Vous êtes sur la page 1sur 129

Università di Pisa

Facoltà di Scienze Matematiche Fisiche e Naturali


Corso di Laurea Specialistica in Fisica Applicata

Anno Accademico 2007-2008

Tesi di Laurea Specialistica

Traffic model uncertainty


for noise mapping
Incertezza associata all’uso di modelli di traffico
per la redazione di mappe acustiche

Candidato Relatore
ELENA ASCARI Chiarissimo Prof. G.LICITRA
ii

Abstract

After the European Parliament published the Environmental Noise Direc-


tive 2002/49/CE (END) and the implementation by Member States in their
own legislation, they had to use the same evaluation methods to analyse
noise pollution and the same indicators suggested by the END (LDEN end
LN ight ). The aim of the END is an international comparison between Eu-
ropean countries, through strategic noise mapping and action planning.
Moreover noise mapping is nowadays the principal way for Italian public
administration to manage noise pollution and to draw up acoustic mitiga-
tion plans. So municipalities ought to have reliable maps and dynamic maps
easily modifiable, to mirror changes over traffic flow circulation. Following
this approach, this work is focused on setting up Pisa road noise map based
on a traffic model.
Because of lots of inputs data requested, it’s necessary to evaluate accuracy
of final product based upon goodness of inputs. This evaluation has been
carried out by the European commission and officially published in 2007 as
the Good Practice Guide. This paper will apply this guide to verify relia-
bility of suggested accuracy in noise mapping and to evaluate uncertainty
associated with predicted noise levels, when measurements are used to vali-
date mathematical models.
iii

Thanks

I want to sincerely thank all members of U.O. IMREC (Mobility Infrastruc-


tures, Electric and Communication Networks) of Pisa ARPAT department
for help, patience and willingness during my stay: special thanks to architect
C.Chiari for her GIS knowledge, to Grad. M.Reggiani to teach me software
IMMI, to Grad. M.Cerchiai, D.Simonetti and A.Panicucci to make possible
the presentation of strategic map. Moreover very special thanks to Fabrizio
Balsini to teach me and help me using instruments to perform both sound
levels and flow measurements. I want also to thank PhD. G.Memoli always
placed at my disposal for explanations. Finally thank to Prof. G.Licitra
and Prof. P.Gallo which support me and this work.
Contents

1 Introduction 1

2 Objectives 3

3 The NMPB method for road traffic noise 5


3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Guide du Bruit: sound emission DB for light and heavy vehicles 5
3.3 NMPB-Routes-96: meteorological correction . . . . . . . . . . 9
3.4 Source characterization . . . . . . . . . . . . . . . . . . . . . 11
3.5 Attenuations due to propagation . . . . . . . . . . . . . . . . 12

4 Traffic and noise models implementation 14


4.1 IMAGINE project . . . . . . . . . . . . . . . . . . . . . . . . 14
4.1.1 Use of traffic models to evaluate road noise levels . . . 15
4.2 Good Practice Guide . . . . . . . . . . . . . . . . . . . . . . . 20
4.2.1 Accuracy evaluation: toolkits . . . . . . . . . . . . . . 21
4.3 European noise mapping experiences . . . . . . . . . . . . . . 23
4.3.1 Noise mapping of Pamplona agglomeration . . . . . . 23
4.3.2 Noise mapping of Scottish agglomerations . . . . . . . 24
4.3.3 Noise mapping pilot project in Portugal . . . . . . . . 24
4.4 Tuscany case studies . . . . . . . . . . . . . . . . . . . . . . . 25
4.4.1 Florence road noise map . . . . . . . . . . . . . . . . . 25
4.4.2 First noise map of Pisa . . . . . . . . . . . . . . . . . 27
4.4.3 Two wheelers sound emission evaluation . . . . . . . . 33

5 A new approach to traffic assessment 35


5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.2 TransCAD characteristics . . . . . . . . . . . . . . . . . . . . 36
5.3 User Equilibrium Method . . . . . . . . . . . . . . . . . . . . 37
5.4 Frank Wolfe algorithm . . . . . . . . . . . . . . . . . . . . . . 40
5.5 Transport network . . . . . . . . . . . . . . . . . . . . . . . . 41
5.5.1 Data collected from first step of TransCAD utilization 43
5.5.2 Road classification and input data . . . . . . . . . . . 43
5.5.3 Intersection delays . . . . . . . . . . . . . . . . . . . . 45

iv
CONTENTS v

5.6 OD matrix calculation . . . . . . . . . . . . . . . . . . . . . . 46


5.6.1 Sample counts . . . . . . . . . . . . . . . . . . . . . . 47
5.6.2 Equivalent vehicles . . . . . . . . . . . . . . . . . . . . 49
5.7 Flow and speed network assignment . . . . . . . . . . . . . . 50
5.8 Model validation . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.9 Uncertainty evaluation . . . . . . . . . . . . . . . . . . . . . . 52

6 TransCAD traffic output elaborations 55


6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Spatial resolution . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.3 Temporal resolution . . . . . . . . . . . . . . . . . . . . . . . 57
6.4 Traffic data resolution . . . . . . . . . . . . . . . . . . . . . . 58
6.4.1 From equivalent vehicles to real vehicles . . . . . . . . 58
6.4.2 From real vehicles to NMPB light and heavy vehicles . 59
6.5 Accuracy of correction coefficients . . . . . . . . . . . . . . . 60

7 Noise model implementation 62


7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
7.2 3D model implementation . . . . . . . . . . . . . . . . . . . . 62
7.2.1 Digital Terrain Model DTM . . . . . . . . . . . . . . . 63
7.2.2 Bridges and viaducts . . . . . . . . . . . . . . . . . . . 63
7.2.3 Sound barriers and walls . . . . . . . . . . . . . . . . . 64
7.2.4 Buildings and streets: GoogleEarth utilization . . . . 64
7.3 Calculation settings . . . . . . . . . . . . . . . . . . . . . . . 68
7.3.1 Source settings and meteorological conditions . . . . . 68
7.3.2 Propagation: reflections and absorption coefficients . . 69
7.3.3 Automated distributed calculation: segmentation . . . 69
7.3.4 Grid resolution . . . . . . . . . . . . . . . . . . . . . . 70
7.4 Façade calculation and population exposure . . . . . . . . . . 72
7.5 Accuracy evaluation . . . . . . . . . . . . . . . . . . . . . . . 72

8 Noise mapping results 74


8.1 Noise road map . . . . . . . . . . . . . . . . . . . . . . . . . . 74
8.2 Population exposure to road noise . . . . . . . . . . . . . . . 77
8.3 Accuracy results . . . . . . . . . . . . . . . . . . . . . . . . . 79
8.3.1 Theoretical accuracy: global uncertainty calculation . 79
8.3.2 Available measurements reliability . . . . . . . . . . . 80
8.3.3 Residuals distributions . . . . . . . . . . . . . . . . . . 82
8.4 Comparison with previous map results . . . . . . . . . . . . . 85
8.5 Strategic noise map . . . . . . . . . . . . . . . . . . . . . . . . 87
8.5.1 People exposure to global levels . . . . . . . . . . . . . 90
8.5.2 Conflicts maps . . . . . . . . . . . . . . . . . . . . . . 90

9 Conclusions and developments 97


CONTENTS vi

A Acoustics basics 99

B Road noise maps 102

C Strategic noise maps 110


Chapter 1

Introduction

After the European Parliament published the Environmental Noise Direc-


tive 2002/49/EC (hereinafter END, [1]) and the implementation by Member
States in their own legislation, they had to use the same evaluation meth-
ods to analyse noise pollution and the same indicators suggested by the
END (LDEN and LN ight ). The aim of the END is an international compar-
ison between European countries especially using strategic noise maps1 and
action plans2 . Member States have to ensure that maps and action plans
will be write up in these different situations: agglomerations with more than
250.000 inhabitants, major roads with more than 6 millions vehicles per year,
major railways with more than 60.000 trains per year and major airports
with more than 50.000 movements per year (maps no later than 30/6/07,
plans no later than 18/7/08); agglomerations with more than 100.000 in-
habitants, major roads with more than 3 millions vehicles per year, major
railways with more than 30.000 trains per year (maps no later than 30/6/12,
plans no later than 18/7/13).
These dead lines have been adopted in Italy by the D.L. n.194 19/8/05 which
transposes the END, establishing endorsements for defaulting authorities [2].
However, noise mapping is nowadays the principal way for public adminis-
trations to manage noise pollution even in context different from the ones
imposed by the END: in fact, an acoustic map is not only a photo of the
acoustic conditions of a single infrastructure, but also a dynamic instrument
for urban planning.
Especially in Italy, strategic noise mapping is the first step of Italian ac-
tion plans (hereinafter PCRA) which must be drawn up by municipalities
in order to attempt DM Ambiente 29/11/00, [3]. Authorities have to verify
if noise levels exceed limits established by local acoustic classification plans
(hereinafter PCCA) defined in DPCM 14/11/97, [4]. Therefore, using re-
1
A map designed for the global assessment of noise exposure in a given area due to all
sources.
2
Plans designed no manage noise issues and effects, including noise reduction.

1
CHAPTER 1. INTRODUCTION 2

sults of an acoustic map, authorities may set up strategic measures which


could modify inhabitants life in terms of streets furniture (e.g. barriers,
absorbent asphalts. . . ) or circulation changes (e.g. limited traffic zones,
lowering of speed limits, traffic circles. . . ).
On the other hand, municipalities ought to have reliable maps and at the
same time dynamic maps easily modifiable to mirror changes in traffic flow
circulation. Following this approach, this thesis is focused on setting up Pisa
noise map based on a traffic model.
However, it’s always been critical to map very large zones as an entire mu-
nicipality, not only because calculation models may take long run time, but
also because we need to collect lots of input data not always available.
We have to do many assumptions and to estimate as many parameters to
carry out acoustic maps (and more have to be done using traffic model):
therefore, we evaluated goodness of final product based on detail and accu-
racy of inputs, according evaluation carried out by the European commis-
sion and officially published in 2007 as the Good Practice Guide (hereinafter
GPG)[5]. This document is fundamental for this work because it provides
accuracy for each modelling choice.
This thesis will present the noise map for road source in the municipality
of Pisa using a traffic model and it will evaluate the uncertainty associated
with estimated noise levels: this work is continuous with previous acoustic
map of Pisa written up in 2006 [6]. Although it wants to improve the level
of deepening and reliability making a different assignment of traffic flows;
this approach should be an improvement of the assignment based on road
classification (fixed flow for each class) used for the first map and it’s fore-
casted by GPG [5] and suggested also in [6].
Both software TransCAD (traffic flows calculation) and IMMI (noise level
calculation) will be presented with special attention to methods implemented
and to all elaborations needed to use the output of the first as input of the
second. Moreover, this work is part of the broader project of strategic noise
map: so results of strategic map will be presented taking into account also
railway and airport noise.
Strategic noise map is a project of regional environmental agency ARPAT
carried on in Pisa department by U.O. IMREC (Mobility Infrastructures,
Electric and Communication Networks Unit): therefore this work has been
done in ARPAT department.
Chapter 2

Objectives

The aim of this study is to write up the road noise map of Pisa municipality
using a traffic model for vehicles flows calculation and its validation through
measurements to verify reliability of accuracy suggested by GPG.
Therefore, we want to evaluate uncertainty associated with residuals be-
tween measurements and calculated levels to show accuracy of used mod-
elling methods.
First of all, we have to estimate vehicles flows on the whole municipality,
that is:

• create town road network including principal roads;

• measure flows on sample roads to calibrate model;

• assign traffic with TransCAD;

• measure flows on sample roads to validate model;

• insert roads not included in traffic assignment;

• evaluate traffic flows on these roads (based on previous map taking


into account recent modifications).

After having number of vehicles per road, we need to estimate sound power
levels: unfortunately, noise emission model imposed by the END establishes
sound power level for two vehicles categories which don’t correspond to cal-
culated vehicles1 . Therefore, we need to elaborate data before being inserted
in sound emission model: disaggregation of results in categories requested
by French sound emission model NMPB (official method, see D.L. 19/8/05)
is performed together with sound levels measurements related to traffic flows
ones to verify the reliability of disaggregation itself.
The next step has been the creation of an IMMI project including necessary
cartography to set up a digital terrain model. Moreover, 3D model has been
1
Causes of this mismatch are fully explained in following chapters.

3
CHAPTER 2. OBJECTIVES 4

prepared through 3D views and eye inspections: we used georeferred photos


from GoogleEarth imported by the software and 3D visualization on Live
Search Map website. Then sound levels calculation has been done with a
5 m step grid through which we assign sound façade levels to residential
buildings with Italian and European indicators.
After that, results have been evaluated performing

• comparison between available measurements and calculated levels to


estimate accuracy;

• evaluation of theoretical accuracy from GPG and other considerations


not included in that paper;

• comparison between theoretical and real accuracy.

Moreover, thanks to population data of 2001 census, it has been possible to


calculate people exposure to road noise levels and the global exposure from
the analysis of strategic noise map.

Before going through each step of this project, we want to explore the state
of art of noise mapping and international documents that must be taken
into account approaching noise mapping according the END:

• The Guide du Bruit and the French method NMPB indicated in the
END as the official method for road traffic noise;

• The IMAGINE project (Improved Methods for the Assessment of


the Generic Impact of Noise in the Environment, see www.imagine-
project.org): this project analysed different techniques of noise map-
ping, included the ones with traffic model, emphasizing general diffi-
culties;

• The Good Practice Guide ;

• European mapping experiences;

• Road noise map in the city of Florence published in February 2008 by


ARPAT and correlated studies on sound emission levels of different
categories of vehicles;

• Previous noise map of Pisa;

• ARPAT study on sound emission of two wheelers vehicles carried out


during evaluation of acoustic clime in Pisa.
Chapter 3

The NMPB method for road


traffic noise

3.1 Introduction
The calculation model for road traffic noise definitively adopted by the END
is the French official method NMPB-Routes-96 (SETRA-CERTU-LCPC-
CSTB), cited in Arrêté du 5 mai 1995 relatif au bruit des infrastructures
routières, Journal Officiel du 10 mai 1995, article 6 and in French law XPS
31-133 and successively adjusted to European indicators in commission
recommendation 2003/613/EC. The NMPB (Nouvelle Méthode de Prévision
du Bruit, [7]) takes into account meteorological effects on long distance
propagation: so it’s useful for modelling big infrastructures (as the ones to
be mapped according the END) in free field propagation conditions (its limit
of application is 800 m far from the source). However, the critical problem
of this model is the sound emission database: in fact, considered vehicles
categories are not always suitable to all European countries (two wheelers
are not considered) and it’s not at all updated because the database is the
same of the Guide du Bruit of 1980 [8]. In next section the sound emission
database is presented.

3.2 Guide du Bruit: sound emission DB for light


and heavy vehicles
Vehicles are treated as a point isotropic source 80 cm height above road
line: point approximation is good for almost all situations but we have to
pay attention when barriers are installed near the source (in this case ad hoc
calculation method has been set up); the isotropic approximation is broken
only by very big heavy vehicles.
With this kind of model, the vehicle sound power W is correlated to the

5
CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 6

sound pressure p through next expression:


p2rms
W = 2πr2
ρ0 c
in which ρ0 is air density, c is sound speed and r is the distance from the
source. The sound power has to be compared to the reference power source
(10−12 Watt) emitted on a sphere so the sound power level is given by:
LW = Lp + 20 log r + 8
The level above is referred to a single vehicle emitting in a semi-sphere
but we can calculate the sound power level for units length considering the
number of vehicles per hour Q and the averaged speed v expressed in km/h:
 
Q
(LW )m = LW + 10 log
1000v
in which LW is the sound power level associated to a point source whose
equivalent length will be fully described in section 3.4.
Database values are emission levels, that is the sound pressure level of a
single passage in an hour measured 30 m far from the source and at 10 m
above road surface.
Sound pressure must be integrated in the referred time interval (i.e. an
hour):
1 t2 W ρo c
Z
 2 t2
p t1 = dt
T t1 2πr2
This relationship can be rewritten considering a point source moving at
speed v and at a distance d under an angle of view θ:
 2 t2 1 W ρo c θ
p t1 =
T 2π d · v
So we can write:
[Leq ]tt21 = LW − 10 log(d · v) − 8 − 10 log T + 10 log θ
Now, using an hour as time interval and π as angle, we obtain that the
equivalent hourly level of a single passage is given by:
Lh,eq = LW − 10 log(d · v) − 38
and for Q vehicles:
Lh,eq = LW − 10 log(d · v) − 38 + 10 log Q
Therefore, if we take into account the reference distance (and light air ab-
sorption) and speed in km/h, the sound emission level is defined by this
equation:
E = LW − 10 log v − 50 = (LW )m − 20
CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 7

These emission values are given by the Guide du Bruite for two vehicles
class, for different type of circulation and type of slope of the road. Vehicles
classes are:
• light vehicles: under 3.5 ton full-load;
• heavy vehicles: over 3.5 ton full-load.
Diversification of circulation type is based on average acceleration:
• fluid and continuous: vehicles number is constant on time and space,
there aren’t accelerations;
• pulsed continuous: vehicles number and speed vary along time al-
though it’s possible to define an average speed;
• pulsed accelerating: majority of vehicles is accelerating;
• pulsed decelerating: majority of vehicles is decelerating.
Types of slopes are:
• horizontal: ramp with inclination under 2%;
• ascending: ramp with inclination over 2% in ascending direction;
• descending: ramp with inclination over 2% in descending direction.
In figure 3.1 the noise emission database from [8] is represented and in the
following table we show values for some typical speeds and for all types of
circulation in an urban context (emission values are given for an horizontal
road).

