Institute in the 2m x 3m x 15m wind
engineering test section. Particle Image Velocimetry and hot-wire anemometry measurements
are performed along a 2D plane parallel to the wind direction as described in Fig. 1.a).

272
One of the main results of this study is the high influence of the upstream ridge on the flow
field on the main mountain. Fig. 1.b) shows the Fractional Speed-up Ratio over the terrain
and enlightens the important speed-down (almost -50% compared to the inlet conditions R1)
at the foot of the mountain (R2) due to the disturbance of the upstream ridge (x = 0.75m).
Measurements show the speed-up and the turbulence intensity at the top of the mountain very
affected by the separation occurring at the ridge. It leads to the conclusion that the area of
influence of the flow up to the mountain is vast and must be taken into account in the model.
On the other hand, the extreme scaling factor leads to some difficulties in reproducing the
inflow conditions. The inlet turbulence profile provided by the VDI guidelines (2000) is hard
to match close to the surface for this scaling factor. Additionally, questions emerge
concerning the relaxation of the Reynolds number and the surface roughness of the model.

(a) (b)
Figure 1: Top view of the Alaiz terrain at the wind tunnel scale (a) and Fractional speed-up ratio at 90m over
the mountain (b).

Very high scales are difficult to work with, however, as shown by the study of the Alaiz
mountain, the flow field over a mountain is influenced by an upstream ridge three times
smaller and situated 7km upstream.

This study aims at illustrating the question of the modelled area size through a parametric
study designed to evaluate the influence of an upstream hill on a major downstream
mountain. Different configurations will be tested by changing the distance from the ridge to
the mountain and inlet conditions.
2 EXPERIMENTAL CONDITIONS
2.1 Experimental set-up
For the parametric study, a suction-type wind tunnel is used with a 0.35m x 0.35m x 2m test
section. The velocity can be adjusted up to 35 m/s. The use of a reduced size tunnel is an
advantage for a parametric study so that several configurations can be tested in a short time
and for limited costs.
To keep the blockage ratio below 10%, the model is scaled down to 1/19 200. At that scale,
the simulation of turbulence scales might not be respected. The objective of this model is a
parametric study; results may not be used for direct application at full scale.
The mock-up is a two-dimensional model of the width of the test section representing the line
shown in Fig. 1-a) from point R1 to P7. The choice of a two dimensional model is supported
by the comparison performed in a previous study showing a good agreement between the
273
wind tunnel test and a 2D CFD computation on this very same line (Munoz-Esparza, 2011).
The front ridge is simplified to a triangular shape and can be displaced upstream and
downstream for the parametric study. In the original configuration: H = 2.5*h; D = 24*h and
h = 13 mm (Fig. 2).

Figure 2: Mock-up in the wind tunnel.

The wind tunnel is equipped for two-dimensional Particle Image Velocimetry (PIV)
measurements. Three planes are necessary to measure all the topography but a reconstruction
of the measurements is possible for average quantities: U, V, Iu and Iv. 500 images are
acquired and averaged for each plane.
2.2 Flow conditions
The atmospheric inflow conditions are modeled thanks to a 1m fetch where roughness
generators are placed to simulate an atmospheric boundary layer (ABL). Three different inlet
conditions (Fig. 3) are tested in the wind tunnel by combining surface roughness (lego floor
with h
lego-floor
= 2mm) and Counihan wings (h
CW
= 90mm) in the following, they are named:
FP = Flate plate, LF = Lego floor and CW = Counihan wings + Lego floor. (Fig. 3)

At the scale of 1/19 200, the boundary layers represent three different terrain roughness (Fig.
4) from slightly rough to rough according to the VDI guidelines (2000).

The two-dimensionality of the flow is estimated by performing PIV planes at four locations
in the Z direction around the middle measurement plane. In a +/- 20mm slice in the middle of
the test section the velocity varies by less than 0.7% and the turbulence by less than 0.4%.

In the range of the wind tunnel velocities, measurements at three different Reynolds number
are recorded at the top of the model. Taking as reference length the front ridge height (h), it
gives Re = 8 300; Re = 12 400 and Re = 16 600. The Reynolds number dependency decreases
with increasing velocity, the maximum discrepancies are at around 400 m with less than 3%
difference in speed. The very high scaling factor may explain this dependency. The highest
velocity (20 m/s) is chosen for the tests.

Figure 3: Boundary layers simulated.
3 PRESENTATION OF THE RESULTS
274
3.1 Flow around the original configuration

Figure 4: Velocity and turbulence intensity fields, velocity streamlines and velocity vector field in the original
Alaiz configuration with low roughness inlet conditions (FP).

When it reaches the front ridge, the velocity increases and creates an over-speed at its top. On
the lee side, the flow separates and creates a recirculation bubble (Fig. 4). This separation is
generating high turbulence and a very important velocity reduction at the height of the ridge.
The velocity recovers after the perturbation but the flow is still disturbed, with lower
velocities and higher turbulence, when it reaches the mountain. The behavior is very similar
to the one described on the three-dimensional model (Conan, 2011).

Two local quantities are used in this study: the Fractional Speed-up Ration (FSR) and the
Turbulence Intensity Ratio (TIR), they represent the ratio of change compare to the inlet
conditions:
(1)
inlet
U U
i
U
FSR
inlet i

inlet
Ti Ti
i
Ti
TIR
inlet i

275

The FSR computed at 90m and compared with the study at scale 1/5357 presented in the
introduction. It illustrates well the behavior of the flow around the ridge and the mountain
(Fig. 5), the front ridge creates a speed-up followed by a strong speed-down in the
recirculation, then the speed increases again until x = 1.25m, at this point the mountain
creates a speed-down. On the mountain, the speed is rising until the top, where the speed-up
is the highest. The mountain speed-up is nearly 100% front the foot to the top but due to the
front ridge, it is reduced to around 50%.

Figure 5: Comparison of the FSR at two scales.

Figure 5 shows the comparison between the test in the big wind tunnel at scale 1/5357 with a
three-dimensional model and the test presented above at 1/19 000 scale with a two-
dimensional model. The two measurements are very comparable, both in terms of behavior
(shape) and quantities; the speed-up at the top of the mountain is very well reproduced.
Discrepancies appear in the recirculation after the ridge (x = 0.75m), this is most probably
due to the simplification of the geometry (2D), therefore, 3D effects are not all reproduced.
The two dimensionality of the flow in the measurement plane was already detected by the
comparison of the wind tunnel tests with 2D CFD computation (Munoz-Esparza, 2011) and
reinforces the validity of the choice of a two-dimensional model.
3.2 Profiles at the top of the mountain (position P4)
Figure 6 presents the results (PIV) from the parametric study with 5 ridge positions and three
inlet conditions (see section 2.2).

Two heights (a.l.g) are defined to describe the influence of the ridge on the flow at the top of
the mountain: z = 2h and z = h/2.
The influence of the ridge is clearly enlightened by comparing the velocity profile for the no
ridge case with any of the ridge distances: the velocity is reduced and the turbulence
increased at, below and above the ridges height.
Concerning the Counihan wings case, the boundary layer is not completely measured in the
PIV field so the normalization of the velocity profile can be a problem: the velocity of the
no ridge case might be underestimated.

Figure 7 presents the ratio of change in velocity and in turbulence intensity with the distance
downstream of the ridge; the reference point is the case without front ridge. This is a great
tool to describe the relative evolution of the values with the distance.

The effect of the ridge distance depends of the height of comparison, at z = h/2, for all inlet
conditions, the velocity deficit, that is very high after the separation, gradually decreases with
276
increasing distance. In the same time, the turbulence intensity decreases. It can be noticed
from Fig. 7-a) that the ratio of velocity is increasing equally for the flat-plate (FP) and the
rough surface (LF) cases. With the Counihan wings, (CW), the velocity increases faster.
Concerning turbulence levels, the ridge induces much higher perturbation for the flat-plate
case than for the CW case. For both FP and LF configurations, the turbulence level decreases
asymptotically to the initial value.

Figure 6: Velocity and turbulence profiles for five ridge distances: 16, 24, 36, 48 and 72 times the ridges
height (h), and three inlet conditions: left: flate plate (FP), middle: rough surface (LF) and right: Counihan
wings + rough surface (CW). Global views and closer look are proposed.

At z = 2h, observations for the FP and the LF inlet conditions are similar: the velocity
decreases with increasing distance and the turbulence increases. This is the opposite behavior
compared to the z = h/2 case. For the CW inlet conditions, the flow follows the same
behavior as the z = h/2 case, the velocity is increasing and the turbulence intensity decreasing
with increasing distance. As previously, the flow is much more affected with FP inlet
conditions than with the CW.

277

Figure 7: Fractional Speed-up Ratio [%] and Turbulence Intensity Ratio [%] evolution with the distance (FP=
Flate plate, LF= Rough surface and CW= Counihan wings).

At a distance of d = 72.h, for z = h/2, the velocity deficit is of the order of 3.4% for the FP
and the LF cases, it is lower for the CW case. The relative turbulence increase is still
important, especially for the FP case (+28% compare to the case without hill).
At z = 2h, the velocity deficit is lower but the relative increase of turbulence much higher and
very dependent on the inlet conditions.
4 INTERPRETATION AND DISCUSSIONS
The flow is highly affected by the ridge and a high velocity deficit is created together with a
great turbulence increase. This happens at the ridge position, for z < h through a recirculation
on the lee side of the ridge. After this, the flow recovers: the velocity and the turbulent level,
affected by the separation, tend to come back to the inlet conditions. Flow conditions at the
top of the mountain are then influenced by the distance of the ridge: the further the ridge, the
more recovered is the flow when it reaches the mountain.

Figure 8: Tentative of a schematic representation of the evolution of the velocity and the turbulence intensity in
the wake of a hill at two altitudes below and above the ridges height.

Above the ridges height, the flow experiences a speed-up due to the relief, the wake
propagates upward creating a velocity deficit and a turbulence increase for z > h. After a
while, the velocity and turbulence level will then tend to come back to the inlet flow
conditions. The observations are coherent with a recovery of the flow after the ridge
separation (Fig.8).
278
From the observations, an important factor influencing the flow on the mountains top is the
inlet turbulence intensity; for the same distance, the influence of the ridge is lower for more
turbulent inlet conditions. This parameter is, in this case, more important than the distance
from the ridge to the mountain. The results summarized in Fig. 9 presents clearly this effect:
the FSR is much more affected by the inlet conditions than by the distance from the hill to the
mountain. Indeed, the wake of the ridge dissipates faster in a turbulent flow.

Figure 9: FSR at the mountains top for the three inflow conditions and the two extreme ridge positions.
5 CONCLUSIONS AND PERSPECTIVES
The study underlines the two-dimensionality of the flow around this part of the Alaiz
mountain, the 1/19000 scale is a valid assumption for parametric study. However, limitation
rises when reproducing the inlet conditions.
The parametric study shows the influence of the upstream ridge on the flow conditions on the
top of the mountain, at 72 times the ridge height, the FSR at h/2 is influenced by -3.5% and
the turbulence intensity is increased by 30% (ratio compare to inlet conditions). The
turbulence is the most important quantity modified.
When changing the roughness of the inlet terrain simulated, the influence on the flow at the
top of the mountain turns out to be more important than changing the ridge position. This is
probably due to the faster wake recovery in more turbulent flows. The FSR is 37% for FP,
60% for LF and 80% for CW at h/2. The inlet condition, in this case, is a predominant
parameter to consider compared to the ridge position.

For further studies, the investigation of the wake of simplified hills is planned with a
particular focus on the far wake flow conditions and on the influence of the inlet flow
conditions. Results can also be compared with available linear models.
6 REFERENCES
Cabezon, D. et al., 2006. Sensitivity analysis on turbulence models for the ABL in complex terrains, in: Proc. of
the EWEC 2006, Athens, Greece.

Cermak, J.E.., 1971.Laboratory simulation of the atmospheric boundarylayer. AIAA journal vol.9 pp1746-1754.

Conan, B et al., 2011. Feasibility of Micro Siting in Mountainous Terrain by Wind Tunnel Physical Modelling,
in: Proc. of the EWEA 2011, Brussels, Belgium.

Munoz-Esparza, D et al., 2011. Sensitivity of Inlet Conditions of wind resource assessment over complex terrain
using CFD solvers and wind tunnel data, in: Proc. of the EWE2 2011, Brussels, Belgium.

VDI guidelines 3783/12, 2000. Physical modeling of flow and dispersion processes in the atmospheric boundary
layer application of wind tunnels. Beuth Verlag, Berlin.
279
Design of an aerodynamically scaled wind turbine
F. Cuzzola
a
, M. Drenkmper
a
,B. Leitl
a
, M. Schatzmann
a

a
Meteorological Institute, University of Hamburg, Germany,
Francesco.cuzzola@zmaw.de
ABSTRACT: For an experimental investigation of the wake of a wind turbine a rotor model
was designed. The scope of the study is to describe the changes in the behaviour of the flow
with respect to the variation of the similarity parameters.
This paper describes the design procedure of a model wind turbine that allows the researchers
to systematically change the similarity parameters and thus simulate consistently the wake of
a full scale wind turbine.
1 INTRODUCTION
Within the scope of the FP7 funded EU project WAUDIT, an investigation of the wake of a
wind turbine is carried out in the Meteorological Institute at the University of Hamburg. The
scope of this study is to deliver high quality data that can be used for the validation of
numerical models. Additionally other topics such as the interference among wakes in a wind
farm or the wind turbine performance in a simulated atmospheric boundary layer will be
investigated. After a literature review (Cuzzola et al., 2010) a model wind turbine was
designed and tested in the Gttingen-type wind tunnel in the Environmental Wind Tunnel
Laboratory (EWTL). The results of these experiments, shown in Drenkmper et al., (2011),
aim to characterize the wind turbine model with respect to the variation of the similarity
parameters that govern the wake behaviour. The aim of this paper is to describe the design
procedure of a small-scale wind turbine and first results with respect to its performance
characteristics in a homogeneous wind tunnel flow.
2 PHYSICAL OVERVIEW OF THE WAKE OF A WIND TURBINE
The wake of a wind turbine is a complex flow. In the field it results from the mixing of
different flow fields. It is affected by a wide range of length scales. The main phenomena
forming the wake are the atmospheric boundary layer flow, the rotation of the blades and the
flow around an airfoil (turbine blade). A simulation in the wind tunnel of this flow requires a
thorough reproduction of each of these phenomena and an appropriate matching of the
corresponding scales.
3 DESIGN PROCEDURE
The aerodynamic design of commercial wind turbine blades is commonly based on the Blade
Element Momentum (BEM) theory. This theory is capable of quickly delivering different
blade geometries used as the base for performance optimization of the turbine. The BEM
theory is well established (Sanderse (2009), Manwell et al.(2009), Burton et al. (2001). As
shown in Fig.1 it describes the blade of the turbine as a sequence of elementary sections.
280
The loss of momentum, in the axial and radial directions, is treated as the result of
aerodynamic lift and drag forces acting on each blade element.

Fig. 1 Illustration of the Blade Element Momentum Theory (BEM) used in wind turbine blade design.

The scope of the theory is to offer a mathematical model that allows the calculation of the
performances of the wind turbine once the momentum loss in the flow is estimated. Since the
problem is highly non-linear, the estimation of the momentum loss needs an iterative
procedure. The design routine implemented with Matlab, as shown in Manwell et al. (2009),
needs as inputs:
the radius of the rotor
the range of operative conditions of interest
a characteristic value of the chord of the blade
the polar of the airfoil used.

The procedure iteratively calculates the induction factors of the blade as well as the angle of
attack of the flow and, from tabulated data, it delivers the relative lift and drag coefficient of
the airfoil. Once convergence is achieved for each blade element it is possible to calculate the
momentum loss and the power coefficient of the turbine. This procedure is repeated for each
operative condition, defined by the dimensionless parameter tip speed ratio , and it is
possible to show the variation of the power coefficient with respect to the tip speed ratio, see
Fig.2.
281

Fig. 2 Power coefficient as a function of tip speed ratio

It is then possible to define the tip speed ratio that delivers the maximum value of the power
coefficient as the design tip speed ratio
D
. This optimal tip speed ratio can be used to define
the geometry of the blade in terms of chord and twist angle of each blade element. As an
example, in Fig.3 the chord length distribution is shown along the span of the turbine blade.
Note, this is an ideal distribution that does not take into account cut-in and cut-off of the
chord. Figures 2 and 3 were obtained using the nominal characteristics of the Vestas V80
wind turbine as inputs, one of the most widely used machines.

Fig. 3 Chord distribution as a function of blade length at the example of a Vestas V80 type blade.

282
4 SMALL SCALE MODEL AND SIMILARITY
Neff et al.(1990) performed a dimensionless analysis and highlighted the dimensionless
parameters that rule the wake of a wind turbine: tip speed ratio , Reynolds
number , power coefficient

and thrust coefficient
.
The latter two depend mainly on the aerodynamics of the blade; gives an indication of the
operative conditions and the distance between two tip vortexes in the wake; Re, since the
working environment of the model is not very different from the field in terms of wind speed,
depends on the characteristic length scale L chosen (note, often an average value of the chord
length is used). A successful simulation of the wake of a wind turbine needs to match all of
these parameters. The design of our small scale model is a result of the iterative procedure
described subsequently.

4.1 Inputs and constraints
As discussed in Sec.3, the Matlab routine requires specific inputs. The choice of the rotor
radius is the first step of the design procedure. Since the final scope of this research project is
to investigate the wake of a wind turbine in the presence of a simulated atmospheric
boundary layer, the radius of the model wind turbine was chosen with respect to the height of
the boundary layer that will be generated in the WOTAN wind tunnel. Therefore, the radius
of the model wind turbine is 20cm. The characteristic value of the chord is 5cm, and this
implies that the resulting Re is 2 orders of magnitude smaller than in full scale conditions
(from 10
6
to 10
4
). It is well known that airfoils are sensitive to changing Re numbers, thus a
geometric scaling of the blade would not be appropriate. The low values of the lift curve will
indeed lower the momentum that can be extracted from the flow, and thus lower C
P
and C
T
.
Therefore, the Jedelsky EJ85 airfoil designed for low Reynolds numbers was chosen for the
whole blade span. The choice of equipping the model wind turbine with the Faulhaber DC
motor 3268 G024 BX4 allowed, together with the available range of wind speeds of the
tunnel, an investigation of a wide range of operative conditions.
As previously shown, ideal BEM procedures deliver a large chord close to the root of the
blade which, for technical reasons, is avoided in practice. This is an advantage when
designing a scale model where loads and economy of the energy extraction are not an issue.
In fact the airfoil EJ85, although among the most performing airfoils at low Re, has an
efficiency ten times lower compared to the airfoils used at full scale which deliver typical
values of 150. Fig. 4 shows polars of EJ85 at different Re.
It was then decided to keep the chord at the ideal design length in order to help achieving
dynamic similarity.
283

Fig. 4 Variation of the EJ85 airfoil polars at different Re.

4.1 Results of the design procedure and constructions of the blades

Given the inputs of the small scale case the design procedure delivers, as previously stated,
thrust and power coefficients (Fig. 5) as well as chord and twist angle distributions (Fig. 6).

Fig. 5 Thrust and Power coefficient plots versus Tip Speed Ratio

Fig. 6 Twist angle and chord distributions along the blade span
284
This geometric information is then used by a second procedure which delivers the three
dimensional Cartesian coordinates of the blade. This helps in the three dimensional drawings
and the construction of the blades using numerically controlled machines. The blades were
then constructed using glass fiber reinforced epoxy resin. The unavailability of a 5-axes
machine led to the construction of a mould made out of 67 wooden sections (blade elements)
which allowed the placement of the fibers and the resin. Fig. 6 shows the blades at an
intermediate stage of the manufacturing process.

Fig. 7 Construction of the blades for the present study.

An outer structure was then designed to hold the motor, allow for wiring of the DC motor and
the placement in the wind tunnel test section. An elongated nacelle was designed to avoid an
increase in turbulence and to allow the rotation of the large blades. A thick tower-nacelle
structure, joint and bolts were used to avoid vibrations in the model turbine. Fig. 8 shows the
model in our Gttingen-type wind tunnel.

Fig. 8 Model Turbine set up in the Gttingen-type wind tunnel.

285
5 Flow Visualization

Fig. 9 Helicoidally structures and tip vortex made visible by using a laser light sheet and fog.

In Fig. 7, the tip vortex and the helicoidally structures in the wake flow can be seen. The
rotational velocity of the turbine was 120rpm and the wind speed 1m/s. The resulting tip
speed ratio is =2.76. The picture was taken during a flow visualization session using a laser
light sheet. An exposure time of 1/200 seconds, a focal length of 17mm at F2.8 and ISO 3200
were used.
6 Conclusions
An experimental study of the wake of a wind turbine can not be performed by using a simple
geometrical scaling of the machine. Due to the low Re number achievable in the wind tunnel,
the simulated flow would not be similar to full scale conditions. Therefore, a geometrically
distorted model had to be used.
The iterative Matlab procedure based on BEM theory delivered the geometry of the blades
optimized with respect to the highest power within the range of operative conditions. It
showed good agreement with theoretical curves from full scale wind turbines.
The design presently in use is a first approach to the problem. Depending on the outcome of
the experiments, the design of the turbine will be modified and further optimized before the
effect of boundary layer sheer and ambient turbulence on the performance of the wind turbine
will be studied.

286
4 REFERENCES
Cuzzola F., Leitl B., Schatzmann M.. Wind turbines in ABL-flow: A review on wind tunnel studies, in
Proceedings of the iTi 2010 Conference on Turbulence, Bertinoro, Italy

Drenkmper M., Cuzzola F., Leitl B., Schatzmann S., Measurements on the wake of an aerodynamically scaled
wind turbine. (under preparation)

B. Sanderse B.. 2009. Aerodynamics of wind turbine wakes. Literature review. ECN-E09-016

Manwell J.F., McGowan J.G., Rogers A.L.. 2009. Wind energy explained- theory, design and application 2
nd

Ed., John Wiley & Sons.

Burton T., Sharpe D., Jenkins N., and Bossanyi E.. 2001. Wind energy handbook. John Wiley & Sons.

Neff D.E., Meroney R.N., McCarthy E., Davis E., Upstream and lateral wake effects on wind turbine
performances, Journal of Wind Engineering and Industrial Aerodynamics 36, pp 1405-1414, 1990.

287
Measurements in the wake of an
aerodynamically scaled wind turbine
Martin Drenkmper
a
, Francesco Cuzzola
a
, Bernd Leitl
a
, Michael Schatzmann
a
,
a
Meteorological Institute/University of Hamburg, Hamburg, Germany,
martin.doerenkaemper@zmaw.de
ABSTRACT: Physical models of wind turbines allow studying the wake flow Due to the
small-scale an aerodynamically scaled wind turbine, rather than a geometrically scaled one,
offers a solution to resolve all phenomena in the near and far wake. For this study a rotor
model was tested in a Goettingen type wind tunnel with homogeneous flow. Measurements in
the near wake (one and two diameter downstream) show a clear defined structure with higher
turbulence close to the tip region. Frequency analysis allows detecting turbulent structures
like the tip vortex that was also visualized using laser lightsheet and fog release.
1 INTRODUCTION
During the last decades the wind energy experienced a boom. Worldwide thousands of
turbines were erected in large (up to several hundred turbines) wind farms. Very little is
known about the structure and development of wind turbine wakes. Boundary layer wind
tunnel studies enable to investigate the flow around a wind turbine in a higher resolution in
time and space than field measurements. The design of a wind turbine for boundary layer
wind tunnels is a complex task due to the much lower Reynolds number. An aerodynamically
scaled wind turbine was designed, built and tested in a Goettingen type wind tunnel. The aim
of this paper is to characterize the wind turbines near wake (1 and 2 rotor diameter
downstream) for different operative conditions.

2 WIND TURBINE MODEL
A wind turbine model in the scale of 1:200 was designed. Table 1 shows the turbines rough
dimensions. The turbine was driven by a Faulhaber brushless DC-motor 3268 G024 BX4
enabling to investigate a large range of tip speed ratios (TSR).
For a detailed description of the design procedure see Cuzzola et al. (2011). In Figure 1 the
wind turbine model as mounted to the wind tunnel is presented.

