Seismic Waveform Tomography at a Test Site with Open Chimneys

Khiem T. Tran1, Michael McVay2, David Horhota3, Michael Faraone4, and Brian W. Sullivan5

ABSTRACT:

The paper presents an application of 2-D time-domain waveform tomography to map the extent
of open chimneys at a variable karstic limestone site. The seismic surface wavefields were
measured next to three open chimneys and inverted using a full waveform inversion (FWI)
technique, which is based on a finite-difference solution of 2-D elastic wave equations and
Gauss-Newton inversion method. Both the compression wave (P-wave) and shear wave (S-
wave) velocities are inverted independently and simultaneously to increase credibility of
characterized profiles. The waveform analysis successfully profiled embedded low-velocity
zones and highly laterally and vertically variable limestone. The inverted results are consistent to
the known open chimneys observed from the ground surface. Poisson’s ratio determined from
the inverted P-wave and S-wave velocities is consistent with soil types; high values of 0.3 to 0.5
for silt and clay and low values of 0.1 to 0.2 for limestone. Results presented here show that the
FWI technique is very applicable to both shallow and deep foundation design where soil
stratigraphy and variability are important. It is also computationally practical, as the results were
all achieved in about three hours of computer time on a standard laptop computer.

_________________________________________________________________________
1
Assistant Professor, Clarkson University, Department of Civil and Environmental Engineering,
P.O. Box 5710, Potsdam, NY 13699-5710, email: ktran@clarkson.edu;
2
Professor, University of Florida, Department of Civil and Coastal Engineering, 365 Weil Hall,
P.O. Box 116580, Gainesville, FL, 32611, email: mcm@ce.ufl.edu;
3
State Geotechnical Engineer, Florida Department of Transportation, 5007 N.E. 39th Avenue
Gainesville, FL 32609, email: david.horhota@dot.state.fl.us;
4
Student, University of Florida, Department of Civil and Coastal Engineering, 365 Weil Hall,
P.O. Box 116580, Gainesville, FL, 32611, email: fara1@ufl.edu;
5
Student, Clarkson University, Department of Civil and Environmental Engineering, P.O. Box
5710, Potsdam, NY 13699-5710, email: sullivbw@clarkson.edu;

1
TRB 2014 Annual Meeting Paper revised from original submittal.
 Introduction
The assessment of soil/rock stratigraphy and associated properties at a site is of significant
importance in the design, construction and maintenance of all civil infrastructure systems.
Unanticipated site conditions such as highly variable soil/rock layers with embedded low-
velocity anomalies (soft soils) cause significant problems during and after construction of
foundations. Knowledge of the low-velocity anomalies is crucial for design of foundations, as
the anomalies can cause structural damages or collapses. Traditional surface-based geophysical
methods have been usually used to identify spatial variation and embedded anomalies. These
include techniques involving wave velocity dispersion, such as spectral analysis of surface waves
(SASW) (Nazarian, 1984), multi-channel analysis of surface waves (MASW) (Park et al., 1999),
refraction microtremor (ReMi) (Louie, 2001); and techniques using travel times, like
conventional seismic refraction and seismic refraction tomography (SRT). However, these
standard techniques employed in current engineering evaluations have limitations in dealing with
challenging profiles, which include anomalies such as low-velocity zones. O’Neill et al. (2003)
identified that conventional methods using wave velocity dispersion may not produce an accurate
inverse shear wave velocity profile, which include a buried low-velocity layer. Similarly, travel
time techniques which only use first-arrival signals to characterize subsurface conditions suffer
other limitations. Specifically, the travel time from a source to a receiver is measured from the
fastest ray that starts from the source and travels through a medium to the receiver. The fastest
ray tends to go through high-velocity zones and avoid low-velocity zones; thus SRT does not
characterize low-velocity zones (Tran and Hiltunen, 2012c). In addition, as discussed by
Sheehan et al. (2005) and others, SRT always models a sharp contrast in velocity with a gradient
in velocity, resulting in poor profiling of layer interfaces with sharp contrast in velocities. For
example, a sharp interface of soil/rock will be modeled with a zone (typically a few meters) of
increasing velocity which could be mistaken as weathered rock.