Table 3.1: Emission Levels in urban context


Circulation speed E light vehicles E heavy vehicles
Type [km/h] [dBA] [dBA]
Fluid Continuous 30-50 29.5 44.0
50-70 30.5 42.5
60-80 32.0 43.0
Pulsed continuous 30-60 31.5 43.5
Pulsed accelerating low 37.0 47.0
high 33.0 43.0
Pulsed decelerating 29.0 36.0 - 38.0

Finally, we define emission of Ql light vehicles (EQl ) and of Qp heavy ones


(EQp ):

EQl = El + 10 log Ql
EQp = Ep + 10 log Qp
CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 8

Figure 3.1: Noise emission database [8]


CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 9

3.3 NMPB-Routes-96: meteorological correction


In previous sections we explained that innovation of this method is how
meteorological effects are treated. Acoustically speaking, meteorological
conditions are divided in three classes:

1. homogeneous conditions: sound energy propagates along straight lines;

2. conditions favourable to sound propagation: sound energy get down


toward ground giving much noise at receivers;

3. conditions unfavourable to sound propagation: sound energy rise to-


ward the sky giving less noise at receivers.

NMPB considers only first and second conditions, not only because the third
is difficult to calculate, but also because this assumption overestimates levels
and so it’s safer.
The origin of meteorological effects on propagation is due to combination
of thermal gradient with aerodynamic factors of wind directions; in homo-
geneous conditions these factors balance their selves, instead in favourable
ones acoustic rays go downwards.
In fact, favourable condition occurs when wind direction is the same as
propagation one and when thermal gradient is positive (hot air is up) that
means sound speed increases with distance from ground (c ∼ = 331.6 + 0.6Tc ,
in which Tc is temperature in Celsius degrees).
We obtain the effect shown in figure 3.2.

Figure 3.2: Favourable Conditions [9]

The model calculates sound level separately for each meteorological condi-
tion, obtaining LF for favourable conditions and LH for homogeneous ones;
after that, long term level is estimated by:
LF LH
h i
LLT = 10 log p10 10 + (1 − p)10 10

in which p is favourable conditions probability.


The NMPB [7] provides this probability for some French cities and suggests
CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 10

to use 50% for day period and 100% for night period for all places where me-
teorological databases are not available; that values of probability descends
from the observation of night temperature inversion.
Calculation process implemented by NMPB is described in figure 3.3.

Figure 3.3: General flow chart of NMPB method


CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 11

3.4 Source characterization


Road source is represented by many point sources located along the centre
line. Sectioning a road infrastructure in point sources needs to identify
acoustic homogeneous arcs: that means to define arcs as road sections with
the same geometrical profiles and with constant traffic flows. Each arc is
then split up into point sources to assign sound power. This splitting up
may be done in different ways:
• Equiangular splitting up1 : the site is scanned from the considered
receiver point by a group of rays whose angle step is constant (the
more a receiver is close to source the more the step is small) and at
each intersection of one of these rays with a source line, a point source
is placed;
• Splitting up with a constant step: each source line is split up into
point sources regularly spaced out (the step between two consecutive
sources does not have to be greater than half the orthogonal distance
between the lane and the closest receiver point and the value of the
step shouldn’t be greater than 20 m)
• Variable splitting up: as the first method but with local variation of
angular step;
Finally, the sound power level for octave band j of each point source is given
by:  
EQ EQ
l p
LAW i = 10 log 10 10 + 10 10 + 20 + 10 log li + R(j)

in which li is length (in meters) of road section represented by the current


point source i and expressed in figure 3.4;

Figure 3.4: Length of road section represented by point source i

and in which R(j) is road noise normalized A-weighted spectrum given in


the next table:

1
This method is the once implemented by IMMI.
CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 12

Table 3.2: Road noise Spectrum


j frequency [Hz] R(j) [dBA]
1 125 -14
2 250 -10
3 500 -7
4 1000 -4
5 2000 -7
6 4000 -12

3.5 Attenuations due to propagation


Sound power level of a point source has to be expressed as sound pressure
level: to do this we have to take into account all attenuations of propagation.
So, we have to calculate a different level for homogeneous and favourable
conditions:

Li,F = LAW i − (Adiv + Aatm + As,F + Adif,F )
Li,H = LAW i − (Adiv + Aatm + As,H + Adif,H )

Values of each source are then long term evaluated and added above all
sound trajectories and octave bands.
We notice that geometrical attenuation Adiv 2 and atmospheric absorption
do not change with propagation conditions3 and are expressed by:

Adiv = 20 log d + 11
Aatm = α(j)d/1000

in which d is distance in a direct line between source and receiver, α is air


absorption coefficient expressed in dB/km:

j frequency [Hz] α(j) [dB/km]


1 125 0.38
2 250 1.13
3 500 2.36
4 1000 4.08
5 2000 8.75
6 4000 26.4

All other attenuations are calculated in different ways for each meteorolog-
ical condition.
In favourable conditions, to determine ground surface attenuation As,F , we
2
It’s divergence attenuation of an isotropic source.
3
They do not change with sound ray direction but the atmospheric absorption coeffi-
cient is given for a specific temperature (15◦ C) and for humidity of 70%.
CHAPTER 3. THE NMPB METHOD FOR ROAD TRAFFIC NOISE 13

need to distinguish three zones of propagation: close to the source, inter-


mediate, close to the receiver. Attenuation will be calculated as a sum of
three contributions considering that in central zone rays are less influenced
by surface attenuation. In homogeneous conditions rays are straight so we
don’t need to section zones. Formulas could be found in [7], however As
depends on ground surface absorption G.
Attenuation due to diffraction is caused by the lengthening of trajectory
compared with direct line one. Moreover, diffracted trajectory is influenced
even by ground absorption: in this case we have to separate calculation be-
cause a diffracted trajectory in homogeneous conditions could be a direct
trajectory in favourable conditions. In fact, ray curvature could make visible
a source and a receiver which couldn’t be in a straight trajectory (see figure
3.5). Therefore, calculation of diffraction for favourable conditions consid-
ers an equivalent height for barriers which is defined based on ray curvature.

Figure 3.5: Diffracted trajectory in favourable and homogeneous conditions


Chapter 4

Traffic and noise models


implementation

4.1 IMAGINE project


IMAGINE project (Improved Methods for the Assessment of the Generic
Impact of Noise in the Environment) was developed between 2003 and 2007:
it was born as a scientific instrument of environmental policies support to
Member States as a natural prosecution of HARMONOISE project. In this
previous project harmonized emission models were developed for the entire
Europe regarding railway and road noise pollution. IMAGINE’s aim was to
produce the same methods to estimate airport and industrial noise requested
by the END. Therefore, the common aim of both projects was to produce
harmonized methods for the implementation of noise mapping in Europe.
Project involved many partners: environmental agencies, research centres,
infrastructures administrators, cars producers etc. and it was divided in
different working packages with the following aims (see [10]):

1. produce guidelines for noise map data management (Work Package 1);

2. produce guidelines and examples for an efficient link between road cir-
culation management and mapping for action plans (Work Package 2);

3. produce guidelines and examples of how and when measurements are


useful to add to reliability of estimated noise levels (Work Package 3);

4. produce harmonized methods for airport noise (Work Package 4);

5. produce noise emission database for different vehicles class based on


HARMONOISE methods and guidelines for vehicles not in standard
classification (Work Package 5);

14
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION15

6. produce noise emission database for passenger and freight trains based
on HARMONOISE methods and guidelines for vehicles not in standard
classification (Work Package 6);

7. produce harmonized methods for industrial noise (Work Package 7);

8. make possible an easy and fast implementation of methods above


(Work Package 8).
This project tried to be a complete guideline to develop methods alterna-
tive to the ones suggested by the END: in fact, END methods are national
models that are not always suitable to all European countries, anyway they
represent the standard by law. So, IMAGINE methods are nowadays not
implementable but the analysis carried out by working groups is still useful,
even for this work with particular regard to comparison between mapping
methods.
The work package WP2 is very interesting because of having reviewed all
kind of available traffic models and tested the capability of each one.

4.1.1 Use of traffic models to evaluate road noise levels


Deliverable 7 of IMAGINE project [11] provides guidelines for traffic mod-
elling and indications about theoretical accuracy due to this kind of method.
Principal parameters affecting goodness of traffic model are: traffic flow (i.e.
number of vehicles), speed, speed distribution (how speed varies within the
same vehicle class), accelerations and fleet composition (number of heavy
vehicles). However, these parameters haven’t the same effect on sound lev-
els: if a doubling of flow is needed to raise levels of 3 dB, it’s sufficient a
variation of average speed of 30 km/h to have the same increasing on levels.
Therefore, not all parameters have to be known with the same accuracy to
obtain a fixed uncertainty; the following significance order is given (i.e. the
order to be proceeded to improve accuracy):
1. speed and fleet composition;

2. flows;

3. accelerations and decelerations;

4. speed distribution;

5. data on low flow roads.


These observations make easy to understand that a traffic model could im-
prove speed accuracy: in fact, speed is usually taken equally to speed limits
(not including congestions).
Furthermore, IMAGINE WP2 carried out a Montecarlo simulation to evalu-
ate which precision is necessary on speed and flow to achieve 0.5 dB or 1 dB
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION16

uncertainty on noise levels1 . In figure 4.2 simulation results are listed.


Apart from accuracy of input data, there are different traffic models that
are suitable to different situations and that differ on data requested and pro-
vided; we exclude demand models (based on survey about travellers charac-
teristics and choices) and we concentrate on different network models which
provide speed and flow on each link of transport network. These kinds of
models are so classified:

• static assignment models;

• dynamic assignment models;

• continuous models;

• microsimulations models.

In document [13], produced by WP2, methods have been described through


SWOT analysis (Strengths, Weaknesses, Opportunities, Threats) to under-
stand which one fits better data available (or deliverable), aim or time of
the study. In the following sections, models characteristics will be briefly
presented.

Static traffic assignment


Static traffic assignment is the process of allocating trips in one or more trip
matrices (origin-destination matrices or OD-matrices) to their routes (paths)
in the network, resulting in flows on links. After calculation, it’s possible to
obtain reasonable link flows and speed and also to identify congested links.
The basic assumption is that travellers choose the route that minimize their
travel cost (cost and time are directly correlated); however, this method
is rarely able to distinguish vehicles classes, so after the assignment it’s
necessary a class distribution on roads to produce vehicles categories.
Moreover, static assignment is generally hour based: the origin-destination
matrix contains for instance the trips of a peak hour. Capacities of the
road network are expressed as number of vehicles per hour. As a result, the
estimated flows are averaged per hour and so we need a daily distribution
to estimate noise indicators.

1
Method used to produce noise from vehicles flows is HARMONOISE’s one, [12].
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION17

Table 4.1: SWOT analysis for static assignment


Strengths Weaknesses
Fast; All results are averages;
Data collection relatively easy; Inaccuracy of results
Easy-to-understand indicators; (especially speeds per link);
Concise results; Usually night periods aren’t
Results easy to use in GIS; modelled;
Opportunities Threats
Very common in local, regional Improvement of output for
authorities; noise calculations requires
More digitized or automatically a large effort;
generated data will become available; Model results can cause a
Relative simplicity of models ensures false sense of accuracy: results
that new developments will usually seem detailed, but not
be tried out in static models first; all indicators are significant;

Dynamic traffic assignment


Dynamic traffic assignment is the process of allocating trips in function of
time in order to achieve time distribution of flows on network. Put into
practice, it means that instead of a single OD-matrix we have one for each
time interval (an hour or a quarter) that we want to analyse. Many models
implementing dynamic assignment request that the temporal step of analysis
is smaller than the shortest travel time over all network links (this means
long run time in urban context).
This assignment produces flow and speed as function of time and it makes
also possible to distinguish between different vehicles classes.

Table 4.2: SWOT analysis for dynamic assignment


Strengths Weaknesses
Correct modelling of demand Long run time;
fluctuations; Few dynamic matrices for
Possibility to model more accurately full day available;
and correctly effects like congestion; Inaccuracy of results
Include multiple vehicle types; Times step size determines
Results easy to use in GIS; accuracy of results;
Opportunities Threats
DTA models offer possibilities in high It needs detailed input data;
demand in impact assessment studies;
More digitised or automatically Difficult modelling technique;
generated will become available;
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION18

Continuous models
Continuous models treat vehicles as a unique entity idealized as a continuous
fluid. Mathematical idea is that fluid delivery (i.e. flow per time reference)
increases till maximum capacity of pipe (i.e. of road) and then decreases as
shown in figure 4.1.
This model is based on mathematical equations (solved sometimes numeri-
cally) to produce flow, speed and density through time (time step from 0.5 s
to 10 s). This type of analysis has an high level of detail and it’s common
in circulation management and not in noise studies: in fact, this model is
suitable for long crowded roads because it doesn’t consider nodes.

Figure 4.1: Fundamental diagram of traffic flow (k is density; q is flow)

Microsimulation models
Microsimulation models attempts to model the progression of individual ve-
hicles: within each time step it uses a number of individual algorithms to
generate decisions for all vehicles on the network. Position and speed of
vehicles are updated at each step. There are two kinds of microsimulation
models, the ones which consider a continuous network and the others that
consider it as discrete (in this case the road could be occupied only if another
vehicle is on the road).
These models require more input data about network itself as signalized in-
tersection data, numbers of lanes, width etc., so they are suitable to little
local studies.

After this short review, it’s clear that some models fit better than others
to calculation of a specific parameter and that not all models are suitable
to urban noise mapping; in the following tables, properties of models are
compared.
Apart from observations about models types, some problems remain critical:
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION19

models often calculate single vehicles class so we have to pay attention to


fleet composition; models are not reliable on low flow roads so we must
decide how to treat them. These difficulties are greatest problems linking
traffic model to noise model: that’s why WP2, in analogy with European
Commission paper GPG, has written up a lot of toolkits to know previewed
level of accuracy2 associated with a modelling choice.
These toolkits will be used together with table in figure 4.2 to estimate
uncertainty due to traffic model on noise levels calculated in Pisa.

Table 4.3: Models & Output


Static Dynamic Continuous Micro simulators
Traffic Flows + ++ - +/-
Speed + + ++ ++
Speed Distribution - + + ++
Acceleration - - + ++
Fleet composition +/- +/- +/- +

++ available and reliable


+ available
- not available

Table 4.4: Models & Contest and reference Time


Static Dynamic Continuous Micro simulators
Study Area
Regional/National ++ + N N
City + ++ N +
Local motorways N + ++ +
Local urban N 0 0 ++
Study period
Peak Hour ++ + + +
Day + ++ 0 +
Year 0 0 N N

++ available and reliable


+ available
0 Neutral
N not available/not common practice

2
In this case is not a decibel accuracy but a quality evaluation.
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION20

Figure 4.2: Montecarlo simulation results - boundary regions of accuracy on


flow and speed to achieve 0.5 dB and 1 dB accuracy on sound power levels

4.2 Good Practice Guide


This document was written up by the European Commission working group
Assessment of Exposure to Noise (WG-AEN) in a first draft in 2003 and
its final version was published in August 2007 after a long international
consultation process. The purpose of this Position Paper is to help Member
States and their competent authorities to undertake noise mapping and to
produce the associated data required by the END3 .
This Position Paper dials with all possible problems applying the END and
it tackles them with toolkits which suggest solutions and estimate associated
accuracy.
3
It’s especially oriented to the first round of sources to be mapped.
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION21

4.2.1 Accuracy evaluation: toolkits


This paper guides authorities through noise mapping suggesting which mod-
elling technique must be used to achieve a certain level of knowledge and
accuracy: GPG carries on a deep discussion about all kinds of choices about
the END requirements. In the toolkits (double-entries tables) different possi-
bilities are listed with associated accuracy, degree of deepening and complex-
ity (i.e. technical effort needed). However, not all tools evaluate accuracy
in the same way: almost all evaluate accuracy in terms of decibel but some
of them express more generally quality of the result. Legend of a general
toolkit is shown in figure 4.3.

Figure 4.3: Toolkits Legend

GPG suggests single steps accuracy but accuracy of final result is anyway
tricky. In fact, uncertainty of each solution have to be combined to obtain
total uncertainty but it’s not always known how them are related: they
could add or subtract each others, besides it’s common practice to admit
data belonging to independent Gaussian distributions and to do a square
sum. However, this practice is not indicated in GPG which purpose is only
to give a scale of goodness for each technique. We will calculate square sum
but we will also carry on a comparison between measurements and calcu-
lated levels to determine discards distribution.
GPG toolkits are shown in figure 4.4: each one is split in different tools
according to available data. These tools will be taken into account for un-
certainty calculation of Pisa road noise map.
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION22

Figure 4.4: Good Practice Guide step by step


CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION23

4.3 European noise mapping experiences


4.3.1 Noise mapping of Pamplona agglomeration
Pamplona agglomeration includes 20 municipalities and it’s part of first
round noise mapping because it takes up an area of 127 km2 with a to-
tal population of 280200 inhabitants. According to the requirements of the
END, strategic map has been performed including industrial sources, air-
craft, railway and a total of 7441 roads [14].
The area was very large so lots of iso-lines, curve lines and about 40000
elevation points were used to build Digital Terrain Model.
Calculation was performed with software Cadna/A with a 10 m step grid (a
smaller step was impossible because of width of the area). Other technical
settings were:

• only first order reflections considered;

• maximum action radius of sources: 2 km;

• building absorption 1 dB;

• ground absorption G = 0.4.