288
Table 1:
3
The experiments were conducted in the Goettingen
Environmental Wind Tunnel Laboratories (EWTL) in
Figure 1 shows a
placed 150 mm behind
below the jet outlet to
mm into the flow and the blades were covered completely by the
A two-
(above the test section) with a small bar
minimized. The setup enabled to measure
diameter
device was mounted in 50 mm distance to the ou
PHYSMOD2011
1: Model wind turbines dimensions
Tower

Bottom diameter
Top diameter
Shaft (length)
Shaft (diameter)
Figure 1:
EXPERIMENTAL SETUP
Figure 1 shows a schematic overview of the experimental setup.
150 mm behind
below the jet outlet to
-dimensional L
diameters downstream.
was mounted in 50 mm distance to the ou
Model wind turbines dimensions
Tower
Height
Bottom diameter
Top diameter
Shaft (length)
Shaft (diameter)
Figure 1:Aerodynamically scaled wind turbine model (about 1:200).
EXPERIMENTAL SETUP
schematic overview of the experimental setup.
150 mm behind the outlet of the wind tunnel, the ground plate was mounted 10 mm
below the jet outlet to reduce the
dimensional Laser Doppler Velocimetry
downstream. A reference wind speed U
Model wind turbines dimensions.
Value
350 mm
Bottom diameter 60 mm
Top diameter 40 m
Shaft (length) 250 mm
Shaft (diameter) 40 mm
Aerodynamically scaled wind turbine model (about 1:200).
EXPERIMENTAL SETUP
the outlet of the wind tunnel, the ground plate was mounted 10 mm
reduce the influence on the flow. This way th
aser Doppler Velocimetry
A reference wind speed U
Value Blades
350 mm Length
60 mm Chord (tip)
40 mm Chord (root)
250 mm Aerofoil
40 mm
influence on the flow. This way th
aser Doppler Velocimetry-system
(above the test section) with a small bar, so the influence
minimized. The setup enabled to measure at two locations
A reference wind speed U
ref
was mounted in 50 mm distance to the outlet extending 150 mm into the flow.
KlimaCampus, University of Hamburg, Germany August 22
Blades
Length
Chord (tip)
Chord (root)
Aerofoil Jedelsky

The experiments were conducted in the Goettingen-type open jet wind tunnel at the
Environmental Wind Tunnel Laboratories (EWTL) in Hamburg.
influence on the flow. This way th
system was mounted to a traverse system
the influence of the assembly on the flow was
at two locations behind the turbine, one and two
ref
was measured with
tlet extending 150 mm into the flow.
August 22-24, 2011
Value
200 mm
10 mm
160 mm
Jedelsky EJ 85

type open jet wind tunnel at the
Hamburg.
schematic overview of the experimental setup. The wind turbine
influence on the flow. This way the hub extended about 340
mm into the flow and the blades were covered completely by the jet.
was mounted to a traverse system
of the assembly on the flow was
behind the turbine, one and two
was measured with a Prandtl tube
EJ 85
The wind turbine model
e hub extended about 340
a Prandtl tube
model was
e hub extended about 340
a Prandtl tube. This
289
At the two measurement locations
system was chosen
enabled
turbine
influence of
4
The recorded time series
addition a frequency analysis was performed on selected time series to see the turbine
influence on the resulting power spectra.
4.1
All parameters show a clearly defined M
The maximum turb
hub region these values are reduced
within one rotor diameter
Figure 3
(right). Profiles recorded for a TSR of
downstream of the turbine
PHYSMOD2011
Figure
system was chosen
enabled to measure
. The model
influence of this parameter
Experimental Results
The recorded time series
Mean Quantities
The maximum turb
within one rotor diameter
Figure 3: Development of the wake within 1 rotor diameter. Mean U
Profiles recorded for a TSR of
Figure 2: Experimental setup in the Goettingen type open jet wind tunnel
system was chosen such that the origin was placed in t
to measure up to a distance of
The model allowed changing
this parameter on the
Experimental Results
The recorded time series (250.000 points)
ean Quantities
The maximum turbulence is found at a distance from the
within one rotor diameter, the profile smoothes
Development of the wake within 1 rotor diameter. Mean U
Profiles recorded for a TSR of
downstream of the turbine, (error bars from 4.2)
Experimental setup in the Goettingen type open jet wind tunnel
At the two measurement locations lateral and vertical profiles were
that the origin was placed in t
up to a distance of around 2 rotor diameters
allowed changing the pitch angle of the blad
on the wake flow

(250.000 points)
ulence is found at a distance from the
hub region these values are reduced. The U
he profile smoothes
Profiles recorded for a TSR of =2.9 (design TSR), under a pitch angle of 20
(error bars from 4.2).
lateral and vertical profiles were
that the origin was placed in t
around 2 rotor diameters
the pitch angle of the blad
wake flow as well
(250.000 points) were analyzed for different mean quantities. In
All parameters show a clearly defined M-shaped structure
ulence is found at a distance from the
The U-componen
he profile smoothes (Fig. 3, right)

9 (design TSR), under a pitch angle of 20

lateral and vertical profiles were
that the origin was placed in the hub of the turbine.
around 2 rotor diameters
the pitch angle of the blad
as well.
were analyzed for different mean quantities. In
shaped structure (e.g.
ulence is found at a distance from the axis of rotation
component turbulence is reduced by about half
(Fig. 3, right).
Development of the wake within 1 rotor diameter. Mean U-component wind speed (left), Urms
August 22-24, 2011
lateral and vertical profiles were measured. The coordinate
he hub of the turbine.
around 2 rotor diameters (in both directions)
the pitch angle of the blades in order
e.g. Fig. 3) in the lateral profiles
axis of rotation
t turbulence is reduced by about half
.
component wind speed (left), Urms

Experimental setup in the Goettingen type open jet wind tunnel.
measured. The coordinate
he hub of the turbine. The setup
(in both directions) from the
in order to quantify
in the lateral profiles
axis of rotation of 0.7 y/R. In the
9 (design TSR), under a pitch angle of 20, 1 and 2 diameter
measured. The coordinate
The setup
from the
quantify the
in the lateral profiles.
0.7 y/R. In the

, 1 and 2 diameter
290
The vertical profiles
change of the direction in t
wake. Several
In the turbulent components (Fig 4, right) the influence of the turbines tower is visible
lower located measurement points (z/R <
fluctuations than
Figure 4

4.1.1
To test the influence of the
were repeated for
Due to the large chord (Table 1
big influence of this parameter is
component.
Figure 5
for three different pitch angles (0,20,30
The biggest influence is found in the inner hub region.
U component is about half the size than for a pitch angle of 0 degree. The large ch
PHYSMOD2011
The vertical profiles
change of the direction in t
wake. Several visualization experiments (Hand et al. 2001) showed this rotational behavior.
fluctuations than those
4: Development of the wake within 1 rotor diameter. Mean V
Influence of the
To test the influence of the
were repeated for p
component.
5: Lateral wind speed profiles (Umean and Urms) measured 2 diameter downstream of the wind
The vertical profiles taken at the center downstream of the turbine
change of the direction in the lateral wind speed component. This is due to the rotation of the
visualization experiments (Hand et al. 2001) showed this rotational behavior.
those above the turbine (z/R > 1).
Development of the wake within 1 rotor diameter. Mean V
downstream of the turbine, (error bars from 4.2)
Influence of the Pitch Angle
To test the influence of the pitch a
pitch angles of 0
Lateral wind speed profiles (Umean and Urms) measured 2 diameter downstream of the wind
taken at the center downstream of the turbine
he lateral wind speed component. This is due to the rotation of the
above the turbine (z/R > 1).
(right). Profiles recorded for a TSR of =2.9 (design T
Pitch Angle
pitch angle
itch angles of 0, 20
Due to the large chord (Table 1 and Figure 1
big influence of this parameter is to be
for three different pitch angles (0,20,30 ) at the design TSR (
lower located measurement points (z/R < -1) have a several times higher V
above the turbine (z/R > 1).

=2.9 (design TSR), with a pitch angle of 20, 1 and 2

ngle () of the blades
and 30.
and Figure 1) of the rotor blades in the inner hub region, a
expected. Figure 5 shows

) at the design TSR ( =2.9)
have a several times higher V
Development of the wake within 1 rotor diameter. Mean V-component wind speed (left), Vrms
SR), with a pitch angle of 20, 1 and 2
of the blades on the wake flow,
) of the rotor blades in the inner hub region, a
expected. Figure 5 shows
=2.9), (error bars from 4.2)
The biggest influence is found in the inner hub region. At =30
August 22-24, 2011
taken at the center downstream of the turbine (Fig. 4
have a several times higher V
component wind speed (left), Vrms
on the wake flow,
expected. Figure 5 shows lateral profiles for the U
(error bars from 4.2)
=30 degrees, the dimensionless
(Fig. 4, left) show the
have a several times higher V-component
on the wake flow, measurements
lateral profiles for the U
degrees, the dimensionless
) show the
In the turbulent components (Fig 4, right) the influence of the turbines tower is visible. The
component

SR), with a pitch angle of 20, 1 and 2 diameter
measurements
lateral profiles for the U-

Lateral wind speed profiles (Umean and Urms) measured 2 diameter downstream of the wind turbine
degrees, the dimensionless
U component is about half the size than for a pitch angle of 0 degree. The large chord
291
combined with a high pitch angle effects a higher blocking in the inner parts of the turbine.
Tests without power supply of the turbine showed a sufficient lower cut
20 and 30
turbine run just driven by the wind. Full
1.
4.1.2
The equipment of the turbine with
large range of tip speed ratios. To quantify the influence of this parameter, different TSR
were tested.
TSRs. The main
well as the mean U
maximum values are found at around 1 y/R, while for the lower TSR of 2.2 and 2.9 it is in the
range of 0.6
Figure 6
diameter downstream of the turbine
Full scale turbines operate in a wide range of TSR from 3
of 5-8 (Gasch 2009).
and mixing while operating the turbine at the design TSR (Cuzzola et. al. (2011)).

4.2
To investigate the reproducibility of the experiments several profiles were measured
Figure 7
The profiles were mea
speed, turbine rotation, pitch angle, etc.)
diameters downstream is high
downstream, this means a
quantities
quantities (Fig
the blades.
PHYSMOD2011
and 30 pitch (3
Influence of the Tip Speed Ratio
ange of tip speed ratios. To quantify the influence of this parameter, different TSR
were tested. Figure 6 shows the mean (left) and RMS (right) U
. The main difference is that
well as the mean U
range of 0.6-0.8 y/R.
6: Lateral profiles of the
diameter downstream of the turbine
8 (Gasch 2009).
Reproducibility
investigate the reproducibility of the experiments several profiles were measured
Figure 7 shows vertical profiles f
The profiles were mea
downstream, this means a
quantities (see Fig
quantities (Fig 7, left) the area where the lowest reproducibility occurs is the inner region
the blades.
pitch (3-5 ms
-1
), the 0
Figure 6 shows the mean (left) and RMS (right) U
difference is that
well as the mean U-component wind speed
0.8 y/R.
Lateral profiles of the mean
diameter downstream of the turbine, for a Pitch angle of 20 and 3 different TSR
8 (Gasch 2009). Visualisation experiments with a light sheet showed intense t
Reproducibility
vertical profiles f
The profiles were measured on
downstream, this means about 110mm behind the nacelle
(see Fig 7, right) is sufficiently lower than in outer regions
, left) the area where the lowest reproducibility occurs is the inner region
), the 0 pitch needed wind speeds of
turbine run just driven by the wind. Full-scale cut
The equipment of the turbine with a brushless DC motor allowed
difference is that with increasing TSR
component wind speed
mean U-component wind speed (left) and Urms
, for a Pitch angle of 20 and 3 different TSR
Visualisation experiments with a light sheet showed intense t
vertical profiles for the lateral wind component,
sured on different days, keeping the same conditions (wind tunnel
diameters downstream is high (less than 10% for all parameters)
bout 110mm behind the nacelle
, right) is sufficiently lower than in outer regions
pitch needed wind speeds of
scale cut-in wind speeds are in the range of 2

a brushless DC motor allowed
with increasing TSR
component wind speed moves outward of the rotor

component wind speed (left) and Urms
or the lateral wind component,
ferent days, keeping the same conditions (wind tunnel
speed, turbine rotation, pitch angle, etc.) constant. The reproducibility for a distance of two
(less than 10% for all parameters)
bout 110mm behind the nacelle
pitch needed wind speeds of
in wind speeds are in the range of 2
a brushless DC motor allowed
with increasing TSR the maximum
moves outward of the rotor
component wind speed (left) and Urms
Full scale turbines operate in a wide range of TSR from 3-10,
or the lateral wind component,
. The reproducibility for a distance of two
(less than 10% for all parameters)
bout 110mm behind the nacelle, the reproducibility of the turbulent
August 22-24, 2011
Tests without power supply of the turbine showed a sufficient lower cut-in wind speed for the
pitch needed wind speeds of around 8 ms
a brushless DC motor allowed running
Figure 6 shows the mean (left) and RMS (right) U-component for three different
the maximum of the turbulence as
moves outward of the rotor
component wind speed (left) and Urms-component
, for a Pitch angle of 20 and 3 different TSR, (error bars from 4.2)
10, with design Tip Speed Ratios
or the lateral wind component, resulting from these tests.
(less than 10% for all parameters). After 1 diameter
the reproducibility of the turbulent
, right) is sufficiently lower than in outer regions (up to 30%)
in wind speed for the
around 8 ms
-1
to let the
the turbine under a
component for three different
of the turbulence as
moves outward of the rotor. For a TSR of 4,
component (right) measured
design Tip Speed Ratios
Visualisation experiments with a light sheet showed intense tip vortices
from these tests.
. After 1 diameter
(up to 30%). For mean
in wind speed for the
to let the
in wind speeds are in the range of 2-3.5 ms
-
the turbine under a
component for three different
of the turbulence as
. For a TSR of 4,

(right) measured 2
.
design Tip Speed Ratios
ip vortices
investigate the reproducibility of the experiments several profiles were measured twice.
from these tests.
. After 1 diameter
. For mean
, left) the area where the lowest reproducibility occurs is the inner region of
292
Figure 7
The low reproducibility of the experiments in the region of the nacelle shows that time series
in this region should be even longer than the chosen limiting value of 250.000 points. The
mixing of the turbine and flow separation at the blades
low values.
4.3
At every measurement location time series of up to 250000 points were
time series can later be analyzed to see the impact of turbulent structures and periodical
oscillations. A
the spectrum.
Figure 8
(0.9 D). Unfiltered
All time series in the tip region of the rotor show a clear peak for the rotational speed of the
rotor. The spectrum in Figure
downstream of the turbine in a height of
turbines TSR
wind tunnel 6 ms
at around 30
where the data is separated
segments
PHYSMOD2011
7: Reproducibility of the experiments. Same profiles measured twice on different days
low values.
Frequency Analysis
oscillations. A fast f
the spectrum.
8: Power spectral density for the V
). Unfiltered spectrum
turbines TSR was in the range of the design
wind tunnel 6 ms
-1
at around 30-40 Hz
where the data is separated
segments. The spectrum shows one major and a second minor spike. The
Reproducibility of the experiments. Same profiles measured twice on different days
Frequency Analysis
fast fourier transformation
Power spectral density for the V
spectrum (left), filtered spectr
was in the range of the design
1
). One clear spike
Hz. The spectrum in Figure
where the data is separated into segments
The spectrum shows one major and a second minor spike. The
ourier transformation
Power spectral density for the V-component of a timeseries of a vertical profile (1D)
, filtered spectrum (Welchs method)
rotor. The spectrum in Figure 8 (left) was calculated for a time series located 1 diameter
was in the range of the design
clear spike can be observed i
The spectrum in Figure
into segments

ourier transformation (FFT) can then show
component of a timeseries of a vertical profile (1D)
, filtered spectrum (Welchs method)
(left) was calculated for a time series located 1 diameter
downstream of the turbine in a height of 0.9 z/R
was in the range of the design tip speed ratio
can be observed i
The spectrum in Figure 8 (right) was
into segments of 2
4
values
mixing of the turbine and flow separation at the blades are the reason
can then show
um (Welchs method) (right)
0.9 z/R above the nacelle (the tip region)
tip speed ratio (DTSR)
can be observed in the unfiltered spectrum (Fig 8
(right) was filtered
values. The FFT is then
August 22-24, 2011
are the reason for these comparably
can then show the frequencies that dominate
(right)
above the nacelle (the tip region)
(DTSR) (turbine rot= 750 rpm,
n the unfiltered spectrum (Fig 8
filtered using Welchs method,
. The FFT is then performed on these
or these comparably
At every measurement location time series of up to 250000 points were monitored
the frequencies that dominate
component of a timeseries of a vertical profile (1D) in the tip region
above the nacelle (the tip region)
(turbine rot= 750 rpm,
n the unfiltered spectrum (Fig 8
using Welchs method,
performed on these
The spectrum shows one major and a second minor spike. The major spike is

or these comparably
monitored. These
the frequencies that dominate

in the tip region
above the nacelle (the tip region). The
(turbine rot= 750 rpm,
n the unfiltered spectrum (Fig 8 (left))
using Welchs method,
performed on these
major spike is
293
found for a frequency of 30-40 Hz. This peak is due to the rotational speed of the turbine,
which gives a frequency of 37.5 Hz. The minor spike is observed in the range of 80 Hz.
Flow visualization experiments (Cuzzola et al. 2011, Alfredsson 1979) showed intense
turbulent structures (= tip vortices) in the tip region of the blades. These structures had length
scales of several cm. The estimated size (by dividing the wind speed by the frequency) of a
coherent structure of this spike lead to a length scale of 7.5 cm.
So, the minor spike could be the signature of the tip vortex.
5 Comparison with literature
Very little is published in this field of research. Two studies did measurements in the near
wake of a wind turbine. Magnusson (1999) compared a simple numerical model with full
scale measurements of a wind farm in Gotland, Sweden. The field data as well as the
numerical model showed a similar M-shaped structure in the velocity deficit as in the wind
tunnel measurements. The spikes were observed for a distance of about 0.5-0.7 y/R, with a
velocity deficit of 0.5-0.7 at a free stream wind speed of 6 ms
-1
, depending on the
stratification. Binghl et al. (2010) used a LIDAR mounted to a wind turbines nacelle to
measure the wind speed in the near wake. The U-components in about 3 diameter distance
from the rotor showed the described structure. The dimensionless U-component wind speed
ranged from about 0.6 in the hub region up to about 1.2 for 0.7 y/R. This compares well with
the observed values in our wind tunnel tests (Fig. 3 and 5). There the highest values can be
found for 0.6-0.8 y/R being in the range of 1.2 U
mean
/U
ref
for 1 diameter downstream and 1.0
in a distance of 2 diameter downstream.
6 Conclusions and Outlook
The wind turbine model allows to be operated in a wide range of operative conditions. For
distances of one and two diameter downstream cross sections of the U and V component
were measured for different pitch angles and tip speed ratios. The best operative condition for
this wind turbine model is the design TSR of 2.9 with a blade pitch of 20. Comparison with
field data is hard, because very little is published in this field of research so far. The U-
component wind speed compares well with field data from a LIDAR measurement campaign.
To quantify the influence of the blade shape other sets of blades with different chord
distributions and aerofoils will be tested. The experiments were all done in a low turbulent
Goettingen type wind tunnel. Experiments in a Boundary Layer wind tunnel are under
preparation. Their objective is investigating the interaction of a clearly defined boundary
layer with the wake of the turbine.
7 REFERENCES
Alfredsson P.H., Dahlberg J-A., 1979. A preliminary wind tunnel study of windmill wake dispersion in various
flow conditions. Technical Note HU-2189, Part5, FFA, Stockholm, Sweden, June 1981.

294
Binghl F., Mann J., Larsen G.C., 2010. Light detection and ranging measurements of wake dynamics Part I:
One-dimensional Scanning. Wind Energy, 13, 51-61

Cuzzola F., Drenkmper M., Leitl B., Schatzmann M., 2011. Design of an aerodynamically scaled wind
turbine, in: Proceedings of the International Workshop on Physical Modelling of Flow and Dispersion
Phenomena, PHYSMOD, Hamburg, Germany

Gasch R. and Twele J. ,2010. Windkraftanlagen Grundlagen, Entwurf, Planung und Betrieb.
Vieweg+Teubner, Wiesbaden.

Hand M., Simms D., Fingersh L., Jager D., Cotrell J., Schreck S. Larwoord S., 2001. Unsteady aerodynamics
experiment phase vi: Wind tunnel test configuration and available data campaigns. Technical report
NREL/TP-500-29955, NREL.

Magnusson M., 1999. Near wake behaviour of wind turbines. Journal of Wind Engineering and Industrial
Aerodynamics, 80, 147-167

Vermeer L.J., Sorensen J.N., Crespo A. 2003. Wind turbine wake aerodynamics. Progress in Aerospace
Sciences, 39, 467-510

295
Study on Air Pollutant Dispersion under a Sky Train
Station in Bangkok
Kyosuke HIYAMA
a
, Tomomi Hoshiko
b
, Shinsuke Kato
a
, Tassanee Prueksasit
c

a
Institute of Industrial Science, The University of Tokyo, Tokyo, Japan
b
Department of Urban Engineering, Faculty of Engineering, The University of Tokyo,
Tokyo, Japan
c

Department of Environmental Science, Faculty of Science, Chulalongkorn University,
Bangkok, Thailand
ABSTRACT: The ventilation performance of a street in Bangkok, Thailand, was investigated
by performing measurements and by conducting a CFD analysis. We focused on a street that
was covered by an elevated train station. It was shown that the ventilation efficiency varied
drastically depending on the angle between the street and the wind direction. When the wind
direction was parallel to the street, the elevated structures had a negative influence, which
made the pollutant concentration higher than in locations without elevated structures.
However, when the wind direction was perpendicular to the street, the pollutant concentration
values for the two situations were similar. From the investigation using a CFD analysis and
ventilation performance indexes, it was shown that the elevated structure directed the wind
flow and enhanced the ventilation efficiency, which positively affected the ventilation
performance.
1 INTRODUCTION
To reduce the density of toxic gases in the urban atmosphere and thereby reduce the health
risks due to air pollution, one option is to limit the generation of toxic gases. Another option
is to dilute the generated gas by mixing urban air with fresh air from higher altitudes. The
ultimate objective of this study is to optimize the mechanism for diluting toxic gases in
residential areas and to enhance urban ventilation efficiency. In particular, this study aims to
optimize the ventilation performance of streets in Asian cities, where air pollution due to
automobile emissions is still a considerable concern.
In Bangkok, Thailand, the government has been taking strong measures to reduce air
pollution, including the adoption of unleaded gasoline and the expansion of mass transport
systems, such as the subway and BTS SkyTrain (hereafter BTS). These measures have been
proven to be effective to a certain degree. However, traffic jams in urban areas still remain a
major problem, and air pollution from automobile emissions is still a health risk to citizens
[1]. Insufficient urban infrastructure coupled with the rapidly growing urban population is
considered the major reason for these problems. For a variety of historical reasons, Bangkok
still maintains its ancient urban structures. In the old days, waterways were the major routes
of traffic, and therefore, Bangkoks percentage of road surface area is extremely low. The
low road surface area and the increase in the urban population are the major reasons for the
lack of improvement in the traffic of Bangkok, despite the expansion of the public
transportation system.
In contrast, some reports address the fact that while elevated structures, such as elevated
highways and elevated train stations, enrich the transportation system, they decrease the air
296
quality under them because they serve as an obstacle to ventilation [3]. In this paper, we
clarify the impact of an elevated structure on the air quality underneath the structure. First,
we performed measurements to study the actual situations qualitatively [4]. Then, we
conducted CFD analyses to observe the wind and pollutant dilution characteristics and to
evaluate the ventilation performance quantitatively. For the quantitative analysis, we used the
purging flow rate (PFR) and the local air change rate (LACR).
2 MEASUREMENT
2.1 Subjects of Measurement
The concentrations of exhaust gas were measured on Rama I Street in Bangkok, which is
partially covered by the BTS railway and passes the BTS National Stadium Station. The
street has two lanes north of the center divider and three lanes south of the divider. Traffic
jams are observed during the day. To determine the effects of elevated structures,
measurements were performed at three different locations: (1) at a point with no elevated
structures, (2) at a point where the BTS railway covers only the center divider, and (3) at a
point where there is a BTS station and all lanes are covered by an elevated structure. Air
samples for concentration measurements were collected at the center divider. Figure 1 shows
the measurement locations, along with photographs of each location. To limit the influence of
traffic volume change, each of the three locations was chosen at specific intervals on the
same street such that there were no major intersections.

2.2 Description of Measurements
The concentration of exhaust gas is represented by the concentration of NO
2
, which has a
strong correlation with traffic volume. At each measurement location, a sample of 10 l of air
was collected in a Tedlar
bag for 10 min (to determine the 10-minute average concentration)

at the center divider. The samples were collected at a height of approximately 160 cm to
approximate the location of breathing while standing upright. The concentration of NO
2
was
determined with a detector tube (Kitagawa gas detector tube 740, supplied by Komyo
Rikagaku Kogyo, measurement range for NO
2
: 0.01 to 0.2 ppm). The concentration of sulfur
dioxide, which hinders NO
2
measurement, was recorded with a detector tube for sulfur
dioxide and was confirmed to be low enough not to adversely affect the NO
2
measurement
during each measurement. Traffic volume was determined by making a video recording at
location (3). The traffic volumes at location (1) and location (2) were assumed to be the same
Figure 1: Description of measurement locations
Location (1):
Without an elevated
structure
Location (2):
With an elevated structure
(BTS railway)
Location (3):
With an elevated structure
(BTS station)
N
Rama1 Street
297
as that at location (3) because there are no major intersections between the locations. The
measurements took place on January 27th and 28th, 2010, and were taken five times each day
at 6:00, 8:00, 12:00, 16:00, and 18:00. For location (2), however, measurements were taken
only on January 27th, 2010. In addition, wind direction, wind speed, temperature, humidity,
and solar radiation were measured approximately 1 km south of the street locations at the
meteorological observation station at Chulalongkorn University. The meteorological
observations were performed at five-minute intervals at a height of 18 m above ground.
2.3 Result
Figure 2-a shows the wind direction and the wind speed on January 27th, 2010. For this wind
direction, an angle of zero degrees represents a northerly wind. The wind was blowing from
the northwest until approximately 7:00, when it shifted to a constant easterly wind. As Rama
I street runs east-west, the day was characterized by wind blowing parallel to the street.
Figure 2-b shows the wind direction and the wind speed on January 28th, 2010. Unlike on the
27th, the wind direction was not constant throughout the day: the wind shifted from easterly
in the morning to westerly in the evening.

0
90
180
270
360
0
1.2
2.4
3.6
4.8
4:00 8:00 12:00 16:00 20:00
W
i
n
d

D
i
r
e
c
t
i
o
n

[
d
e
g
]
W
i
n
d

S
p
e
e
d

[
m
/
s
]
Time
Wind Speed Wind Direction

0
90
180
270
360
0
1.2
2.4
3.6
4.8
4:00 8:00 12:00 16:00 20:00
W
i
n
d

D
i
r
e
c
t
i
o
n

[
d
e
g
]
W
i
n
d

S
p
e
e
d

[
m
/
s
]
Time
Wind Speed Wind Direction

a) January 27th b) January 28th
Figure 2: Wind direction and speed

0
300
600
900
1200
1500
0
0.04
0.08
0.12
0.16
0.2
6:20 8:00 12:0016:0018:00
T
r
a
f
f
i
c

[
U
n
i
t
/
1
0
m
i
n
]
N
O
2

[
p
p
m
]
Time
Traffic Location (1)
Location (2) Location (3)

0
300
600
900
1200
1500
0
0.05
0.1
0.15
0.2
0.25
6:00 8:00 12:0016:0018:00 T
r
a
f
f
i
c

[
U
n
i
t
/
1
0
m
i
n
]
N
O
2

[
p
p
m
]
Time
Traffic Location (1)
Location (3)

a) January 27th b) January 28th
Figure 3: NO
2
concentration and traffic volume
Figure 3-a shows the relationship between traffic volume and NO
2
concentration on January
27th, when the wind direction was stable. At 6:00, when there was little traffic, the
concentration of NO
2
was relatively low. Because the traffic volume increased after 8:00, the
concentration tended to increase. The tendencies at location (1) and (2) were the same
throughout the day, which shows that the relatively small elevated structure that covers only
the center divider did not influence the ventilating efficiency. In contrast, the concentrations
were higher at location (3) relative to the two other locations. This is an indication of a
reduction in ventilating efficiency due to the elevated structures that cover the road area. The
298
maximum concentration was found at location (3) at 16:00 with a value of 0.164 ppm; this
value was slightly higher than the concentrations found at locations (1) and (2). Figure 3-b
shows the relationship between the traffic volume and the NO
2
concentration on January 28th
(wind direction shifted throughout the day). The peak value exceeded 0.2 ppm, the upper
limit of the measurement range of the detection tube, at 12:00 and 16:00 on January 28th.
This is above the level stipulated in the environmental guidelines of Thailand (maximum
mean value for 1 h: 0.17 ppm). The concentration at all three locations was similar. Uehara et
al. showed that the lid effect from an elevated road does not significantly increase ground-
level concentrations when the wind direction is perpendicular to the road [5]. Thus, the lid
effect was not prominent on January 28th.
3 CFD ANALYSIS
3.1 Subjects of the CFD analysis
In the measurements, the sky train station had an obvious impact on the air quality only when
the wind direction was parallel to the street. To analyze the cause of this phenomenon, we
conducted a CFD analysis. Figure 4 shows an outline of the calculation target. We used two
models, Model 1 and Model 2, for the comparison. Model 1 is a model with a cavity that
represents a street canyon, such as a roadway surrounded by buildings. Model 2 is a model
with a cavity that represents a street canyon and that has an elevated structure present, such as
a sky train station in Bangkok. To generalize the problem, we use simplified models. The
height of the cavity was 17.5 m, corresponding to the height of buildings with 5 floors. The
width of the cavity was 25 m, corresponding to the width of a 5-lane roadway with sidewalks.
The length of the cavity was 200 m. In Model 2, an object corresponding to a sky train station
was installed in the cavity. The bottom surface was located 7 m above the bottom surface of
the cavity. The height of the object was 7 m, which matches the height of a sky train station
with 2 floors. The width was 20 m, and the length was 100 m. With the models, we analyzed
the wind and pollutant dilution characteristics in two cases. The first case has a wind
direction parallel to the street canyon, and the other case has a wind direction perpendicular
to the street canyon.
1
7
.
5
m
7
m
3
.
5
m
7
m
8
0
m
3
0
0
m
2
5
m
5
0
m
2
.
5
m
2
.
5
m
2
0
m

Figure 4: Calculation mode and control volume for concentration generation
3.2 Calculation Conditions
299
Table 1 shows the calculation conditions of inflow. The outflow condition was solved by
applying no gradient condition to the normal line. The lateral faces were treated as symmetric
boundaries, and the top face was treated as a free-slip boundary, while the surface wall
equation was solved by applying the generalized logarithm law. The standard k- model was
employed for the turbulence model. The first-order upwind differential scheme was used for
the advection term, and the second-order upwind differential scheme was used for the
concentration. According to the concentration transport calculation, once the flow field was
calculated, the scalar equation was solved as a passive contaminant. The concentration field
is analyzed when the contaminant is discharged uniformly from the control volume set in the
cavity. This idea is based on the fact that there is a massive amount of clean air flowing in the
atmosphere above cities, and the ventilation efficiency can be evaluated by approximating the
airflow intake into the cavity from higher altitudes. Figure 4 shows the control volume. The
control volume was located at the center of the cavity. The source rate was 1 kg/m
3
s. The
calculation areas were divided into approximately 1,000,000 meshes. The mesh size in the
cavity under the elevated structure was 0.5 m x 0.5 m x 0.5 m. The mesh sizes grow outward
from the cavity. The maximum size was 5.0 m x 5.0 m x 5.0 m.
Table 1: CFD calculation condition
Inflow condition
0
4 / 3 4 / 1
0
2 / 1 2 / 3 2 4 / 1
0 0
/ ) ( 4 , / , 1 . 0 , ) ( 5 . 1 , ) / ( U Z Z k C l l k C I U I k Z Z U U = = = = =