As reviewed by Plessix (2008) and Vireux and Operto (2009), full-waveform inversion (FWI)
offers the potential to produce more detailed resolution of the subsurface by extracting
information contained in the complete seismic waveforms. Nasseri-Moghaddama et al. (2007),
for example, have clearly shown that the recorded responses at the surface can carry valuable
information regarding the presence and characterization of anomalies, e.g., voids below the
surface, as well as embedded weaker layers. Many algorithms for waveform inversion have been
developed and applied to synthetic and real data in large-scale (kilometric scale) domains (Shipp
and Singh, 2002; Ravaut et al., 2004; Sheen et al., 2006; Cheong et al. 2006; Brenders and Pratt,
2007; and Choi and Alkhalifah, 2011). Generally, for the large scale, surface waves are easily
separated from body waves and removed from the inversion. However, at small scale (0-100 m),
it is difficult to separate body waves from surface waves, and only a few studies of waveform
inversion involving both body and surface waves have been performed for near-surface
investigations on synthetic data (Ge´lis et al., 2007; and Romdhane et al., 2011)

Recently, a full waveform inversion (FWI) technique (Tran and McVay, 2012) was reported
which inverted both body and surface waves in the case of real experimental data. The technique
employs a Gauss-Newton technique to invert the seismic full wave fields of near-surface velocity
profiles by matching the observed and computed waveforms in the time domain. Virtual sources
and a reciprocity principle are used to calculate partial derivative wavefields (gradient matrix) to
reduce the computer time. Observed and estimated wavefields are convolved with appropriate

2
TRB 2014 Annual Meeting Paper revised from original submittal.
 reference traces to remove the influence of source signatures, i.e. the inversion technique is
independent of sources or source signatures are not required to be measured during field testing.
The technique has been successfully applied to a real experimental data set of a laterally uniform
profile (horizontal layers). Herein, another application of the FWI is presented for a site with
high lateral and vertical variations (i.e. site variability) with embedded low-velocity anomalies.
The inversion was carried out independently for the S-wave and P-wave velocities in each cell
with the mass density of the medium assumed constant.

Full Waveform Inversion

The full waveform inversion technique has been presented in details by Tran and McVay (2012).
For completeness, it is briefly described herein. The technique includes forward modeling to
generate synthetic wavefields and Gauss-Newton inversion method to update model parameters
until residual between predicted and measured surface velocities are negligible. For the forward
modeling, the classic velocity-stress staggered-grid finite difference solution of 2-D elastic wave
equations in the time domain (Virieux, 1986) is used in combination with perfectly matched
layer boundary conditions (Kamatitsch and Martin, 2007). For model updating, Gauss-Newton
method involves minimizing the residual between the estimated responses obtained by forward
simulation and the observed seismic data. Implementation of our full waveform inversion
scheme involves the following seven steps:
1) Filter measured (observed) seismic data to remove high-frequency noises, and transform
the observed data to reduce 3-D and near-field effects.
2) Select good channels (from observed data) to be reference traces for cross convolutions
to remove influence of source signatures during inversion.
3) Calculate residual between estimated and observed data for the i-th shot and j-th receiver,
'di , j Fi , j (m) di ,k  di , j Fi ,k (m) ,
(1)
where di,j and Fi,j (m) are the observed data and the estimated data associated with the model m.
The model m includes all unknowns (S-wave and P-wave velocities of cells). Fi,k (m) and di,k are
the reference traces from the estimated and observed data, respectively, at the k-th receiver
position that is selected from step 2 for the i-th shot. The symbol * denotes the convolution.
4) Calculate the least-squares error E(m):
'd 'd, where 'd ^'di , j , i 1,... NS , j 1... NR`
1 t
E(m) (2)
2
where the superscript t denotes the matrix transpose. NS and NR are the numbers of shots and
receivers, and Δd is a column vector, which is the combination of residuals Δdi,j for all shots and
receivers. The change of E(m) from one to the next iteration is used to determine if the
convergence criterion has been reached (less than 1%).
5) Calculate gradient of data from the i-th shot and j-th receiver with respect to model mp
using virtual source and reciprocal wavefields (Sheen et al., 2006):
wFi , j (m) wFi ,k (m)
J i,j d i ,k  d i , j , i 1...NS , j 1...NR (3)
w mp w mp
6) Update model m at iteration n+1 from iteration n
mn1 mn  D n [J t J  O1 Pt P  O2 I t I @ J t 'd, (4)
where I is the identity matrix, and P is a matrix, whose elements are either 1, -4, or 0,
determined by a 2-D Laplacian operator. Step length ( D n ) is simply kept as a constant value of