Input data for road acoustic map have been taken from national agency
(only main roads) based both on measurements and predictions; data in-
cluded both average speed and traffic composition. For the rest of not
considered roads several criteria were implemented. One of them was to
extrapolate from measurements to other similar roads. Another criterion
was to evaluate the traffic density as function of the density of population
to which the road is serving.
In some cases great deviations were found but a pragmatic decision was
finally taken and acceptable correlations with measurements were found.
Average values of Hourly Average Intensity (HAI) expressed as vehicles per
hour (v/h) for the six types of roads were as follows:

Table 4.5: Pamplona road classification


Type Name HAI(day)
1 Outlying roads 10 v/h
2 Quiet Residential Areas 50 v/h
3 Residential 100 v/h
4 Residential-Commercial 200 v/h
5 Commercial 350 v/h
6 Industrial 200 v/h

The hourly distribution of traffic for day, evening and night periods were
obtained from data continuously recorded at 66 measurement stations. The
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION24

percentages of heavy vehicles for different types of roads were obtained from
real measurements in several roads and subsequent extrapolation to similar
types of roads. Several noise measurements were carried out with the aim
to configure the calculation parameters according to NMPB model, mainly
to adjust the type of asphalts. Correlation between predicted and measured
values was quite high (differences were less than 1 dB for receiver points
near to emission lines). Nevertheless, differences increased with distance.
Results showed percentage of people exposed to high levels: on day period
9% and 37% population are exposed to levels bands of 55-65 dB and 50-
60 dB respectively and 13% and 44% during night period.

4.3.2 Noise mapping of Scottish agglomerations


Scottish government disposed maps of Edinburgh and Glasgow agglomera-
tions and major roads: these maps cover a very large area (only Glasgow
agglomeration area is 766 km2 ) and include roads with more than 1000 ve-
hicles daily passages. A specific traffic model, TMfS (Transport Model for
Scotland) was implemented to perform calculation [15]: TMfS is a multi
modal traffic model which provides peak hour flows for three periods (AM,
intermediate, PM) with heavy goods vehicles percentage. Instead, streets
with low flows inside agglomerations have been considered apart, assigning
a typical flow. They implemented an automatic procedure based on arcs
and nodes recognition to adapt network links of TMfS to real cartography;
also road width has been automatically detected from cartographic data.
Road surface and speeds have been taken from national database and cor-
rected by local authorities.
Digital terrain model was calculated by steps [16], considering both two
dimensional cartography and LIDAR information about height: cartogra-
phy resolution was 5 m (i.e. based on 1:10.000 cartography); instead height
information was given rounded to meter. Moreover, they assigned a fixed
minimum height of 5 m for all buildings.
Regarding ground absorption they used information about rural areas to
define hard ground.
Scottish government (and whole United Kingdom) made a strong effort to
adapt their indicators to European ones and to update maps already done;
it’s nowadays managing action plans to be drawn up according the END.

4.3.3 Noise mapping pilot project in Portugal


Portugal environmental agency Apambiente carried out in 2004 a study
about noise mapping in two different areas [17]: Carregado zone (Alenquer
municipality and suburbs) and urban area Linda-a-Pastora in the munici-
pality of Oeiras. Noise estimation was performed with software MITHRA
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION25

using NMPB method: meteorological conditions were deduced from local


data, so probability of favourable conditions was taken as 30%.
Different input/output resolution was chosen for each scenario: municipal-
ity scale scenario was reproduced with 1:10000 cartography and noise grid
step was set to 18 m; instead urban scenario used 1:2000 cartography and
an 8 m grid step. Moreover, municipality scenario considered second order
reflections and urban one considers third order reflections.
All road sources were considered with an action radius of 2 km and traffic
data (flow and heavy vehicles percentage) were produced with automatic
counters (from local highway administrator), manual counts and by com-
paratives analysis. Speed was estimated over all roads.
A validation of noise maps was performed: average absolute differences be-
tween measured levels and estimated ones was 1.5 dB (with maximum de-
viation of 2.9 dB) over municipality scenario and 1 dB (with maximum
deviation of 2.2 dB) over urban scenario.
This good result is partially due to reduced number of considered streets
(less than 50) and smallness of area. Furthermore, the area counts about
1650 inhabitants that are not very annoyed (27% are affected by diurnal
levels higher than 60 dB and nocturnal ones higher than 55 dB).
Nevertheless, this project shows Portugal willingness of improving environ-
mental noise policies which didn’t exist before European Environmental
Noise Directive (see [18]).

4.4 Tuscany case studies


4.4.1 Florence road noise map
Published in February 2008, noise mapping of Florence municipality followed
an agreement between municipality, province and ARPAT regarding noise
façade calculation to provide PCRA. This study was carried out by U.O.
IMREC of Florence ARPAT department and financed by municipality and
province.

Traffic flow evaluation method


Traffic assignment was done through road classification (according PUT,
Urban Traffic Plan) and using a standard flow per road class. Public Trans-
port buses flow was assigned to links according data given by local transport
administrator ATAF.
Standard flow on links was estimated from following data:

• automatic counts at 16 city gates (data from SILFI);

• automatic counts at 16 ZTL gates (Limited Traffic Zones);


CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION26

• counts campaign carried on by ARPAT on 25 streets (time period


48h);

• manual counts on ZTL streets.

Standard flows are here listed per road class:

Figure 4.5: Traffic flows assignment [19]

Sound Emission of vehicles classes


Sound emission levels, as defined in the official method, don’t correspond to
Italian fleet; therefore, it was necessary to transform real vehicles in light
and heavy ones of NMPB database. Put into practice, this leads to the use
of an acoustic weight of real flows to produce equivalent (noisily speaking)
light and heavy flows.
Weights were been calculated in a previous study carried out by ARPAT
[20]: so the following emission values were assumed.

Table 4.6: Emission values in Florence at 30 m distance and 10 m height


Cars (C) Two wheelers (TW) Buses (B) Heavy vehicles (HGV)
28.6 dB(A) 30.1 dB(A) 37.3 dB(A) 33.4 dB(A)

Then emission values were compared with NMPB ones (supposing 50 km/h
speed) and a mathematical relationship was defined to obtain equivalent
flows: 
Ql,eq = nC ∗ 0.61 + nT W ∗ 0.87
Qp,eq = nB ∗ 0.29 + nHGV ∗ 0.12
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION27

Calculation reliability
Before treating reliability, we want to underline calculation settings: traffic
source was considered as a source on centre line with standard asphalt and
road gradient was supposed null for all roads. Moreover, calculation took
into account first order reflection and 10 m step grid is used. Façade levels
were extrapolated subtracting 3 dB from grid interpolated values.
Accuracy evaluation of diurnal levels was carried out and reported in [21]
comparing results with 47 continuous measurements4 : these validation points
were yet available from older campaigns (1995-2003). Results showed that
only 22 measured points presented a difference from calculated values less
than 3 dB. Main reasons of these errors are problems on 3D model realiza-
tion, measurements reliability (too old to represent current situation) and
mismatches between assigned traffic category and real flow on the road. This
last problem results greater on night levels so that probability of residuals
over 3 dB is enlarged.
Therefore, accuracy of 3 dB was achieved only over 60% of control points.

4.4.2 First noise map of Pisa


First road noise map of Pisa was written up following an agreement with
Pisa municipality for the PCRA. This action plan shall identify buildings
where façade levels are higher than the ones established in the acoustic clas-
sification plan (PCCA) to manage reduction measures according to priority
index defined by DM Ambiente 29/11/2002 in annex 1.
In fact, maps simplify the identification of hot spots (i.e. critical areas) and
the comparison with limits established by law. In September 2006 day lev-
els map was produced[6] and at the beginning of 2007 the night levels map
followed[22].
Moreover, exposed population was evaluated according to END exposure
bands and European indicators (population from 1991 census).

Source characterization
Traffic flow on local network was assigned identifying homogeneous classes
of streets and then a standard flow was assigned. Standard flow was based
on the following data:

• flow on province roads (report 2003, [23]);

• flow from PUT of neighbouring municipality San Giuliano Terme;

• hourly counts carried out by ARPAT (2005-2006);

• routes and scheduling of urban and suburban public transport.


4
Measurements devices without direct control
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION28

Roads classification was based on suggested GPG classification (see tools 2.5,
4.5) and adapted taking into account PUT indications; to estimate night lev-
els, a different coefficient per class was applied. Speed values were estimated
assigning limits by law. Table in figure 4.6 summarizes these classes.

Figure 4.6: Standard Flow by road class

Notice that identification of class 60 (Borgo, Corso Italia, Piazza dei Mira-
coli) and classes 40, 30-32 it’s immediately given by intersecting roads and
polygonal shape of ZTL and 30 km/h zones; instead other classes needed a
comparison with PUT maps. Roads classification is shown in figure 4.7.

Figure 4.7: Roads classifications [6]


CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION29

As like in Florence, this work used an acoustic weighting to obtain NMPB


equivalent light and heavy flow: same weight of ARPAT study [20] was used
regarding cars, two wheelers and buses but not for heavy vehicles. In fact,
heavy vehicles flow was set to zero on urban roads and it wasn’t weighted
on other roads.

Propagation and calculation settings


Sound levels calculation was provided with software IMMI version 5.2 that
implements NMPB method. All streets were inserted in the IMMI project
with a good asphalt; streets were set as double direction (the one-way ones
too) and total flow was placed on centre line.
Terrain model was developed through regional cartography CTR 1:10000
using altitude points, iso-lines and curve lines, bridge altitude lines and hy-
drological lines.
At the same way, walls and buildings were taken from regional cartography:
it was necessary to modify some walls and building whose height was clearly
mistaken (buildings minimum height was set to 3 m).
To manage calculation, simplified method was used which takes into account
reflections from surfaces: buildings were considered completely reflective in-
stead of ground whose absorption was set to 0.5 (according to residential
areas absorption suggested by GPG in tool 13.1).
Meteorological correction considered only favourable condition (p = 1) ac-
cording ISO 9613-2 and temperature of 25◦ C and humidity of 50% were
set.
Calculation was done dividing municipality in 200-300 m large zones and
performing a grid with a step of 10 m.

Uncertainty evaluation
An accurate analysis was performed through GPG toolkits and other doc-
uments ([20] and [24]). Therefore, accuracy was evaluated for each step of
noise mapping process and table in figure 4.9 summarizes theoretical accu-
racy for day period map (a similar one was made for night period map). So,
a square sum was done and it resulted that day levels were affected by a
global theoretical uncertainty of 4.3 dB and night levels of 4.6 dB.
These uncertainty values were validated through a comparison with mea-
surements: differences were calculated between modelled values and contin-
uous (full day without direct control) and spot (an hour with direct control)
measurements.
Measurements were done by ARPAT to define Acoustic Clime in 2005-2006
or more generally to define road noise: day levels comparison was done on
54 continuous positions and 106 spot ones; night comparison was done on
51 continuous and 94 spot. Notice that night levels of spot measurements
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION30

were estimated according ARPAT guidelines [25] based on flow distribution


per road class. Table from [25] follows, reporting correction factors for night
levels and suggested measurement time interval for each road class:

Road Type Measure Days Time period Correction


Urban or local road with Mon.-Sat. 9.00 - 11.00 8 dB
low flow and low %HGV
Inter-district or suburban Mon.-Sat. 10.00 - 12.00 6 dB
road with low %HGV
Main suburban roads Thu.-Fr. 12.00 - 15.00 5 dB
and highways

Comparison evidenced that:

• 90% day levels residuals were between 4.6 dB and -4.8 dB with median
0.11 dB;

• 80% night levels residuals were between 5.3 dB and -4.1 dB with me-
dian 0.78 dB.

By the end, we could say that residuals were distributed as previewed by


GPG, i.e. a better result couldn’t be possible with available input data.
Differences between calculated values and measured ones are shown in figure
4.8.

Figure 4.8: Residuals Distributions


CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION31

Final considerations
Improvement of accuracy is possible only with an improvement on traffic
input data: already conclusions of first noise mapping report presented the
possibility of a traffic model (recently bought by ARPAT) to estimate noise
levels and a first example was shown in [26].
Furthermore, an improvement could be achieved with a better weighting:
an estimation of local weights for two wheelers and buses was recommended
after the first map.
Finally, conclusions of that report emphasized that mapping is a dynamic
instrument to be updated and verified: in fact, only if it’s updated, it could
be a strategic help to protecting policies. So ARPAT continued working in
this direction: this paper updates acoustic map with a traffic model and it’s
part of the larger project of producing strategic map (including railway and
aircraft noise).
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION32

Figure 4.9: Day model steps and accuracy [6]


CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION33

4.4.3 Two wheelers sound emission evaluation


This study is part of a three-years master thesis of 2004 [27]: it’s a com-
plementary study of the evaluation of acoustic clime in Pisa committed by
municipality to ARPAT.
The aim was to evaluate two wheelers contribution to environmental noise in
some city streets: particular attention was given to ZTL streets of historical
centre (here two wheelers circulate free in spite of cars). Moreover, it tried
to determine average spectrum of two wheelers (divided between 50 cm3 and
the ones higher powered).
Sound levels were measured with spectrum analysers: levels were post-
processed to quantify two wheelers contribution. Single passages were iden-
tified within time history and total contribution was evaluated per type of
street.
After this analysis critical streets were evidenced to manage PCRA whose
aim is an efficient noise reduction.

Achievements
Counts and SEL calculation was made on 27 positions; counts per measure-
ment point were between eight and twelve vehicles.
From data acquired, it was possible to establish two wheelers contribution
to global noise, i.e. to establish the maximum reduction cutting down on
them: in ZTL noise levels could be lowered from a minimum of 1 dB to a
maximum of 3.7 dB, instead on other streets contribution was less than a
decibel.
Measured SEL were extremely large distributed: SEL varied not only be-
tween streets with the same geometrical structure, but also within the same
street (maximum variation was 3.4 dB). However, 50 cm3 power class was
usually noisier than higher ones.
This large dispersion of values was due to distribution of speed and accel-
eration which are also very broad in city context. In figure 4.10 is shown
a section of summary table taken from [27] where SEL are listed. These
values will be used in this paper to estimate sound emission for this vehicles
class. Notice that measurements were done at an average distance of 3 m
from the source and 4 m height above street surface; therefore, in following
chapters, divergence correction will be applied to obtain emission values.
CHAPTER 4. TRAFFIC AND NOISE MODELS IMPLEMENTATION34

Figure 4.10: Measured SEL from [27]


Chapter 5

A new approach to traffic


assessment

5.1 Introduction
This work innovation consists in traffic assignment method: traffic model
is implemented with software TransCAD version 4.8 (powered by Caliper
Corporation and licensed to ARPAT). Utilization of this software started
in 2006 to support data elaborations to be done for PCRA agreement with
Pisa municipality; unfortunately, no traffic network analysis was carried out
because there wasn’t enough time within dead line of PCRA. Traffic project
restarts in September 2007 with this thesis: the first job has been to collect
all data of first attempt, then to understand, update and enlarge data to
mirror current situation.
In fact, Pisa has been recently modified by:
• insertion of many traffic circles;
• development of north-east viability (Via Paparelli - Via Moruzzi);
• definition of south ZTL (S.Antonio and S.Martino zones);
• insertion of reserved lanes for public transport especially for new high
mobility routes (LAM verde, rossa, blu).
Moreover, viability next to railway station is actually influenced by con-
struction of underground parking area. So, we have to decide how to model
circulation between the following possibilities:
1. adopt old circulation to use previous measurements;
2. model temporary circulation to do calibration measurements;
3. model future circulation (according to approved project) to estimate
future flow and impact.

35
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 36

We decided to take the third possibility because map ought to evaluate long
term noise and mustn’t mirror old or temporary situations.
During predisposition of traffic network, we tried to create a project as up-
dated as possible providing necessary data and trying to obtain them from
authorities: we collected counts of ZTL gates from PisaMo (mobility agency
of Pisa) and we requested lights cycles to municipality which didn’t thought
convenient to furnish them1 .