Z
0
: referential height 10.0 m, U
0
: referential velocity 1.0 m/s, U: wind velocity, Z: vertical coordinate [m],
k: turbulence energy [m
2
/s
2
], : energy dissipation [m
2
/s
3
], I: turbulence intensity, l: turbulence length [m]
3.3 Calculation Results
Table 2 shows the results of the averaged concentration of source in the control volume, as
shown in Figure 7. The concentration in Case 3 where the wind direction was parallel to the
street and the elevated structure was present is obviously higher than that in Case 1 without
the elevated structure. When the wind direction was parallel to the street, the obvious impact
of the elevated structure was observed. In contrast, the concentrations in Case 2 and Case 4
have no such obvious differences. The results show the same tendencies that were observed
in the experimental measurements. Figure 5 shows the velocity fields of Case 2 and Case 4 in
the cavity for a cross section in the Y direction. In Case 2, a vortex was observed in the
cavity. The pollutant concentration is transported by the vortex and diluted by the wind
passing above the cavity. In this case, even when an obstacle was installed at the center of the
cavity, it did not have an obvious influence on the flow field made by the vortex.
Furthermore, the obstacle could play a role in arranging the flow made by the vortex and
enhancing the airflow in the cavity. As a result, the concentration in Case 4 became smaller
than the concentration in Case 2. In contrast, the obstacle played a different role when the
wind direction was parallel to the street. Figure 6 shows the velocity fields for Case 1 and
Case 3 in the cavity for a cross section in the X direction. In Case 1, the airflow passing
above the cavity was blowing down into the cavity and diluting the pollutant concentration
generated in the control volume. In contrast, the airflow passing above the cavity in Case 3
becomes part of the airflow circulating around the obstacle and also contributes to the
phenomenon observed in Case 4. Furthermore, the wind direction at the bottom of the cavity
in Case 3 is opposite of the direction seen in Case 1. In these contexts, the characteristic of
the concentration dilution is dramatically changed by the existence of the elevated structure.
While the values of the concentrations in Cases 2, 3 and 4 at which circulation flow fields
such as vortexes are observed are the same, the concentration in Case 1 is one-tenth smaller
than in the other cases. In other words, the efficiency of the ventilation in Case 1 is more than
ten times higher than in the other cases.
300
Table 2: Averaged concentration in the control volume
Case No. Case 1 Case 2 Case 3 Case 4
Existence of elevated structure No No Yes Yes
Wind direction to street Parallel Perpendicular Parallel Perpendicular
Concentration [kg/m
3
] 68 430 400 360

a) Case 2 b) Case 4
Figure 5: Velocity field

a) Case 1

b) Case 3
Figure 6: Velocity field ( the length scale of vector is differ from figure 5)
4 Discussions
From the results of the CFD analyses, we evaluated the ventilation efficiency on each
measurement date. The generated amount of pollutant had a great influence on the
concentration. To study the ventilation efficiency, we have to exclude the pollutant emission
rate. Then, we use PFR and LACR as generalized indices to evaluate the ventilation
efficiency.
PFR was originally defined as the effective airflow rate required to remove/purge the local
pollutions. Equation (1) shows the definition of PFR. In this paper, we use the units of PFR
m
3
/h.
PFR=q/c (1)
where q is the spatially uniform generation rate of the pollutant source [kg/h] and c is the
average concentration of the entire target domain [kg/m
3
].
PFR is also defined by Equation (2).
PFR=V/(VF x T) (2)
where V is the volume of the area [m
3
], VF is the visitation frequency [-], and T is the average
length of time the pollutant stays in the area [h].
The idea of PFR includes the idea of VF. The VF describes the average frequency with which
pollutants generated in the local domain return to the local domain after being transported
301
outside it once [6]. Therefore, the PFR indicates the airflow rate itself, which works
effectively to dilute the air pollution in the targeted domain. The airflow that returns to the
local domain after being transported outside once is excluded from the PFR. Furthermore, the
LACR in the investigated area is calculated by dividing the LACR by the volume of area V.
This is shown in Equation (3).
LACR=PFR/V (3)
Sandberg[7] first proposed the idea of the PFR. It has been widely used as an index to
evaluate the ventilation performance and air quality in indoor airflow problems [8]. At
present, its usage has been expanded to evaluate the ventilation performance in outdoor
airflow problems [6, 9, 10, 11].
Table 3 shows the PFR and the LACR in each case. In Case 1 and Case 3, the indices were
recalculated with a wind velocity of 1.7 m/s at the height of 18 m, assuming the average wind
velocity on January 27th when the measurement was performed. In Case 2 and Case 4, the
indices were recalculated with a wind velocity of 1.3 m/s, assuming the average wind
velocity on January 28th when the measurement was performed. Kato et al. note that over 60
times the LACR outdoors should be satisfied to reduce the health risk of residents in the
buildings along a street [12]. We should note that the area with the sky train station on Rama
I street did not satisfy the criterion on January 27th, while the area without any elevated
structures satisfied this criterion. However, there were no obvious effects from the sky train
station on January 28th when the wind direction was perpendicular to the street. In contrast,
the CFD results show the possibility that the sky train station enhances the ventilation by
diluting the pollutants effectively. Therefore, we cannot simply conclude that the sky train
station dramatically reduces the air quality under it. Kato et al. also noted that the ventilation
efficiency must be evaluated with an exceedance probability of hours year-round. Using the
full-year wind profile and an exceedance probability analysis using the indices, we can
optimize the location of elevated structures. The optimization can minimize the impacts of
elevated structures on the air quality beneath them, for example, by preventing structures
from being built above the streets parallel to the main wind direction of the target city.
Furthermore, it could possibly enhance the ventilation efficiency, for example, by positioning
the elevated structure above the street perpendicular to the main wind direction and thus
enhancing the creation of wind vortex as shown in the CFD analysis. With the accumulation
of these devices, we expect that we can achieve an optimized city planning where the
exceedance probability is satisfied and the health risk to pedestrians is minimized.
Table 3: PFR and LCAR
Case No. Case 1 Case 2 Case 3 Case 4
PFR [m
3
/h] 680000 82000 120000 98000
LACR [1/h] 78 9.4 13 11
5 Conclusion
To reduce the health risks of urban environments, it is necessary to reduce the pollutant
concentration outdoors. To accomplish this, urban ventilation to dilute the pollutants
generated in city areas efficiently has gained a great deal of attention, as has source control to
reduce the actual release of pollutants. To address these issues, we performed measurements
in Bangkok, Thailand, and conducted a CFD analysis to evaluate the ventilation efficiency. In
this paper, we focused on the impact of an elevated structure on the ventilation efficiency. In
the experimental measurements, the phenomenon in which the ventilation efficiency in an
302
area with an elevated structure obviously declined when the wind direction was parallel to the
street was observed. In contrast, there were no obvious impacts from the elevated structure
when the wind direction was perpendicular to the street. Through CFD analysis, we analyzed
the wind and pollutant dilution characteristics. The wind pushed the pollutants and removed
them from the street in cases without an elevated structure when the wind direction was
parallel to the street. In the case involving an elevated structure, the pollutants were removed
by wind circulation along the elevated structure. The existence of the elevated structure had
an obvious influence and decreased the efficiency by almost one-tenth when the wind
direction was parallel to the street. However, there were no obvious influences when the wind
direction was perpendicular to the street. In contrast, the elevated structure directed the wind
flow and enhanced the ventilation efficiency. To exclude the effect of the amount of pollutant
emissions and focus on the ventilation efficiency itself, we used PFR and LACR as
generalized indices. With the year-round exceedance analysis using the indices and the full-
year wind profile, we can plan optimized elevated structures. The results reveal the
possibility that the optimization of elevated structure locations could contribute not only to
minimizing the impact of the structures on the ventilation efficiency but also could increase
the ventilation efficiency by directing the circulation flow in street canyons.
REFERENCES
[1] Thailand State of Pollution Report 2005, 2005. Pollution Control Department, Ministry of Natural Resources
and Environment
[2] S. Shimada; Motorization in Thailand, 1997. An analysis of its stages and social influences. A dissertation,
Hitotsubashi University
[3] U. Charusombat, 1994. Air Pollution Distribution under an Elevated Train Station (A Case Study of Silom
Station in Downtown Bangkok), Thesis submitted to the Faculty of Virginia Polytechnic Institute and State
University in partial fulfillment of the requirements for the degree of Master of Science in Environmental
Engineering
[4] K. Hiyama, T. Hoshiko, S. Abe, S. Kato and T. Prueksasit, 2011. Field measurement to assess the correlation
between wind ventilation performance of and wind structure inside a street with an elevated structure in
Bangkok, Thailand, Procedia Environmental Sciences, Volume 4, Pages 239-244, Urban Environmental
Pollution 2010
[5] K. Uehara et al., 2003. Wind Tunnel Experiments on Roadside Air Pollution around an Actual Major Road
in an Urban Area : Lid Effect of an Elevated Road and Prediction of Concentrations within the Street
Canyon, Journal of Japan Society for Atmospheric Environment 38(6), 358-376
[6] Z. Bu, S. Kato, Y. Ishida, H. Huang, 2009. New criteria for assessing local wind environment at pedestrian
level based on exceedance probability analysis, Building and Environment, Volume 44, Issue 7, 1501-1508
[7] M. Sandberg, M. Sjoberg, 1983. The use of moments for assessing air quality in ventilated rooms, Building
and Environment 18:181197.
[8] SH. Peng, L. Davidson, 1997. Towards the determination of regional purging flow rate. Building and
Environment 32:513525.
[9] Z. Bu, S. Kato, T. Takahashi, 2010. Wind tunnel experiments on wind-induced natural ventilation rate in
residential basements with areaway space, Building and Environment, Volume 45, Issue 10, 2263-2272
[10] Z. Bu and S. Kato, 2011. Wind-induced ventilation performances and airflow characteristics in an
areaway-attached basement with a single-sided opening, Building and Environment, Volume 46, Issue 4,
Pages 911-921
[11] M. Bady, S. Kato, Y. Ishida, H. Huang, T. Takahashi, 2011, Application of exceedance probability
based on wind kinetic energy to evaluate the pedestrian level wind in dense urban areas, Building and
Environment, Volume 46, Issue 9, 1817-1826
[12] S. Kato, 2010. Urban Wind Environment and it's Impact On Indoor Environment: Acceptable Wind
Features of Void Space, in Proceedings of the 7th International Conference on Indoor Air Quality
Ventilation and Energy Conservation in Buildings, Syracuse, New York, USA

303
Wind tunnel experiments on flow field
in real urban canyons
Takeaki Katsuki
a
, Ayumu Sato
b
, Takenobu Michioka
c
and Aya Hagishima
d

a
Central Research Institute of Electric Power Industry, Chiba, Japan,
katsuki@criepi.denken.or.jp
b
ayumu@criepi.denken.or.jp
c
michioka@criepi.denken.or.jp
d
Interdisciplinary Graduate School of Engineering Science, Kyushu University,
Fukuoka, Japan, ayahagishima@kyudai.jp
ABSTRACT: Outdoor ventilation is affected by flow behaviors such as turbulence and
canyon vortices within urban canyons. In this study, wind tunnel experiments for a real
apartment complex were conducted to confirm the reproducibility of the flow field and to
comprehend the behaviors of turbulence and mean characteristics of canyon vortices within
real urban canyons. The wind velocities and turbulent intensities obtained by both Particle
Image Velocimetry (PIV) and Laser Doppler Velocimeter (LDV) fairly agree with the field
measurement values within the urban canyon, although small-scale surface roughnesses such
as balconies are not modeled in the present wind tunnel experiments. It is found that, within a
real canyon, a helical vortex is generated by the interaction between the flows separated from
the side and top edges of an upstream building.
1 INTRODUCTION
Improvement of an outdoor-ventilated environment contributes not only to energy saving
and suppression of CO
2
but also to indoor air quality and comfort. There are already some
studies on well-ventilated environment using wind tunnel experiments (e.g., Yoshie et al. [1],
Kubota et al. [2]). These papers mainly focus on wind velocity within urban canyons.
However, outdoor ventilation is affected by flow behaviors such as turbulence and canyon
vortices within urban canyons.
To investigate those properties within urban canyons, there are already many studies by
field measurements (e.g., DePaul and Shieh [3], Louka et al. [4], for the canyon vortex within
an urban canyon; Eliasson et al. [5], Longley et al. [6], Takimoto et al. [7], for the wind
direction above an urban canyon). In the case of wind tunnel experiments, the detailed flow
fields within ideal urban canyons were investigated (e.g., Brown et al. [8][9], for 2D and 3D
canyons; Sekine and Umino [10], for regular arrays; Meng and Oikawa [11], for staggered
arrays; Hagishima et al. [12] for various geometries in an urban area; Rafailidis [13], for
various roof shapes). Although flow fields within ideal urban canyons have been investigated
(Kanda [14]), the detailed flow fields such as turbulence and mean characteristics of canyon
vortices within real urban canyons are still not fully understood (Klein and Rotach [15], for
parameterization of Reynolds stress; Robin et al. [16] and Carpentieri et al. [17] for flow
field around an intersection in London).
304
In this study, wind tunnel experiments for a real apartment complex were conducted to
confirm the reproducibility of the flow field and to comprehend the behaviors of turbulence
and mean characteristics of canyon vortices within the canyons.
2 WIND TUNNEL EXPERIMENTS
2.1 Wind tunnel and Measurement instruments
Wind tunnel experiments were conducted in section I of the wind tunnel facility (Twinnel:
twinned wind tunnel) of Central Research Institute of Electric Power Industry (CRIEPI). The
dimensions of the wind tunnel are 17 m (length) x 3.0 m (width) x 1.7 m (height). The mean
and instantaneous velocities within the canyons were measured using a Laser Doppler
Velocimeter (LDV) and Particle Image Velocimetry (PIV), respectively. The seeding using
oil-based smoke particles released from the inlet of the wind tunnel. The averaging time of
LDV measurement was 1 min. The flow fields were also measured by PIV, which were
illuminated using a pulsed Nd:YAG laser (120 mJ/ pulse). Double-pulsed images were
obtained for 3 min using a CCD (Charge coupled devise) camera (1344 x 1024 pixels) at a
frequency of 15 Hz. The interval between each pair of particle images was 1.2 ms.
2.2 Study area and Experimental setup
The apartment complex is located at Musashino City, Tokyo, where the field measurement
was conducted. The buildings with heights of more than two-stories were resolved within a
200 m radius under the model length ratio of 1/150. Fig. 1 shows the view inside the wind
tunnel. Fig. 2 shows a schematic view of the experimental setting. The wind velocity was set
at 3.0 m/s (U
) at 1m height (Z
) above ground level. The surface roughness comprising five

rows of L-shaped cross sections was set on the wind tunnel floor from 2.87 to 8.15 m from
the tunnel inlet at equal intervals of 1.32 m to generate turbulence. The mean wind velocity
profile is expressed as UZ
1/5
at the center of the study area. The wind direction used in the
experiment is north, which is the dominant direction. The distribution of wind velocity was
measured on a cross section of canyon C1 (in Fig. 3). These measurements at five points from
C1-1 to C1-5 within the urban canyon were compared with the field measurements. Fig. 4
shows a 3D schematic view. X and Y axes were from north to south and from west to east
directions, respectively. In Fig. 5, the urban canyons consist of four apartment buildings
located in the center [hereafter, it is written as central canyon (suffix is c or none)], three
apartment buildings located in the east and six apartment buildings with slanted roofs located
in the west of the view. To comprehend the behaviors of turbulence and mean characteristics
of canyon vortices within the canyons, the experiments were measured at 9-10 points within
each canyon, e.g., in the case of point 1 in canyon C1, it is written as C1-1.

Fig. 1: The view inside the wind tunnel Fig. 2: A schematic view of
the experimental setup
Spire
1.55m 1.32m 1.32m 1.32m 1.32m 1.32m 0.6m
8.75m
Buildings Model
Nd-YAG Laser
LDV probe
Traverse
CCD Camera
Wind velocity
U
= 3.0m/s
Wind
L type angle
305

Fig. 3: The cross section for the comparison Fig. 4: A 3D schematic view
of LDV, PIV and Field measurement

Fig. 5: Plan view for study area
3 RESULTS
3.1 Comparison between field measurements and wind tunnel experiments
The results of wind tunnel experiments for a real apartment complex were compared with
the field measurement values obtained using hyper sonic anemometers in canyon C1. The
measurements were conducted from 22 Sep 2 Dec 2004 (Sugawara et al. [18]). For
comparison, the part of the field measurement data were picked up under both almost neutral
stability (Richardson number was -0.09 0.08) and northerly wind. The sampling frequency
and averaging time were 10 Hz and 10 min, respectively. The comparison was conducted
from C1-1 to C1-5 except C1-3, because the data on C1-3 were not measured during the field
measurement. The height above ground level, Z, and the measured statistical values were
normalized using H and streamwise mean wind velocity ensemble averaged at the height of
1.8H from C1-1 to C1-5, respectively.
Fig. 6 shows the vertical distributions of streamwise mean wind velocity U/U
1.8H
. Fig. 7
shows the vertical distributions of vertical mean wind velocity W/U
1.8H
. The measurement by
LDV on C1-1 was not conducted because of measurement problems. The reversed flow near
the ground is formed in the streamwise direction, which is similar to the results of Brown et
al. [9] and Sato et al. [19]. The absolute values of W/U
1.8H
on C1-1 and C1-5 near the
building walls are large. Therefore, it is found that a canyon vortex is formed within a real
urban canyon. The wind velocities obtained by both PIV and LDV fairly agree with the field
measurements within the urban canyon, although small-scale surface roughnesses such as
balconies are not modeled in the present wind tunnel experiment.
118mm
1 2
59mm
69mm
48mm
Canyon C1
Reference
(Z/H = 1.8)
3 4 5
10mm
Wind
6 7 9 8 3
4
2
1
5
306
C1-1 C1-2 C1-3 C1-4 C1-5

Fig. 6: The vertical distributions of streamwise mean wind velocity U/U
1.8H

Fig. 7: The vertical distributions of vertical mean wind velocity W/U
1.8H

Fig. 8 shows the vertical distributions of the rms values of streamwise velocity fluctuation
u
/U
1.8H
. Fig. 9 shows the vertical distributions of the rms values of vertical velocity
fluctuation
w
/U
1.8H
. For reference, the measurement value at the height of 1.8H (marks of
are shown in Figs. 8 and 9) above a leeward building is plotted from C1-1 to C1-5. Above the
canyon (Z/H > 1.0), the results obtained from the wind tunnel experiment are twice smaller
than those from the field measurements. That is the reason why the wind tunnel experiment
could not reproduce the relatively large vortices such as mesoscale turbulence. On the other
hand, within the canyon (Z/H < 1.0), turbulent motions generated by buildings dominate
more than those generated by mesoscale turbulence, because the turbulent intensities fairly
agree with the field measurements within the urban canyon.

C1-1 C1-2 C1-3 C1-4 C1-5

Fig. 8: The vertical distributions of the rms values of
streamwise velocity fluctuation
u
/U
1.8H

Fig. 9: The vertical distributions of the rms values of
vertical velocity fluctuation
w
/U
1.8H
0
1
2
-0.4 0 0.4 0.8 1.2
Z
/
H
(
-
)
U/U1.8H(-)
0
1
2
-0.4 0 0.4 0.8 1.2
U/U1.8H(-)
0
1
2
-0.4 0 0.4 0.8 1.2
U/U1.8H(-)
0
1
2
-0.4 0 0.4 0.8 1.2
U/U1.8H(-)
0
1
2
-0.4 0 0.4 0.8 1.2
U/U1.8H(-)
0
1
2
-0.5 0 0.5
Z
/
H
(
-
)
W/U1.8H(-)
0
1
2
-0.5 0 0.5
W/U1.8H(-)
0
1
2
-0.5 0 0.5
W/U1.8H(-)
0
1
2
-0.5 0 0.5
W/U1.8H(-)
0
1
2
-0.5 0 0.5
W/U1.8H(-)
0
1
2
0 0.2 0.4 0.6
Z
/
H
(
-
)
u
/U1.8H(-)
0
1
2
0 0.2 0.4 0.6
u
/U1.8H(-)
0
1
2
0 0.2 0.4 0.6
u
/U1.8H(-)
0
1
2
0 0.2 0.4 0.6
u
/U1.8H(-)
0
1
2
0 0.2 0.4 0.6
u
/U1.8H(-)
0
1
2
0 0.2 0.4
w
/U1.8H(-)
0
1
2
0 0.2 0.4
w
/U1.8H(-)
0
1
2
0 0.2 0.4
Z
/
H
(
-
)
w
/U1.8H(-)
0
1
2
0 0.2 0.4
w
/U1.8H(-)
0
1
2
0 0.2 0.4
w
/U1.8H(-)
PIV
LDV
Field
PIV
LDV
Field
PIV
LDV
Field
Field
(Z/H =1.8)
PIV
LDV
Field
Field
(Z/H =1.8)
307
This tendency fairly agrees with the previous results obtained from comprehensive outdoor
scale model (COSMO) [19][20].
The results show that the properties of mean and turbulent flow field can be reproduced
using wind tunnel experiments within a real urban canyon.
3.2 Differences in the properties of flow field at various canyon positions
3.2.1 Differences in the properties of flow field at various canyon positions along the
streamwise direction (cross section along streamwise direction (X-Z cross section))
The properties of flow fields within the central canyon were investigated and mainly
measured by using LDV from here, although the U-W components of C1-1 were only
measured PIV because of measurement problems.
Figs. 10 and 11 show the distributions of streamwise mean wind velocity U and the
distributions of lateral mean wind velocity V, respectively. The vectors within canyon C1 are
U-W components. The reversed flow of U appears near the ground. Therefore, the large
canyon vortex is formed. The distribution of U within each canyon (canyons C1, C2 and C3)
becomes almost the same as the results in the direction perpendicular to buildings. The
distribution of V within each canyon is particularly large near the ground, and the flow moves
from west to east in the positive direction. That is the reason why the direction of the
northerly wind is anticlockwise at 5-8 from the direction perpendicular to the long side of the
building, i.e., within these urban canyons, although the averaged flow moves in the lateral
direction, the large canyon vortex is formed in the streamwise direction (X-Z cross section),
as well as in the previous results in the direction perpendicular to the buildings (Sato et al.
[19]).

Fig. 10: The distributions of Fig. 11: The distributions of
streamwise mean wind velocity U lateral mean wind velocity V
(X-Z cross section) (X-Z cross section)
3.2.2 Differences in the properties of flow field at various canyon positions along the
streamwise direction (cross section along lateral direction (Y-Z cross section))
Figs. 12 and 13 show the distributions of the streamwise mean wind velocity U and lateral
mean wind velocity V, respectively. Although reversed flow of U appears near the ground
because of the canyon vortex (Fig. 12), the areas of the reversed flow on the side edge of the
canyons (C1-6, 2-6 and 3-6) are smaller than those near the center of the canyons (C1-3, 2-3
and 3-3). This tendency is almost the same as the results for ideal urban models obtained
from wind tunnel experiment and CFD (Soulhac et al. [21]). The areas of the reversed flow
decrease on the canyon edge (C1-6, 2-6 and 3-6), which are affected by the flow separated
from the side edge of a upstream building. On the other hand, the areas of the reversed flow
increase on another canyon edge (C1-9, C2-9 and C3-9). The reason is that spiral flow
(helical vortex [5], helical-type streamline [22]) is formed within the canyon. The spiral flow
308
within the canyon is discussed in detail later. The flows for inlet (V component) on the
canyon edges (C1-6, C2-6 and C3-6) move toward other edges (C1-9, C2-9 and C3-9), and V
is large near the ground. The maximum values of V within the canyons appear in the vicinity
of point 7 (C1-7, C2-7 and C3-7). The reason is that the flows separated from top and side
edges of upstream building assemble at this point.
To investigate these behaviors, the vectors of V-W components (contour is U component)
within canyon C1 are shown in Fig. 14. The flows separated from top and side edges of the
upstream building (C1-6) move toward both another canyon edge (C1-9) and the upper area
within the canyon. This tendency is similar to the helical-type streamline in the wind
direction inclined from the direction perpendicular to the building (Xie and Castro [22]). In
the experiments, the flows separated from top and side edges of the upstream building
assemble at point 7 and move spirally within the canyon, as well as the helical-type
streamline shown in the results of a previous paper [22]. It is considered that the spiral vortex
(hereafter, helical vortex) is formed by complicated elements, which are the wind direction
inclined from the direction perpendicular to buildings and complicated building
configurations. Fig. 15 shows the concept of the flow field within the canyon. The reversed
flow of U becomes small on the canyon edge (inlet side) because of the flow separated from
the side edge of the upstream building, and becomes large on the canyon edge (outlet side)
because of the helical vortex.

Fig. 12: The distributions of streamwise mean Fig. 13: The distributions of lateral mean
wind velocity U (Y-Z section) wind velocity V (Y-Z section)

Fig. 14: The distributions of vectors of Fig. 15: The concept of the V-W components
flow field (contour is U component) within the canyon
within canyon C1 (Y-Z section)
309
4 CONCLUSIONS
In this study, wind tunnel experiments for a real apartment complex were conducted to
confirm the reproducibility of the flow field and to comprehend the behaviors of turbulence
and mean characteristics of canyon vortices within the canyons.