3
TRB 2014 Annual Meeting Paper revised from original submittal.
 0.5 through all the iterations. The choice of coefficients O1 and O2 between 0 and infinity is a
compromise result. Larger values provide more optimization stability but produce smoother
inverted velocity models. Several computations are conducted in this study and it is found that
the values of 0.05 for O1 and 0.0005 for O2 suggested by Sheen et al (2006) are reasonable. For
more details, see Sheen et al. (2006) and Tran and McVay (2012).
7) Repeat steps 3 to 6 until the convergence is achieved. Herein, each inversion run is
simply stopped after 20 iterations when the change of the least-squares error E(m) is
usually less than 1% from one to the next iteration.

Application
The 2-D FWI scheme has been applied to a real test site to investigate the capability of the FWI
in characterizing highly variable subsurface profiles and embedded anomalies. The test site was a
Florida Department of Transportation (FDOT) retention pond located in Newberry, Florida.
From invasive tests, the site consisted of medium dense, fine sand and silt 2- to 5-m thick,
overlying a highly variable limestone deposit; the top of limestone varied from 2 m to 10 m in
depth (Tran and Hiltunen, 2011). The site was divided into 26 parallel north–south survey lines
equally spaced 3.0 m apart. The lines were labeled A through Z from west to east across the site,
and each line was 85.3 m long, with station 0 m located at the southern end of each line. Two
test lines were conducted next to open chimneys as shown in Fig. 1. Line 1 was at the grid line
G20–G140, next to open chimneys 1 and 2. Line 2 was at the grid line 95A–95K, perpendicular
to line 1, and next to open chimneys 2 and 3. Due to safety concerns, the two test lines were
conducted about 1 m away from the chimneys.

Line 1 – G6–G42
Line 1 was conducted using a linear array of 24 4.5-Hz vertical geophones at a spacing of 1.5 m,
for a total receiver spread of 34.5 m (station 0.75 m to 35.25 m). The seismic energy was created
by striking a 150-mm square metal plate with a 90 N sledgehammer. Twenty-five shots at 1.5 m
spacing were recorded, for a total shot spread of 36.0 m (station 0.0m to 36.0 m).

As the FWI analysis is two dimensional and wavefields are modeled with a plain strain
condition, the active source should be modeled as a line source (not a point source as a hammer
blow). This discrepancy creates near-field effects/3-D effects that may limit capability of the
technique in characterization of small 3-D anomalies. To reduce the near-field effects/3-D
effects, geometrical spreading corrections (Schäfer et al., 2012) are implemented into the FWI.
The purpose of the geometrical spreading corrections is to transform a point-source wave field
into an equivalent line-source wave field. The corrections include amplitude and phase
transformations. The amplitude transformation is mandatory due to the misfit definition with the
L2 norm. Because the amplitude-decay of waves excited by a point (measured) and a line source
(modeled) for surface waves are different, the modeled wavefield is corrected by an offset
dependent correction factor of the form y(r) = A· rD with offset r. The determination of the factor
A and the exponent Dis done with an iterative least squares inversion by minimization the
energy of waveform residuals. The values of A and D are estimated at the beginning of the
inversion, and kept constant during inversion process. The phase transformation is simply to
convolve the point-source wave field with t-1/2. In addition, it is difficult to separate the
amplitude-decay due to geometrical spreading and material damping, the determined correction

4
TRB 2014 Annual Meeting Paper revised from original submittal.
 factor, y(r), accounts for all kinds of attenuation (geometrical and material damping) of the
wavefields.