5.2 TransCAD characteristics


TransCAD puts together GIS capabilities and traffic flow management pos-
sibilities: it is useful both to traffic managers of administrations and to farms
which want to know citizens routes.
Any model producing flow on roads must start from a network and an Origin
Destination matrix to know how many travellers go from origin Oi to des-
tination Dj , that means to calculate flow path vector p or flow links vector
f . The software provides many methods to evaluate matrix based on citizen
characteristics (number of cars per family, income, population density. . . ):
the matrix created during the first attempt was probably calculated in a
similar way, but it considered too large OD areas making impossible to esti-
mate traffic inside them. We decided to define a new matrix with more OD
pairs.
However, there were no data available about attraction or production of flow
per area, so we used another estimation method. In fact, TransCAD is able
to estimate matrix using following data:
• flow counts on sample roads equally distributed on the network;

• any OD matrix of desired dimension to initialize values;

• road network of links and nodes with attributes requested by utilized


assignment.
Put into practice, this method creates a matrix according to counts and
then it assigns flows to all other links through the selected method.
Assignments implemented in TransCAD are all static assignments (the ones
suitable to strategic purposes):
• All or Nothing: all traffic demand is assigned to shortest path without
capacity restraints;

• STOCH assignment: calculation of path choice probability is per-


formed by a proportion between travel times;
1
New lights cycles management was going to be furnished to municipality by a private
agency but it wasn’t tested yet.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 37

• Incremental assignment: at each step only a fraction of traffic demand


is assigned through All or Nothing assignment, then travel times are
updated as function of flow already assigned;
• Capacity restraint: as previous one, but travel times are updated tak-
ing into account maximum capacity on links (this method may not
converge to equilibrium);
• User Equilibrium (UE): it use an iterative algorithm that reaches equi-
librium when no traveller can improve own travel time changing route
(it considers both traffic volume and capacity);
• Stochastic User Equilibrium (SUE): as previous one, but travellers
don’t know exactly network conditions and they have different per-
ceptions of travel times;
• System Optimum Assignment (SO): as UE, but it reaches equilibrium
when total travel time is minimized (it’s not a realistic condition but
minimize congestions).
We selected UE method because is quite realistic and converges in a limited
number of iterations: this method has a precise mathematical definition
and we can demonstrate that utilized algorithm converges. This algorithm
is Frank-Wolfe iterative process. In fact, at municipality scale, we need
a model which takes into account congestions (i.e. capacity) and which
converges to equilibrium solution; otherwise we could have long run time
without reaching a correct solution.

5.3 User Equilibrium Method


Let’s precise the User Equilibrium concept: UE is usually called also DUE
because it assigns a deterministic utility, i.e. cost, to links. Equilibrium
conditions are expressed by the Wardrop’s First Principle presented in 1952:

Theorem 1 (Wardrop’s First Principle) IF traffic demand is constant


on project time, capacity restraints are not active and travellers behaviour is
deterministic at full knowledge, THEN at equilibrium point, all paths between
an OD pair with not zero flows have all the same travel cost, instead all the
others have equal or higher costs.
That is:
p∗k > 0 ⇒ Ck∗ ≤ Ch∗ ∀h 6= k
in which h, k are possible paths and C is cost.
We want to show that p∗ is flow paths equilibrium vector if and only if it
follows next equation, called variational inequality:
C(p∗ )T · (p − p∗ ) ≥ 0 ∀p ∈ Sp
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 38

Demonstration
In fact, if p∗ is an equilibrium vector only minimum cost paths are used
and any other distribution p should use other paths with equal or higher
costs so equation follows. Vice versa if p∗ verifies the equation it verifies
also Wardrop equilibrium condition: in fact, if it would exist a positive flow
on p which is not minimal, then we could obtain a vector p which wouldn’t
respect Wardrop condition. For example, let’s consider to translate minimal
flow on path k to path h: ph = p∗h + p∗k and pk = 0, then we would obtain

C(p∗ )T p < C(p∗ )T p∗

this is against hypothesis.

QED

Let’s now consider a succession of project times t; at t it’s given the dis-
tribution pt , and relative costs vector C(pt ). At next time t + 1 flow path
vector will change only if exists a path with a lower cost than the actual
one. So flow vector evolves to equilibrium:

C(pt )T · (pt+1 − pt ) < 0

Wardrop equilibrium condition may also be expressed as links (from node


i to j) flow fij , instead of using flow path vector, through the following
substitutions:
P
fij = k∈OD pk δij,k
∗ T ∗
1 if ij ∈ k =⇒ C(f ) · (f − f ) ≥ 0 ∀f ∈ Sf

δij,k =
0 if ij ∈
/k

Vectorf ∗ is called links flow equilibrium vector and it exists and it’s unique
under following conditions.

Theorem 2 (Existence) If C(f ) for all links is continuous then equilib-


rium vector exists.
Demonstration
To demonstrate this theorem, we have to introduce T (f ) = f − C(f ) defined
in Sf . Let’s consider a generic f 0 and the vector f 00 ∈ Sf which minimizes
distance to T (f 0 ):
00 T 
f = min{H(f, f 0 ) = f − T (f 0 ) f − T (f 0 ) }
 

in which H(f, f 0 ) is a scalar function only of f . Hessian of this function is


a positive definite matrix (double of identity matrix), so H(f, f 0 ) is strictly
convex.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 39

Since Sf is closed, then it’s sure that f 00 exists and it solves minimum prob-
lem and it’s also virtual minimum for H(f, f 0 ), i.e it follows next equation:

∇H(f 00 , f 0 )T (f − f 00 ) ≥ ∀f ∈ Sf

Moreover, thanks to strictly convexity minimum point is unique. Now let’s


consider a function which associates f 00 to f 0 : this function is defined in
Sf and its image is still in Sf so it’s continuous only if it’s continuous
C(f ). Browers’s theorem asserts that a continuous function, with image
contained in definition set, has a fixed point f ∗ : this point verify virtual
minimum condition for H(f, f ∗ ). This condition is true for all f therefore
we could apply it to f ∗ itself; considering that:

∇H(f ∗ , f ∗ ) = 2 [f ∗ − T (f ∗ )] = 2C(f ∗ )

we obtain variational inequality.

QED

Theorem 3 (Uniqueness) Equilibrium vector is unique if C(f ) is mono-


tonic strictly crescent, that is if:

[C(f1 ) − C(f2 )]T (f1 − f2 ) > 0 ∀f1 , f2 ∈ Sf

Demonstration
If we would have two different equilibrium vectors, then we could consider
one as f and the other as f ∗ to write equilibrium condition or vice versa and
we would obtain a contradiction with hypothesis of monotonic function:

C(f2∗ )T (f2∗ − f1∗ ) = C(f1∗ )T (f2∗ − f1∗ ) + [C(f2∗ ) − C(f1∗ )]T (f2∗ − f1∗ )

which takes to C(f2∗ )T (f1∗ − f2∗ ) < 0. This leads to contradiction of equilib-
rium condition for f2∗ .

QED

The sufficient condition for cost functions monotonicity is that Jacobian


matrix J[C(f )] is positive defined over whole Sf : elements of this matrix
are partial derivatives of cost function of link i respect to flow of link j.
If cost functions are dissociable, then J[C(f )] is diagonal: in this situation
Jacobian matrix is positive defined if cost functions increase with flows.
Therefore, whenever cost functions are dissociable, equilibrium vector f ∗
exists and is unique.
Of course not dissociable functions might lead to unique vector but detailed
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 40

calculation should be performed.


However, if Jacobian is symmetric, equilibrium vector is calculated as:
I f

f = min C(x)dx
f ∈Sf 0

Moreover, if cost functions are dissociable we can express f ∗ as:


X Z fij
f ∗ = min Cij (xij )dxij
f ∈Sf 0
ij

In the next section Frank Wolfe solution algorithm is explained.

5.4 Frank Wolfe algorithm


We can find equilibrium vector if we solve a constraint minimum problem
minimizing the following quantity:
X Z fij
Z(f ) = Cij (xij )dxij
ij 0

with next constraints:


• not flows negativity: fij ≥ 0;

• all transport demand must be assigned.


The Frank Wolfe algorithm (published in 1956) minimizes Z(f ) looking for
a descendant direction of Z(f ) (for a convex function descendant direction
means minimum one). If at step k a solution f k is given, it looks for a
distribution f k+1 closer to minimum. To find this distribution it’s necessary
to expand linearly Z(f )2 :

Z(f ) = Z(f k ) + ∇Z(f k )T (f − f k )

Notice that a minimal distribution for Z(f ) is minimal also for ∇Z(f k )T f ;
because cost functions are dissociable, Jacobian is diagonal and so distribu-
tion is minimal for C(f k )T f . So descendant direction is identified by flow
auxiliary vector fck defined as minimum for C(f k )T f :

C(f k )T (fck − f k ) < 0

Then calculation of f k+1 is performed looking for a point between f k and


fck :
f k+1 = f k + λk (fck − f k )
2
Calculation is taken from [28], [29] and [30].
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 41

Put into practice, we look for λk Lagrange multiplier which minimizes


Z(f k+1 ): fck minimizes total cost over the network so it’s the vector ob-
tained from All or Nothing assignment.
Therefore we can describe steps of Frank Wolfe algorithm:
1. we define an acceptance threshold ε and we make an All or Nothing
assignment with costs of zero flows;
2. we update costs;

3. we calculate fck performing All or Nothing with updated costs;


4. we calculate λk performing bisection method: partial derivative of Z
with respect to λ is performed in central point and we iterate on right
portion with positive derivative or left otherwise;
∂Z(f k+1 ) X
= Cij fijk+1 (fck k
ij − fij )
∂λ
ij

5. we update flow vector f k+1 ;


6. we decide how to stop algorithm with a convergence test through
threshold ε:
k+1 k
f
ij − fij
max k

ij fij
if it’s not verified we increase k and turn back to step 2.

5.5 Transport network


Before assigning flows to network, we need to define it: a network is a
collection of links and nodes with quantitative attributes. Transport network
is the one whose quantitative attribute is cost function.
We already said that Frank Wolfe algorithm converges to equilibrium if
we use separable cost functions: this condition means that the function
referring to link i is not influenced by flow on other links. Therefore, it’s
a big approximation in urban context because we are asserting that travel
time on a street is not influenced by flows of intersecting roads.
With this hypothesis, cost function used by TransCAD is a function that
provides travel time t:   v β 
t = tf 1 + α
c
in which
• tf is free flow travel time calculated as link length divided by free speed
(law limit speed);
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 42

• v is traffic volume assigned;


• c is link capacity (maximum number of vehicles which can transit
during time interval);
• α, β are calibration parameters that are treated in following sections.
So, within this model, cost is the same as time and is not influenced by other
parameters (beauty of landscapes, monetary cost of travel etc.).
To define OD matrix it’s necessary to distinguish OD nodes from intersec-
tion ones: OD nodes are called centroids and they are traffic accumulation
points. Therefore, traffic on centroids is not necessary balanced and they
idealize parking areas. Moreover, centroids aren’t directly linked to real
network but through connectors that do not correspond to any real street:
these connectors are in place of local streets whose traffic is not possible
to be estimated. Local streets traffic depends on inhabitants distribution
and cannot be calculated from the global network. Suburban, urban and
inter-district streets have been inserted in the network excluding some local
streets; nodes have been divided into centroids and intersection nodes.
In figure 5.1 is shown the network with centroids and connectors empha-
sized.
TransCAD allows inserting delays based on turn type (crossing, left or right
turns) as a function of links classification and delays on specific nodes. These
delays will be added to travel time t on path to make more reliable calcu-
lations. Following sections describe road classes parameters and used inter-
section delays.

Figure 5.1: Traffic Network: centroids and connectors


CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 43

5.5.1 Data collected from first step of TransCAD utilization


First attempt produced a network with major roads: principal parameters
were also been defined like capacity, lanes directions, free speed (law limits),
parking places presence etc. This network has been the base of the final one
but it has been modified to mirror changes.
In addiction counts data already available have been collected from different
measurements campaigns:

Table 5.1: Traffic data available


detection year time temporal n◦ vehicles n◦ and type
technique interval detail categories of places
manual 1999 7.30-10.30 150 eight 17 boundary
roads
manual 2000 7.15-9.30 150 unique 30 junctions
laser detectors 2006 24h 1h five 10 streets
laser detectors 2007 24h 1h five 1 street
video cameras 2006 24h 1h four 7 gates

Obviously time interval and vehicles classes differences make immediately


difficult to use these data: by the end, we decided to discard intersections
data because it was impossible to obtain vehicles categories. All other counts
have been reviewed to verify possible invalidation due to circulation changes.
An OD-matrix was also prepared probably based on PUT data (published
in 2000) providing movements between areas; anyway we decided to ignore
it because we cannot achieve information about how it was created.

5.5.2 Road classification and input data


With the first part of the project a network was already prepared; roads
were classified according PUT classification and following parameters were
already assigned:

• numbers of lanes and directions;

• link type (a code relating to class);

• free speed;

• capacity per lane;

• calibration parameters α and β.

Calibration parameters were assigned changing default values of α = 0.15


and β = 4 to include the approximate effect of intersecting flows and in-
tersection delays associated with a link. Therefore, these values have been
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 44

taken as correct.
These parameters have been estimated using values suggested by Highway
Capacity Manual [31] for bidirectional roads: notice that HCM suggests val-
ues for a not dissociable cost function where v is flow of both links ij and
ji. Values of HCM are:

Table 5.2: Parameters from HCM


Type of road α β
divided highway 0.1 1
3.75 m large per lane 1 2.5
3.00 m large per lane 3 4

Moreover, classes have been enlarged to include highways with link type 0
and to assign different parameters within the same class according real con-
text of roads. In fact, this first classification didn’t include 30 km/h zones
and all ZTL, so the new one is summarized in following table.

Table 5.3: Road link types and inputs


Road type Link type Capacity Free Speed α β
Highways 0 3600 90 0.1 1
Inter-district 1 2600 50 2.5 4
large roads 1300 50 2.5 4
Suburban 3 1600 70 1.5 3
roads 1500 60 1.5 3
Connectors 5 9999 30 1 1
1000 30 1 1
District and 6 1000 40 3.5 4
local roads 500 30 3.5 4
Inter-district 7 2600 50 3 4
roads 1300 50 3 4

This classification mirrors new speed values; in addiction capacity has been
limited for ZTL local roads and connectors. Link classification is shown in
figure 5.2.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 45

Figure 5.2: Traffic Network: road types

5.5.3 Intersection delays


Intersection delays have been inserted as turn penalties without including
considerations about congestion, that is a fixed delay has been assumed.
This kind of delay depends on link type because we considered that, apart
from connectors, links with higher capacity are major roads respect to ones
with a lower one. These delay values have been taken as the critical time
gap needed in major road flow to do a specific movement from the minor
one (left, right turn and crossing).
Critical gaps are the ones described in High Capacity Manual [31] for stop
signal (see figure 5.3): in this manual critical gaps are used to define a delay
based on assigned flow and to perform an assignment with volume-dependent
turning delays, so taking them as fixed produce an underestimation of delays.
In fact, equation for volume dependent delay (seconds per vehicle) is the
following:
 v   
vx
u 2 3600
3600  vx t vx
u
cp,x cp,x 
d= + 900T ·  −1+ −1 + +5
cp,x cp,x cp,x 450T

in which T is analysis time period and cp,x is potential capacity for movement
x which is calculated as function of critical gap tc , follow up time tf (time
needed to do single movement) and conflicting volume vc :
exp(−vc,x tc /3600)
cp,x = vc,x
1 − exp(−vc,x tf /3600)
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 46

Figure 5.3: Critical gap criteria for unsignalized intersections

TransCAD provides a tool performing assignment with volume depen-


dent turning delays: despite having inserted all inputs requested, tool breaks
without assigning flows. This problem could not be solved, so fixed turn
penalties have been used for both unsignalized and signalized intersections.
In addition to turning delays, specific penalties have been inserted at ZTL
gates to simulate low flows.

5.6 OD matrix calculation


As already mentioned, OD matrix has been estimated on traffic sample
counts: this method needs that counts refer to the same time period. In
fact, flow on network refers a specific time interval and, as presented in
section 4.1.1, static models usually refer to peak hour. Moreover, available
counts were done in the morning, so we decided to perform assignment on
morning peak between eight and nine o’clock3 .
Following sections describe how many counts have been used to estimate
matrix and how vehicles categories have been treated.
3
An attempt to perform PM peak was done but calculations produced values similar
to morning ones.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 47

With explained conditions it’s possible to estimate matrix from traffic flow
considering that base OD matrix elements dbi differ from real demand di :
dbi = di + θi . Moreover, also flows are not perfectly assigned so if H is the
OD matrix, d is demand, we obtain estimated flow as: fb = Hd + ε.
To estimate demand matrix we can solve an optimum constraint problem us-
ing Bayesian estimator d∗ which combine prior demand db with experimental
flow fb:  
X (di − dbi )2 X (fbj − P hlj dl )2
d∗ = min  + l 
d≥0 var(θi ) var(εj )
i j

This problem can be solved using an iterative algorithm called steepest de-
scent or gradient descent method:

1. we define an acceptance threshold ε and we initialize dk = d;


b

2. we calculate objective function D(dk ) and not restraint direction:


g k = −∇D(dk );

3. we calculate steep direction hk :


( k
hi = gik if dki > 0 or gik ≥ 0
hki = 0 otherwise

4. we look for λk along hk direction:

λk = min D(dk + λhk )


0≤λ≤λ̄

in which λ̄ is maximum values which ensure not negativity of dk +λhk ;

5. we calculate new demand as dk+1 = dk + λhk and we increase k;

6. we decide how to stop algorithm with a convergence test through


threshold ε:
D(dk+1 ) − D(dk )

D(dk )
if it’s not verified we increase k and turn back to step 2.

5.6.1 Sample counts


Many counts have been collected and performed: 52 of 70 available counts
have been used to estimate matrix. In figure 5.4 a map of sample links is
shown. However, counts are useful also to estimate day and night traffic
distribution so many counts along the whole day have been done to validate
the model.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 48

Figure 5.4: Traffic Network: sample counts

Counts are collected in different ways:

• manual counts: distinguishing cars, two-wheelers, heavy goods vehicles


and buses;

• video cameras: distinguishing cars, two-wheelers, light goods vehicles,


heavy ones;

• laser traffic counter: with Viacount we could have speed distribution


and counts distinguished by vehicles length (following section will de-
scribe categories through length).