1. The wind velocities and turbulent intensities obtained by both PIV and LDV fairly agree
with the field measurements within the urban canyon, although small-scale surface
roughnesses such as balconies are not modeled in the present wind tunnel experiments.
2. Above the canyon (Z/H > 1.0), the results obtained from the wind tunnel experiments are
twice smaller than those from the field measurements. That is the reason why the wind
tunnel experiments cannot reproduce the relatively large vortices such as mesoscale
turbulence. On the other hand, within the canyon (Z/H < 1.0), it is found that turbulent
motions generated by buildings dominate more than those generated by meso-scale
turbulence, because the turbulent intensities fairly agreed with the field measurement
values within the urban canyon.
3. Within the canyons, a helical vortex is generated by the interaction between the flows
separated from the top and side edges of an upstream building.
Acknowledgements
We would like to thank Professor Kenichi Narita at Nippon Institute of Technology,
Professor Hirofumi Sugawara at National Defense Academy in Japan, Professor Jun
Tanimoto at Kyushu University for providing their data on field measurements and Professor
Jun Tanimoto at Kyushu University and Professor Manabu Kanda at Tokyo Institute of
Technology for their suggestions and many informative discussions.
5 REFERENCES
1) R. Yoshie, H. Tanaka and T. Shirasawa, 2008; Experimental study on air ventilation in a
built-up area with closely-packed high-rise buildings, The 4th International Conference on
Advances in Wind and Structures (AWAS'08), jeju, Korea.
2) T. Kubota, M. Miura, Y. Tominaga, A. Mochida, 2008; Wind tunnel tests on the
relationship between building density and pedestrian-level wind velocity: Development of
guidelines for realizing acceptable wind environment in residential neighborhoods, Building
and Environment, Vol.43, pp.1699-1708.
3) F.T. DePaul and C.M. Sheih, 1986: Measurements of wind velocities in a street canyon,
Atmospheric Environment, Vol.20, pp.455-459.
4) P. Louka, S.E. Belcher and R.G. Harrison, 2000: Coupling between air flow in streets and
the well-developed boundary layer aloft, Atmospheric Environment, Vol.34, pp.2613-2621.
5) I. Eliasson, B. Offerle, C.S.B. Grimmond and S. Lindqvist, 2006: Wind fields and
turbulence statistics in an urban street canyon, Atmospheric Environment, Vol.40, pp.1-16.
6) I.D. Longley, M.W. Gallagher, J.R. Dorsey, M. Flynn, and J.F. Barlow, 2004: Short-term
measurements of airflow and turbulence in two street canyons in Manchester, Atmospheric
Environment, Vol.38, pp.69-79.
310
7) H. Takimoto, A. Sato, J.F. Barlow, R. Moriwaki, S. Onomura and M. Kanda, 2011: PIV
measurements of turbulent flow within an outdoor urban scale model and flushing motions in
urban canopy layers, Boundary Layer Meteorology, Vol.140, pp.295-314.
8) M.J. Brown, R.E. Lawson, D.S. Decroix and R.L. Lee, 2000: Mean flow and turbulence
measurement around a 2-D array of buildings in a wind tunnel, Report LA-UR-99-5395, Los
Alamos National Laboratory, Los Alamos, 7pp.
9) M.J. Brown, R.E. Lawson, D.S. DeCroix and R.L. Lee, 2001: Comparison of centerline
velocity measurements obtained around 2D and 3D building arrays in a wind tunnel, Third
Int. Soc. of Environmental Hydraulics conference, Tempe, AZ, USA.
10) T. Sekine, K. Unno, 1976; Experimental study on special structure of airflow around
buildings in urban area: Part I. Investigations on the characteristics of airflow around
regularly arranged city models, Transactions of AIJ, Vol. 245, pp.113-122 (in Japanese).
11) Y. Meng, S. Oikawa, 1997; A Wind-tunnel study of the flow and diffusion within model
urban canopies: Part 1.Flow measurements, Journal of Japan Society for Atmospheric
Environment, Vol.32, pp.136-147, 1997 (in Japanese).
12) A. Hagishima, J. Tanimoto, K. Nagayama and S. Meno, 2009; Aerodynamic parameters
of regular arrays of rectangular blocks with various geometries, Boundary Layer Meteorology,
Vol.132, pp.315-337.
13) S. Rafailidis, 1997: Influence of building areal density and roof shape on the wind
characteristics above a town, Boundary layer Meteorology, Vol.85, pp.255-271.
14) M. Kanda, 2006: Progress in the scale modeling of urban climate Review, Theor. Appl.
Climatol., Vol.84, pp.23-33.
15) P.K. Klein and M. Rotach, 2004: Mean flow and turbulence characteristics in an urban
roughness sublayer, Boundary-Layer Meteorology, Vol.111, pp.55-84.
16) A. Robins, H. Cheng, P. Hayden and T. Lawton, 2004: Flow visualisation studies I,
Note DAPPLE EnFlo 04.
17) M. Carpentieri, A.G. Robins and S. Baldi, 2009: Three-Dimensional mapping of air flow
at an urban canyon intersection, Boundary Layer Meteorology, Vol.133, pp.277-296.
18) H. Sugawara, A. Hagishima, K.Narita, H. Ogawa and M. Yamano, 2008; Temperature
and wind distribution in an E-W-oriented urban street canyon, Scientific Online Letters on
the Atmosphere (SOLA), Vol.4, pp.53-56.
19) A. Sato, T. Michioka, H. Takimoto and M. Kanda, 2010; Field and wind tunnel
experiments about flow and dispersion within street canyons, The Fifth International
Symposium on Computational Wind Engineering (CWE2010), Chapel Hill, North Carolina,
USA.
20) A. Inagaki and M. Kanda, 2008: Turbulent flow similarity over an array of cubes in near-
neutrally stratified atmospheric flow, Boundary layer Meteorology, Vol.615, pp.101-120.
21) L. Soulhac, V. Garbero, P. Salizzoni, P. Mejean and R.J. Perkins, 2009: Flow and
dispersion in street intersections, Atmospheric environment, Vol.43, pp.2981-2996.
22) Z.T. Xie and I.P. Castro, 2009: Large-eddy simulation for flow and dispersion in urban
streets, Atmospheric environment, Vol.43, pp.2174-2185.
311
Optimization of urban structures for city ventilation and pedestrian level
winds
Frieso Kipsch
a
, Bernd Leitl, Frank Harms, Stephan Werk
Meteorological Institute, KlimaCampus, University of Hamburg, Hamburg, Germany
a
Frieso.Kipsch@zmaw.de
ABSTRACT:

In the context of the research network KLIMZUG-Nord adaptation strategies for the
metropolitan region Hamburg to climate change are investigated. To simulate the effects of
various conditions concerning urban planning development a rising suburb (Hamburg-
Wilhelmsburg) was selected. An urban model (scale 1:350) has been built, consisting of over
1000 individual model buildings in the wind tunnel of the EWTL. The effects of large block-
like buildings on the pedestrian level wind field in a city quartier and city ventilation is
evaluated in wind tunnel experiments. The model is systematically investigated with regard
to different urban planning configurations and with a focus on distinctive places and strongly
affected regions in the wake of buildings.

1. Introduction
In recent years there has been an increase in population in urban areas leading to compact,
densely populated city centers requiring city planners, sociologists, architects and engineers
to accommodate this growth by developing attractive living conditions. In addition to socio-
economic factors, key design factors include climatic effects which can be investigated with
physical modeling. In wind tunnel studies, parameters such as air quality (regarding urban
heat islands), urban ventilation or the impact of local wind fields at pedestrian level are
studied. Since city centers are characterized by the building structure, comparative studies at
various stages of urban development are carried out. By means of systematic wind tunnel
testing insight can be gained regarding building structures counteracting negative urban
climate affects such as the urban heat island phenomenon or poor air quality in poorly
ventilated city quarters.
A suburban region was selected and modeled in the wind tunnel to investigate the impact of
building placement on the pedestrian level wind field. This work is being done within the
interdisciplinary KLIMZUG-Project where sociologists, town planners, architects and
engineers will work together on the challenges that arise due to climate change. For this
purpose, different climate scenarios were outlined and discussed.

312
2.1 The field site
The district of Hamburg-Wilhelmsburg is examined in the course of the KLIMZUG-Nord
project. Since the area will be restructured extensively, climatic issues are examined, taking
into account various possible scenarios. As a result, the area may be adapted properly to
climate change. The suburban area is located near industrial facilities and a harbor, with a
mixture of open areas, parks and urban street canyons. In addition, a significant North-South
street configuration marks the district Hamburg Wilhelmsburg.

Figure 1: Sketch of the investigated area

2.2 Experimental setup
A model of the suburban region Hamburg-Wilhelmsburg was built at a scale of 1:350. The
predominate winds in the area are southwesterly (data taken from the NDR transmitting
pole), therefore the model was built to investigate winds coming from a direction of 235 in
the wind tunnel. The wind tunnel model was 9m long and 4m wide, which represents an
investigation area of about 12.6 km in natural scale. The modeled approach flow was
designed to incorporate upstream building effects, therefore Styrofoam buildings were placed
upstream. The local orography was constructed out of wood with a thickness of 1mm.

Figure 2: Investigated model in the wind tunnel: a) approach flow, b) model area

313
For the exact placement of the building models, the corners were marked by means of a CNC
milling cutter with one millimeter accuracy on the model base. By using defined positions the
possibility of misalignment could be determined and corrected in the process of the
measuring campaign. During the measurement campaign, at least every 2
nd
day the correct
position of the traverse system was checked and adjusted if needed.

The model was placed in WOTAN, the large boundary layer wind tunnel of the
Meteorological Institute of the University of Hamburg. The WOTAN tunnel has an
investigation section of 18m length, 4m width and 3m height. The boundary layer is
generated using a setup of 5 different kinds of bluff body obstacles as roughness elements
(Figure 2a) and counihan-type spires at the entrance behind the rectifier. The roughness
elements were placed in the first 7.5m in front of the model section. The ceiling of the wind
tunnel was adjusted to compensate for the different pressures due to the blockage effects of
the model, according to VDI-guideline 3783/12. For flow measurements a 2D fiber-optic
Laser-Doppler-Anemometer (Fa. Dantec) was used; the flow was seeded with fluid particles
about 2m in diameter by a Hazer (Fa. Tourhazer). The measuring period was determined on
the basis of convergence analyses at different height levels to ensure a sufficient statistical
reliability of the measurements.

2.3 Approach flow
The characteristics of the approach flow are well-known due to the continuous collection of
the climatic data at the NDR transmitting pole. After statistical analysis the inflow conditions
in the wind tunnel could be modeled according to the natural conditions. The characteristics
of the boundary layer flow were examined within the front range of baseplate 14 (e.g.
Figure 1). Data was collect under inhomogeneous conditions inside the modeled area, hence
the flow conditions measured in the model area differ from the inflow conditions [according
to C. Peeck]. The spectra of the turbulent fluctuations at three heights are presented in Figure
3a, with the empirical model spectra from Simiu and Scanlan also shown. Each spectra shows
a slight increase against the empirical model. The longitudinal integral length Lu
x
(Figure 3b)
was calculated at various heights above the ground. The results fit to the inflow conditions
over urban area. According VDI-guideline 3783/12, the turbulence intensity (Figure 3c) fits
likewise to the assumed boundary layer. The power law exponent = 0.16 and the roughness
length z
0
= 0.02 seem slightly too small, but since they were determined in the model area
under inhomogeneous flow conditions, they possess only comparable character.

2.4 Uncertainties of the measurements
To estimate the confidence interval of the collected data, several measurements were made
daily at a defined horizontal position and height. The velocity measurements take place at no
prescribed time, partly at the beginning, at the end, or between other measurements.
From this data a reliable statement about the uncertainty range of the collected data can be
made after the measurement campaign. Due to an exchange of the experimental setup
(replacement of the LDA-measurement system with reorientation), the uncertainties include
the coincidental fluctuations and the biased error (Felderhoff et al, 2007). Two relative
uncertainties can be determined from the entire measurement campaign, where the ranges of
uncertainties are 2.51 and 3.05%, adding up to a total uncertainty of 5.20%.
314

3.1 Determining the pedestrian level wind field
For a statement about urban ventilation the ground-level wind field was examined. For this
purpose over 230 positions were measured at the pedestrian level (height above ground:
1.75m). Measurements were taken at locations indicative of human exposure, e.g.,
marketplaces, church forecourts, school yards and sidewalks. Additionally, the profile
development was investigated over the model area on the basis of 28 full sized profiles as
well as 7 partial profiles within street canyons. Since the effect of a high-rise building with a
large aspect ratio was examined, the wind field information was taken for both conditions
with and without the building. To determine the reference velocity the wind speed at 175m
was chosen from the NDR transmitting pole with a mean velocity of 7.6 m/s.
To carry out a quantitative estimate of the pedestrian level wind field, the measured data are
evaluated according to established wind comfort criteria. In this paper the effects of the mean
velocities and the turbulent characteristics are considered, while considerations such as
casting of shadow (for high-rise buildings), air humidity, thermal conditions and stratification
impacts (e.g. Stathopoulos, 2004) were neglected. The wind comfort situation is characterized
by Equation 1, based on the mean velocity where a gust factor is added to include the
turbulent character of the local flow conditions. The gust factor is characterized by the
standard deviation and a peak factor k (Table 1), which defines the intensity range of the
supposed gust wind speed. In case of k = 0, the mean wind speed is regarded without the
fluctuating processes.
u
eff
= u
mean
+ k
u
(Eq. 1)
Figure 3: Flow characteristics in front of the city center
a) Spectral density distribution of the approach flow
b) Longitudinal integral length scale
c) Turbulence intensity I
u

315
Table 1: Values for k used in applied criteria (according to Koss, 2006)

k Criteria by
0 NEN 8100, Isyumov & Davenport
1 Gandemer
1.5 Isyumov & Davenport
3.5 Melbourne

The factor k can be interpreted as an estimation of an exceedance probability (Koss et al,
2004). Another point of view in wind comfort uses these exceedance probabilities to draw
conclusions about excess abundances in the effective wind speed limit within a defined
period of time. From this the in situ conditions for this area could be derived (Eq. 2). Since
there is no general rule which criteria or exceedence probability should be taken, different
methodologies can be applied. Due to the different approaches behind these criteria, they are
difficult to compare with each other.
u
eff
= u
mean
+ k
u
u
lim
(Eq. 2)

In the current case we are considering the relative change of wind conditions in the city area,
therefore we limit ourselves to two criteria to determine the wind field conditions. The two
criteria are the mean velocity approach presented in the NEN 8100 and the peak gust wind
speed approach of Melbourne which focuses on the standard deviation of the wind speed.
Figure 4a+c shows the measured velocities in the inner city with each proposed gust factor.
There are no comfort zone classifications shown in the figure because this is not the objective
for this investigation. Figure 4b+d show the differences in the investigated wind field with
and without the large block-like building. As seen in Figure 4b, there is a slight decrease of
the mean velocity in the inner city in the wake of the block-like building while the standard
deviation increased significantly in this region (Figure 4d). The standard deviation is used to
determine the momentum fluxes and evaluate the effects of urban ventilation. The mean
wind speed and criteria for the effective wind speed shown above for wind comfort
investigations was not analyzed. In the future, gust wind speeds may be considered for this
analysis.

3.2 Effects of urban greening on the pedestrian level wind field
Another point to consider is changes in wind fields due to tress and planting in urban areas.
An area with small garden allotments (i.e., common garden area) is regarded, which is
incidentally declared as a forest. In order to determine the effects of forest areas on the wind
field, a section was aerodynamically modeled and investigated. The subregion was selected
due to that approximately homogeneous height of the tree population. For the modeling of the
forest area aerodynamically similar models (metallic mesh with a mesh size of 2.8mm formed
as open rings with a specific height of 42mm - e.g. Figure 5a) were used according to Aubrun
et al (2004).
316

Figure 4: Comparison of applied comfort criteria to the wind field (from upper left to lower right):
a) wind field, mean velocities (according to NEN 8100), b) comparison w/o the block-like building,
c) wind field, peak gust wind speed approach, d) comparison w/o the block-like building

Figure 5: a) aerodynamically similar model of a forest, b) location of the forest in the model area

317
As shown in Figure 6, the turbulence intensity for the velocity differs in the wake of the
forest model. The effects can be determined up to some hundred meters in the far wake of the
model. For an impression of the turbulent wind field in the inner city the street canyons and
parks must be modeled accurately, therefore model tress must be included in the approach
flow and the city center. In a second measuring campaign these results as well as the effects
of urban planting in street canyons and the drag coefficient of individual trees [according to
A. Schrn] will be considered.

Figure 6: Effects on the wind field in its wake: a) decrease of velocity, b) increase of turbulence

4 Future work
In a subsequent measurement campaign conclusions will be drawn on the urban ventilation
aspects on the basis of dispersion tests. The air interchange rates in street canyons are to be
determined at several significant points inside the model area. With the use of detectable
gases, the urban ventilation will be regarded by correlating the wash out time period and the
turbulent wind conditions. Similar to the air exchange rate used for designing indoor climate
in buildings there may be a similar ventilation rate that exists in urban areas with regard to
building configuration.

5 REFERENCES
Aubrun, S., Leitl, B., 2004. Development of an improved physical modeling of a forest area in a wind tunnel.
Atmospheric Environment 38, 2797-2801

Felderhoff, R., Freyer, U., 2007. Elektrische und elektronische Messtechnik. Carl Hanser Verlag, Mnschen

NEN 8100:2006. Wind comfort en wind danger in the built environment. TU Delft, Delft.

Koss, H., Sahlmen, J.,2004 Methods in pedestrian wind comfort assessment: theoretical and practical
comparissons. Proceedings of the International Conference on Urban Wind Engineering and Building
Aerodynamics.

Koss, H., 2006. On differences and similarities of applied wind comfort criteria. Journal of Wind Engineering
and Industrial Aerodynamics 94, 781-797

318

Stathopoulos, T.,2004. Outdoor human comfort in an urban climate. Building and Environment 39, 297 305

VDI-guideline 3783/12, 2000. Physical modeling of Flow and Dispersion Processes in the Atmosperic
Boundary Layer Application of Wind Tunnels. Beuth Verlag, Berlin.

319

LES of dispersion in a street canyon with tree planting
P. Moonen
a,b
, C. Gromke
c,d
, V. Dorer
a
, J. Carmeliet
a,b

a
Dbendorf, Switzerland
b
ETH, Swiss Federal Institute of Technology Zrich, Zrich, Switzerland
c
WSL, Institute for Snow and Avalanche Research SLF,
Davos Dorf, Switzerland
d
Institute for Hydromechanics, Karlsruhe Institute of Technology, Karlsruhe, Germany
ABSTRACT: We assess the potential of numerical simulations to reliably predict pollutant dis-
persion. Detailed unsteady numerical simulations of coupled flow and dispersion are conducted
for a street canyon with tree planting. Different crown porosities are considered. The simulation
results are compared to wind tunnel measurements, and it is shown that the numerical model can
accurately capture the experimentally observed trends, both qualitatively and quantitatively.
These observations suggest that quantitative information for assessment, planning and implemen-
tation of pollutant mitigation strategies can be obtained based on numerical simulations.
1 INTRODUCTION
Vehicular emission is one of the major sources of anthropogenic pollutants in cities (Liu et al.
2005). The majority of these emissions occur in the urban street canyon. Because of their adverse
impact on human health, numerous studies have been conducted to enrich our understanding of
scalar removal mechanisms in urban street canyons.
Wind tunnel experiments for perpendicular approaching flow showed that avenue-like tree
planting causes increased exhaust concentrations at the leeward wall and reduced concentrations
at the windward wall (Gromke and Ruck 2007). It was shown by Gromke et al. (2008) that nu-
merical simulations based on steady RANS (Reynolds Averaged Navier Stokes) could qualita-
tively reproduce the main aspects of the flow, even though the flow velocities were underesti-
mated. A good agreement with wind tunnel concentration data was achieved by increasing the
diffusivity through lowering the turbulent Schmidt number. Recent investigations by Moonen et
al. (2011a) revealed the importance to explicitly account for turbulent fluctuations on the air ex-
change rate of street canyons: in certain cases, differences up to a factor ten were found between
RANS results and time-averaged Large Eddy Simulations (LES). Other studies have pointed out
the importance of turbulent fluctuations on pollutant removal processes (Baik and Kim 2002;
Letzel et al. 2008). Based on these observations, it can be expected that LES simulations are ca-
pable of better reproducing the actual flow and dispersion patterns.
The goal of this study is to assess the potential of numerical simulations to make quantitative
predictions of dispersion phenomena. To that extent, we perform detailed unsteady simulations of
a number of dispersion experiments, conducted in the atmospheric boundary layer of the Univer-
sity of Karlsruhe (Karlsruhe Institute of Technology), Germany. The results of these experiments
are available to the scientific community via the online database CODASC - COncentration DAta
of Street Canyons (CODASC 2008). We focus on one of the investigated configurations, namely
a 1:150-scaled model of an isolated urban street canyon, with avenue-like tree planting and near-
ground released traffic emissions, under perpendicular approach flow (Section 2). The numerical
model mimics the experimental setup as closely as possible and aims at maximizing accuracy
(Section 3). By comparing the predicted and simulated time-averaged concentration fields, we
assess the model performance (Section 4). The good agreement illustrates the potential of LES to
make quantitative predictions of dispersion phenomena.
320

The wind tunnel model consists of two parallel buildings of 0.12 x 0.12 x 1.2 m (L
x
x L
y
x L
z
),
made out of Plexiglas, and positioned 0.12 m apart in streamwise direction (x). In this way, a
street canyon with W/H = 0.12/0.12 = 1 and L/H = 10 is formed (H = 0.12 m), see Figure 1.
The avenue-like tree planting was modeled by means of a beam-shaped metallic lattice cage
construction, filled with a fiber-like synthetic wadding material (Gromke 2011). The lattice cage
with dimensions 0.5 H x 0.66 H 10 H was aligned along the street axis, with the top surface posi-
tioned at roof level. The tree trunks were not modeled as they are fairly small compared to the
tree crown and have a negligible influence on the flow. Different crown porosities were realized
by filling the lattice cage with defined masses of wadding material. The permeability of the wad-
ding material was experimentally determined and characterized by means of a pressure loss coef-
ficient [m
-1
], defined as the difference in static pressure p
st
[Pa] between the windward (p
st,ww
)
and leeward (p
st,lw
) side of a porous sample in forced convection with uniform streamwise thick-
ness d [m], additionally normalized by the dynamic pressure p
dyn
[Pa]
( )
, ,
2
1
2
st ww st lw
st
dyn
p p
p
p d u d
A
= = (1)
with u [m/s] the mean streamwise velocity and [kg/m
3
] the density of the fluid (i.e. air). Two
different crown porosities were investigated ( = 80 m
-1
and = 200 m
-1
), as well as the extreme
cases = 0 m
-1
(i.e. without trees) and = + m
-1
(i.e. non-porous crown).
The release of traffic exhausts was simulated by emitting a tracer gas (sulphur hexafluoride or
SF6) from four line sources of equal source strength, two for each traffic direction, designed ac-
cording to the method described in Meroney et al. (1996). In order to render the gas flow insensi-
tive to local pressure fluctuations in the model street canyon, the line source consists of a series
of flush-mounted and equidistantly-spaced hypodermic tubes over which a substantial pressure
drop occurs. In order to account for the traffic exhausts released on the sidewise street intersec-
tions, the line source exceeded the street canyon by 0.11 m on either side. The line source
strength was monitored and controlled, ensuring a constant tracer gas supply during the mea-
surements. Both model building walls facing the street canyon were covered with 98 concentra-
tion measurement taps. An electron capture detector (ECD) was used for analyzing the mean
tracer gas concentration. The measured concentrations were normalized according to:
H
cu H
c
Q l
+
= (2)
with c [-] the measured concentration, u
H
[m/s] the mean streamwise flow velocity at height H
[m] in the undisturbed approach flow and Q/l [m
2
/s] the tracer gas source strength per unit length,
with Q being 6.5 cm
3
/min SF6 mixed with 7000 cm
3
/min dry air and l = 1.42 m.
The street canyon model was exposed to an urban atmospheric boundary-layer type flow per-
pendicular to the canyon length axis. The vertical profile of mean flow speed u(y) can be well
approximated by a power-law with roughness exponent k
u
= 0.3 [-]:
( )
( )
u
k
H H
u y
y
u y y
| |
=
|
\ .
(3)
with u(y
H
) = 4.65 m/s the mean flow speed at reference height y
H
= H. A comprehensive docu-
mentation of the boundary-layer flow, involving measurements of the integral length scale profile
and spectral distribution of turbulent kinetic energy, can be found in Gromke and Ruck (2005).
The streamwise pressure gradient was made to vanish by adjusting the wind-tunnel ceiling when
the street canyon model was set up in the test section.
321

3 NUMERICAL MODEL
3.1 Domain dimensions
The numerical model mimics the experimental setup as closely as possible (Figure 1). Therefore,
the cross sectional dimensions of the numerical domain are taken equal to the dimensions of the
wind tunnel, namely 2 m wide by 1 m high, hereby accurately reproducing the experimental
blockage ratio of 7.2%. The total streamwise length of the domain is 25H (3 m), and consists of
an approach flow region of 7H (0.84 m), a model region of 3H (0.36 m) and a wake region of
15H (1.8 m). The position of the upstream domain boundary coincides with the beginning of the
wind tunnel test section, i.e. where velocity profiles have been experimentally acquired. The
downstream distance is chosen in accordance with the best practice guidelines formulated as out-
come of COST Action 732 (Franke et al. 2011).

Figure 1. Layout of the computational domain: (a) perspective view and (b) close-up of the canyon model.
3.2 Boundary conditions
We simulate the flow in an atmospheric boundary layer wind tunnel, in absence of temperature
effects. Boundary conditions (BCs) are chosen accordingly.
A no-slip wall BC is imposed along the top, bottom and lateral boundaries of the domain, cor-
responding to flow along the smooth wind tunnel walls. The same boundary condition is applied
on the surfaces of the canyon model. All walls are considered impermeable for species transport.
On the inflow boundary, time-dependent inlet conditions are generated from the measured
mean profiles of velocity (Eq. 3) and turbulence intensity by means of Ansys FLUENTs inbuilt
spectral synthesizer (Ansys Inc. 2009). In this method, the fluctuating velocity components are
computed by synthesizing a divergence-free velocity-vector field from the summation of (100)
Fourier harmonics. The inflowing fluid is dry air at 293.15 K with zero background tracer con-
centration.
On the outflow boundary, a zero diffusive flux is imposed for all flow variables in the direc-
tion normal to the exit plane. This means that the conditions of the outflow plane are extrapolated
from within the domain and have no impact on the upstream flow. This assumption is valid for
fully-developed flows.
(a) (b)
3 m
2 m
1 m
inflow
outflow
vegetation
line sources
smooth wall
322

The line source is modeled as a series of equidistantly-spaced point sources at ground-level, as
in the experiment. The individual hypodermic tubes have been approximated by square inflow
boundaries, each 0.005 x 0.005 m
2
. The imposed flow rate and concentration are equivalent to the
experimental conditions, namely a mixture of 7000 cm
3
/min dry air ( = 1.225 kg/m
3
,
= 1.7894 10
-05
kg/(ms), molecular weight = 28.966 kg/kgmol) and 6.5 cm
3
/min SF6 ( = 6.164
kg/m
3
, = 1.42 10
-04
kg/(ms), molecular weight = 146.05499 kg/kgmol). The properties of air
are taken from the inbuilt material database (Ansys Inc. 2009). The ones of SF6 are compiled
from various literature sources. SF6 is considered as an inert gas (i.e. chemical reactions are not
modeled). Because of the larger surface area of the individual point sources in the CFD model,
the exhaust momentum is lower than in the experiment. This is compensated by adding a source
term to the vertical momentum equation in the cells in direct contact with the source. Note that
other researchers (Gromke et al. 2008, Salim et al. 2011a,b) proposed more simplified modeling
strategies for the line source and reported a qualitatively good agreement with the experimental
results. Since SF6 is about 6 times heavier than air, gravitational effects need to be included.
The effect of vegetation on the flow is included in the model by adding a source term (actually
a sink term) to the momentum equations within the region occupied by the tree crowns. The for-
mulation of the source term is based on Eq. (1) and requires the experimentally determined pres-
sure loss coefficient [m
-1
] as only input parameter.
3.3 Computational grid
In LES, the computational grid acts as a filter. The turbulent structures larger than the grid size
(i.e. the large eddies) are resolved, while the smaller ones are modeled. Small eddies are more
isotropic. Cubic elements have an equal filter length in the three principal directions, and are thus
the most optimal element type for resolving turbulent structures with LES. The grids employed in
this study are therefore entirely built up with this element type.
Since uniform grid refinement rapidly results in a prohibitively large number of elements for
grids consisting of cubic elements, we developed and validated an algorithm that systematically
refines the grid in all three dimensions towards regions of interest (Moonen et al. 2011b). The
final grid consists of 1226062 perfectly cubic elements (Figure 2). The element size gradually
reduces from H/3 (= 0.04 m) in the bulk up to H/24 (= 0.005 m) in the region of interest. The em-
ployed element size in the region of interest is about two times smaller than the one obtained by
sensitivity analysis (Salim et al. 2011a) and about 2.5 times smaller than the recommended max-
imal grid size for isolated buildings (Franke et al. 2011) and street canyons (Bartzis et al. 2004).