Total counts are summarized in the following table:

Table 5.4: Traffic Counts


n◦ counts Period Time Method Notes
13 Sept 06/Sept 07 24h video cameras by PisaMo
12 Oct 07/Feb 08 1h manual this project
17 Nov 99 3h manual acoustic clime
9 Jul 06 24h Viacount I PCRA project
19 Oct 07/Feb 08 24h Viacount II this project
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 49

5.6.2 Equivalent vehicles


TransCAD provides also multi modal assignment: to perform multi modal
assignment, an OD matrix for each category is needed and there’s a tool
which allows estimating multiple matrices. So, we tried to estimate heavy
vehicles, cars and two wheelers matrices but heavy vehicles and two wheelers
counts were too low to have e reliable assignment.
Therefore, equivalent vehicles have to be used: as like as NMPB vehicles
we have to define a weight but instead of acoustic power it must relate on
road occupation. In fact, a road that reaches maximum capacity for x cars
(nC ) will reach it with fewer heavy goods vehicles (hereinafter HGV) and
with more two wheelers (nT W ); so we decided to define factors to convert
two wheelers and heavy in passenger cars equivalent flow as shown in the
following equation.

neq = nC + 0.5 · nT W + 2.5 · nHGV

Equivalent vehicles have been calculated from manual and cameras counts
taking buses as HGV and light goods vehicles as cars.
Counts performed with laser automatic traffic counters must be analysed
with more attention: counters identifies vehicle length from axis distance,
more exactly a laser beam starts from counter and see different lengths ac-
cording distance between vehicle and counter (see figure 5.5).

Figure 5.5: Detector shadow on two-ways streets

Therefore, different length boundaries values have been established for each
lane (the one closer to counter has shorter boundary) according detector
distance from road less than 0.5 m:
Notice that automatic detectors could underestimate or give unreliable counts
when closer lane traffic obscures opposite lane one. The resulting effect is an
high number of short vehicles which corresponds to pieces of normal vehicles
not completely detected: this means that opposite lane values are reliable
only if traffic is quite low.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 50

Table 5.5: Automatic Counter categories from [32]


category Length in closer lane Length in opposite lane
two wheelers < 322 cm < 400 cm
cars 322 − 750 cm 400 − 920 cm
heavy vehicles > 750 cm > 920 cm

5.7 Flow and speed network assignment


We already explained how Frank Wolfe algorithm performs UE assignment,
so we want to summarize requested inputs and show results. Inputs for OD
matrix estimation and assignment are:

• ID unique identifier of each link;

• DIR code number which indicates if link is one or two way;

• base OD matrix;

in addiction we have to define for each direction available:

• TIME free flow travel time on each link;

• COUNT sample counts of peak hour;

• CAPACITY, α,β;

• PRELOAD.

Pre-load is a fixed background link flow that is always assigned: public


transport buses have been considered as pre-loads and assigned to links
where they haven’t a reserved lane. Bus scheduling has been taken from
CPT website and LAZZI website; urban lines routes have been identified by
map on CPT website, instead suburban lines routes have been asked to bus
drivers4 . Buses assignment has been performed in collaboration with Grad.
A.Panicucci.
Moreover, we added information including penalties.
Resulting outputs for each direction follow:

• equivalent flow;

• travel time;

• speed at average flow;

• Volume to Capacity ratio.


4
In spite of written requests no data were furnished by CPT.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 51

Time and speed are the ones assigned at last iteration that is time corre-
sponds to travel cost and speed is calculated as link length divided by time.
It’s clear that at peak hour speed reflects congestions so also very low value
(under 20 km/h) can be found near intersections or on traffic circles.
Volume to capacity ratio V/C is level of congestion of links and it’s very use-
ful to identify hot spots through creation of thematic coloured maps: Tran-
sCAD provides maps in which colours scale follows V/C ratio and thickness
of links follows volumes. Such a map, produced on last assignment, is shown
in figure 5.6.

Figure 5.6: Traffic Network Assignment

5.8 Model validation


Of course many traffic assignments were done before the last one: in fact
many changes to network have been done before achieving a good balance
between number of streets mapped and flows on them. In fact, if we consider
too many streets, model is not able to assign flows correctly and we see effects
of All or Nothing assignment: let’s consider two local streets going in the
same direction (e.g. Porta a Lucca, CEP, Pisanova areas) connecting to the
same major roads. In this scenario, model assigns all traffic to the shortest
path and nothing to the other because flow is low. Therefore, such streets
should be represented by only one connector or have an high flow. Other
problems occur when too few roads are in the projected network (respect to
the real one) and calculated flow increases extremely.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 52

Finally, we obtain a result that gives flow quite correctly: in figure 5.7
distribution of residuals is shown according uncertainty on noise levels from
IMAGINE paper [11].

Figure 5.7: Residuals between modelled and measured flow

At the end of the work it’s clear that we should have achieved a better
result because only 73% of residuals differ less than 22% which was minimum
accuracy to obtain levels uncertainty of 1 dB according [11].
We were unable to verify directly speed results because too few counts have
been achieved with reliable speed: anyway we considered speed calculation
correct for day period and we assumed speed limits for night period. Notice
that speed calculation is conditioned by limits: more the flow is higher, more
the speed is small but often free speed assumed is too low because many
travellers don’t respect limits (especially on longer links).

5.9 Uncertainty evaluation


Traffic flow has been assigned to network for peak hour and equivalent ve-
hicles.
We want to evaluate decisions according IMAGINE toolkits to give an esti-
mation of accuracy; principal aspects are:

• Speed;

• Acceleration;

• Traffic composition;

• Diurnal and long-time patterns;

• Low flow roads;


CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 53

• Intersections;

• Gradient.

Figure 5.8 summarizes decisions and shows associated accuracy from [11]:
notice that higher number of polygons angles means higher accuracy and
toolkits provides triangle to hexagon degrees. If we assign an accuracy fac-
tor from 1 to 4 to different polygons we could say that adding all accuracy,
total could be within 7 and 28 points: we achieved only 15 points so ac-
curacy is quite poor. As already discussed, it would be unlikely to achieve
1 dB uncertainty over all links.
To evaluate numerical accuracy we will consider GPG source-related toolk-
its (Toolkits 2-7). However, GPG toolkits include implementation of noise
model so here only traffic model features will be analysed related to IMAG-
INE toolkits.
The use of a traffic model on major roads to estimate flow is considered by
GPG (Tool 2.5) having an accuracy of 0.5 dB. However, we have to remem-
ber that estimation on other roads will use first map flows so with 2 dB
accuracy technique.
Tool 3.5 provides accuracy for speed data: speed from traffic model has been
used for day period on major urban roads; speed limits have been used for
night period and for all other low flow roads. Therefore, 1 dB is associated to
major roads day levels, instead night and low flow roads levels are affected
by 2 dB uncertainty. Moreover, tool 6.1, regarding junctions, suggests a
1 dB uncertainty associated to ignore acceleration and deceleration.
Toolkits 4, 5 and 7 regard traffic composition, road type, road gradient;
these toolkits will be treated later because they are associated with mod-
elling measures not explained yet.
CHAPTER 5. A NEW APPROACH TO TRAFFIC ASSESSMENT 54

Figure 5.8: Traffic model decisions


Chapter 6

TransCAD traffic output


elaborations

6.1 Introduction
Traffic models are not born to perform noise calculation so many problems
occur when output of traffic model is used as input of noise models. Principal
reasons are explained in strategic maps guidelines [33] and here listed:
1. spatial resolution of traffic network;
2. temporal resolution of time period considered;
3. traffic data resolution (detail level).
In fact, traffic network not only doesn’t cover all streets but also is mis-
matched with the real one. This means that road profiles should be verified
before inserting in noise model and all other streets should be introduced
(see section 6.2).
Time reference period has been assumed as peak hour but we have to esti-
mate LDEN and LN ight , so we have to extrapolate daily distribution to have
average day evening and night traffic data (see section 6.3).
Finally, traffic data don’t distinguish between vehicles categories, so we have
to apply distributions to obtain NMPB vehicles (see section 6.4).
All these problems are solved with a level of accuracy which is treated at
the end of this chapter.

6.2 Spatial resolution


To complete network we merged the new one with the one of first acoustic
map. Traffic values of first map (see section 4.4.2) were estimated on sample
counts for Italian diurnal (6.00h-22.00h) and nocturnal (22.00h-6.00h) peri-
ods, therefore we considered day and evening traffic equal to Italian diurnal

55
CHAPTER 6. TRANSCAD TRAFFIC OUTPUT ELABORATIONS 56

period.
After having adapted periods, we substituted calculated roads to the old
ones and we left the old flow on ignored links: this geographic merge is not
easy because links don’t match exactly so a manual control on each zone
has been done1 . All connectors and centroids have been deleted; figure 6.1
highlights old and new links.

Figure 6.1: Merging of new and old network

It’s important to repeat that roads of first map didn’t model south ZTL
(started in September 2006), so we adapted classes of links now included in
this area and we modified flow on them; a lower flow has been assigned also
to dead end streets whose traffic was unrealistic.
Spatial resolution problem due to the traffic model causes streets to lie too
close or under buildings on noise model: all streets lines have been modi-
fied to lie at the correct distance from buildings by means of GoogleEarth
integrated with IMMI software. This procedure regarded not only streets
control, but also building destination and other adjustments of the noise
model, so it will be discussed later.
1
This geographic passage has been performed by arch. C.Chiari of ARPAT department
with a sequence of automatic and manual operations on layers.
CHAPTER 6. TRANSCAD TRAFFIC OUTPUT ELABORATIONS 57

6.3 Temporal resolution


Before merging traffic data with the old ones, we needed to produce traffic
flow representative for day, evening and night. Many counts were performed
with automatic detectors for an entire day: so, it was possible to derive flow
distribution on sample roads and apply it on similar roads. Then we esti-
mated coefficients to be applied to peak flow to derive flow on day, evening
and night period. GPG suggested such values distinguishing between main
roads and inter-district ones. GPG values are shown in figure 6.2.

Figure 6.2: Flow from Peak flow for different time periods from [5]

These values have been taken as example and values for Pisa have been
calculated from measurements; a new classification of roads has been done
based on day-peak traffic ratio. In fact, flow time distribution doesn’t nec-
essarily follow link type but it’s an easier classification between:

1. highways (no data available, major roads from GPG used);

2. suburban roads (5 samples);

3. urban inter-district and district roads (20 samples);

4. local and ZTL roads (13 samples).

These classes are homogeneous not because roads have the same traffic pro-
file, but more generally because percentages of day, evening, night flow re-
spect to peak one are similar. Through these percentages flow periods co-
efficients are established for each class and time period; if α is percentage
listed in the following table, average hourly flow for referenced period T is
obtained as follows:
QT = αT ∗ QP eak
In figure 6.3 examples of distribution of classes 2, 3 and 4 are shown; hourly
flows are divided by peak hour flow.
CHAPTER 6. TRANSCAD TRAFFIC OUTPUT ELABORATIONS 58

Table 6.1: Percentage of flow for different time periods in Pisa


Road class Italian day Day Evening Night
6.00h-22.00h 6.00h-20.00h 20.00h-22.00h 22.00h-6.00h
1 1 1 0.7 0.2
2 0.82 0.87 0.45 0.16
3 0.84 0.86 0.54 0.16
4 0.71 0.78 0.44 0.12

Figure 6.3: Examples of time distributions

6.4 Traffic data resolution


6.4.1 From equivalent vehicles to real vehicles
Traffic model produces equivalent vehicles data that must be transformed
into originally measured categories: cars, two-wheelers and HGV. Just like
time distribution, also fleet composition has been calculated based on mea-
surements applying the same road classification. Sample measurements were
so distributed: 11 counts class 2, 23 counts class 3 and 14 counts class 4.
Average percentages assumed are shown in following table.

Road class Two Wheelers (TW) Cars (C) HGV∗


1 0% 80% 20%
2 8% 89% 3%
3 15% 82% 3%
4 45% 52% 3%

∗ Bus not included


CHAPTER 6. TRANSCAD TRAFFIC OUTPUT ELABORATIONS 59

If we would have estimated total vehicles flow, then a simple ratio could be
applied to obtain real vehicles: instead we have equivalent flow so a regres-
sion must be applied. In fact, if p is percentage, number of vehicles is given
by Q = p ∗ T where T is total flow but we have to express T as a function
of total equivalent flow E.

E = 0.5 ∗ pT W T + 1 ∗ pC T + 2.5 ∗ pHGV T


=⇒ T = E ∗ (0.5 ∗ pT W + 1 ∗ pC + 2.5 ∗ pHGV )−1 ≡ E ∗ β

So flows per category are:



 QT W = pT W βE ≡ γT W E
Q = pC βE ≡ γC E
 C
QH = pHGV βE ≡ γHGV E

γ values are listed in the next table for each road category.

Road class γT W γC γHGV


1 0 0.62 0.15
2 0.08 0.89 0.03
3 0.15 0.85 0.03
4 0.56 0.63 0.04

Notice that we never consider buses: as presented in previous chapter bus


routes and scheduling are known as hourly flow so it would be wrong to
estimate them. Buses flows are included in traffic models as pre-loads so we
could, after the assignment, subtract them. Buses are considered as special
category whose distribution is completely known: no time period correction
is needed.
Finally, we have an estimated flow per link at each reference period divided
in four categories: two wheelers, cars, heavy vehicles and buses.

6.4.2 From real vehicles to NMPB light and heavy vehicles


An acoustic weight has to be assigned to each category to adapt NMPB
emission values to real ones: we wanted to verify real emissions in Pisa tak-
ing as reference values the Florence’s ones.
Following this approach, four measurements have been done: sound levels
have been memorized together with time of single vehicles passages. Mea-
surements were performed in different road classes, choosing streets with
high public transport flow to evaluate their contribution. Of course single
passages are very difficult to identify, however we identified 46 cars, 53 buses
and 31 heavy vehicles passages. Instead, two-wheelers emission values have
been calculated from ARPAT study described in section 4.4.3.
For each single passage, SEL has been calculated from time history, then
divergence and time correction has been applied to obtain emission values
CHAPTER 6. TRANSCAD TRAFFIC OUTPUT ELABORATIONS 60

at 30 m distance and 10 m height (measurements were performed at 3 m


distance and 4 m height). Notice that emission values of NMPB model are
calculated in a free field condition but we were always near façades so we
decided to subtract 3 dB.
Same thing has been done with values furnished by [27]: we must consider
that all categories have been averaged together, so about 250 two-wheelers
passages have been considered.
Following table shows measurements results compared to NMPB values2 :

Table 6.2: Pisa emission values [dB(A)]


Two-Wheelers Cars Buses Heavy vehicles
Florence 30.1 28.6 37.3 33.4
Pisa 31.6 28.6 35.7 35.4
NMPB 30.7 30.7 43.7 43.7

Observed values shows that car emission value is equal to Florence one and
so the same weight has been assumed; differences between bus and heavy
vehicles are very low (considering that measured SEL3 vary about 4%) so
we assumed the same weight (from higher values).
Weights are here listed:

Ql,eq = QC ∗ 0.61 + QT W ∗ 1.2
Qp,eq = QB ∗ 0.16 + QHGV ∗ 0.16

Comparing Florence values with Pisa ones, principal differences are:

• two wheelers are noisier in Pisa; this can be explained considering that
Pisa links are very short and acceleration and deceleration influence is
higher, anyway experience teaches that two wheelers are really noisier
than cars so a coefficient higher than 1 is correct;

• buses are noisier in Florence; this is probably due to Florence bus fleet
that is older. Moreover, Pisa fleet counts many methane engine buses.

6.5 Accuracy of correction coefficients


Time coefficient, estimated like GPG example, is affected by two kind of
uncertainty:

• it considers week-end flow as weekday one and so 1 dB uncertainty is


given in tool 2.3;
2
According to average speed, cars and two-wheelers have been compared to 50 km/h
values, instead buses and heavy vehicles to 40 km/h values.
3
Explanation of this indicator is given in appendix A.
CHAPTER 6. TRANSCAD TRAFFIC OUTPUT ELABORATIONS 61

• it’s possible that road categories are too simple and don’t mirror real
time traffic distribution.
However, measured time coefficients vary within the same class less than
25% so equivalent flow varies changing levels less than 1 dB.
Distribution method to estimate traffic composition is the second option of
GPG tool 4.5: uncertainty is estimated to be less than 0.5 dB. Moreover,
if we consider also measurements we can observe that, within these classes,
maximum varying parameter is cars percentage on local roads and it’s about
20%: therefore uncertainty expressed in dB is less than 0.8 dB. Consider-
ing that majority of network roads are classified as inter-district or district
(which uncertainty is less than 0.5 dB), we can consider correct an uncer-
tainty due to composition of 0.5 dB.
Weight accuracy was estimated in first map work [6]: 1 dB accuracy was sug-
gested but no clear explanation was given. Therefore, a test was managed
to verify this accuracy: test was done measuring sound pressure levels and
flows at the same time and then verifying estimated values. Flow and sound
levels were averaged on a week and then light and heavy vehicles were calcu-
lated using weights on measured categories; day, evening and night measured
flows have been inserted in the noise model and levels calculation on recep-
tion point (i.e. measurement position) has been performed. Measured and
calculated values in Via di Gello (test location) are compared in following
table:
Day Evening Night
Measured 65.4 dB(A) 65.7 dB(A) 58.7 dB(A)
Calculated 66.6 dB(A) 64.6 dB(A) 56.6 dB(A)
Residuals -1.2 dB 1.1 dB 2.1 dB

Differences increase during night period because speed change: in fact as-
sumed speed was equal to all periods instead the real one increases. Uncer-
tainty of speed distribution has already been considered, so we accept 1 dB
accuracy as suggested by [6]. Finally, we summarize all accuracy due to the
use of TransCAD output as IMMI input:

Problem Solution Accuracy


Traffic apply distribution 0.5 dB
Composition based on measurements
Emission weight from [27] 1 dB
values and measurements
Week-end flow same as weekdays 1 dB
Long term values apply distribution 1 dB
from peak flow based on measurements
Chapter 7

Noise model implementation

7.1 Introduction
Noise mapping has been performed with prevision software IMMI version
6.3.1 (powered by Woelfel and distributed by Microbel). This version is
very different from the one used for the first map. Not only bridge struc-
tures were added, but also calculation is implemented in a more efficient way.
Moreover, it’s now possible to distribute calculation on many PC without
manually sectioning the project (see section 7.3.3). All these modifications
together with new streets and building elements induced us to create a new
project: therefore, all cartographic elements have to be inserted and con-
trolled.
After that, calculation parameters have to be set: notice that calculation
parameters ought to be a balance between accuracy and run time possi-
bilities, so we try to improve calculation respect to previous possibilities.
Nevertheless, we have to take into account that a maximum of three PC can
be used to perform calculation.
All assumptions and modelling choices will be described in the last section
to evaluate accuracy of noise model.

7.2 3D model implementation


Regional cartography archive provides a digital reproduction with resolu-
tion 1:10.000 of the entire municipality and one with resolution 1:2.000 of
residential areas. First map used only 1:10.000 cartography, instead we de-
cided to cover municipality with a mixed cartography: 1:2.000 layers have
been used to cover residential areas and 1:10.000 layers have been used to
complete the remaining areas.
Digital cartography provides much information but not all elements have
the same accurate height: for example buildings and walls have a relative
height, instead of altitude points and geodesic lines, rivers and sea lines

62
CHAPTER 7. NOISE MODEL IMPLEMENTATION 63

which have absolute one. Moreover, it provides also street lines, foot-path
lines, viaducts and bridge lines but their absolute height is not always reli-
able.
In following sections we will analyse how elements have been treated.

7.2.1 Digital Terrain Model DTM


To create digital elevation model all possible elements have been used: that
means all available lines and ground points with an absolute height greater
than zero have been inserted. So principal elements are:

• altitude points;

• geodesic lines (iso-lines and curve lines);

• streets and foot-path lines;

• railway lines;

• all hydrological lines (river Arno, sea, Navicelli canal and all watering
canals).

Of course random errors on cartography have been noticed and corrected,


examples are here listed:

• altitude points on viaducts and bridges with the same height of viaducts
or bridges instead of ground one: points have been cancelled;

• geodesic, railway, streets or foot-path lines and altitude points near


buildings with gutter instead of terrain elevation: points and lines
have been cancelled;

• hydrological lines only on one river side: hydrological lines have been
supposed equal to the other side and copied;

• lines partly wrong because of previous errors: lines have been split and
the correct part has been inserted.

All these errors have been noticed and selected thanks to 3D IMMI viewer:
the whole project has been analysed with 3D viewer which is able to select
features from the project during the 3D view and it put them on a separate
collection to be elaborated.

7.2.2 Bridges and viaducts


Bridges and viaducts lines have absolute height: these lines have been con-
sidered as suspended obstacles to propagation and classified into element
class BRUCK. Width of each BRUCK structure can be set and also vertical
CHAPTER 7. NOISE MODEL IMPLEMENTATION 64

barriers can be associated. However, not all viaducts or bridges are well de-
fined in regional cartography; in fact, some bridges and viaducts have been
created based on visual consideration.
A list of principal created elements is given:

• San Giusto bridge over railway lines;

• Airport and Darsena FI-Pi-Li turn off;

• Viaducts next to Bocchette bridge (Oratoio);

• Via Livornese bridge over A12 (San Piero);

• S.S.Aurelia bridge over via Conte Fazio.

7.2.3 Sound barriers and walls


In the first noise mapping project, all walls have been inserted including
garden ones. Unfortunately, regional cartography gives relative height that
is not always reliable; furthermore, it distinguishes between concrete, dry
wall and fences but also this characteristic is not reliable.
Therefore, we decided to consider and control only principal walls in the
town:

• city wall;

• military zones walls;

• botanical gardens walls;

• prison walls;

• Cottolengo walls;

• river parapets.

Moreover, almost all sound barriers have been considered: they are classified
as wall whose absorption coefficient is very high. Some barriers height were
available, others have been estimated by visual inspection with the aid of
3D viewer compared with 3D view of Pisa on Live Search Map website.

7.2.4 Buildings and streets: GoogleEarth utilization


Streets from created network have been inserted into the project adapting
them to terrain height. This means that streets profile is the same of DTM:
however, if too few nodes are inserted, streets could sink into the ground
(see figure 7.1). All these situations have been identified with 3D viewer and
modified inserting nodes to let the streets fit terrain profile.
CHAPTER 7. NOISE MODEL IMPLEMENTATION 65

Figure 7.1: Streets adaptation to DTM

Of course streets on bridge or viaducts don’t follow terrain profile but


bridge one: software provides an automatic tool to adapt sound sources on
BRUCK elements. This tool has been used ensuring that a sufficient number
of nodes are included and well positioned in the considered street, otherwise
this tool is not able to adapt the street correctly (see figure 7.2).

Figure 7.2: Streets adaptation to BRUCK

Buildings have been originally taken from regional 1:10.000 cartography,


instead of 1:2.000 as in the first map, because their relative heights are
more reliable; moreover, detailed cartography includes balconies and other
spreading elements that could complicate calculation. However, looking at
the project, we realized that this layer lacks of many new buildings so we
started using GoogleEarth to create new buildings. IMMI provides import
of GoogleEarth georeferred maps on the project so that it’s possible to see
CHAPTER 7. NOISE MODEL IMPLEMENTATION 66

both real photo and project elements. Therefore, we start adding new build-
ings (with height based on number of floors seen with Live Search Map) but
during this process a new 1:2.000 layer was produced: we decided to insert
new identified buildings from new layer merging with the old ones.
Notice that we didn’t substitute the new layer with the 1:10.000 layer be-
cause this new layer is still less reliable than the other, especially in historical
centre where no new buildings have to be added.
In addition to new buildings identification, GoogleEarth tool has been used
to update buildings use: regional cartography provides different buildings
codes to identify their use but sometimes this could be different from real
one (e.g. cemetery is coded as residential building). Codes distinguish be-
tween residential, industry, religious, and also other types of elements like
hothouses, penthouses, huts that haven’t been considered.
Finally, GoogleEarth has been used also to set streets on centre line: es-
pecially San Piero Fi-Pi-Li turn off and Via Moruzzi at the limits of the
town were designed through this tool because no updated cartography was
available.
Other settings regarding buildings and streets are:

• road surface has been considered as normal asphalt for all streets (ex-
cept streets were absorbent asphalt has been recently installed);

• circulation on road has been considered fluid and continuous with


speed from traffic model for day and evening period and speed limit
during night period (emission sound in these conditions is highlighted
in figure 7.3);

• minimum buildings height has been set to 3 m;

• buildings with area less than 20 m2 have been cancelled.


CHAPTER 7. NOISE MODEL IMPLEMENTATION 67

Figure 7.3: Sound emission levels for fluid continuous circulation


CHAPTER 7. NOISE MODEL IMPLEMENTATION 68

7.3 Calculation settings


Calculation settings menu is divided into global parameters tab, elements
parameters tab and calculation model tab. Elements whose parameters have
to be set are streets, buildings and walls. Global parameters include tem-
perature and meteorological conditions and other settings useful to manage
population exposure without having data per building (it’s not this sit-
uation). Anyway, it’s more useful to analyse settings divided into sources
and propagation settings because parameters of different tabs influence each
other. Finally, calculation model issues will be treated.

7.3.1 Source settings and meteorological conditions


Most important setting regards how line source is transformed into point
sources: as presented in chapter 3, French splitting methods are three
(equiangular, constant and variable step), however this software implement
only equiangular one.
Equiangular step depends on receivers position and the equivalent length
associated with point source is l ≤ αd, in which d is direct distance source-
receiver and α is a parameter whose standard value is 0.5 but could be
changed to increase number of sources. This distance criterion factor has
been increased to manage free field calculation: in that case, otherwise,
buildings distance would be too large and sources too much far from each
other (point sources effect would be visible, see figure 7.4).

Figure 7.4: Different distances of receivers, effect on sources

Moreover, each source has an action radius so that at a grater distance source
is not considered: this radius has been set to 500 m for all streets apart from
A12 highway whose radius has been set to 1.5 km. In fact, urban streets are
very close to buildings and their noise cannot reach greater distance without
loosing much power (due to reflection and absorption); instead, A12 is in
free field condition and its power is very high so it could be heard at very
CHAPTER 7. NOISE MODEL IMPLEMENTATION 69

long distances.
Finally, source meteorological correction has been taken from GPG because
no specific data were available. So probability of favourable conditions has
been set to 50% − 75% − 100% respectively for day, evening and night peri-
ods. Other global meteorological settings regard temperature and humidity
to calculate atmospheric absorption (a propagation issue): average temper-
ature has been taken as 15◦ C and humidity as 70%.

7.3.2 Propagation: reflections and absorption coefficients


Only first order reflections have been considered: we tried to perform second
order but run time increases exponentially and it was impossible to perform
second order calculation on the entire municipality.
Reflections have been activated for all vertical and horizontal elements, so
an absorption coefficient has been chosen for buildings, walls and bridges.
Tool 16 of GPG has been used and absorption coefficient has been set to 0.2
(i.e. lowering levels of 0.97 dB) for all elements apart from noise barriers
whose coefficients have been expressed in dB and calculated from available
measurements.
Finally, ground absorption has to be set: tool 13.1 indicates 0.5 factor for
residential areas and the whole municipality has been supposed to be like
that.

7.3.3 Automated distributed calculation: segmentation


The program system of IMMI [34] provides the possibility of efficiently edit-
ing very large models and comprehensive grid calculations by means of the
“AUDINOM – distributed grid calculation” module. The module is able to
distribute calculation over several computers: in fact, project is automati-
cally distributed over various computers and after calculation is completed,
combination of partial grids is performed to obtain total grid.
This module is the innovation, respect to the old version, which allows build-
ing and running a unique project over the entire municipality.
Calculation has been distributed over three PC (maximum of licenses avail-
able): project is divided in several segments overlapping each other in all
directions. Segmentation 9x9 has been carried out with an overlapping buffer
of 500 m: buffers role is to include in the calculation area of each segment
also other sources of neighbour segments whose power influences levels of
considered segment. Notice that, for each segment, model doesn’t estimate
noise levels in buffers but only inside the segment; after all segments are
calculated, global grid is automatic assembled.
Segmentation is a physical division of the project, so it’s possible that a
segment does not include any sources in the buffers: here levels could not be
calculated because there is no knowledge of other far sources. This problem
CHAPTER 7. NOISE MODEL IMPLEMENTATION 70

occurred in the area of S. Rossore Park: however, traffic noise is not zero be-
cause A12 and S.S.Aurelia have high power levels that reach very far zones.
To solve this problem a particular segmentation 3x5 with overlap of 1 km
has been defined to include both areas and sources and only one segment
has been calculated to cover this specific area.
Finally, we must underline that few areas haven’t been calculated because
sources action radius ended before reach them: in these cases interpolation
has been performed.

7.3.4 Grid resolution


First map implementation performed a 10 m step grid: at the beginning of
the project we would obtain 2 m grid step to improve resolution. Before
trying this step over the whole project, we tested differences between grid
steps: we consider a segment in the area of Tirrenia (about 2.8 km x 8.5 km
large) where there is average buildings density but quite a plane DTM.
Number of grid points and run time per step are listed in following table:

Step Points Time (minuts)


20 m 60489 65
10 m 241674 255
5m 964419 1020
2m 6019086 6316

So 20 m and 10 m are very fast steps, instead 5 m grid needs about 17


hours and 2 m grid needs more than 4 days. Of course time for each point
calculation is approximately constant (about 0.06 sec): so doubling step,
time become a quarter. Particularly if we plot steps against time we notice
a power trend.
CHAPTER 7. NOISE MODEL IMPLEMENTATION 71

Of course slow steps are more reliable in urbanized context: a street is


usually 6-8 m large, so 10 m step may calculate a point on the street or
on the buildings instead 5 m one can calculate both. In following figures
we compare grids in a specific area: we notice that 20 m grid is unable to
represent noise correctly (an average area of 400 m2 is too large) instead
other grids are quite similar. However, 10 m grid is unable to estimate
correctly back façades levels, so it’s been excluded.
From this example, we could assume that performing 2 m calculation over
all municipality (15 km x 18.8 km) would take about 49 days: of course this
time could be lowered considering segmentation on many PC; however, city
context is more complicated so we decided to perform 5 m step grid (which
should take 8 days on a single processor).
Finally, 5 m step grid has been performed with AUDINOM on three PC (2
dual core with Windows XP, and a Pentium 4 with Windows 2000) and it
takes about a week so a much longer time than expected.

Figure 7.5: Differences between grids levels (Diurnal)

Figure 7.6: Differences between grids levels (Nocturnal)


CHAPTER 7. NOISE MODEL IMPLEMENTATION 72

7.4 Façade calculation and population exposure


Population data of census 2001 have been elaborated by arch. C.Chiari and
Grad. A.Panicucci to obtain number of inhabitants per building: in fact,
they had to establish priority index defined by DM Ambiente 29/11/00 to
assign action plans priorities. This elaboration is based on guidelines APAT
[35] method b) which is the same suggested in deliverable 8 of IMAGINE
project [36] (procedure F.b): population per building is taken proportional
to the volume considering the use of each building (only residential ones)
and knowing the number of inhabitants in a topological area (ISTAT cell).
Notice that accuracy of this method could be low if no check of building use
is performed, so an accurate control of schools and hospitals has been done.
This elaboration considered residential buildings of new 1:2.000 layer. As
we already explained that’s not the same used in IMMI project; however,
population per building was there available and façade calculation have been
performed only for these sensitive buildings (excluding industrial).
This layer has been inserted into the project with population information to-
gether with codes to distinguish schools, hospitals from houses. Then IMMI
performs façade calculation interpolating grids: levels have been calculated
on a points ring around each building (2 m far from façade and 2.5 m far
from each other).
Maximum and minimum levels are evaluated to establish if a quiet façade1 is
present. Then population exposure is calculated considering all inhabitants
exposed to maximum level as established by the END.
Notice that façade calculation according the END doesn’t consider reflected
sound, but only incident one: reason is that people are not affected by sound
reflected from the building, so 3 dB are subtracted by default when grids
are interpolated. This subtraction suppose that façade are completely re-
flective: despite this, our settings are different because absorption coefficient
is not zero. Therefore, for an incident level of 60 dB we have 59 dB reflected
and total is 62.5 dB and so subtraction of 3 dB underestimate exposure of
0.5 dB.

7.5 Accuracy evaluation


Accuracy of noise levels is estimated with GPG toolkits. Source related
toolkits not yet considered are toolkits 5 and 7 regarding road surface and
gradient. Surfaces of Pisa streets are all normal asphalt apart from streets
near two schools were absorbent asphalt is in use: so we assume an accuracy
equal to 1 dB from tool 5.2 in which surface correction is based on physical
properties. Road gradient is estimated from DTM because streets lines fit
1
By definition from annex 6 of DL 19/08/05 a quiet façade is one whose level is 20 dB
lower than the higher façade level.
CHAPTER 7. NOISE MODEL IMPLEMENTATION 73

terrain profile: accuracy of this method is greater than 0.5 dB (tool 7.1).
Propagation issues are treated in toolkits 11-18.
Toolkits 11-12 regard how cuttings and embankments are inserted in DTM:
our cartography already includes them, so the once uncertainty is given by
necessity to check them (accuracy greater than 0.5 dB, tool 12.1).
Toolkit 13 regards how surface absorption is defined: we consider the whole
municipality as residential area so tool 13.1 gives 1 dB accuracy.
Toolkits 14-15 regards elements height: building height is known, so no
uncertainty is given. However, height data from cartography comes from
aerial photos so 1 dB accuracy should be considered according tool 15.2;
instead, barriers (more generally walls) height is not always known correctly
so we verified them by visual inspection. This last method accuracy is 1 dB
according tool 14.2; however, considering that most walls are correct and
known from regional cartography, we assumed 0.5 dB accuracy.
Toolkit 16 regards absorption of vertical elements: we assumed suggested
absorption coefficients so 1 dB accuracy has to be considered.
Toolkits 17-18 regard meteorological condition but they express quality eval-
uation: we have low accuracy due to lack of local data.
Finally, GPG gives qualitative accuracy of inhabitants estimation per build-
ing and per dwelling: as already said, according IMAGINE evaluation, our
procedure is the best with available data. Furthermore, quality is good be-
cause we distribute population considering use of building: in addiction to
this, 1:2.000 cartography distinguish between dwellings units so if we have a
building which use is both commercial and residential, people are distributed
proportional to residential volume.
Chapter 8

Noise mapping results

8.1 Noise road map


In previous chapters we explained how segmentation has been done and how
we obtained global map. We will present all maps in appendix B: notice
that there is a map for each period (day evening and night) and also one
with LDEN indicator and one with Italian diurnal indicator. Here we show
only maps of LDEN and LN ight indicators (figures 8.1 and 8.2).
Together with whole municipality maps some detailed maps in appendix B
show city centre, hospitals areas and an example of residential area.
From whole municipality maps could be identified quiet area of S. Rossore
Park on north-west and military areas on south-west; it’s clear that higher
levels are due to highways A12 (north-south) and Fi-Pi-Li (east-west).
Finally, we can observe that noise levels are quite low in south part of the
municipality because population density is low too.