Figure 2. (a) Surface mesh and (b) part of the vertical cross section of the employed computational grid.
323

3.4 Solution method
The three-dimensional Navier-Stokes equations, the continuity equation and the species transport
equation were solved using the commercial CFD code Ansys Fluent 12.1 that employs the con-
trol-volume technique. The staggered pressure-based solver was selected. For the unsteady LES
simulations, closure was obtained by means of the dynamic Smagorinsky-Lilly subgrid-scale
model (Lilly 1992). The subgrid-scale turbulent Schmidt number was dynamically updated, ob-
viating the need to prescribe a default value. Near-wall modeling was performed using the law-
of-the-wall. It was verified that the instantaneous y+ values in the region of interest did not ex-
ceed 60, i.e. the limit of validity (Ansys Inc. 2009). Pressure-velocity coupling was taken care of
by the SIMPLEC algorithm (Vandoormaal and Raithby 1984). Pressure interpolation was per-
formed by means of the PRESTO! scheme (Peyret 1996). Green-Gauss cell-based gradient eval-
uation was employed. The third-order MUSCL discretization scheme (Van Leer 1979) was used
for the convection terms of the governing equations. By default, second-order accuracy is used
for the viscous terms. Second order implicit time integration is employed.
3.5 Initialization and termination criterion
The simulations were initialized from a preliminary RANS simulation. A time step of 0.00125 s
was selected, resulting in cell Courant numbers below 2.5. Typically 100 iterations were required
per time step, until the desired level of convergence was reached, i.e. constant residuals of 10
-4
or
less for all equations (Franke et al. 2011). We reached 10
-4
for the species transport equation, 10
-9

for the continuity equation and machine precision (<10
-12
) for the three momentum equations.
The residuals are defined as the imbalance of the conservation equations, summed over all ele-
ments. These values are scaled and the scaled values are used in the code as a measure of the ite-
ration convergence (Ansys Inc. 2009).
All simulations were run on the ETH Brutus cluster (17000 cores, peak performance 180 TF).
Typically, one time step (0.00125 s) required 1.6 hours of CPU time. A stable flow pattern is
reached after 10 s (8000 time steps), corresponding to more than 15 domain flow-through times.
In this paper, we report statistics, collected during the subsequent 800 time steps (1 s). Statistics
are not yet fully converged at this point, which is also visible in some of the presented results.
Nevertheless, some meaningful tendencies and trends can already be observed.
4 RESULTS
We present a comparison between the simulations, based on the model described in section 3,
and the corresponding wind tunnel measurements. The comparison focuses on the mean near-
wall pollutant concentration. First, results are presented for the street canyon without vegetation.
Afterwards, the influence of vegetation on the mean near-wall pollutant concentration is dis-
cussed.
4.1 Street canyon without tree-planting
Figure 3 depicts mean profiles of normalized concentration (Eq. 2) at three different locations
near (a) leeward and (b) windward canyon wall. All locations are situated 5 mm in front of the
wall, corresponding to the position of the measurement taps in the wind tunnel experiment. y/H
and z/H correspond to vertical and lateral dimensions, respectively. All locations with y/H=0 are
situated on the ground. The vertical plane of symmetry of the model has z/H=0.
From Figure 3, it is clear that there is a good agreement between simulated and measured con-
centrations, indicated with solid lines and circular markers respectively. Some minor discrepan-
cies can be observed, e.g. a small overestimation of the normalized concentration at z/H=0 on
both walls, but these are within an acceptable range.
324


Figure 3. Mean profiles of normalized concentration at three different locations near (a) leeward and (b) windward
wall of an unobstructed street canyon, as obtained by LES (solid line) and wind tunnel (markers). The dashed lines
indicate the spread of the instantaneous normalized concentration, as obtained by LES.

The dashed lines in Figure 3 indicate the extreme values of normalized concentration, encoun-
tered during the simulation (after steady-state was reached). The profiles are not entirely smooth,
since statistics are still incomplete. Nevertheless, it is seen that peak concentrations largely ex-
ceed the average concentration. This is of major importance when discussing human short-time
exposure to emissions. Steady-state simulations, such as RANS, do not provide this type of in-
formation.
4.2 Street canyon with avenue-like tree-planting
In this section we analyze to which extent the near-wall concentration field is affected by the
presence of avenue-like tree-planting in the street canyon. Figure 4 depicts the mean profiles of
dimensionless concentration at the same locations near (a) leeward and (b) windward canyon
wall for four different crown porosities. Results are given for two realistic crown porosities
( = 80 m
-1
and = 200 m
-1
), as well as for the extreme cases = 0 m
-1
(no tree) and = + m
-1

(non-porous tree).
In comparison with the street canyon without tree-planting, larger discrepancies between si-
mulated (lines) and measured results (markers) can be observed. This is most likely due to the
(too) short averaging time of the simulated data, resulting in a non-representative ensemble aver-
age. Current research focuses on testing this hypothesis.
Both experiments and simulations show that crown porosity has a distinct effect on the near
wall concentrations. For the leeward wall, increased crown density generally leads to increased
concentrations, i.e. scalars are more trapped underneath trees with a denser crown. The concen-
tration field at the windward wall is less affected by the crown density.
(a)
(b)
325


Figure 4. Mean profiles of normalized concentration at three different locations near (a) leeward and (b) windward
wall of a vegetated street canyon, as obtained by LES (lines) and wind tunnel (markers), for four different crown po-
rosities: = 0 m
-1
(-;), = 80 m
-1
(--;), = 200 m
-1
(-
.
-
.
;) and = + m
-1
(
.
;).
5 CONCLUSIONS
Unsteady simulations of dispersion inside a street canyon with avenue-like tree planting were
performed and compared to wind tunnel measurements. The model mimics the experimental se-
tups as closely as possible:
The dimensions of the computational domain are equal to the dimensions of the wind tunnel.
Boundary conditions for velocity and turbulent kinetic energy are extracted from the wind tun-
nel measurements.
The line source is modeled as a series of point sources, instead of a continuous line, area or vo-
lume source. Care is taken to match both the exhaust momentum and the chemical composition
of the tracer gas.
The vegetation model has been validated by simulating the experimental setup, used to deter-
mine the pressure loss coefficient, and excellent agreement was found.
All remaining modeling choices, such as the grid structure and resolution, the temporal discreti-
zation, the employed algorithms and the termination criterion, aim at reaching maximum accura-
cy, at the cost of being computationally expensive simulations take about 1150000 times
longer than reality on a state-of-the-art quad-core pc. The latter is the main reason why statistics
were not yet fully converged at the time of writing this paper. In view of applying more simpli-
fied models in the future, it is important to have a reference solution, against which the impact of
simplifications can be tested.
Although based on incomplete statistics, the model shows very good performance. The resi-
duals approach machine-precision within 100 iterations for most equations, indicative for the
(a)
(b)
326

quality of the computational grid. Predicted mean scalar concentrations inside a non-vegetated
street canyon match the experimentally observed values with good accuracy. For street canyons
with trees, some discrepancies between the model and the corresponding wind tunnel experiment
could be observed, but general trends were still captured. It is expected that the agreement will
improve as the averaging period is extended.
The results in this paper illustrate the potential of numerical simulations in reliably predicting
pollutant dispersion. The main advantage of numerical simulations over wind tunnel measure-
ments is the availability of full-field data of all quantities of interest over the considered period of
time. Furthermore, simulations can be performed at full scale, hence obviating flow similarity is-
sues. The main disadvantage is however the extremely large computational cost, both in terms of
memory and time. Future research will therefore focus on assessing the impact of model simula-
tions on prediction accuracy.
6 REFERENCES
Ansys Inc., Ansys Fluent 12.0 Users Guide, 2009.
Ansys Inc., Ansys Fluent 12.0 Theory Guide, 2009.
Baik, J.-J., Kim, J.-J., 2002. On the escape of pollutants from urban street canyons. Atmospheric Environment 36,
527-536.
Bartzis, J.G., Vlachogiannis, D. and Sfetsos, A., 2004. Thematic area 5: Best practice advice for environmental
flows. The QNET-CFD Network Newsletter 2 (4), 34-39.
CODASC, 2008. Concentration Data of Street Canyons. Internet Database. Karlsruhe Institute of Technology KIT,
Germany. www.codasc.de.
Franke, J., Hellsten, A., Schlnzen, H. Carissimo, B., 2011. The COST 732 Best Practice Guideline for CFD simula-
tion of flows in the urban environment: a summary, International Journal of Environment and Pollution 44(1-4),
419-427.
Gromke, C., Ruck, B., 2005. Die Simulation atmosphrischer Grenzschichten in Windkanlen. Fachtagung Laser-
methoden in der Strmungsmesstechnik, 6-8 September 2005, BTU Cottbus.
Gromke, C., Ruck, B., 2007. Influence of trees on the dispersion of pollutants in an urban street canyon
Experimental investigation of the flow and concentration field, Atmospheric Environment 41, 32873302
Gromke, C., Buccolieri, R., Di Sabatino, S., Ruck, B., 2008. Dispersion study in a street canyon with tree planting
by means of wind tunnel and numerical investigations Evaluation of CFD data with experimental data, Atmos-
pheric Environment 42, 86408650.
Gromke, C., 2011. A vegetation modeling concept for Building and Environmental Aerodynamics wind tunnel tests
and its application in pollutant dispersion studies, Environmental Pollution 159 (2011) 2094-2099.
Letzel M.O., Krane M., Raasch S. 2008. High resolution urban large-eddy simulation studies from street canyon to
neighbourhood scale, Atmospheric Environment 42(38), 8770-8784.
Lilly, D. K., 1992. A proposed modification of the Germano subgridscale closure method, Physics of Fluids A 4,
633-635.
Liu, C.-H., Leung, D.Y.C., Barth, M.C., 2005. On the prediction of air and pollutant exchange rates in street canyons
of different aspect ratios using large-eddy simulation. Atmospheric Environment 39, 1567-1574.
Meroney, R., Pavageau, M., Rafailidis, S., Schatzmann, M., 1996. Study of line source characteristics for 2-D physi-
cal modelling of pollutant dispersion in street canyons, Journal of Wind Engineering and Industrial Aerodynam-
ics 62, 279290.
Moonen, P., Dorer, D., Carmeliet, J., 2011a. Effect of local wind climate on the ventilation potential of urban street
canyons. Journal of Wind Engineering and Industrial Aerodynamics 9(4), 414-423.
Moonen, P., Dorer, D., Carmeliet, J., 2011b. On the relevance of incorporating flow unsteadiness in simulations of
wind flow in the urban environment. 13
th
International Conference on Wind Engineering (ICWE13), Amsterdam,
The Netherlands, July 10-15, 2011.
Peyret, R., 1996. Handbook of Computational Fluid Mechanics. Academic Press Limited, USA.
Salim, S.M., Buccolieri, R., Chan, A., Di Sabatino, S., 2011. Numerical simulation of atmospheric pollutant disper-
sion in an urban street canyon: Comparison between RANS and LES. Journal of Wind Engineering and Industri-
al Aerodynamics 99(2-3), 103-113.
Salim, S.M., Chan, A., Cheah, S.C., 2011. Numerical simulation of atmospheric pollutant dispersion in tree-lined
street canyons: Comparison between RANS and LES. Journal of Building and Environment 46(9), 1735-1746.
Vandoormaal, P., Raithby, G.D., 1984. Enhancements of the SIMPLE Method for Predicting Incompressible Fluid
Flows. Numerical Heat Transfer 7, 147-163.
Van Leer, B., 1979. Towards the ultimate conservative difference scheme. V - A second-order sequel to Godunov's
method. Journal of Computational Physics 32, 101-136.
327
Modelling Probability Density Function of passive
scalar concentrations within a Turbulent Boundary
Layer
Nironi C., Salizzoni P., Mejean P., Grosjean N., Soulhac L., Cierco F.X.
Laboratoire de Mcanique des Fluides et dAcoustique, Universit de Lyon,
CNRS, Ecole Centrale de Lyon, Universit Claude Bernard Lyon I, 36,
Avenue Guy de Collongue, 69134 Ecully, France, chiara.nironi@ec-lyon.fr
ABSTRACT: The impact assessment of an accidental release of toxic or inflammable
substances in the atmosphere requires the prediction of the Probability Density Function
(PDF) of their concentration. This can be estimated by means of a large variety of strategies
(Pope, 2000). In this study we consider meandering plume model with internal fluctuations,
which provides an analytical relation for the concentration PDF and its moments. The model
predictions have been compared to experimental results concerning the dispersion of a
passive tracer in a simulated neutral boundary layer developing over a rough surface. This
allows us to critically discuss the accuracy of the model when applied to dispersion
phenomena in atmospheric flows.
1 INTRODUCTION
The prediction and characterisation of concentration fluctuations of a passive scalar in a
turbulent flow is important in a number of applications. This is required for example for the
impact assessment of odours, toxic or inflammable substances released in the atmosphere. In
these cases the knowledge of the mean concentration only is not sufficient, and it is required
the computation of the Probability Density Function (PDF) of the pollutant concentration, or
at least the estimation of its low order moments. A significant body of research work in the
last years has been devoted to improve analytical models for the estimate of the moments of
the pollutant concentration PDF in turbulent flows.
A starting point was given by Gifford (1959) who derived a meandering plume model for the
PDF of concentration of a passive scalar emitted from a point source. Evolutions of Giffords
two-dimensional isotropic meandering plume model are ascribed to several authors. Among
these we cite Yee at al. (1994) who included the effects of internal fluctuations in a one-
dimensional meandering plume model, and validated their model to field measurements in
the atmospheric boundary layer for a fixed distance from the ground. A further development
of the model was then provided by Yee and Wilson (2000) who developed a two dimensional
isotropic fluctuating plume model for the dispersion of a passive scalar in grid generated
turbulence.
The aim of this study is to verify the reliability of a meandering plume model for applications
in a neutral atmospheric boundary-layer. This is done comparing the model predictions with a
set of measures of dispersion from a point-source plume of scalar tracer.

328
2 THE FLUCTUATING PLUME MODEL
The fluctuating plume model (Yee et al., 1994) is a meandering plume model incorporating
internal fluctuations which allows the determination of all one-point statistics of
concentration. In particular, it provides explicit analytical expressions for all the
concentration moments and the concentration Probability Density Function (PDF).
Tracer dispersion is regarded in terms of two contributions depending on the size of the
turbulent eddies involved in relation to the instantaneous plume width. One is given by eddies
which are larger than the plume width and the other by eddies which are smaller than the
plume width. The large scale sweeping action of the turbulence causes an irregular
meandering movement of the plume as a whole. The plume mass centre is thus transported by
these large-scale eddies, that produces low frequency concentration fluctuations and that do
not affect the internal structure of the plume. Simultaneously, an internal fine-scale structure
is developed as a result of the turbulent stirring and mixing action of the eddies of smaller
size, that entrain clean air contributing to the plume internal fluctuations. The predominance
of one of the two phenomena depends on the stage of plume development and on the source
size. The hypothesis on which the model is based is that each of the two mechanisms is
bounded to a characteristic spatial scale and that the two phenomena are therefore statistically
independent.
2.1 Model formulation
The fluctuating plume model formulation presented is essentially the one proposed by Yee et
al. (1994) and Yee and Wilson (2000). The first (Yee et al., 1994) is a one dimensional model
and has been used for predictions in a boundary layer in neutral conditions, for a fixed
distance from the ground. The second (Yee and Wilson, 2000) has a slightly different
formulation and applies to two-dimensional and isotropic dispersion. It was validated for
dispersion in grid generated turbulence.
The instantaneous concentration c(x,y,t) is a random variable and can be completely
characterized by its PDF f(,x,y)

( ) * ( ) + ( )
. (1)

Since it is assumed that the two mechanisms responsible of the scalar dispersion are
statistically independent, the PDF of the concentration in each transverse cross section in the
dispersing plume is obtained from the convolution of two PDFs: f
c
(x,y
c
) and f
r
(,x,y
r
)

( )
, (2)

where f
c
(x,y
c
) is the PDF of the instantaneous plume centroid location, characterizing the
large scale random crosswind displacements of the mass centre (meandering), while f
r
(,x,y
r
)
is the PDF of the concentration in relative coordinates, meaning the random instantaneous
concentration profile in the moving frame attached to the centroid position (relative
diffusion). The functional form of the two PDFs have to be defined a priori. It is generally
assumed that f
c
is well approximated by a Gaussian distribution, considering that the plume
meandering takes place only in the horizontal direction,

()
(
()
) (3)

where
y,c
is the standard deviation (spread) of the plume centroid position in the y-direction.
The functional form of f
r
differs according to the authors: log-normal distribution (Csandy,
1973), exponential distribution (Sawford, 1987), a combination of exponential and
329
generalized Pareto distribution (Lewis and Chatwin, 1995), clipped Gaussian distribution
(Lewellen and Sykes, 1986, Mylne and Mason, 1991, Reynolds, 2007). Following Yee et al.
(1994) and Yee and Wilson (2000) we assume here a Gamma distribution

()
(
) (4)

where is the gamma function, C
r
denotes the mean concentration on a transverse plane of
the dispersing plume, measured relative to the centroid location at a downwind distance x. C
r

is assumed to be Gaussian with zero mean and standard deviation (spread)
y,r
(x) having C
r,0

as its centreline value

( )
() (
()
) . (5)

The parameter k is equal to the inverse of the fluctuation intensity in relative coordinates, i
r
2
.
It is assumed that k depends on x only. As a consequence k is constant in the transverse
planes, meaning that we are assuming isotropy along the y-direction. Resolving Equation 2
with the mentioned assumptions leads to an analytical expressions for the concentration PDF
f(;x,y) and therefore to the concentration shape parameters, namely, the fluctuation intensity,
i
2
<c
2
>/C
2
, skewness, S<c
3
>/<c
2
>
3/2
, and kurtosis, K<c
4
>/<c
2
>
2
. All these quantities
are therefore dependent on two parameters, which admit a physical interpretation. The first
one is the meandering ratio M=
y,c
/
y,r
, defined as the ratio between the plume meander
variance and the instantaneous plume width variance, which is related to the larger scale
fluctuations. The second is the previously defined k=1/i
r
2
, which is related to the effect of the
smaller scale fluctuations. In the 1D model of Yee et al. (1994) and isotropic model of Yee
and Wilson (2000) these parameters are assumed to be dependent on the distance from the
source only, which implies constant values along the y and z axes. In the next paragraphs we
investigate the spatial evolution of M and k in order to check this hypothesis in a boundary
layer flow and eventually find empirical relations describing their variability in the y-z
planes. The aim is verify the reliability of these models for predictions in boundary layer
flows which may be useful for operational purposes.
3 EXPERIMENTAL SET-UP AND MEASUREMENT TECHNIQUES
The experiments were performed in the atmospheric wind tunnel of the Laboratoire de
Mcanique des Fluides et dAcoustique de lEcole Centrale de Lyon, France. This is a
recirculating wind tunnel with a test section 14 m long, 2.5 m high and 3.7 m wide. A
neutrally stratified boundary layer was generated by combining the effect of a row of spires at
the beginning of the test section and floor roughness elements (Irwin 1981). The spires were
500 mm high and the floor roughness consisted of cubes of side H=20 mm, distributed in
rows along the entire working section (Fig. 1). This experimental set-up allowed us to
reproduce a boundary-layer whose depth was approximately 0.8 m. The reference free-
stream velocity U
at the boundary-layer height was set at 5 m s

-1
. The Reynolds number Re
= U
/ 2.610
5
is sufficiently high to ensure the adequate simulation of a fully rough
turbulent flow (Jimenez, 2004). The flow dynamics above the obstacle array were
investigated by measuring vertical profiles of mean and fluctuating wind velocities with hot
wire anemometry, using an X-wire probe with a sampling frequency of 5000 Hz.
Dispersion phenomena above the array have been studied by injecting ethane (C
2
H
6
), a
passive tracer neutrally buoyant in air, from a point source placed at height 2.5H and at a
distance 7.5 from the beginning of the test section. This position is taken as the origin for the
330
reference system of concentration measurements (XYZ). Concentrations were measured by a
Flame Ionization Detector (FID). At the source, the injected ethane flow had the same
velocity of the surrounding flow. Calibration was performed by subjecting the FID probe to a
series of known and controlled concentrations over an appreciable range.

Figure 1: experimental set-up in the wind tunnel. XYZ: reference system of the concentration measures (x
origin of the source), XYZ: reference system of velocity mesures (x origin at the entrance of the test section).
For each measurement point 300 s time series were acquired, sampled at a frequency of 1000
Hz. This averaging time allows the convergence to be achieved for all the moments of
concentration, thus ensuring reliable statistics. Different statistics were derived from the
measurements, namely the mean, variance, third and fourth moments of concentration. The
experimental results constitute a data-set for the validation of dispersion models that aim in
estimating the higher order concentration moments.
4 RESULTS
4.1 Flow field
To characterise the dynamics of the velocity field we recorded velocity profiles starting from
a distance of 6.25 (equal to 5 m) downstream the entrance of the test section. At this
distance we assume that the development of coherent structures in the wake of the vortex
generators has already reached an equilibrium condition and the dynamics of the flow depend
only on the scales imposed at the wall and on the boundary-layer depth (Salizzoni et al.,
2008). Figure 2 shows the vertical profiles of the mean stream-wise velocity U, standard
deviation
u
,
v
,
w
of the three velocity components and Reynolds stress <uw>. All
quantities are plotted in normalized form. Overall agreement is shown with data from
Raupach et al. (1991), Fackrell and Robins (1982) and Garbero et al. (2010) from flows over
different roughness. The profiles measured at different downstream positions do not differ
significantly one from the others. In this sense we can assert that, in this region, the flow can
be assumed to be homogeneous in the horizontal planes. According to similarity theory
(Tennekes and Lumley, 1972) the mean velocity profile is well modelled by a logarithmic
law

()
) (6)

where k is the Von Karman constant, u* = 0.185 m s
-1
is the friction velocity, z
0
= 1.1410
-4

m is the roughness length and d = 0.0129 m is the displacement height.
331

Figure 2: Normalized velocity profiles. a) mean velocity; b)
u
, c)
v
, d)
w
, e) <uw>. Distances are given from
the entrance of the wind tunnel.
4.2 Specification of the model parameters
The model parameters M and k have been computed by means of the experimental data of
concentration statistics. The meandering ratio M is related to the spread
ny
of the spatial
distribution of the nth moment of the concentration PDF (for n=2,3,4) and to the lateral
spread
y
of the mean concentration, by means of the relation (Yee et al.,1994; Yee and
Wilson, 2000)

. (7)

The values of
ny
and
y
are inferred by fitting a Gaussian law to the measured crosswind
profiles of concentration moments about zero, for n=2,3,4.
The parameter k is chosen so that the model value of i
2
(Eq. 8) matches the measured one at
the mean-plume centreline (with C/C
0
=1, where C is the mean concentration and C
0
is its
centreline value):

()
(
)
(
)
. (8)
4.2.1 Spatial variations of the model parameters
Firstly, we focus on the stream-wise evolution of M and k, computed at a fixed distance z
from the ground and by means of the transversal profiles of the first four moments of the
concentration PDF (n=1,2,3,4). The parameter M (Fig. 3a) decreases along the stream-wise
axis x, showing how the meandering motion dampens moving away from the source. The
variability of the values of M computed by means of different nth-moments of the PDF is
high close to the source and decreases away from it attaining a value of 1-2%. The parameter
k increases with the distance from the source, meaning that the internal fluctuations fade
when the plume becomes more diluted (Fig. 3b). The variability of k with n is of the order of
2%. As a second step we analyse the dependence of M and k on the height z with the aim of
studying their variation with the distance from the ground. To that purpose, we will retain
only M
2
and k
2
, the values inferred from the transversal profile of the concentration variance.
332

Figure 3: The evolution of the parameters M and k. a), b) Variability along the stream-wise axis x. Losangles
represent M
2
, k
2
, viz. the parameters inferred from the spread
2y
of the variance profile (for n=2). Error bars
give a measure of the variability with n. c), d) Evolution of M
2
and k
2
along the vertical direction for two
distances from the source: 0,5 m (triangles); 2 m (losangles).
The results, which are shown in Figure 3c and 3d, indicate that the trend in the z-direction,
for both M and k, is far from being constant. Furthermore, the variability in the z-direction for
each parameter depends on the crosswind position. M grows faster close to the source
(x=0,5m) compared to the slower increase of its z-profile measured at a distance of 2 m. k,
instead, slightly decreases at 0,5 m and shows a drop at 2 m from the source.
The parameter k was calculated experimentally (k
exp
) from the profiles of turbulent intensity
obtaining crosswind profiles of k
exp
at various distances from the source. As revealed from
Figure 4, the transverse profile of k
exp
at 1 m from the source is fairly constant. The same
behaviour is observed closer to the source, meaning that k can actually be considered constant
along the transverse direction at a given crosswind distance, near the source. Starting from
x=2 m, the variability grows and k cannot realistically be considered constant anymore.

Figure 4: Distribution of k=1/i
r
2
in the transverse profile at x=1 m from the source.
This first analysis shows that the model parameters vary significantly with the distance from
the ground, and that this dependence changes with increasing stream-wise distances. This
feature makes difficult the adoption of simple functions defined empirically to model this
dependence.
4.2 Comparisons with experimental results
Finally we compare the predictions of the model from Yee et al. (1994) with the measured
crosswind profiles of fluctuation intensity (i
2
), skewness (S) and kurtosis (K). In these
comparisons we consider only concentration profiles measured at the source height
(z=50mm). In doing that, we will adopt the values of M and k related to the plume centreline
and estimated with the 2
nd
moment transversal profiles. Results are displayed in Fig. 5 a1, b1,
c1. The relative error (ER=(exp.value modelled value)/exp.value*100) calculated in the
transverse profile is shown for each shape parameter in a2, b2, c2. An overall agreement can
be observed between the model and the observations, yet the relative error grows increasing
the moment order, reaching the 30% in the kurtosis profile. Figure 5 (a3, b3, c3) shows the
mean and maximum ER along the stream-wise direction. Speaking about i
2
and S, the mean
ER in the near field (viz. from 0.25 to 2 m from the source) is 5 and 10% respectively, but it
333
rises starting from 2 m from the source. In particular, the maximum ER grows in the far field
differing significantly from the experimental measurements. An opposite behaviour is
observed for the kurtosis K, for which the greater errors are located close to the source. This
augmentation of the relative error can be explained with the emergence of off-centreline
peaks in the moments profiles, as show in Figure 6 for a distance of 3 m from the source. As
explained by Yee and Wilson (2000), the exact quantitative conditions required for the
emergence of off-centreline peaks have not been elucidated. Nevertheless this phenomenon is
present in our profiles starting from a distance of 2 m from the source. Comparing the
expected profiles and the measurements, we observe that the fluctuating plume model fails in
the prediction of the size of the off-centreline peaks but is able to qualitatively forecast their
position as regards to the centreline.

Figure 5: profiles and Relative Error (ER) for a) fluctuation intensity, b) skewness and c) kurtosis. For each
shape parameter: 1) comparison between the measured and modelled values in a crosswind profile at a distance
of 0.5 m from the source; 2) ER along the crosswind profile at x=0.5 m; 3) mean and max ER along the stream-
wise axis x, referring to the value at the plume centreline.