74
CHAPTER 8. NOISE MAPPING RESULTS 75

Figure 8.1: Roads traffic LDEN levels


CHAPTER 8. NOISE MAPPING RESULTS 76

Figure 8.2: Roads traffic LN ight levels (22.00-6.00)


CHAPTER 8. NOISE MAPPING RESULTS 77

8.2 Population exposure to road noise


Population exposure has been evaluated according annex 6 of END: number
of inhabitants exposed to levels of LDEN higher than 55 dB and LN ight
higher than 50 dB has been plotted with 5 dB bands (see figure 8.3).

Figure 8.3: Exposed population to road traffic noise

Moreover, inhabitants number with quiet façades is shown (see figure 8.4).

Figure 8.4: Inhabitants with quiet façades (road noise)

Schools and university departments exposure has been analysed (see figure
8.5) divided into four categories according speech equivalent level: in fact,
speech emission is about 55 dB and if incident sound is quite the same,
people start to speak louder (open windows) and hearing becomes hard.
Furthermore, hospital buildings distribution is shown in figure 8.6. We must
consider that most exposed buildings belong to S. Chiara Hospital which is
going to be dismissed, so exposure is going to be better.
Finally, also percentage distribution is shown for Italian indicators in figure
8.7: this kind of graphic underlines also good levels to have an idea of the
entire municipality.
CHAPTER 8. NOISE MAPPING RESULTS 78

Figure 8.5: School buildings exposed to diurnal levels

Figure 8.6: Hospital buildings exposure

Figure 8.7: Distribution of exposed population


CHAPTER 8. NOISE MAPPING RESULTS 79

8.3 Accuracy results


As already said, accuracy of maps presented is evaluated both theoretically
and through measurements comparison. In following sections we summarize
choices and we calculate global uncertainty, then we compare calculated and
measured levels to validate model and to establish reliability of GPG.

8.3.1 Theoretical accuracy: global uncertainty calculation


We summarize choices divided into two groups: ones related to traffic model
and more generally to input data which are different for TransCAD roads
(flow from model), classified roads (flow from first map) and for night esti-
mation; others related to noise model implementation. These choices and
their accuracy are shown in following figure, together with toolkits used.
CHAPTER 8. NOISE MAPPING RESULTS 80

Notice that in GPG is given the following explanation related to uncer-


tainty of each tool:
“The quantified accuracy statements presented within the toolkits represent
the likely level of acoustic uncertainty introduced into the result by the use
of that toolkit option, with a 95% confidence level. It must be noticed, that
this represents the uncertainty of the total results only if all other input data
is accurate. If there is uncertainty in any, or all, other input datasets, then
the research concludes that total uncertainty in the receptor result level will
be larger than any of the individual uncertainties.”

This means that we have to establish how to calculate global uncertainty.


We used the same method of the first mapping project: a square sum of un-
certainties has been performed. In fact, distribution of possible errors due
to a single choice is supposed to be independent1 from other ones: there-
fore superposition of distributions should be a Normal distribution with σt
obtained squaring single uncertainties2 :
sX
σt = σi2
i

So global obtained theoretical uncertainties for different periods and streets


are listed below:

Table 8.1: Uncertainties of noise levels


TransCAD Classified roads
Day 3.1 dB 4.0 dB
Night 3.5 dB 4.4 dB

8.3.2 Available measurements reliability


Comparison between estimated levels and measurements has been performed:
measurements have been chosen within available ones including only traf-
fic noise measurements. This means that we included only measurement
campaigns whose aims were to detect noise from road infrastructures or
environmental noise campaigns whose positions were far from other sound
sources. Moreover, measurements have been excluded whenever traffic con-
ditions have changed (new one way streets, new ZTL, new traffic circles. . . ).
Of course also measurements have an uncertainty that ought to be included
before analysing residual distribution. This kind of uncertainty is due to
1
Speed and flow for modelled roads are obviously not independent but their covariance
is negligible.
2
Square sum corresponds to 95% boundary so 1.96σ values is calculated.
CHAPTER 8. NOISE MAPPING RESULTS 81

instrumental chain, to operating conditions (time variability of source),


whether conditions and residual sound (according ISO 1996-2, [37]): we can
neglect all uncertainties but instrumental because measurements are just
long term evaluated and flows are high so that operating conditions aren’t
influential. Therefore, we have a σm equal to 1 dB and to acquire a 95%
confidence level on residual distribution, 2 dB have to be squared together
with global uncertainty on levels:
q
σtot = σt2 + σm 2

Global uncertainty at 95% of differences distribution is evaluated and listed


below:

Table 8.2: Uncertainty of residuals distributions


TransCAD Classified roads
Day 3.7 dB 4.5 dB
Night 4.0 dB 4.8 dB

Measurements include both continuous and spot positions because night


levels for spot ones have been calculated according ARPAT guidelines [25].
Figures 8.8 and 8.9 show considered control points.

Figure 8.8: Measurements north positions


CHAPTER 8. NOISE MAPPING RESULTS 82

Figure 8.9: Measurements south positions

8.3.3 Residuals distributions


We explained that roads have been modelled in two different ways and accu-
racy has been evaluated separately, so also residual distribution is supposed
to be different for each kind of road. We extracted two data sets from avail-
able measurements: points in proximity of TransCAD modelled roads and
ones performed on classified roads3 .
So residual distributions have been calculated for each sample set:
• 147 day and 110 night levels on modelled roads positions;
• 63 day and 51 night levels on classified roads positions.
We want to verify if 95% of samples are within theoretical values. In the
first set 95% of diurnal and 92% nocturnal samples are within expected
values (95% is within 4.3 dB). The first data set is quite large so we could
also fit the distribution with a Normal distribution and analyse estimated
parameters.
Therefore, we applied following function to data set:
x−b 2
f (x) = ae−( c )

being Normal function:


 2
1 − x−µ

N=√ e 2σ
2πσ
√ √
then σ = c/ 2 and 95% fitted confidence level is given by 1.96/ 2c. Fits
are shown in figure 8.10.
3
Measurements near junctions of different types have been here ignored.
CHAPTER 8. NOISE MAPPING RESULTS 83

Figure 8.10: Diurnal and nocturnal distributions of TransCAD data set

Fits show that 95% confidence level value of diurnal distribution is 3.7 dB
and night distribution fit returns 4.0 dB that are the same as previewed, so
we have a perfect correspondence between fits and GPG forecasts.
Instead of 4.5 dB, 4.7 dB is estimated value for day distribution on classified
roads (for nocturnal it’s 5.5 instead of 4.8): however, this data set is quite
small and distribution is not really a Gaussian because it’s not symmetric.
Therefore, real dispersion is smaller than fitted one: in fact, 89% of diurnal
data and 80% of nocturnal one are within expected values and only few
measurement points are outside expected boundary.
Moreover, distributions are not central: diurnal levels calculations of both
sets seem to overestimate measured values (medians are −0.1 and −0.4);
instead nocturnal sets have opposite medians (−0.8 and 0.8). This will
produce a broader distribution on global levels: in particular for diurnal
distribution we expect uncertainty similar to classified data set (it’s the
larger one and it includes the other one); instead, nocturnal distribution
will be broader than previous ones. Fitting global data (including ones at
junctions) we obtain 4.3 dB for diurnal distribution and 5.5 dB for nocturnal
that is what we were expecting (see fits in figure 8.11). We show also fits
results in terms of estimated parameters, 95% convidence level and adjusted
root square. Notice that b is average of the fitted curve and a is percentage
of discards which are less than 1 dB: we can observe that levels of classified
rods are overestimated (averages are negative), instead of TransCAD ones
whose distributions averages are positive.
CHAPTER 8. NOISE MAPPING RESULTS 84

Figure 8.11: Diurnal and nocturnal global distributions

Table 8.3: Fits results


Fit global global TransCAD TransCAD classified classified
diurnal nocturnal diurnal nocturnal diurnal nocturnal
a 0.34 0.28 0.42 0.38 0.31 0.27
b [dB] 0.06 0.49 0.07 0.84 -0.32 -1.03
c [dB] 3.1 4.0 2.6 2.9 3.4 4.0
95% 4.3 5.5 3.7 4.0 4.7 5.5
r-sq. 0.99 0.99 1.00 0.98 0.91 0.78

Therefore, we can assert that GPG is able to predict accuracy in a re-


liable way but we have to consider uncertainties due to sound power levels
and to flow period distribution. We want to underline that accuracy on cal-
culated levels are the ones in table 8.1. Moreover, we could estimate global
accuracy subtracting measurements uncertainty from obtained values: we
have about 3.8 dB for diurnal values and 5.1 dB for nocturnal.
Accuracy has been improved (see next section) on levels but main problem
remains nocturnal values: in fact, we obtained that one method underes-
timates and the other overestimates producing a broader distribution. A
possible improvement is to estimate correctly time coefficients for road traf-
fic flow during night especially on classified roads (overestimation of values
on classified roads is due to an overestimation of flow); furthermore, we could
improve accuracy measuring speed during night because, especially on main
roads, speed limits are often exceeded (that explains underestimation of
night levels on TransCAD roads).
CHAPTER 8. NOISE MAPPING RESULTS 85

8.4 Comparison with previous map results


If we compare first map with this one, we notice that residual distribution
has been improved: in fact, distribution width is smaller. Notice that this
improvement is not only due to traffic model (previous distribution is larger
even than classified one), but also to a more accurate 3D model. Despite
previewed nocturnal accuracy is quite the same of first map, residual distri-
bution is better (see figure 8.12): that’s another evidence of a more accurate
model.

Figure 8.12: Residual distributions

Comparison between population estimations has been performed: at that


time, IMMI version wasn’t able to manage automatic population exposure,
so calculation was carried out with GIS software (Arcview 3.2). Grid values
were assigned to residential buildings according nearest neighbour technique;
CHAPTER 8. NOISE MAPPING RESULTS 86

then residential population was calculated based on volume proportion be-


tween the considered building and all residential buildings in that census
zone (ISTAT cell, census 1991).
Population exposure has been calculated both with Italian and European
indicators: notice that LDEN and LN ight consider only incident sound. We
prefer to compare distributions according diurnal and nocturnal Italian in-
dicators (figure 8.13) because they are not corrected for reflections and so
correspond to grid values.
People exposure is more critical according new project: this fact could be
due to walls modelling (garden walls not included), but also to more real-
istic traffic flows. Anyway, if we consider number of exposed citizens, this
is quite the same because city population decreased. In fact, population
exposed to nocturnal levels higher than 50 dB increased of 4100 inhabitants
and to diurnal levels higher than 55 dB decreased of 5700 inhabitants.

Figure 8.13: Population comparisons


CHAPTER 8. NOISE MAPPING RESULTS 87

8.5 Strategic noise map


Strategic noise map comes from superposition of other two maps to road one:
aircraft noise has been calculated by Grad. Simonetti with software INM
and railway noise has been calculated by Grad. Cerchiai and Panicucci with
software SoundPlan. These maps have been exported as ASCII grids and
then transformed into IMMI format so that energy sum could be performed
according next equation:
 La Lr Ls

L = 10 log 10 10 + 10 10 + 10 10

in which:
La is aircraft noise level; Lr is railway noise level; Ls is streets noise level.
We have to underline that railway noise hasn’t been calculated over the
whole municipality, but only where it’s audible (i.e. till 2 km far from rail-
way lines).
Railway noise has important effects on west part of municipality where there
is Torino-Palermo line, but it’s not dominant because it’s parallel to high-
way A12 and to S.S. Aurelia. Other important effects are due to Pisa-
Florence line on east part because many buildings are close to railway.
Aircraft noise is prevalent on south-west part of municipality because de-
partures and arrivals go along the same routes so that most populated part
of the city is not affected. Of course road noise is prevalent in all other
situations in the town.
Accuracy of strategic map has been verified by Grad. Panicucci in [38]:
global accuracy depends obviously on which source is prevalent (according
next equation) but, if all sources are comparable, uncertainty is the biggest
one between sources.
p
100.2La ∆L2a + 100.2Lr ∆L2r + 100.2Ls ∆L2s
∆LG =
100.1La + 100.1Lr + 100.1Ls
in which:
∆LG is global levels uncertainty;
∆La is aircraft levels uncertainty;
∆Lr is railway levels uncertainty;
∆Ls is streets levels uncertainty.
Railway accuracy is the lowest and aircraft one is the higher: therefore, we
can assert that near railway levels accuracy is about 4.5 dB (estimated in
[38] using DEFRA4 position papers) and in all other locations it’s the same
as road map.
Strategic noise maps and single sources maps are shown in appendix C, here
we show only LDEN and LN igth maps (figures 8.14 and 8.15).

4
Department for Environment, Food and Rural Affairs of United Kingdom.
CHAPTER 8. NOISE MAPPING RESULTS 88

Figure 8.14: Strategic LDEN levels


CHAPTER 8. NOISE MAPPING RESULTS 89

Figure 8.15: Strategic Night levels (22.00-6.00)


CHAPTER 8. NOISE MAPPING RESULTS 90

8.5.1 People exposure to global levels


Global levels exposure graphics (see figures 8.16, 8.17) show that road noise
is dominant and people distribution is quite the same as road noise: despite
this, there are high annoyed people affected from aircraft and railway noise in
proximity of these infrastructures. The evidence of this fact is decreasing of
people with quiet façades. Finally, we should underline that annoyance due
to different sources could be perceived in different ways [39]: many studies
are going to identify indicators to evaluate correctly the contribution of each
source so as they are perceived (see [40], Silence project [41]).

Figure 8.16: Inhabitants exposure

Figure 8.17: Inhabitants with quiet façades

8.5.2 Conflicts maps


In addition to people exposure, we elaborated conflicts maps: these maps
are an efficient instrument to show critical areas. In fact, they show dif-
ferences between law limits and calculated levels. Nowadays there aren’t
limits for European indicators, so in Italy we elaborate conflicts maps for
Italian indicators [42] whose limits are established not only by PCCA, but
also by DPR n.142, 30/3/04 (road noise, [43]) and DPR n.459, 18/11/98
(railway noise, [44]); maps in figures 8.18 and 8.19 show these limits for Pisa
CHAPTER 8. NOISE MAPPING RESULTS 91

municipality. Conflicts maps in figures 8.20 and 8.21 report differences.


A detailed area is shown in figures 8.22 and 8.23 of the most critical area:
in fact, south-east area is affected by Fi-Pi-Li, railway and aircraft noise so
it’s a very annoyed area especially during night period. Other critical areas
are a sector of S. Rossore Park, because of A12 highway noise in a very low
limits zone, and south area next to railway and A12 which has low popu-
lation density. These maps will be useful to manage action plans: PCRA
establishes priority index based upon receiver type (school, hospital, house)
and how much limits are exceeded, therefore is essential to know differences
at receivers.
CHAPTER 8. NOISE MAPPING RESULTS 92

Figure 8.18: Diurnal limits


CHAPTER 8. NOISE MAPPING RESULTS 93

Figure 8.19: Nocturnal limits


CHAPTER 8. NOISE MAPPING RESULTS 94

Figure 8.20: Diurnal differences between levels and limits


CHAPTER 8. NOISE MAPPING RESULTS 95

Figure 8.21: Nocturnal differences between levels and limits


CHAPTER 8. NOISE MAPPING RESULTS 96

Figure 8.22: Diurnal south-east differences between levels and limits

Figure 8.23: Nocturnal south-east differences between levels and limits


Chapter 9

Conclusions and
developments

This work has been carried out in Pisa ARPAT department and it produced
Pisa road noise map according European Noise Directive 2002/49/EC which
has been transposed in Italy by D.L. n.194, 19/8/05.
A new approach for traffic flow estimation has been performed: TransCAD
software has been used and an innovative technique has been developed to
implement it in noise mapping procedure. We verified that noise mapping
accuracy has been improved, according forecasts of Good Practice Guide
and we produced new information about speed and flow starting from a
limited number of measurements.
Traffic flow measurements have been used to calibrate model and then to
verify reliability of estimated flow. A procedure has been tested and veri-
fied to use passenger car equivalent flow into the traffic model and then to
provide NMPB vehicles categories requested by the END.
Moreover, traffic measurements have been used together with sound levels
ones to calibrate noise model; finally, we validated noise estimation with
available measurements and theoretical considerations.
This calibrated model is able to predict both traffic and noise hot spots with
good accuracy: actually it’s possible to obtain sound levels over the whole
municipality with following accuracy: approx 66% of diurnal and nocturnal
values are far from measured values less than 2.6 dB. These values confirm
GPG suggested uncertainty: it means that future improvements should try
to modify modelling methods which have an high influence over global un-
certainty (i.e. the ones which contribute with more than 1 dB). In fact, it’s
clear from GPG toolkits that low flow roads and especially speed estimation
are critical problems.
Possible improvement is to include night flow estimation into traffic model;
this choice it’s possible only through traffic measurement campaigns on sam-
ple roads during night time. These future campaigns should include speed

97
CHAPTER 9. CONCLUSIONS AND DEVELOPMENTS 98

measurements: Pisa mobility agency has recently installed speed and flow
detectors at town boundary streets, so they could provide input data for a
more accurate traffic model in which free speeds might exceed law limits.
In fact, principle obstacle to produce night flow model was lack of night
data over a sufficient number of streets; furthermore, a night traffic model
should use a transport network different from diurnal one, including only
main roads (too low flows are not estimable).
Another improvement of traffic model, which will produce an improvement
on sound levels, is to include traffic lights cycles and use an assignment able
to predict volume-dependent intersections delays. This would produce not
only more reliable flows, but also more reliable speeds on links. Furthermore,
if we consider traffic lights, we could also identify links with acceleration and
deceleration.
Traffic modelling could be also used for atmospheric emissions estimation
in order to find a key synergy to tackle environmental issues as an holistic
approach.
The strong effort performed to produce an accurate 3D model will be use-
ful for future studies: for example, if new census data will be available or
new law limits, it will be sufficient to update values without building new
projects.
This work is also a useful instrument for future action plans because of
provided results: it will allow authorities to test solutions in which traffic
circulation change is a management tool to reduce noise levels.
So this paper, together with the one of strategic map, will allow munici-
pality to draw up action plans. Pisa has to draw up Italian local action
plans PCRA, whose aim is to manage protecting measures and to promote
policies oriented to noise lowering. Protecting measures for highly annoyed
people (identified through conflicts maps) should be financed by responsi-
ble infrastructures. In fact, strategic noise map is essential for action plans
because it allows identifying contribution of each source and establishing
which infrastructure is responsible for specific limits overcoming.
Finally this thesis produced data requested by the END to draw up Euro-
pean action plans: in fact, together with neighbour municipalities, Pisa is
an agglomerate with more than 100.000 inhabitants and should therefore
manage action plans. This thesis tackle both European and local policies
providing technical management solutions.
Appendix A

Acoustics basics

Sound Pressure level (SPL) Lp is a logarithmic measure of the root mean


square (rms) sound pressure of a sound relative to a reference value. Sound
pressure is the local pressure variation from the atmospheric room pressure
caused by a sound wave at a given location and given instant of time.
Let’s consider an infinitesimal volume V0 = dx dy dz with ρV0 mass mov-
ing with air at speed ux and a pressure gradient ∂p/∂x along x direction;
therefore, force along x is given by pressure difference multiplied by surface:
∂p
fx = − dx(dydz)
∂x
This force is balanced by acceleration so motion law is given by:
∂p dux
V0 = −ρV0
∂x dt
Density doesn’t usually vary too much so we can write previous equation as:
∂p ∂ux
= −ρ0
∂x ∂t
Moreover, if we consider all directions, we obtain Eulero equation:
∂~u
grad p = −ρ0
∂t
However, motion induces a mass variation inside the volume but according
mass conservation we can assert that time variation of density δ = (ρ−ρ0 )/ρ0
is related to speed divergence by: ∂δ/∂t = −div ~u. These variations are very
fast so process is considered adiabatic so that following equation is given:
1 ∂p ∂δ
=γ = −γdiv ~u
p0 ∂t ∂t
in which p0 is static pressure and γ = cp /cv is specific heat ratio. Therefore,
wave equation is calculated:
1 ∂2p
∇2 p =
c2 ∂t2

99
APPENDIX A. ACOUSTICS BASICS 100

in which c is sound speed defined by:


r
p0
c= γ
ρ0

RMS sound pressure is the root mean square of instantaneous one over a
given interval of time. SPL is measured in decibels (dB) and reference sound
pressure is 20µP a which is considered as the threshold of human hearing.
So SPL is given from the following expression:
 2 
prms
Lp = 10 log
p20

At the same way, sound power level is given for a reference sound power of
10−12 W:  
W
LW = 10 log
W0
Sound power level (measured one meter far from source) is related to sound
pressure level so that for a point source in free field condition and for room
temperature (i.e. when ρ0 c is 400P a · s/m) SPL is given from:

Lp = LW − 10 log 4π

instead for a linear source:

Lp = LW − 10 log 2π

However, sound perception varies with frequency that is SPL at different


frequencies is heard at different loudness: therefore, it’s been defined equal
loudness contours (expressed in Phon) which are a family of curves functions
of frequency.
In particular the 40 dB curve is called A-weighting and it’s used to correct
SPL to mirror real human sensation. So, environmental sound pressure level
is usually1 expressed in dB(A) where A-weighting correction is applied to
each octave band.
Environmental noise is estimated through A-weighted equivalent level which
corresponds to a constant hypothetical source whose sound energy is the
same of real time-varying sound:
 Z T 2 
1 pA (t)
LAeq = 10 log dt
T 0 p20

Notice that T is time reference interval and it’s usually an hour so we speak
about LAeq,h .
Another important indicator is Sound Exposure Level SEL which is used
1
Aircraft noise is weighted with 100 dB countours.
APPENDIX A. ACOUSTICS BASICS 101

to identify contribution of single events: it’s the level that the event would
assume if all his energy would be concentrated in one second
Te
SEL = LAeq + 10 log
1sec
in which Te is real event time length.
Law limits in Italy are expressed as average diurnal and nocturnal values
of LAeq,h : this means that we estimate average hourly equivalent level over
diurnal period (6.00:22.00) and nocturnal period (22.00:6.00).
" #
1 X LAeq,hi
LD = 10 log 10 10
16
i
" #
1 X LAeq,hi
LN = 10 log 10 10
8
i

in which i varies along hours.


The END instead establishes different indicators: in addiction to LN called
LN ight , it defines LDEN as a more complete indicator defined by the following
expression (time periods adapted according DL n.194, 19/08/05):
 
1 Lday Levening +5 Lnight +10
LDEN = 10 log 14 · 10 10 + 2 · 10 10 + 8 · 10 10
24

in which Lday , Levening , Lnight are the A-weighted long-term average sound
level as defined in ISO 1996-2: 1987, determined over all day (6.00:20.00),
evening (20.00:22.00) or night (22.00:6.00) periods of a year.
These levels should be estimated at 4 m height and they should consider
only incident sound, this means subtraction of 3 dB must be done measuring
façade levels.
Appendix B

Road noise maps

We show whole municipality maps of day, evening and diurnal time period
(figures B.1-B.3). Then detailed maps of LDEN and LN ight levels are shown:

• City centre zone with principal roads and ZTL zones


(figures B.4 and B.5);

• Porta a Lucca residential area at the north limits of the town


(figures B.6 and B.7);

• S. Chiara hospital area


(figures B.8 and B.9);

• Cisanello hospital area


(figures B.10 and B.11).

Notice that S. Chiara hospital is going to be dismissed so whatever kind of


buildings will be built, it will lay in a quiet area only if actual hospital walls
(or similar) would not be destroyed: in fact, Via Bonanno has high power
levels and it might affect the area.
Moreover, Cisanello hospital lies actually in a quiet area but with the future
enlargement a new viability is previewed: therefore, administrations should
pay attention to position of hospital rooms and major access roads.

102
APPENDIX B. ROAD NOISE MAPS 103

Figure B.1: Road traffic noise Day levels (6.00-20.00)


APPENDIX B. ROAD NOISE MAPS 104

Figure B.2: Road traffic noise Evening levels (20.00-22.00)


APPENDIX B. ROAD NOISE MAPS 105

Figure B.3: Road traffic noise Diurnal levels (6.00-22.00)


APPENDIX B. ROAD NOISE MAPS 106

Figure B.4: Road traffic noise LDEN levels

Figure B.5: Road traffic noise LN ight levels (22.00-6.00)


APPENDIX B. ROAD NOISE MAPS 107

Figure B.6: Road traffic noise LDEN levels

Figure B.7: Road traffic noise LN ight levels (22.00-6.00)


APPENDIX B. ROAD NOISE MAPS 108

Figure B.8: Road traffic noise LDEN levels

Figure B.9: Road traffic noise LN ight levels (22.00-6.00)


APPENDIX B. ROAD NOISE MAPS 109

Figure B.10: Road traffic noise LDEN levels

Figure B.11: Road traffic noise LN ight levels (22.00-6.00)


Appendix C

Strategic noise maps

Strategic maps are energy sum of three maps: aircraft noise, railway noise
(both 10 meters step grids interpolated to obtain 5 m) and road noise. We
show day, evening and diurnal strategic maps (not shown in previous chap-
ters) in figures C.1-C.3 and maps of aircraft (figures C.4-C.6) and railway
noise (figures C.7-C.9) in terms of Diurnal, DEN and Night levels.

110
APPENDIX C. STRATEGIC NOISE MAPS 111

Figure C.1: Strategic Day levels (6.00-20.00)


APPENDIX C. STRATEGIC NOISE MAPS 112

Figure C.2: Strategic Evening levels (20.00-22.00)


APPENDIX C. STRATEGIC NOISE MAPS 113

Figure C.3: Strategic Diurnal levels (6.00-22.00)


APPENDIX C. STRATEGIC NOISE MAPS 114

Figure C.4: Aircraft Diurnal levels (6.00-22.00)


APPENDIX C. STRATEGIC NOISE MAPS 115

Figure C.5: Aircraft LDEN levels


APPENDIX C. STRATEGIC NOISE MAPS 116

Figure C.6: Aircraft LN ight levels (22.00-6.00)


APPENDIX C. STRATEGIC NOISE MAPS 117

Figure C.7: Railway Diurnal levels (6.00-22.00)


APPENDIX C. STRATEGIC NOISE MAPS 118

Figure C.8: Railway LDEN levels


APPENDIX C. STRATEGIC NOISE MAPS 119

Figure C.9: Railway LN ight levels (22.00-6.00)


Bibliography

[1] European Parliament and Council. Directive 2002/49/EC of 25 June


2002. Official Journal of the European Communities, 17-07-2002.

[2] Presidente della Repubblica Italiana. Decreto Legislativo 19 Agosto


2005, n.194. Gazzetta Ufficiale n.222 del 23-9-2005.

[3] Ministero dell’Ambiente. Decreto 29 Novembre 2000. Gazzetta Ufficiale


n.285 del 6-12-2000.

[4] Presidente del Consiglio. DPCM 14 Novembre 1997. Gazzetta Ufficiale


n.280 del 1-12-1997.

[5] European Commission Working Group Assessment of Exposure to


Noise. Good practice guide for strategic noise mapping and the pro-
duction of associated data on noise exposure, 2007.

[6] G.Memoli. Mappatura acustica del comune di Pisa (Parte I ). Technical


report, IPCF, 2006.

[7] CSTB-SETRA-LCPC-LRPC. Bruit des Infrastructures routières


NMPB-Routes-96. CERTU, 1997.

[8] CETUR. Guide du bruit des transports terrestres: prevision des niveaux
sonores. CETUR, 1980.

[9] J. Defrance et al. Harmonoise,NMPB-96 - Road Traffic Noise New


French Method for Outdoor Sound Propagation: complete document.
Technical report, IST, 2002.

[10] M.Beuving; B.Hemsworth. Final synthesis report. Technical report,


IMAGINE, 2007.

[11] WP2 partners. Guidelines for the use of traffic models for noise mapping
and noise action planning. Technical report, IMAGINE, 2006.

[12] H. Jonasson. Harmonoise, work package 1.1: Source modelling of road


vehicles. Technical report, IST, 2003.

120
BIBLIOGRAPHY 121

[13] WP2 partners. Review of the suitability of traffic models for noise
mapping. Technical report, IMAGINE, 2004.

[14] M.Arana et al. Noise map of Pamplona, Spain. Main results. Acústica,
October 2008.

[15] D.Connolly. Roads input data. www.scotland.gov.uk, 2007.

[16] B.McKell. Noise mapping process. www.scotland.gov.uk, 2007.

[17] F.R.Pinto. Projecto-piloto de demonstração de mapas de ruı́do - escalas


municipal e urbana. Technical report, Istitudo do Ambiente, 2004.

[18] J.Patricio. Environmental noise. Legislation and experience in Portugal.


In Progettare il risanamento acustico, March 2006.

[19] T.Verdolini; D.Casini; A.Poggi. Rapporto tecnico sulla metodologia se-


guita per l’elaborazione della mappa acustica dell’agglomerato urbano
di firenze in adempimento alla direttiva europea 49/02. Technical re-
port, ARPAT, 2008.

[20] L.Moran et al. Fattori correttivi per i dati di emissione da utilizzare


nei modelli previsionali di rumore stradale in ambito urbano. Technical
report, ARPAT, 2005.

[21] D.Casini; L.Moran; A.Poggi; T.Verdolini. Accuratezza dei livelli sonori


stimati a firenze mediante un modello acustico di tipo deterministico.
Technical report, ARPAT, 2008.

[22] G.Memoli. Mappatura acustica del comune di Pisa (Parte II ). Tech-


nical report, IPCF, 2006.

[23] G.Licitra; M.Cerchiai; C.Chiari. Mappatura acustica e piani di azione


nell’attività dell’ARPAT. In Progettare il risanamento acustico, March
2006.

[24] G.Memoli. Testing the acoustical corrections for reflections on a façade.


Master’s thesis, Università di Pisa, 2005.

[25] ARPAT. Linee guida tecniche per la predisposizione dei piani di clas-
sificazione acustica. www.arpat.toscana.it, 2004.

[26] G.Licitra; G.Memoli. Limits and advantages of Good Practice Guide


to noise mapping. In Acoustic08-Paris, 2008.

[27] A. Panicucci. Studio del contributo di rumorosità prodotto dai mezzi a


due ruote nell’ambito della valutazione del clima acustico del Comune
di Pisa. Master’s thesis, Università di Pisa, 2005.
BIBLIOGRAPHY 122

[28] E.Cascetta. Metodi quantitativi per la pianificazione dei sistemi di


trasporto. CEDAM, 1990.

[29] Caliper Corporation. Travel Demand Modeling with TransCAD 4.8,


2005.

[30] P.Ferrari. Pianificazione dei trasporti. Pitagora, 2001.

[31] National Research Council. Highway Capacity Manual. Transportation


Research Board, 1985.

[32] Via traffic controlling gmbh. Viacount II Apparecchio per il monitor-


aggio del traffico, 2004.

[33] Progetto di norma U20 00 1441: Linee guida alla mappatura acustica
e mappatura acustica strategica (Parte I), December 2008.

[34] W ÖLF EL. IMMI 6.2 Druck, 2007.

[35] G. Brambilla et al. Indicazioni operative per la costruzione


dell’indicatore “Popolazione esposta al rumore” in riferimento alla Di-
rettiva Europea 2002/49/CE. Technical report, APAT CTN-AF, 2005.

[36] WP1 partners. Guidelines and good practice on strategic noise map-
ping. Technical report, IMAGINE, 2007.

[37] ISO 1996-2 Acoustics-Description, measurement and assessment of en-


vironmental noise - Part2: Determination of environmental noise levels,
March 2007.

[38] A. Panicucci. Definizione della mappatura acustica strategica del Co-


mune di Pisa ai sensi della Direttiva europea 49/2002/EC. Master’s
thesis, Università di Pisa, 2009.

[39] G.Licitra; M.Nolli. Mappatura del territorio e disturbo da rumore. In


Il disturbo da rumore, April 2008.

[40] B.Griefahn. SILENCE Report to partners: Annoyance of residents


living in urban areas. Technical report, European Commission DG
Research, 2007.

[41] Bonn Office. Night noise guidelines (NNGL) for Europe. Technical
report, European Centre for Environment and Health, 2007.

[42] G.Licitra. Implementazione della direttiva europea nel sistema nor-


mativo nazionale: opportunità e problemi aperti. In La direttiva
2002/49/CE: determinazione e gestione del rumore ambientale e suo
impatto sulla legislazione italiana, November 2004.
BIBLIOGRAPHY 123

[43] Presidente della Repubblica Italiana. DPR 30 Marzo 2004, n.142.


Gazzetta Ufficiale n.127 del 1-06-2004.

[44] Presidente della Repubblica Italiana. DPR 18 Novembre 1998, n.459.


Gazzetta Ufficiale n.2 del 4-01-1999.

[45] Parlamento Italiano. Legge 26 Ottobre 1995, n.44. Gazzetta Ufficiale


n.254 del 30-10-1995.

[46] WP2 partners. Review of data needs for road noise source modelling.
Technical report, IMAGINE, 2004.

[47] M.Arana. Spanish legislation against noise. Some recent studies. In


Progettare il risanamento acustico, March 2006.

[48] B.McKell. Facilitation of strategic noise mapping for the Environmental


Noise Directive 2002/49/EC implementation. Technical report, Scottish
Executive Central Research Unit, 2005.

[49] D.Casini (ARPAT). Note on traffic noise modelling by means of line


type sources. Rivista italiana di acustica, Gen-Mar 2008.

[50] G.Licitra; F.Bertellino. Modellli di previsione del rumore determinato


dalle attività produttive e dal traffico veicolare, aereoportuale e fer-
roviario. In Noise mapping, June 2001.

[51] R.Spagnolo. Manuale di acustica applicata. UTET, 2001.

[52] M.Beuving. State of the art. Technical report, IMAGINE, 2004.

[53] G. Brambilla et al. Procedure per la conversione dei dati esistenti


sul rumore ambientale nei descrittori previsti dalla Direttiva Europea
2002/49/CE. Technical report, APAT CTN-AF, 2005.

Vous aimerez peut-être aussi