Figure 7: transverse profiles of the central moments ( a) second moment, b) third, c) fourth) at a distance x=3 m
from the source.
334
5 CONCLUSIONS
We have tested the meandering plume model with internal fluctuations (Yee et al., 1994; Yee
and Wilson, 2000) with experimental results of pollutant dispersion within a turbulent
boundary layer. To that purpose we have preliminary estimated the model parameters M and
k within the domain from the experimental results, and evaluated their spatial variability.
This analysis showed that, for a pollutant plume dispersion within a turbulent boundary layer,
both parameters are highly dependent on the vertical coordinate. However, considering
concentration at a fixed distance from the ground only, comparisons between model
predictions and experimental results show overall good agreement in the near field.
We therefore conclude that the model can be applied for operational purposes to predict
pollutant concentration moments at a fixed distance from the ground. This may be the case
for example of ground level concentrations. On the other hand, if the purpose is to achieve a
complete description of the concentrations field, the model should be reformulated to
consider the variability of the fundamental parameters in the vertical direction.
6 REFERENCES
Csanady, G.T., 1973. Turbulent diffusion in the environment. D. Reidel, Dordrecht, The Netherlands.
Fackrell, J. E., Robins, A. G., 1982. Concentration fluctuations and fluxes in plumes from point sources in a
turbulent boundary layer. J. Fluid Mech., 117, 1-26.
Garbero V., Salizzoni, P., Soulhac, L., 2010. Experimental study of pollutant dispersion within a network of
streets. Boundary-Layer Meteorol., 136, 457-487.
Gifford, F.A., 1959. Statistical properties of a fluctuating plume dispersion model, Adv. Geophys., 6, 117-137.
Irwin, HPAH, 1981. The design of spires for wind simulation. J Wind End Ind Aerodyn, 7, 361-366.
Jimenez, J., 2004. Turbulent flows over rough wall. Ann. Rev. Fluid Mech. 36, 173-196.
Lewellen, W. S., Sykes, R. I., 1986. Analysis of concentration fluctuations from lidar observations of
atmospheric plumes. J. Clim. Appl. Meteorol., 25, 1145-1154.
Lewis, D. M., Chatwin, P. C., 1995. A new model probability density function for contaminants dispersing in
the atmosphere. Environmetrics, 6, 583, 593.
Mylne, K. R., Mason, P. J., 1991. Concentration fluctuations measurements in a dispersing plume at a range up
to 1000 m. Q. J. R. Meteorol. Soc., 117, 177-206.
Pope S.B., 2000. Turbulent flows. Cambridge university Press, Cambridge, UK.
Raupach, M. R., Antonia, R. A., Rajoplan, S., 1991. Rough-wall turbulent boundary layers. Appl. Mech. Rev.,
44-1, 1-25.
Sawford, B.L., Stapountzis, H., 1986. Concentration fluctuations according to fluctuating plume modes in one
and two dimensions. Boundary-Layer Meteorol., 37, 89-105.
Salizzoni, P., Soulhac, L., Mejean, P., Perkins, R., 2008. Influence of a two-scale surface roughness on a neutral
turbulent boundary layer. Boundary-Layer Meteorol.,127(1), 97-110.
Tennekes, H., Lumley, J. L., 1972. A first course in turbulence. MIT press, Cambridge.
Yee, E., Chan, R., Kosteniuk, P. R., Chandler, G. M., Biltoft, C. A., Bowers, J. F., 1994. Incorporation of
internal fluctuations in a meandering plume model of concentration fluctuations. Boundary-Layer Meteorol.,
67, 11-39.
Yee, E., Wilson, D. J., 2000. A comparison of the detailed structure in dispersing tracer plumes measured in
grid-generated turbulence with a meandering plume model incorporating internal fluctuations. Boundary-
Layer Meteorol., 94, 253-296.

335
Study of urban effects on the dispersion process in a
wind tunnel
Adrin Roberto Wittwer
a
, Acir Mrcio Loredo-Souza
b
, Edith Beatriz Camao
b

a
Universidad Nacional del Nordeste, Resistencia, Argentina,
a_wittwer@yahoo.es
b
Universidade Federal de Rio Grande do Sul, Porto Alegre, Brazil.
ABSTRACT: A single emission source representing the conditions at a low height chimney
was modeled. The atmospheric turbulent flow was physically simulated using roughness
elements and mixture devices, and a neutrally stable boundary layer was obtained in the wind
tunnel of the Universidade Federal de Rio Grande do Sul (UFRGS). An urban environment
was modeled to evaluate the effects of a building group on the plume emission.
Complementary, the effects of simple building model leeward to the emission was analyzed.
The analysis of the dispersion process and concentrations was performed using a fast-
response, aspirated concentration sensor based on a hot-wire anemometer. The mean and
fluctuating components of the plume concentration were obtained. These preliminary results
allowed a partial characterization of the concentration fields.
1 INTRODUCTION
Studies of dispersion process and analysis of pollutant concentration levels discharged in the
atmosphere are required by environmental control agencies. Experimental data must be
obtained to validate the recently developed numerical models related with this type of
problem. The use of laboratory is very usual by the expensive cost of field experimental
works. Wind tunnel tests could be used to analyze dispersion process; however, the flow of
atmospheric boundary layer and the characteristics of the plume emission must be modeled
adequately (Isymov & Tanaka, 1980).
Cermak & Takeda (1982) proposed different similarity numbers like an alternative to the
exact simulation of the emission process. On other hand, similarity criteria may be modified
according to the plume region to be modeled. According to Robins (2003), different aspects
are studied by wind tunnel dispersion modeling to improve the understanding of phenomena
and to solve specific problems. In this work, basic dispersion processes near the emission,
urban environment effects and concentration field fluctuations are analyzed.
A single emission source was modeled and three situations of local dispersion were analyzed.
Roughness elements and mixture devices were used to reproduce the atmospheric turbulent
flow. A neutrally stable boundary layer was obtained in the wind tunnel of the Universidade
Federal de Rio Grande do Sul (UFRGS). The isolated source model in a homogeneous terrain
was firstly analyzed. Then, an urban environment was modeled to evaluate the effects of a
building group on the plume emission. Finally, the case of a simple building located leeward
to the emission was analyzed.
The analysis of the dispersion process and concentrations was performed using a fast-
response, aspirated concentration sensor based on a hot-wire anemometer. The mean and
fluctuating components of the plume concentration were obtained and these results allowed
336
the characterization of the concentration fields. The intermittency of the concentration field is
also shown.
2 EXPERIMENTAL TECHNIQUES
These tests were performed at Prof. Joaquim Blessmann boundary layer wind tunnel of
UFRGS (Blessmann, 1982). Roughness elements mixture devices were used to reproduce a
neutrally stable boundary layer. Low velocities are used in these experiments for similarity
criteria. A power law velocity profile with exponent 0.23 was fit to the experimental values
and the corresponding values of turbulence intensity were also obtained (Wittwer et
al., 2007). The reference velocity U
ref
is approximately 1.91 m/s at low speed. The
corresponding values of turbulence intensity and low velocity spectra were obtained.
Introduction of urban neighborhood modify strongly the velocity profile characteristics. The
power law exponent of mean velocity profile is greater than 0.45 and the turbulence
intensities takes values upper than 50 % at height z below 130 mm.
The emission source model was built with a circular tube of 20 mm diameter. Pure helium
was used to simulate the emission. To evaluate urban effects three cases were considered; the
isolated chimney in a homogeneous terrain, the same emission source with a non-
homogeneous urban environment defined by several prismatic buildings, and the case of a
simple building located leeward to the emission (Figure 1). Details of the urban environment
and the simple building near the source emission are shown in Figure 2.

Figure 1: Isolated source model, urban environment model and simple building model leeward to the emission.

Figure 2: Urban environment and simple building details.
337
The adopted non-dimensional parameters used in the plume characterization are the ratio
between the emission velocity w
0
and the approaching flow velocity U
ref
, the emission
momentum and the buoyancy parameter defined by equations (1) and (2) . The values used in
these experiments are indicated in Table 1. The values of the velocities w
0
and U
ref

correspond to the scale model values and D
0
is the emission source internal diameter.
2 2
0 0
/
ref a
U w (1)
3
0 0 0 0
/ ] ) [(
ref a
U D gw (2)
Table 2: Emission characteristics parameters.
w
0
[m/s] U
ref
[m/s] w
0
/U
ref
r
0
w
0
2
/ r
0
U
ref

2
[(r
0
-r
a
)gw
0
D
0
]/ r
0
U
ref

3

1.26 1.91 0.66 0.060 -0.031

Concentration values were measured leeward from the emission source. Measurements were
realized using an aspirating probe connected to a hot-wire anemometer (Halstead & Wood,
1982, Wilson, D. J., Netterville, 1981). This probe is composed by the hot wire and a 0.3 mm
internal diameter tube aspirated by a vacuum pump. The system allows the measurement of
fluctuating concentrations (Harion et al., 1996). One minute samples were digitized in each
measurement point using an acquisition frequency of 1024 Hz.
3 RESULTS
The results obtained in the tests are presented as concentration coefficient K and intensity of
the concentration fluctuations I
c
profiles.
0
2
/ Q H CU K
ref
= (3)
max
/ C c I
c
= (4)
where C and c are the mean concentration and the standard deviation of the fluctuations,
respectively, Q
0
is the flow emission, and U
ref
is the wind velocity corresponding to the
emission height. The vertical coordinate in these profiles is measured from the wind tunnel
floor.
Concentrations vertical profiles were measured at several distances x/H from the emission.
The concentration mean and rms values were obtained for each point. Figure 3 presents the
vertical profiles of the concentration coefficient K and the I
C
for isolated emission related to
the positions x/H = 0.60, 1.20 and 1.80. Due to buoyancy effects, the profiles present an
asymmetry, tending to deflect the plume upwards.
Profiles of K and I
C
at positions x/H = 0.47, 0.95 and 1.43 for the source in the urban
environment are shown in Figure 4. The greater turbulence produces the increasing in the
vertical dispersion parameter
z
. In this case, the figure indicate a decline of the plume at
position x/H = 1.43. In this leeward position, the rms values of the concentration fluctuations
are much lower although this is not clearly shown in the figure where the fluctuation local
intensity profiles are indicated.
The profiles corresponding to simple building model leeward to the emission are indicated in
Figure 5. The position is x/H = 0,54, and the ratio of the heights chimney/building is 1,00 (a)
338
and 0,93 (b), respectively. The inertial effects diminishing the plume effective height and
therefore increases the concentration level in the lower part (close to the roof). The intensities
of the concentration fluctuations have a great drop in this region of the profile.
Finally, Table 2 indicates maximum values of concentration coefficient K. The smaller
absolute values of K obtained in the urban environment case with respect to isolated emission
indicate a greater dilution effect in the concentrations caused by flow conditions imposed by
the presence of building models. A new evaluation would be realized with respect to the
influence of parameters used to obtain dimensionless concentrations and comparison of
results for building model leeward to the emission and isolated source.

0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0.00 0.50 1.00 1.50
K/K
max
z
/
H
x/H=0,60
x/H=1,20
x/H=1,80

0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0.00 0.50 1.00 1.50
Ic=c/C
max
z
/
H
x/H=0,60
x/H=1,20
y/H=1,80

Figure 3: Concentration profiles for isolated emission model.

339
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0.00 0.50 1.00 1.50
K/K
max
z
/
H
x/H=0,47
x/H=0,95
x/H=1,43

0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0.00 0.50 1.00 1.50
Ic=c/C
max
z
/
H
x/H=0,47
x/H=0,95
y/H=1,43

Figure 4: Concentration profiles for the emission in urban environment.
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0.00 0.50 1.00 1.50
K/K
max
z
/
H
x/H=0,54 (a)
x/H=0,54 (b)

0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
0.00 0.50 1.00 1.50
Ic=c/C
max
z
/
H
x/H=0,54 (a)
x/H=0,54 (b)

Figure 5: Simple building model leeward to the emission.
340
Table 2: K
mx
values.
Position (x/H) 0,60 1,20 1,80
Isolated emission 7,18 1,87 1,21
Position (x/H) 0,47 0,95 1,43
Urban environment 2,45 0,78 0,49
Position (x/H) 0,54 (a) 0,54 (b) -
Leeward building 4,78 4,69 -

Digitized samples of the concentration fluctuations as a function of time are shown in Figures
6 and 7. The time series obtained near the plume centre (z/H = 1.12) is indicated in Figure 6
and the corresponding series obtained near the plume upper edge (z/H = 1.26) is indicated in
Figure 7. These series were measured for isolated emission at the downstream position x/H =
0.60. Lower intermittency is observed near the centre plume and the peak concentration
values are about are about 1.75 10
5
ppm. Near the plume edge, the concentration field is
highly intermittent with concentration peaks of 1 10
5
in ppm separated by intervals of zero
concentration. The intermittency of a process could be analyzed using probability
distributions (Cheng & Melbourne, 2000).
0.00E+00
5.00E+04
1.00E+05
1.50E+05
2.00E+05
0 5 10 15 20 25 30
t [s]
C

[
p
p
m
]
z/H=1.12

Figure 6: Digitized sample of concentration fluctuations at plume centre (without intermitency).
0.00E+00
4.00E+04
8.00E+04
1.20E+05
1.60E+05
0 5 10 15 20 25 30
t [s]
C

[
p
p
m
]
z/H=1.26

Figure 7: Digitized sample of concentration fluctuations at plume upper edge (high intermitency)..

341
4 CONCLUSIONS
The urban effects on the concentration field of a plume emission were analyzed using a
reduced scale model in a boundary layer wind tunnel. A greater dilution effect caused by the
presence of the urban environment was detected. The fluctuation intensity of concentration
also indicates minor values than the isolated emission case. An increasing of the
concentration level in the lower part close to the building roof was observed in the case of the
simple building model leeward to the emission.
From the partial analysis of concentration fluctuations profiles, the general behavior is similar
to that obtained by other authors. An interesting urban effect is the plume fall at downstream
positions, and consequently greater concentration levels at lower positions.
Finally, the intermittent behavior of the concentration field at different dispersion plume
regions was shown only in preliminary form. The intermittency will be analyzed using
cumulative probability distributions. Comparison between different cases and regions of the
dispersion plume must be realized.
5 REFERENCES
Blessmann, J., 1982. The Boundary Layer Wind Tunnel of UFRGS, J. Wind Eng. Ind. Aerodyn. 10, 231-248.

Cermak, J. E., Takeda, K., 1985. Physical modeling of urban air-pollutant transport, Journal of Wind
Engineering and Industrial Aerodynamics, 21, 51-67.

Cheung, J. C. K., Melbourne, W. H., 2000. Probability distribution of dispersion from a model plume in
turbulent wind, J. Wind Eng. Ind. Aerodyn. 87, 271-285.

Halstead, A.GJ., Wood, C. J., 1982. The calibration of an aspirating conductivity probe for thermal pollution
experiments, J. Wind Eng. Ind. Aerodyn. 9, 237-249.

Harion, J., Favre-Marinet, M., Schettini, E. B. C., 1996. An Improved Method for Measuring Velocity and
Concentration by Thermo-Anemometry in Helium-Air Mixtures. Experiments in Fluids, v. 22, 174-182.

Isymov, N., Tanaka, H., 1980. Wind tunnel modelling of stack gas dispersion Difficulties and aproximations,
in: Proceedings of the fifth International Conference in Wind Engineering,, Fort Collins, Colorado, USA, Ed.
by J. E. Cermak, Pergamon Press Ltd..

Wilson, D. J., Netterville, D. D. J., 1981. A fast response, heated-element concentration detector for wind-tunnel
applications, J. Wind Eng. Ind. Aerodyn. 7, 55-64.

Wittwer, A., De Paoli, F., Camano Schettini, E. B., Loredo-Souza, A. M., 2007. Wind tunnel study of the
concentration fields in a plume emission, in: Proceedings of International Workshop on Physical Modelling
of Flow and Dispersion Phenomena (PHYSMOD 2007), Orlans, France, 2007.

342
Investigation on Characteristics of Building Wake
Dispersion for the Nearby Source by using Wind
Tunnel Experiment
Jiro Yoneda
a
, Kazuki Okabayashi
a
, Eiichi Hori
b
, Tomohiko Inui
c
, Kensuke
Yoshihara
c

a
Fluid Dynamics Laboratory, Nagasaki Research and Development
Center/Mitsubishi Heavy Industries, Ltd., NAGASAKI, JAPAN, E-mail:
jiro_yoneda@mhi.co.jp
b
Radiation Safety Engineering Section, Nuclear Energy Systems/Mitsubishi
Heavy Industries, Ltd., HYOUGO, JAPAN, E-mail: eiichi_hori@mhi.co.jp

c
Nuclear Power division/The Kansai Electric Power Co., Inc., FUKUI,
JAPAN, E-mail: inui.tomohiko@b5.kepco.co.jp
ABSTRACT: For environmental assessment, analytical model like Gaussian plume model
such as ISC-PRIME (EPA, 1997) is usually applied to predict the concentration of the gasses
released from stacks like power plant. However, it is difficult to predict concentration behind
buildings by using the analytical model because the dispersion phenomenon is very complex
behind the buildings. In this study, we suggested to modify the configuration coefficient of
the Gaussian plume model which was the simplest proposed by Slade (1968) in order to
exactly predict the concentration behind building by using the wind tunnel experiment. The
model derived from wind tunnel experiment is applied for near buildings (i.e. X/H < 3, X is
the horizontal distance from release point and H is the height of building.)
1 INTRODUCTION
Some analytical models such as Gaussian plume model are generally used for environmental
assessment to predict the concentration of the gasses released from stack like power plant. It
is usually applied for the evaluation points where are outside of plants and far from stack. It
needs to predict concentration of the near stack area like behind buildings for workers, who
are stayed inside the plant site particularly when some gasses are released from short stacks.
One of the reasons why analytical models are used for environmental assessment is simplicity
for handling. However, it is difficult to predict the concentration behind buildings for
analytical model because the dispersion phenomenon is very complex behind the buildings.
For example, ISCST3 (EPA, 1995) can calculate only beyond three building heights
downwind and ISC-PRIME express building effect as cavity (i.e. constant concentration) at
the region where close to building. The concentrations near release point are very large and a
significant impact for environmental assessment. Therefore the exact and conservative
prediction is needed near release point for our purpose.
In order to apply the Gaussian plume model to the above-mentioned dispersion phenomenon,
the configuration coefficient c to consider the effect of buildings, which was proposed by
Slade(1968), was introduced again in this study.
Then we tried to apply the Gaussian plume model, which is modified by using the wind
tunnel experiment with the representative plant shape models, to building wake diffusion.
343
That means the modification of the configuration coefficient c for prediction of the
concentration behind a building (i.e. X/H < 3).
The method of the wind tunnel experiment is shown in Section 2, the result of the wind
tunnel experiment is shown in Section 3 and in Section 4 the development of modeling is
proposed.
2 WIND TUNNEL EXPERIMENT
For the wind tunnel experiments in the current study, the wind tunnel in the Mitsubishi
Heavy Industries, Ltd. (MHI) facility was used (Fig.1). The test section of our blow down
wind tunnel is 3(W)*2(H)*25(L) m. The wind speed range is from 0.2 to 15 m/s. The
simplified model was scaled by 1/200. The model used in the experiments was a simplified
model of 2 buildings as shown in Table 1. The release points were located on the top of
building A as a stack (elevated release) and on the roof of the building B (low-level release).
By using ammonia (NH
3
) as a tracer gas, the concentration was measured on the building
surface (wall and roof) on the buildings.
In evaluating the concentration of 1 hour average should be considered because the 1 hour
averaged concentration is normally used for the environmental assessment. However the
plume width indicated in Pasquill-Gifford (P-G) chart that corresponds to averaging time
from a few minutes to ten minutes. Therefore, the plume width was corrected to equivalent to
the value at 1 hour averaged horizontal plume width by 1/5 power law (Hanna et al, 1977,
MHLW, 1986).
To predict the value at 1 hour average by using the wind tunnel experiment, the turn table
equipment of our wind tunnel as shown in Fig.1 was rotated depend on the probability
density of wind direction (it is called overlapping method). The detailed method is described
in Ide (1994, 1988) and Okabayashi (1991, 1996).

The experimental cases were tabled in Table 1 and Fig 2 shows the locations of release point
and wind directions.

Fig 1 MHI large scale diffusion wind tunnel

25m
3m
10m
Wind
6.75m
6m
344
Table 1 Experimental case
Building type Source
H
A
(m)
D
A
(m)
H
B
(m)
W
B
(m)
L
B
(m)
Standard
deviation of
wind
directional
fluctuations

(degrees)
Wind
direction
a, b, c, c,
e, f
80 50 30 100 132 8 I
a, b 80 50 30 100 132 3 I
b 80 50 30 100 132 15 I
Standard

a, b 80 50 30 100 132 8 II
Tall
building
A

a, b 90 50 30 100 132 8 I
Tall
building
B

a, b 90 50 65 100 132 8 I
Without
building
B

a 50 50 - - - 8 I
Wide
building
B

b 80 50 30 180 132 8 I
Without
building
A
b - - 30 100 132 8 I
H
A
, D
A
: Height and Diameter of building A.
H
B
, W
B
, L
B
: Height, Width, Length of building B.

Fig 2 Locations of release points and wind directions.
3 RESULT OF WIND TUNNEL EXPERIMENT
From the result of the wind tunnel experiment, the maximum standardized concentrations
dependent on the downwind distance from the release point were plotted as Fig 3 (elevated
50m
50m
66m 66m
25m
50m
100m
132m
41m
16.7m
5m
12.5m
12.5m
25m
a
b
c d
e f
Wind direction I
(0 degrees)
Wind direction II
(45 degrees)
90m
30m
35m
25m
132m
a
b,c d,e f
Building B
Building A
Building A
Building B
a : Elevated release
b, c, d, e, f : Low-level release
Wind
345
release) and Fig 4 (low-level release).The concentration measured in the wind tunnel
experiment C (m
3
/m
3
) was standardized by wind speed U (m/s), release rate Q (m
3
/s) and
building height H
A
or H
B
(m). The downwind distance from release point X (m) was also
standardized by H
A
or H
B
. Fig 3 and Fig 4 also contained the calculation result of ISC-
PRIME. The calculations of ISC-PRIME also included the buildings same as wind tunnel
experiment, however ISC-PRIME couldnt be strictly applied to the cylinder or hemisphere
shape like the building A. Then the building A was described as rectangular in ISC-PRIME
calculations.
As shown Fig 3(a) the concentration of wind tunnel were varied by each building type and
the concentrations decreased according to X/H. On the other hand, the concentration of ISC-
PRIME were constant behind the building A for each case. Except for the hemisphere shape
of the building A (tall building B), the maximum concentration of ISC-PRIME was mostly in
agreement with the experimental one. However, the relation of the concentration value
between ISC-PRIME and the experiment was changed depend on X/H.
Similar to Fig 3(a), Fig 4(a), (b) and (c) show the comparison between wind tunnel and ISC-
PRIME for low level release. When the building effect was considered in calculation of ISC-
PRIME, the concentrations decreased. As shown in Fig 4(b), even the results of the wind
tunnel were similar with each other, the results of source c and d of the ISC-PRIME were
smaller than source b. In addition, From Fig 4 (c), the concentrations of wind tunnel were
different according to building types, although the characters of concentration were similar.
(i.e. concentrations decrease according to distance). However the characters of concentration
of ISC-PRIME were totally different for without building A case and partially different for
tall building B case. It means that the ISC-PRIME couldnt sufficiently express the building
effect in detail for our purpose.
The concentrations of
= 3 degrees were a little larger than the case of
= 8 degrees (Fig
3(b) and Fig 4(d)). As shown in Fig 3(b), the effect of meandering was appeared gradually in
the downwind distance of more than X/H = 1.4, because the building effect was dominant for
dispersion phenomena just only behind the building. However the effect of the meandering
was small in comparison with the effect of the building shape (or size). On the other hand, the
effect of wind direction was little for both sources.

. Fig 3 Result of elevated release (Source a)
(Maximum concentration value of each downwind distance X/H on the roof of building B)
a) Variation of building types b) Variation of wind direction and meandering
H
A
2
U
C
/
Q
*
1
0
6

Building type Source Wind tunnel ISC-PRIME
Standard a

Tall building A a

Tall building B a

Without building B a

Building
type
Source
Wind
direction

Wind
tunnel
a I 8

a I 3

Standard
a II 8

H
A
2
U
C
/
Q
*
1
0
6

X/H
A
X/H
A

0.0E+00
2.0E+05
4.0E+05
6.0E+05
8.0E+05
1.0E+06
1.2E+06
0 1 2 3
0.0E+00
2.0E+05
4.0E+05
6.0E+05
8.0E+05
1.0E+06
1.2E+06
0 1 2 3
346

Fig 4 Result of Low-level release (Source b through f)
(Maximum concentration value of each downwind distance X/H on the roof of building B)

Fig 5 Distribution of concentration (Source a with tall building A case)
(The low concentrations around 0 are not exactly because of interpolation for drawing)

Fig 5(a) shows the shaded contour of concentration as an example. Although the value of
the concentration was changed according to the change of the building shape, the distribution
of the concentration was similar to all elevated release cases. From the image analysis of
smoke visualization (Fig 5(b)), the plume axis which means high concentration dropped
behind the building A by the effect of the building.
Wind
tunnel
ISC-PRIME
e

Standard
f

Building
type
Source
Wind
tunnel
ISC-PRIME
b

c

Standard
d

Building type Source Wind tunnel ISC-PRIME
Standard b

Without building A b

Tall building B b

Wide building B b

Building
type
Source
Wind
direction

Wind
tunnel
b I 3

b I 8

b I 15

Standard
b II 8

a) Variation of release point behind the building B b) Variation of release point from
the side of the building A
c) Variation of building types
d) Variation of wind direction and meandering
H
A
2
U
C
/
Q
*
1
0
6

H
B
2
U
C
/
Q
*
1
0
6

H
B
2
U
C
/
Q
*
1
0
6

H
B
2
U
C
/
Q
*
1
0
6

X/H
A
X/H
B

X/H
B
X/H
B

UC/Q10
6

(1/m
2
)

a) Measured concentration b) Image analysis of smoke visualization

Brightness (%)

1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+08
1.0E+09
-2 -1 0 1 2
1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+08
1.0E+09
-2 0 2 4
1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+08
1.0E+09
-2 0 2 4
1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+08
1.0E+09
-2 0 2 4
347
4 MODELLING OF THE CONFIGURATION COEFFICIENT
For environmental assessment, a detailed prediction of concentration is needed. Then it was
tried to modify the Gaussian plume model (Equation (1)), which is generally used for
environmental assessment in Japan. The building effect is significant for dispersion behind
building. Therefore the dispersion parameters
y
and
z
are small compared with building
effect (cA/ of Equation (1)) and Equation (1) is simplified as Equation (2).

( ) ( )
+
+

=
z z z y
H z H z Q
C
2
exp
2
exp
2
2 2
(1)

C : Concentration (m
3
/m
3
)
Q : Release rate (m
3
/s)
z y
, : Horizontal and vertical dispersion parameter (plume width) with building effect (m)

cA cA
z z y y
+ = + =
2 2

y
,
y
: Horizontal and vertical dispersion parameter (m)
c : Configuration coefficient (-)
A : Cross-sectional area of buildings (m
2
)

z : Receptor height (m)
H : Release height (m)
( )
( ) Q UC H
Q UC A
c
cA
cA cA
Q UC
b
z y
2
2 2
1 1
1 1
=
+ +
=

(2)

b
H : Building height (m)

In this study, Equation (2) is used for the modification of the configuration coefficient c. As
the result of Fig 3 and Fig 4, the concentrations on the roof of the building B was dependant
on the distance from the release point. Then the configuration coefficient c, which defined as
Equation (2), for each case was plotted on Fig 6. The downwind distance X was normalized
by the representative length (i.e. Height of building A, H
A
, for elevated release and height of
building B, H
B
for low-level release.).

Then the case which indicates the smallest configuration coefficient (i.e. the maximum
concentration) on the roof was selected for modeling. The fitting curve of the configuration
coefficient from the maximum concentrations on the roof of the building B was described as
following Equation (3-1) and (3-2).

348

Fig 6 Modeling of configuration coefficient c

For elevated release,
( ) ( ) B building of roof D X H X c : 15 . 2 83 . 0 7037 . 0 / 0373 . 0 + = (3-1)

For low-level release,
( ) ( ) ( ) B building of roof H X H X H X c : 90 . 2 / 32 . 0 011 . 0 / 0523 . 0 0337 . 0
2
+ = (3-2)

In addition, the distributions of the concentration for Y direction (crosswind direction) were
similar for all elevated release cases, because the most affecting building was the building A
of cylindrical shape. Therefore, it was possible for fitting to the distribution curve normalized
by the diameter D (m) of the building A as shown in Fig 7(a). Fig 7(b) shows the /D ( is
normalized with diameter D of building A, such as X/D) and the fitting normalized
distribution curve for each distance. In a similar way, the /D could be approximated on roof
of building B as Equation (4).

For elevated release,
( ) ( ) B building of roof D X D X D : 15 . 2 83 . 0 1638 . 0 1847 . 0 + =
(4)

Fig 7 Distribution of concentration on roof of building B
Y/D
(
U
C
/
Q
)
/
(
m
a
x
i
m
u
m

U
C
/
Q
)

a) Distribution of concentration (/D) at X/D=1.45
/D = 0.1874X/D) + 0.1638
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.0 0.5 1.0 1.5 2.0 2.5
/D Approximated function
X/D
/
D

b) /D depend on X/D
H for
X/H
Wind
tunnel
Standard a H
A

Tall building A a H
A

Tall building B a H
A

Without building B a H
A

Building type Source H for X/H Wind tunnel
b H
B

c H
B

d
H
B

e
H
A

Standard
f H
A

Tall building B b H
B

Wide building B b H
B

Without building A b
H
B

C
o
n
f
i
g
u
r
a
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t

c

X/H
a) Elevated release
b) Low-level release
C
o
n
f
i
g
u
r
a
t
i
o
n

c
o
e
f
f
i
c
i
e
n
t

c

X/H
0.0
1.0
2.0
3.0
4.0
5.0
6.0
0 1 2 3
1.0E-04
1.0E-03
1.0E-02
1.0E-01
1.0E+00
1.0E+01
1.0E+02
-2 0 2 4
H
A
: Height of building A above building B (m)
H
B
: height of building B (m)
0.0
0.2
0.4
0.6
0.8
1.0
-2 0 2
Building type
Wind
tunnel
8

Standard
3

Tall building A 8

Tall building B 8

Without building B
8

: /D=0.42
349
5 CONCLUSION
For environmental assessment, it is needed to predict the detail concentration behind
building by analytical model, although the detailed prediction of concentration behind
building is difficult for existing analytical model such as ISC-PRIME etc. for the purpose of
this study.
Therefore, it was tried to modify the configuration coefficient c of the Gaussian plume
model by using wind tunnel experiment. The building effect depend on the shapes of the
buildings had more significant influence for the dispersion behind the building compared
with the effect of the meandering or wind direction. As the result, it was found that the
configuration coefficient c and the Y directional distributions of the concentration were
modeled by experimental formula of the maximum concentration case which envelops all
other cases.
By using the result of this study, it becomes the detailed prediction of concentrations behind
building where the conventional analytical models treat them as constant concentration. This
modified model can calculate the concentrations of behind building easily and conservatively.
Therefore we could indicate the possibility of application of this modified model for
prediction of the concentration behind building for environmental assessment.
4 ACKNOWLEDGMENT
The current study was achieved with cooperation of Kansai Electric Power Co., Inc.,
Hokkaido Electric Power Co., Inc., Tohoku Electric Power Co., Inc., Tokyo Electric Power
Co., Inc., Chubu Electric Power Co., Inc., Hokuriku Electric Power Co., Inc., Chugoku
Electric Power Co., Inc., Shikoku Electric Power Co., Kyushu Electric Power Co., The Japan
Atomic Power Company, J-POWER/Electric Power Development Co., LTD., Hitachi-GE
Nuclear Energy, Ltd. and Toshiba Corporation Power Systems Company. We express our
sincere thanks for their cooperation.
5 REFERENCES
EPA, 1995. Users guide for the industrial source complex (ISC3) dispersion models.
EPA, 1997. Addendum to ISC3 users guide the prime plume rise and building downwash model.
Hanna S. R., et al., 1977. Meeting review ; AMS work shop on stability classification schemes and sigma
curves.,
Ministry of Health, Labour and Welfare of Japan, 1986. The manual of environmental assessment of garbage
Incineration facility, (in Japanese).
K.Okabayashi.et al, 1996. Effects of wind directional fluctuations on gas diffusion over a model terrain.
Atmospheric Environment, Vol. 30, No 16, pp. 2871-2880.
K.Okabayashi and Y. Ide. et al, 1991. A new wind tunnel technique for investigating gas diffusion behind a
structure, Atmospheric Environment, Vol. 25A, No 7, pp. 1227-1236.
Slade, D. H, 1968. Meteorology and Atomic Energy, TID-24190.
R.Fukuda, et.al., 2009. Evaluation of main control room habitability in Japanese LWR (2) Evaluation for
applicability of existing atmospheric dispersion models to building wake dispersion by using wind tunnel
experiment, ICAPP 2007 proceeding.
Y. Ide, R. Ohba and K. Okabayashi, 1994. Development of Overlapping Modelling for Atmospheric Diffusion,
Atmospheric Environment, Vol. 28, No 11, pp. 1925-1932
Y. Ide. et al. 1988 Simulation of Wind Direction Fluctuation in the Diffusion Wind Tunnel with Overlapping
Technique, J. Japan Soc. Air Pollut, Vol. 23, No 4, pp. 199-208.
350

Pressure measurements of the detachment bubble on the
Bolund Island
Tee Seong Yeow
a
, Alvaro Cuerva
b
, Javier Prez
c

abc
Instituto Ignacio Da Riva, Universidad Politcnica de Madrid, Madrid,
Spain. Corresponding author: teeseongyeow@upm.es

Pressure measurements on the surface of a 1:230 scale model of Bolund Island are presented.
The model is smooth and no boundary layer generation has been considered since the
experiment is designed as the simplest possible reference case. Measurement have been taken
for a range of Reynolds numbers based on the average undisturbed wind speed U
and the
maximum height of the island, h [1.710
4
, 8.510
4
], and for a range of wind directions. Four
minutes time series of pressure in more than 400 points have been acquired and analysed to
obtain the spatial distribution of both the time average and the variance of the pressure signal.
The horizontal extension of the detachment bubble for the different Reynolds numbers and
wind directions is identified by isobars and curves of constant value of pressure variance. The
applicability of this technique for evaluating the horizontal topology of high turbulence
regions associated to detachment bubbles after escarpments in potential wind farm sites is
analysed. The results obtained shows that the behaviour of the mean pressure coefficient, C
p
,
the std. pressure coefficient, C
p
, and the skewness of the pressure, S
p
can be used to study the
bubble over the island to a certain extent. This experiment is part of the set of different
analysis on the Bolund test case that is being undertaken within WAUDIT project by the
different scientific groups.
1 INTRODUCTION
Recently, two new test cases have been proposed for benchmarking of numerical and
physical modelling of complex terrain flows, these are the Alaiz test case (see Conan et al.,
2011) and the Bolund experiment (see Bechmann et al., 2009). The Bolund experiment was
initiated by Ris DTU as a blind comparison of different numerical models (including linear,
RANS and LES simplifications of Navier-Stokes equations). More detailed description of the
blind comparison can be found in Bechmann et al (2009). The main conclusions from the
initial analysis are: a) a great scatter of the numerical results exist (mainly in the vicinity of
the island escarpment, b) the mean velocities are better predicted than turbulent kinetic
energy (TKE), and c) the best models predicting both, mean wind speed and TKE are the
RANS models with two closure equations.
One of the main geometric characteristics of the Bolund island is the escarpment facing
approximately the wind directions 200 to 295 (see figure 3). It can be idealised as a
combination of a 50 ramp extending from the sea level to 0.5h, plus an almost vertical step
from 0.5h to h, being h the total height of the escarpment. The escarpment height varies
slightly in the interval 200-295 being roughly the maximum height of the island (11.73 m).
This geometry guarantees that flow detachment at the edge (with a sufficiently large
Reynolds number) while the flat top ensures reattachment of the flow on the island. This flow
pattern on Bolund island by smoke visualizations in wind tunnel in Bechmann et al., 2009,
and quantified by direct measurement of very high values of TKE on the real field in the met
masts close to the escarpment for heights below 2 m (met masts M2 and M6 for 239 and
270 wind directions respectively).
351

The detached-reattaching flow structures, in Bradshaws wordings, can provoke a weak,
strong or an overwhelming perturbation in the approaching flow (see Bradshaw & Wong,
1972) depending on the relation /h (>>1, O(1), <<1, respectively) being the size of the
incoming shear layer, producing from a slight change to a complete mutation of the original
flow structure. The preliminary studies of the incoming boundary layer at Bolund island (for
239 and 270 wind directions) indicate a minimum value for the ratio of the boundary layer
thickness, , to the escarpment height, /h 1.5, and therefore /h O(1).
The most relevant references for understanding the detached-reattaching zone on the Bolund
island have been considered the existing studies on blunt flat plates with right-angled corners
(BFP, see, for instance, Kiya & Sasaki, 1983 or Nakamura & Ozono, 1987) and forward-
facing steps with right-angled corners (FFS, see for instance Tachie et al. 2001 or Largeau &
Moriniere, 2007). The geometry of BFP and FFS are defined by the BFP thickness, a=2h, or
the FFS height, h; and for both geometries, their width, d, and length, l. The studies on FFS
and BFP have been focused on three main aspects, a) the flow topology at the detached-
reattaching zone, b) the relaxation process of the new boundary layer after reattachment and
c) the surface pressure topology beneath the bubble region and upstream-downstream it and
its relation with the flow field in its vicinity, both inside and outside of the detachment
bubble. The study of the surface pressure topology just under and close to the detachment
bubble on simple geometries (such as the mentioned FFS and BFP) has been an issue of
direct concern in the understanding of wind loads on bluff bodies like buildings (Li &
Melbourne, 1995).
Additionally, the study of the pressure field topology on the top surface of BFP and FFS has
been established as an adequate diagnostic means for the determination of the size and the
intensity on the detached-reattaching zone (Hillier & Cherry, 1981, Castro & Dianat, 1983,
Kiya & Sasaki, 1983, Saathoff & Melbourne, 1989, Li & Melbourne, 1995, Li & Melbourne,
1999, Largeau & Moriniere, 2007, Sherry et al., 2010) as well as a methodology to determine
the influence of scale parameters such as Re
h
, /h or
x
u
L /h or pure inflow parameters such as
I
u
on the bubble characteristics.
Therefore the methodology could be quite appropriate to determine the fidelity of wind
tunnel tests. Taking this into account, the authors has considered interesting to analyse the
pressure behaviour on the surface of a model of the Bolund Island in order to gain insight in
the bubble topology likely formed on the flat part to the island. The paper is organised in two
main parts: the first covering background study, a description of the bubble phenomenon and
the non-dimensional numbers describing it and the second covering instrumentation and wind
tunnel setup followed by results obtained and finally conclusions.
2 MAIN CHARACTERISTICS OF THE DETACHED-REATTACHING BUBBLE
Let us consider as a reference flow configuration for our test a rectangular body of
dimensions ldh, located on the surface on a channel flow and immersed in a neutrally
stratified boundary layer with thickness , free mean wind speed, U
, square root of the free

wind speed variance,
u
and the longitudinal integral length scale
x
u
L . The dimensional
analysis describes the ensemble averaged flow topology by non-dimensional parameters l/h,
d/h, Re
h
, /h,
x
u
L /h and finally I
u
=
u
/U
. Of course any ensemble mean parameter will also

depends on the non-dimensional coordinates x/h, y/h and z/h but not on time, since the flow
field is stationary. The aspect ratios l/h and d/h affect considerably to the flow topology in
front of the body and downstream the edge (see Sherry at al., 2010). l/d must be large enough
to assure that the forward-step is truly isolated (Sherrys wording) meaning that the flow field
at the front is not affected by the downstream wake of the body. Regarding the span-wise
352

aspect ratio d/h, works like Hillier & Cherry (1981) or Kiya & Sasaki (1983) establish values
d/h >10 to assure 2D mean conditions at the central part of the body (see Largeau &
Moriniere, 2007). Obviously the Bolund topography is not fulfilling the geometric
requirements to be considered as a 2D configuration with a FFS truly isolated but the analysis
of such an idealised configuration can provide certain insight on the topology of the
detachment. In the figure 1 an schematic on the ensemble mean velocity field on a 2D
(d/h>>1), truly isolated (l/d>>1) FFS is shown
Fig X2a
h
z
x
, ,
x
u u
U I L
o
l
R
l
F
R
D
d
r

Figure 1: Schemantic of the ensemble mean flow velocity field arroud a 2D-truly isolated FFS (adapted from
Sherry et al., 2010 and Bradshaw & Wong, 1972). d: mean position of the detachment point in front of the FFS,
r: mean position of the reattachment point on the vertical wall, D: position of the dettachment point on top of the
FSS (edge), R: mean position of the reattachment on top of the FFS. l
F
is the mean length of the front separation
bubble and l
R
is the mean length of the separation bubble on top of the FFS. The shear layer originated at D is
schematised with grey dots.
It is quite well established that when the flow approaches the FFS, the blockage provoked by
the vertical wall generates a severe adverse pressure gradient. As a consequence a first
detachment bubble (see figure 1) is initiated at a mean position [x/h, z/h] [( 0.8, 1.5),0],
so l
F
/h ( 0.8, 1.5). The flow reattaches to the vertical wall at a mean position [x/h, z/h]
[0,(0.6,0.65)], (see Sherry et al., 2010 and Huiyin & Yanhua., 2011). The dynamics of this
first detachment bubble is complex even for a 2D- truly isolated FFS in both laminar and
turbulent conditions (see results and analysis for Re
h
[940,8400] in Stuer et al., 1999 and
the analysis in Largeau & Moriniere, 2007). This first detachment bubble acts a fluid ramp
and its width seems to depend on /h.
On the top surface of the FFS, just at the edge, a second separation of the flow is produced
(therefore at [x/h, z/h] = [0,1]). A shear layer with high TKE evolves from the edge adjusting
the velocity from the reverse flow region inside the bubble to the free conditions. The ratio
/h influences notably the length of the bubble, l
R
/h, when /h>1, since, depending on the
specific value of /h, the FFS can interact with the viscous layer, with the log layer or the
outer layer of the incident boundary layer (see Tachie et al., 2001 and Sherry et al., 2010).
The Reynolds number, Re
h
, is another important parameter. There is a growing dependence
of the non-dimensional length of the bubble, l
R
/h, with Re
h
, In Sherry et al. (2010) a
mechanism of detachment at the FFS edge is proposed as responsible for the high sensitivity
of l
R
/h with Re
h
for Re
h
<8500. For very low Re
h
, laminar separation occurs at the edge of the
FFS followed by laminar to turbulent transition and finally a turbulent reattachment at a
certain distance on top of the FFS. The transition distance which can be used to describe the
topology of the bubble decreases to zero as Re
h
increase till Re
h
8500 where the
detachment is fully turbulent. For Re
h
> 8500 there is not any change in the characteristics of
the detachment (this is fully turbulent) so that the sensitivity of l
R
/h on Re
h
, although still
positive, is much smaller in opinion of Sherry et al. However the data from Largeau &
Moriniere indicate that the sensitivity is still important. Both data sets where obtained for
353

similar I
u
(0.015) but for different value of /h, much smaller in the case of Largeau &
Moriniere (what is in accordance with a larger value of l
R
/h). In the case of Camussi et al., the
remarkably lower value of l
R
/h could be explained both by a higher value of /h and I
u
. The
ratio
x
u
L /h is not indicated in any case, being this a source of uncertainty.
The formation of the bubble on the top of the FFS and BFP can be briefly described by the
accumulation of vorticity giving growth to the bubble (see Sherry et al., 2010 and Kiya &
Sasaki, 1983). After a certain time, a large scale vortex is ejected, and the bubble size
decreases, the process of vortices accumulation repeating again. This dynamic process is
responsible of the instantaneous change of the reattachment line (so that l
R
/h as a defined line
can be only determined as a mean characteristic) and it is known as flapping (Camussi et
al., 2008). Finally, several authors (see Kiya & Sasaki, 1983, Largeau & Moriniere, 2007,
Camussi et al., 2008) have identified that the mean convection velocity of vortex structures
on the FFS in the bubble ranges from 0.3U
close to the front, up to 0.7U
well after the

reattachment zone, being in the order of 0.5 U
at the reattachment. All these studies (Hillier

& Cherry, 1981, Nakamura & Ozono, 1986, Saathoff & Melbourne, 1989, Li & Melbourne,
1995, 1999) agree in a length of the bubble l
R
/(2h) (3.5,5) for the reference smooth case. A
general agreement exists on the influence of I
u
on the longitudinal size of the bubble l
R
/h.
Larger values of longitudinal turbulence intensity lead to smaller bubbles, so l
R
/h decreases as
I
u
increases. At the same time the bubble is more intense, for instance, the minimum mean
pressure coefficient C
p
under the detached zone is smaller (more negative), or the maximum
standard deviation pressure coefficient, C
p
, is larger as I
u
grows.
3 STATISTICS DERIVED FROM SURFACE PRESSURE AS DIAGNOSTIC
PARAMETERS OF THE BUBBLE EXTENSION
It has been mentioned that a largely negative mean pressure coefficients surface pressure
beneath the bubble are associated to the accumulation of vortices originated at the edge in the
bubble (Kiya & Sasaki, 1983, Largeau & Moriniere, 2007, Camussi et al., 2008). So the
existence/intensity of the bubble can be diagnosed from the exploration of the C
p
.
Additionally, different authors have identified the reattachment region with that part on the
FFS where the variation of mean C
p
with x/h is largest (Hillier & Cherry, 1981, Li &
Melbourne, 1999) being this region roughly coincident with the location where the standard
deviation of the pressure coefficient, C
p
, is maximum. Most of the referred studies have
analysed the influence of parameters such as I
u
, or
x
u
L /h on the functions C
p
(x/h) or C
p
(x/h)
(Nakamura & Ozono, 1986 and Li & Melbourne, 1995, 1999 are representative examples)
inferring characteristics of the bubble such as its length l
R
/h from the location of
max[dC
p
/d(x/h)] or max(C
p
).
Kiya & Sasaki (1983b) and Saathoff & Melbourne (1989) argue that the entrainment of outer
energetic fluid into the bubble at the reattachment zone provokes a larger probability of
positive peaks of the pressure fluctuation originating an asymmetry around the mean of the
PDF of the pressure fluctuation which is quantified by large positive values of the pressure
skewness, S
p
. Camussi et al. (2008) also agree with this reasoning. The authors identified, by
means of PIV, a bubble length l
R
/h 2.1 for Re
h
= 2.6310
4
, and calculate the PDF of the
pressure fluctuation at x/h = 0.45, 1.95 and 2.7, finding that the PDF is skewed negatively at
x/h = 0.45 and positively at x/h = 1.95 and mainly at x/h = 2.7. In Camussi et al. (2008) are
also determined largely positive values of S
p
upstream of the front of the FFS also in
agreement with the findings of Steinwolf & Rizzi (2006). In figure 6, and schematic of the
process for the determination of the bubble length l
R
/h based on the behaviour of the mean
354

pressure coefficient, C
p
, the std. pressure coefficient, C
p
p

is shown.
Fig X7
p
C
o
p
C
0.0
1.0
1.0
1.0
p
C
p
C
o
p
S
0.0
p
S
max
m

Figure 2: Schematic on the diagnostic process on the bubble length l
R
/h based on the behaviour of the mean
pressure coefficient, C
p
, the std. pressure coefficient, C
p
p
.
4 EXPERIMENTAL SET-UP AND DATABASE
4.1 Wind tunnel description
The test was conducted in the A9 wind tunnel in IDR/UPM which is an open circuit suction
type Eiffel tunnel with a closed test chamber. The convergent section of the wind tunnel is bi-
dimensional with a length of 5.25m and an input section of 4.8m wide and 1.8m high. The
test chamber has the following dimensions: Length: 3m Width: 1.5 Height: 1.8m. The
wind tunnel is driven by nine eight-bladed variable speed fans with nominal output of 10kW
capable of producing winds of 5-35m/s. The mounting of the model is made using a turntable
installed in the sidewall of the test section of the wind tunnel. See figure 3.
4.2 Instrumentation and sampling rate
A total of 475 pressure tabs were installed on the model of the island and an additional 12
pressure tabs were installed as a reference in a straight line on the ramp forward of the
Bolund model. The pressure tabs on the island are distributed at a distance of 0.02m (0.4h)
while the 12 reference tabs were installed 0.04m (0.8h) apart along a straight line in front of
the model. Tabs 15-18 were installed along the 239 line. The distribution of the pressure tabs
can be seen in figure 3. Tabs 2-5 were installed with a spacing of 0.01m (0.2h) along the 270
line with tab 4 being located on the edge of the Bolund island. Each pressure tabs consists of
a brass tube, flushed on the model surface, connected to the data acquisition system by a
plastic tube with both having 0.001m inner diameter. The plastic tubes are connected to two
64-ports pressure scanners from Scanivalve Corp. (ZOC33).
Measurements were taken at a rate of 100Hz over 180s for each pressure tab. The
measurements for all the pressure tabs were done in 2 block measurements of 256 pressure
tabs with the 12 reference pressure tabs always present in each block. A pitot was installed
upstream of the model to measure the instantaneous static pressure, p
and the instantaneous

total pressure,
T
p . For each block of 256 pressure tabs, a first set of 128 pressure tabs is
measured simultaneously during a first interval of 180s, and a second set of 128 pressure tabs
is measured in a second-consecutive interval of 180s. Both the acquisition time of 180s and
the sampling frequency of 100Hz are chosen taking into account the limitation of the
equipment (maximum buffer size) and technical issues. The acquisition time is selected to
assure a good convergence of the statistics. The sampling frequency, which gives rise to
associated non-dimensional acquisition frequencies St = fh/U
= 1.02 (5 ms
1
), 0.34 (15
ms
1
), 0.204. (25 ms
1
) was selected anticipating the occurrence of energetic pressure
355

fluctuations for St = 0.01 and St = 0.2 (see point 2) at the detachment. Obviously non-
dimensional sampling frequencies St = 0.34 and 0.204 do not fulfil the Shannon theorem for
characterising pressure fluctuations with St = 0.2.
4.3 Calculations
The instantaneous pressure coefficient is calculated from by ( )
p
c p p q

= where p is the
static pressure measure locally by the pressure tabs and q
is the dynamic pressure at the

pitot. Since the installation is normally oriented to determine mean pressure coefficients, C
p
,
and the scanivalve is a differential scanner, the difference
T
r p p = is recorded, and the
instantaneous pressure coefficient is then calculated 1
p
c r r
= , where
T
r p p

= .
The mean pressure coefficient, C
p
, the standard deviation pressure coefficient, C
p
, and the
skewness of the pressure, S
p
are calculated respectively
,
1
1
N
p p n
n
C c
N
=
=

. (0.1)
1 1
2 2
1 2 1 2
1 1
1 1
N N
p n n
n n
C Q p Q r
N N
o

= =
| | | |
= ~ | |
| |
\ . \ .

. (0.2)
3 3
2 2
3 2 3 2
1 1 1 1
1 1 1 1
N N N N
p n n n n
n n n n
S p p r r
N N N N

= = = =
| || | | || |
= ~ | | | |
| | | |
\ .\ . \ .\ .

. (0.3)
For the previous calculation of second and third moments of the pressure fluctuation, p, from
the readings, r, it is considered that typically the fluctuation of the total pressure at the pitot is
typically lesser than the fluctuation of the static pressure on the model, mainly at the
detachment.
The Bolund model was manufactured with a scale of 1:230 giving a maximum height for the
model of 0.0512m. 3D data for the Bolund Island was obtained from the data provided by
Bechmann et al. (2009). The scaled model was manufactured using Necuron400 material in
an automated 3D milling machine.
Line 270
Line 239
Tabs
15-18
Tabs 2-
5 & 9
Ramp
0.0 0.2 0.4 0.6 0.8 0.2 0.4 0.6 0.8 1.0 1.2
U
0.0
0.2
0.4
0.6
0.8
0.2
0.4
0.8
0.6
[m]
WT
x
[m]
WT
y

Figure 3: Pressure tabs distribution on the island model and schematic of the mounting in the A9 wind tunnel.

356

4.5 Measurement Campaign
A total of four angles of wind direction were chosen for the inflow direction. Three angles,
270, 239 and 255 correspond to the test cases specified in Bechmann et al. (2009). While
the final angle 180 was to be an extra case. The three wind velocities are 5m/s, 15m/s and
25m/s giving a Reynolds number range of 1.710
4
8.510
4
with h=0.0512m. The wind
tunnel has a boundary layer thickness, 0.04m and longitudinal turbulence intensity at the
reference position I
u
2.5%.

0.4 0.6 0.8 1 1.2 0.00 0.05 0.10 0.15 0.20
WT
z
h
0.0
1.0
2.0
3.0
4.0
WT
z
h
0.0
1.0
2.0
3.0
4.0
u
U o U U

Figure 4: Vertical profiles of mean wind speed and turbulence intensity at at x
WT
/h=0, at y
WT
/h=0. Conditions
measured without the mock-up.
5 RESULTS
The evolutions for the mean pressure coefficient, C
p
, the standard deviation of the pressure
coefficient, C
p
p
, along the line B (270) are presented in
figure 5. Results for two Reynolds numbers, Re
h
=5.110
4
and Re
h
=8.510
4
are shown. It can
be observed that, as expected, a large overpressure occurs just in front of the escarpment due
to the blockage produced. Just on the top of the escarpment (x/h=0) a high suction pressure
occurs, probably indicating the presence of separation. The suction pressure reaches
maximum values C
p
2 for the higher Reynolds case and C
p
1.75 for the lower
Reynolds one. Except for the variation on the maximum value of C
p
the Reynolds number
seem not to affect the distribution of mean pressure coefficient, what is in agreement with the
conclusions presented in Nakamura & Ozono (1987) for a BFP for Re
(2h)
> 1.410
4
. The
values of C
p
at the edge are larger than the ones found for right-angled corners-BFP for
similar values of Re
(2h)
(see for instance Li & Melbourne, 1999, where maximum values C
p

1.1 are declared), probably due to the specific geometry of the Bolund escarpment at 270
(with an initial ramp and with a slightly rounded edge).
The evolution of the standard deviation of the pressure coefficient, C
p
, grows dramatically at
the top of the escarpment, reaching a maximum for both Re
h
cases slightly downstream of the
edge, at x/h 0.8. This is an evidence of the presence of separation. It has been established
that location of max(C
p
) roughly coincides with the mean position of the reattachment,
therefore in this case it can be established that l
R
/h 0.8 0.1. For similar Reynolds number
and /h, but lower I
u
, Largeau & Moriniere (2007) determined values l
R
/h 3. Skewness of
the pressure shows large positive values up-stream of the escarpment (what is in agreement
with the coclusions for FFS shown by Camussi et al., 2008 and Steinwolf & Rizzi, 2006).
After the edge, there is a region 0.1<x/h<1.1 with S
p
<0, recovering positive values roughly at
x/h 1.1 0.1. This change of sign of S
p
(from negative to positive values) has been
357

propossed above as an indication of the reattachment region, so the location of such
reatachment based on maxC
p
, and the sign change of S
p
roughly coincide around x/h1.

Figure 5: Distribution of mean pressure coefficient, C
p
, expression (0.1), standard deviation pressure
coefficient, C
p
, expression (0.2), Skewness of pressure coefficient, S
Cp
, and skewness of the pressure, S
p
,
expression (0.3) along line B (270), for two Reynolds numbers, Re
h
, indicated in the figure. The profile of the
island for the mentioned direction is shown in the figure in continuous line.
In figure 6 the surface distributions of C
p
, C
p
, and S
p
(left-right columns) for line 270 and
the two Reynolds numbers (top,bottom). The white color for the two figures in the first
column, indicate values C
p
>1, therefore the mean topology of the detached bubble. It is
evidenced the 3D-like character of such topology and the existence of local patterns that can
be easily identified with the geometry of the escarment. For intance the detached region
seems to be more intense at the center of the escarpment where the vertical portion of it is
larger and faces perpendicular to the 270 wind. The results shown in figures at the second
column (C
p
) corroborates the 3D character of the separated region. The analysis of S
p
(third
column) reveals a similar conclusion. The white dotted line in the figure marks the loci on the
surface where S
p
= 0. It must be remarked that S
p
values are trustable only when C
p
is high
(as it was argumented above) so the values of S
p
for x/h>3 must not be used to extract any
conclusion without a further analysis.

Figure 6: Surface distribution of mean pressure coefficient, C
p
, (first column), standard deviation pressure
coefficient, C
p
(second column), and skewness of the pressure, S
p
, (third column) along line B (270), for two
Reynolds numbers (top row, Re
h
= 5.110
4
, bottom row, 8.510
4
). The dark lines indicate isolevel curves, and
the white lines indicates the loci where S
p
= 0 on the surface.
358

The same analysis is showed in figure 7 for the 239 case. It is obvious that the region of
detachment now appears at regions perpendicular to the 239 wind. It is remarked the
appearance of a region of high values of C
p
and C
p
at the uppper region of the topography
probably incating a backward-facing step-like detached region.

Figure 7: Surface distribution of mean pressure coefficient, C
p
, (left), standard deviation pressure coefficient,
C
p
(middle), and skewness of the pressure, S
p
, (right) along line A (239), for a Reynolds number Re
h
=
5.110
4
. The dark lines indicate isolevel curves, and the white lines indicates the loci where S
p
= 0 on the
surface

Figure 8: Surface distribution of the instantaneous pressure coefficient,
p
c
, for the case 270 and a Reynolds
number Re
h
=8.510
4
. The color bar is the same as for the C
p
figures in figure 15 and 16.

In order to illustrate the non-steadyness of the separation proccess, in figure 8, a sequence of
21 shots of the instantaneous preassure coefficient,
p
c , are shown for the case 270 and Re
h

= 8.510
4
. The time interval between shots is t=0.01s corresponding to a nondimensional
time interval T=4.8 (T = tU
/h). It is reminded that flapping process for a FFS takes part (in
a mean sense) every T
F
100. It can be realised that the region where is instantaneously
p
c >1 changes its shape remarkably.
6 CONCLUSIONS
The study has shown that surface pressure distribution can provide a description of the
detaching-reattaching flow topology over the Bolund Island. By studying the statistics of the
mean ensemble averages of the pressure distribution, an estimation of the horizontal
extension of the bubble size can be obtained. Although this is limited to simple or simple-to-
moderately complex topology such as the Bolund island, this method, coupled with other
measuring technique can provide a good description of the flow topology providing quick
359

and important validation results for other modellers. Evidently, the 3D effects of the terrain
clearly expose the weakness of using 2D geometry study as a validation as can be seen from
the skewness plot of x/h>3. This shows a need for more benchmark studies to be carried out
in order to study more complex terrain.
Acknowledgments.
Carlos Pascual, Luis, Enrique Vega of IDR, UPM.
4 REFERENCES
Bechmann, A. 2010. Presentations from "The Bolund Experiment Workshop" 3-4th December 2009 Risoe-R-
1745(EN).
Bechmann, A., Berg, J., Courtney, M.S., Joergensen, H.E., Mann, J., Soerensen, N.N. 2009. The Bolund
Experiment: Overview and Background Risoe-R-1658(EN).
Bradshaw, P., Wong, F. 1972. The reattachment and relaxation of a turbulent shear layer. Journal of Fluid
Mechanics 52, 113.
Camussi, R., Felli, M., Pereira, F., Aloisio, G., Di Marco, A. 2008. Statistical properties of wall pressure
fluctuations over a forward-facing step. Physics of Fluids 20.
Castro, I. 1979. Relaxing wakes behind surface-mounted obstacles in rough wall boundary-layers. Journal of
Fluid Mechanics 93, 631-659.
Castro, I., Dianat, M. 1983. Surface flow patterns on rectangular bodies in thick boundary-layers. Journal of
Wind Engineering and Industrial Aerodynamics 11, 107-119.
Conan, B., Buckinham, S., van Beeck, J., Aubrun, S., Sanz-Rodrigo, J. 2011. Feasibility of Micro Siting in
Mountainous Terrain by Wind Tunnel Physical Modelling. European Wind Energy Conference Scientific
Proceedings, 136-140.
Hillier, R., Cherry, N. 1981. The effects of stream turbulence on separation-bubbles. Journal of Wind
Engineering and Industrial Aerodynamics 8, 49-58.
Huiyin, R., Yanhua, W. 2011. Turbulent boundary layer over smooth and rough forward-facing steps. Physics of
Fluids 23.
Kiya, M., Sasaki, K. 1983. Free-stream turbulence effects on a separation bubble. Journal of Wind Engineering
and Industrial Aerodynamics 14, 375-386.
Kiya, M., Sasaki, K. 1983. Structure of a turbulent separation bubble. Journal of Fluid Mechanics 137, 83-113.
Largeau, J.F., Moriniere, V. 2007. Wall pressure fluctuations and topology in separated flows over a forward-
facing step. Experiments in Fluids 42, 21-40.
Li, Q., Melbourne, W. 1999. The effect of large-scale turbulence on pressure fluctuations in separated and
reattaching flows. Journal of Wind Engineering and Industrial Aerodynamics 83, 159-169.
Meroney, R.N. 1993. Wind Tunnel Modelling of Hill And Vegetation Influence On Wind Power Availability.
Task 1: Literature Review.
Nakamura, Y., Ozono, S. 1987. The effects of turbulence on a separated and reattaching flow. Journal of Fluid
Mechanics 178, 477-490.
Saathoff, P., Melbourne, W. 1989. The generation of peak pressures in separated reattaching flows. Journal of
Wind Engineering and Industrial Aerodynamics 32, 121-134.
Sherry, M., Lo Jacono, D., Sheridan, J. 2010. An experimental investigation of the recirculation zone formed
downstream of a forward facing step. Journal of Wind Engineering and Industrial Aerodynamics 98, 888-
894.
Steinwolf, A., Rizzi, S. 2006. Non-Gaussian analysis of turbulent boundary layer fluctuating pressure on aircraft
skin panels. Journal of Aircraft 43, 1662-1675.
Stuer, H., Gyr, A., Kinzelbach, W. 1999. Laminar separation on a forward facing step. European journal of
mechanics.B, Fluids 18, 675-692.
Tachie, M., Balachandar, R., Bergstrom, D. 2001. Open channel boundary layer relaxation behind a forward
facing step at low Reynolds numbers. Journal of fluids engineering 123, 539-544.
Taylor, P., Teunissen, H. 1987. The Askervein hill project - Overview and background data. Boundary - layer
meteorology 39, 15-39.
Teunissen, H., Shokr, M., Bowen, A., Wood, C., Green, D. 1987. The Askervein hill project - Wind-tunnel
simulations at 3 length scales. Boundary - layer meteorology 40, 1-29.
Yu, X., Weng, W., Taylor, P., Liang, D. 2011. Relaxation Factor Effects in the Non-Linear Mixed Spectral
Finite Difference Model of Flow Over Topographic Features. Boundary - layer meteorology 140, 23-35.

360
Helpful Information
Location
PHYSMOD2011 will take place at the Geomatikum (Bundesstrasse 55, 20146 Hamburg), located near the
Schlump underground station which can be accessed from lines U2 and U3, see map below.
Local PHYSMOD2011 Contact Information
Address: Meteorological Institute, Bundesstrasse 55, 20146 Hamburg, Germany
Phone: +49 (0) 40 42838 5093
Fax: +49 (0) 40 42838 5452
Email/Website: physmod2011@zmaw.de / www.physmod.org
Local Transportation
Public Transit: Information for public transportation can be found online at: www.hvv.de/en/ One way ticket
prices depend on the length of journey, but start at 1,80 e for local distance and 2,80 e from the airport to
central Hamburg. If you plan to do several trips throughout the day there is a day ticket available (Tageskarte)
for 6,80 e. The same ticket will work on the bus, underground train (U-bahn) and suburban train (S-bahn). A
map of the local train lines can be found on the next page.
Taxi: Please ask someone at the registration desk to call a taxi for you.
Arrival by train
If you plan to arrive by train you should get o at the Hamburg Main Train Station (Hamburg Hauptbahnhof),
to get to the Geomatikum take the U2 underground train in the direction of Niendorf Nord/Markt and get o
at Schlump (trip length: 7 minutes, service frequency: every 5 minutes).
Arrival by plane
The Hamburg Airport (HAM) is connected to public transit via the S1 suburban train. To reach the Geomatikum
take the S1 in the direction of Ohlsdorf, change at Ohlsdorf to the U1 underground train in the direction of
Ohlstedt, change a second time at Kellinghusenstrae to the U3 toward Schlump and get o at Schlump (trip
length: 23 minutes, service frequency: every 10 minutes).
Internet
Wireless internet access will be available to all participants of the PHYSMOD conference. Access requires an
individually assigned login and password, available at registration.
Copyright Notice
Authors retain the copyright for all submissions published in the PHYSMOD2011 proceedings.
361
A
l
s
t
e
r
B
e
r
l
i
n
e
r
T
o
r
R
i
t
t
e
r
s
t
r
a
e
W
a
r
t
e
n
a
u
L
a
n
d
w
e
h
r
B
a
r
m
b
e
k
F
r
i
e
d
r
i
c
h
s
-

b
e
r
g
K
o
r
n
w
e
g
(
K
le
in
B
o
r
s
t
e
l)
H
o
h
e
n
e
i
c
h
e
n
W
e
l
l
i
n
g
s
b
t
t
e
l
P
o
p
p
e
n
b
t
t
e
l
H
a
b
i
c
h
t
s
t
r
a
e
R
b
e
n
k
a
m
p
(
C
it
y
N
o
r
d
) A
l
t
e

W
h
r
(
S
t
a
d
t
p
a
r
k
)
O
h
l
s
d
o
r
f
S
e
n
g
e
l
m
a
n
n
s
t
r
a
e
(
C
it
y
N
o
r
d
)
A
l
s
t
e
r
d
o
r
f
L
a
t
t
e
n
k
a
m
p
(
S
p
o
r
t
h
a
lle
)
H
u
d
t
w
a
l
c
k
e
r
s
t
r
a
e
K
e
l
l
i
n
g
h
u
s
e
n
-

s
t
r
a
e
N
o
r
d
e
r
s
t
e
d
t

M
i
t
t
e
R
i
c
h
t
u
n
g

N
e
u
m
n
s
t
e
r
C
h
r
i
s
t
u
s
k
i
r
c
h
e
E
m
i
l
i
e
n
s
t
r
a
e
O
s
t
e
r
s
t
r
a
e
L
u
t
t
e
r
o
t
h
s
t
r
a
e
H
a
g
e
n
b
e
c
k
s
T
i
e
r
p
a
r
k
H
a
g
e
n
d
e
e
l
N
i
e
n
d
o
r
f

M
a
r
k
t
J
o
a
c
h
i
m
-
M
h
l
-
S
t
r
a
e
S
c
h
i
p
p
e
l
s
w
e
g
N
i
e
n
d
o
r
f

N
o
r
d
S
t
e
r
n
s
c
h
a
n
z
e
(
M
e
s
s
e
)
F
e
l
d
s
t
r
a
e
(
H
e
ilig
e
n
g
e
is
t
f
e
ld
)
S
t
.
P
a
u
l
i
S
c
h
l
u
m
p
H
o
l
s
t
e
n
s
t
r
a
e
A
l
t
o
n
a
W
e
d
e
l
R
i s
s
e
n
S
l l d
o
r
f
I s
e
r
b
r
o
o
k
H
o
c
h
k
a
m
p
B
l
a
n
k
e
n
e
s
e
K
l e
i n
F
l o
t
t
b
e
k
( B
o
t a
n
i s
c
h
e
r G
a
r t e
n
)
O
t
h
m
a
r
s
c
h
e
n
B
a
h
r
e
n
f
e
l d
R
e
e
p
e
r
b
a
h
n
K
n
i g
s
t
r
a
e
L
a
n
d
u
n
g
s
b
r
c
k
e
n
E
i
d
e
l
s
t
e
d
t
Z
e
n
t
r
u
m
P
r
i
s
d
o
r
f
E
l
m
s
h
o
r
n
L
a
n
g
e
n
m
o
o
r
S
p
a
r
r
i
e
s
h
o
o
p
B
o
k
h
o
l
t
V
o
l
o
c
h

B
a
r
m
s
t
e
d
t
B
r
u
n
n
e
n
s
t
r
a
e
B
a
r
m
s
t
e
d
t
P
i
n
n
e
b
e
r
g
T
o
r
n
e
s
c
h
R
i
c
h
t
u
n
g

N
e
u
m
n
s
t
e
r
/
K
i
e
l
R
i
c
h
t
u
n
g

I
t
z
e
h
o
e
/
H
u
s
u
m
D
a
u
e
n
h
o
f
H
o
r
s
t
A
l
v
e
s
l
o
h
e
L
a
n
g
e
l
n
B
u
c
k
h
o
r
n
H
o
i
s
b
t
t
e
l
O
h
l
s
t
e
d
t
V
o
l
k
s
d
o
r
f
A
l
t
e
r
T
e
i
c
h
w
e
g
S
t
r
a
b
u
r
g
e
r

S
t
r
a
e
W
a
n
d
s
b
e
k
M
a
r
k
t
W
a
n
d
s
b
e
k
-
G
a
r
t
e
n
s
t
a
d
t
T
r
a
b
r
e
n
n
b
a
h
n
F
a
r
m
s
e
n
B
e
r
n
e
M
e
i
e
n
d
o
r
f
e
r
W
e
g
B
u
c
h
e
n
k
a
m
p
A
h
r
e
n
s
b
u
r
g
W
e
s
t
A
h
r
e
n
s
b
u
r
g
O
s
t
S
c
h
m
a
l e
n
b
e
c
k
K
i e
k
u
t
G
r
o
h
a
n
s
d
o
r
f
B
a
d

O
l
d
e
s
l
o
e
K
u
p
f
e
r
m
h
l
e
B
a
r
g
t
e
h
e
i
d
e
A
h
r
e
n
s
b
u
r
g
R
e
i
n
f
e
l
d
R
i
c
h
t
u
n
g

L
b
e
c
k
W
a
h
l
s
t
e
d
t
A
l
t
e
n
g
r
s
B
a
d

S
e
g
e
b
e
r
g
W
a
k
e
n
d
o
r
f
F
r
e
s
e
n
b
u
r
g
R
i
c
k
l
i
n
g
F
a
h
r
e
n
k
r
u
g
R
i
c
h
t
u
n
g

N
e
u
m
n
s
t
e
r
R
o
t
h
e
n
b
u
r
g
s
o
r
t
B
i
l
l
w
e
r
d
e
r
-
M
o
o
r
f
l
e
e
t
M
i
t
t
l
e
r
e
r

L
a
n
d
w
e
g
B
e
r
g
e
d
o
r
f
R
e
i n
b
e
k
W
o
h
l t
o
r
f
A
u
m
h
l
e
F
r
i e
d
r
i c
h
s
r
u
h
S
c
h
w
a
r
z
e
n
b
e
k
M
s
s
e
n
R
i
c
h
t
u
n
g
S
c
h
w
e
r
i
n
/
R
o
s
t
o
c
k
B
c
h
e
n
R
i
c
h
t
u
n
g
D
a
n
n
e
n
b
e
r
g
R
i
c
h
t
u
n
g

U
e
l
z
e
n
L
n
e
b
u
r
g
H
a
r
b
u
r
g
L
a
u
e
n
b
u
r
g
E
c
h
e
m
T
i
e
f
s
t
a
c
k
M
e
c
k
e
l
f
e
l
d
M
a
s
c
h
e
n
S
t
e
l
l
e
W
i
n
s
e
n
R
a
d
b
r
u
c
h
B
a
r
d
o
w
i c
k
V
a
s
t
o
r
f
B
a
v
e
n
d
o
r
f
D
a
h
l e
n
b
u
r
g
N
e
e
t
z
e
n
d
o
r
f
G
h
r
d
e
W
e
n
d
i s
c
h
E
v
e
r
n
H
i
t
t
f
e
l
d
K
l
e
c
k
e
n
B
u
c
h
h
o
l
z
S
u
e
r
h
o
p
H
o
l
m
-
S
e
p
p
e
n
s
e
n
B
s
e
n
b
a
c
h
t
a
l
H
a
n
d
e
l
o
h
S
p
r
t
z
e
T
o
s
t
e
d
t
R
i
c
h
t
u
n
g

S
o
l
t
a
u
H
a
r
b
u
r
g
R
a
t
h
a
u
s
H
e
i m
f
e
l d
N
e
u
w
i e
d
e
n
t
h
a
l R
i
c
h
t
u
n
g

B
r
e
m
e
n
B
a
r
g
s
t
e
d
t
B
r
e
s
t
-
A
s
p
e
K
u
t
e
n
h
o
l
z
A
p
e
n
s
e
n
R
u
s
c
h
w
e
d
e
l
H
a
r
s
e
f
e
l d
R
i
c
h
t
u
n
g
B
r
e
m
e
r
h
a
v
e
n
N
e
u
k
l
o
s
t
e
r
H
o
r
n
e
b
u
r
g
D
o
l
l
e
r
n
A
g
a
t
h
e
n
b
u
r
g
S
t
a
d
e
H
a
m
m
a
h
H
i
m
m
e
l
-
p
f
o
r
t
e
n
R
i
c
h
t
u
n
g
C
u
x
h
a
v
e
n
N
e
u
g
r
a
b
e
n
N
e
u W
u
l m
s
t
o
r
f
T
h
e
s
d
o
r
f
E
l
b
g
a
u
s
t
r
a
e
M
n
c
k
e
b
e
r
g
-
s
t
r
a
e
M
e
b
e
r
g
S
t
e
i n
s
t
r
a
e
R
a
t
h
a
u
s
R
d
i n
g
s
m
a
r
k
t
B
a
u
m
w
a
l l M
e
s
s
e
h
a
l
l
e
n
G
n
s
e
m
a
r
k
t
(
O
p
e
r
)
S
t
e
p
h
a
n
s
p
l
a
t
z
(O
p
e
r
/
C
C
H
)
S
t
a
d
t
h
a
u
s
-
b
r
c
k
e
E
p
p
e
n
d
o
r
f
e
r

B
a
u
m

H
o
h
e
l
u
f
t
-

b
r
c
k
e
K
l
o
s
t
e
r
s
t
e
r
n
H
a
l
l
e
r
s
t
r
a
e
D
a
m
m
t
o
r
(
M
e
s
s
e
/
C
C
H
)
B
o
r
g
w
e
g
( S
t a
d
t p
a
r k
)
S
a
a
r
l a
n
d
-
s
t
r
a
e
S
i e
r
i c
h
s
t
r
a
e
H
a
m
m
e
r
b
r
o
o
k
(
C
it
y
S
d
)
W
i
l
h
e
l
m
s
b
u
r
g
L
o
h
m
h
l
e
n
s
t
r
a
e
H
a
m
b
u
r
g
e
r

S
t
r
a
e
M
u
n
d
s
b
u
r
g
L
b
e
c
k
e
r

S
t
r
a
e
U
h
l
a
n
d
s
t
r
a
e
D
e
h
n
h
a
i
d
e
R
a
h
l
s
t
e
d
t
T
o
n
n
d
o
r
f
W
a
n
d
s
b
e
k
R
i
c
h
t
u
n
g

L
b
e
c
k
M
l
l
n
R
a
t
z
e
b
u
r
g
H
a
l
s
t
e
n
b
e
k
K
r
u
p
u
n
d
e
r
A
l
l
e
r
m
h
e
N
e
t
t
e
l
n
b
u
r
g
H
a
s
s
e
l
b
r
o
o
k
G
r
o
b
e
r
e
i
c
h
H
a
m
b
u
r
g
G
r
o
b
e
r
e
i
c
h
H
a
m
b
u
r
g
S
t
a
n
d
/
D
a
t
e

o
f

i
s
s
u
e
1
2
.

D
e
z
e
m
b
e
r

2
0
1
0

H
V
V
M
o
o
r
b
e
k
h
a
l
l
e
(
S
c
h
u
lz
e
n
t
r
u
m
N
o
r
d
)
B
n
n
i
n
g
s
t
e
d
t
A
s
h
a
u
s
e
n
O
c
h
s
e
n
z
o
l
l
F
u
h
l
s
b
t
t
e
l

N
o
r
d
K
l
e
i
n

B
o
r
s
t
e
l
K
a
l
t
e
n
k
i
r
c
h
e
n

S
d
N
t
z
e
n
B
a
d

B
r
a
m
s
t
e
d
t

K
u
r
h
a
u
s
B
o
o
s
t
e
d
t
R
i
c
h
t
w
e
g
G
a
r
s
t
e
d
t
K
i
w
i
t
t
s
m
o
o
r
L
a
n
g
e
n
h
o
r
n

N
o
r
d
L
a
n
g
e
n
h
o
r
n

M
a
r
k
t
F
u
h
l
s
b
t
t
e
l
E
l
l
e
r
a
u
Q
u
i
c
k
b
o
r
n
M
e
e
s
c
h
e
n
s
e
e
H
r
g
e
n
s
w
e
g
B
u
r
g
w
e
d
e
l
T
a
n
n
e
n
e
c
k
F
r
i
e
d
r
i
c
h
s
g
a
b
e
Q
u
i
c
k
b
o
r
n
e
r

S
t
r
a
e
H
a
s
l
o
h
f
u
r
t
h
H
a
s
l
o
h
Q
u
i
c
k
b
o
r
n

S
d
S
c
h
n
e
l
s
e
n
G
r
o
e
n
a
s
p
e
W
i
e
m
e
r
s
d
o
r
f
B
a
d

B
r
a
m
s
t
e
d
t
L
e
n
t
f
h
r
d
e
n
d
o
d
e
n
h
o
f
K
a
l
t
e
n
k
i
r
c
h
e
n
U
l
z
b
u
r
g

S
d
H
e
n
s
t
e
d
t
-
U
l
z
b
u
r
g
H
o
l
s
t
e
n
t
h
e
r
m
e
D
i
e
b
s
t
e
i
c
h
L
a
n
g
e
n
f
e
l
d
e
E
i
d
e
l
s
t
e
d
t
S
t
e
l
l
i
n
g
e
n
A
r
e
n
a
F
i s
c
h
b
e
k
V
e
d
d
e
l
(
B
a
llin
S
t
a
d
t
)
W
a
n
d
s
b
e
k
e
r
C
h
a
u
s
s
e
e
H
a
m
b
u
r
g
A
i
r
p
o
r
t
(
F
lu
g
h
a
f
e
n
)
G
a
r
t
e
n
h
o
l
z
B
u
r
g
s
t
r
a
e
H
a
m
m
e
r
K
i r
c
h
e
R
a
u
h
e
s
H
a
u
s
H
o
r
n
e
r
R
e
n
n
b
a
h
n
L
e
g
i e
n
s
t
r
a
e
B
i
l
l
s
t
e
d
t
M
e
r
k
e
n
s
t
r
a
e
S
t
e
i n
f
u
r
t
h
e
r
A
l l e
e
M
m
m
e
l
m
a
n
n
s
b
e
r
g
J
u
n
g
f
e
r
n
s
t
i
e
g
H
a
u
p
t
-
b
a
h
n
h
o
f
C
e
n
t
r
a
l
S
t
a
t
i
o
n
N
o
r
d
S
d
S
c
h
n
e
l
l
b
a
h
n
-
/
R
e
g
i
o
n
a
l
v
e
r
k
e
h
r
R
a
p
i
d

T
r
a
n
s
i
t
/
R
e
g
i
o
n
a
l

R
a
i
l
I
n
f
o
s

F
a
h
r
p
l
n
e

S
e
r
v
i
c
e
w
w
w
.
h
v
v
.
d
e

0
4
0
-
1
9

4
4
9
B
u
x
t
e
h
u
d
e
H
a
l
t
e
s
t
e
l
l
e
n
S
c
h
n
e
l
l
b
a
h
n
-
/
R
e
g
i
o
n
a
l
v
e
r
k
e
h
r
R
a
p
i
d
T
r
a
n
s
i
t
-
/
R
e
g
i
o
n
a
l
R
a
i
l
-
S
t
a
t
i
o
n
s
G
r
o
b
e
r
e
i
c
h
s
g
r
e
n
z
e
G
r
e
a
t
e
r
H
a
m
b
u
r
g
b
o
u
n
d
a
r
y
n
i
c
h
t
i
m
H
V
V
n
o
t
i
n
s
i
d
e
H
V
V
n
e
t
w
o
r
k
P
a
r
k
e
n
u
n
d
R
e
i
s
e
n
p
a
r
k
a
n
d
r
i
d
e
N
u
r

z
e
i
t
w
e
i
l
i
g
/
l
i
m
i
t
e
d

s
e
r
v
i
c
e
s
B
a
r
r
i
e
r
e
f
r
e
i
e
r
Z
u
g
a
n
g
z
u
d
e
n
S
c
h
n
e
l
l
b
a
h
n
e
n
I
n
f
o
s
z
u
m
b
a
r
r
i
e
r
e
f
r
e
i
e
n
Z
u
g
a
n
g
z
u
d
e
n
R
e
g
i
o
n
a
l
b
a
h
n
e
n
u
n
t
e
r
T
e
l
. 0
4
0
-
1
9
4
4
9
B
a
r
r
i
e
r
-
f
r
e
e
a
c
c
e
s
s
t
o
t
h
e
r
a
p
i
d
t
r
a
n
s
i
t
t
r
a
i
n
s
I
n
f
o
r
m
a
t
i
o
n
o
n
b
a
r
r
i
e
r
-
f
r
e
e
a
c
c
e
s
s
t
o
t
h
e
r
e
g
i
o
n
a
l
t
r
a
i
n
s
o
n
t
e
l
0
4
0
-
1
9
4
4
9
N
i
c
h
t
a
l
l
e
R
e
g
i
o
n
a
l
v
e
r
k
e
h
r
s
-
z
g
e
h
a
l
t
e
n
a
n
j
e
d
e
m
B
a
h
n
h
o
f
N
o
t
a
l
l
r
e
g
i
o
n
a
l
t
r
a
i
n
s
s
t
o
p
a
t
a
l
l
s
t
a
t
i
o
n
s
F
e
r
n
b
a
h
n
h
o
f
M
a
i
n
-
L
i
n
e
S
t
a
t
i
o
n
Conference Dinner
The PHYSMOD conference dinner will take place on Tuesday evening at a beautiful location on the Al-
ster Lake. Accompanying people are welcome to join, but additional dinner tickets cost 60 e and must be
purchased at registration by Monday afternoon. Diner will start at 18:00, but a group will meet at 17:15
in front of the Geomatikum (Bundesstrasse 55) to walk to dinner (distance: 2km, time: 25 minutes).
Location: Ruderclub Germania an der Alster
Address: Alsterufer 21, 20354 Hamburg
Website: www.alstergastronomie.de
Time & Date: 18:0022:00, Tuesday, 23 August 2011
GEOMATIKUM
RUDERCLUB
DistancefromGeomatikumtoRuderclubabout2km
363

PHYSMOD2011 Proceedings

Transféré par

Informations du document

Description originale:

Copyright

Formats disponibles

Partager ce document

Partager ou intégrer le document

Options de partage

Avez-vous trouvé ce document utile ?

Ce contenu est-il inapproprié ?

Droits d'auteur :

Formats disponibles

PHYSMOD2011 Proceedings

Transféré par

Droits d'auteur :

Formats disponibles

PHYSMOD2011

3-dimensional flow in a volume (V3V)

3-dimensional flow in a plane (Stereo PIV)

2-dimensional flow in a plane (2D PIV)

Scalar measurements in a plane (PLIF)

Stereo PIV + PLIF

-9000 probe, robust

x2/L0=-0.75 x2/L0=-0.5 x2/L0=-0.25

average advection turbulent diffusion production dissipation

: generated mean bulk velocity above the

: generated mean bulk temperature above the

| is 36 C. This is used as the reference temperature difference in this

, by both case-a and case-b well matched

= , where Sn is the spectrum in standard units and f is the normalised

= , the skewness is calculated as

> > < <

> > < <

, = z/, z/H, were used in this study,

expresses viscous length scale calculated as

/H is 0.005 (t: time step). The maximum

/H is 250. In the present LES model, the building Reynolds

, two subset of measurements

is depending on the Reynolds

downstream of the cube

in the cubes wake (Fig. 6) it is even more evident

downstream of the cube

show consistent differences in the range

on the top of the cube

Figure 9: Vertical profile of

(Fig. 14) tend to the

on the top of the cube

downstream of the cube

downstream of the cube

of was found to be within 3%, 20% and 16%

bag for 10 min (to determine the 10-minute average concentration)

) above ground level. The surface roughness comprising five

at the boundary-layer height was set at 5 m s

= 3 degrees were a little larger than the case of

, square root of the free

. Of course any ensemble mean parameter will also

close to the front, up to 0.7U

well after the

at the reattachment. All these studies (Hillier

and the instantaneous

is the dynamic pressure at the

Vous aimerez peut-être aussi