To avoid the inversion being trapped in local minima, a proper initial model is required. The
initial model could be generated by using global inversion techniques, such as genetic algorithm
(Tran and Hiltunen, 2012a) or simulated annealing (Tran and Hiltunen, 2012b) on full
waveforms. This approach likely produces a global solution but requires significant computer
time. Alternatively, from the study on synthetic model (Tran and McVay, 2012), a 1-D model
(i.e. no horizontal variability) is usually enough for inversion of a 2-D model. For simplicity, an
estimate of the initial model was established via a spectral analysis of the measured data. Figure
2 presents a normalized power spectrum obtained using the cylindrical beam-former technique
(Tran and Hiltunen, 2008) from the measured shot data at station 0 m. The observed energy in
the measured wave field concentrated in a narrow band, with Rayleigh wave velocities in the
range of 200 to 350 m/s. The S-wave velocity, which is slightly larger than Rayleigh wave
velocity, was therefore taken in the range of 200 to 400 m/s. Also, the depth of the medium
analyzed is assumed to be one half the length of total geophone layout (ensures signals pass
through domain of interest). Consequently, with a range of S-wave velocities, depth of the
model, and initial 1-D assumption (i.e. no initial variation in horizontal direction), a linear
increasing S-wave velocity (Fig. 4a, top) from 200 m/s at the surface to 400 m/s to a depth of
18m over a length of 36 m was considered. The initial P-wave velocity (Fig. 4a, bottom) for the
domain was calculated from the S-wave velocities assuming that the initial Poisson’s ratio
throughout the domain was 0.25. The mass density throughout the model was kept constant at
1800 kg/m3 for all inversions. Efforts to invert the mass density of medium from the measured
wavefields have been shown to be unsuccessful. This can be explained that most of the energy in
the full wave field measured on the surface are Rayleigh waves (Richart et al. 1970) that are not
very sensitive to mass density (Nazarian, 1984).

Three inversion runs were performed for frequency ranges with central frequencies of 10, 15,
and 20 Hz, beginning from the lowest frequency range. The medium of 18 m u 36 m was divided
into about 1200 cells of 0.75 m u 0.75 m. During the inversion, S-wave and P-wave velocities of
cells were updated independently, and each run was stopped after 20 iterations when the
observed waveform data and the estimated waveform data were similar. The observed surface
waveforms, the estimated surface waveforms, and residuals (difference between observed and
estimated) associated with the final inverted model are shown horizontally in Fig. 3 for the shot
at station 0 m. Evident, the observed and estimated data are very similar across the entire range
of offsets, and the residuals are small.

The inverted results of the three runs are shown in Figs. 4b, c and d for analysis of the data at 10,
15 and 20 Hz, respectively. Locations of chimneys 1 and 2 are also shown in Fig. 4d for
comparison. The final inverted S-wave profile (Fig. 4d, top) shows two low-velocity zones at
distances 12 m and 21 m, along with high lateral and vertical variations in limestone boundaries
(Vs > 800 m/s) at the bottom of profile. Evidently these anomaly locations were the same as
those of the chimneys (i.e., 12 m and 21 m). The inverted P-wave profile (Fig. 4d, bottom) was
consistent with the estimated S-wave profile. Chimney 1 of about 1.5-m diameter was also
characterized in both S-wave and P-wave images. Chimney 2 of about 1-m diameter was not
shown, due to 3-D effects. To characterize the smaller chimney, the test line may have needed to

5
TRB 2014 Annual Meeting Paper revised from original submittal.
 be closer to the chimney, and data at higher frequencies (20–40 Hz) may also have been
required.

The Poisson’s ratio profile was also calculated from the independently determined S-wave and
P-wave velocities (Fig. 4d) and shown in Fig. 5. The computed Poisson’s ratio is consistent with
soil/rock profiles. High values (0.30 to 0.50) were found for shallow materials of silt and silty
sands, and low velocity zones (blue area in Fig. 4d), which may be voids filled with raveled soil
and water. Low values (0.1 to 0.2) were found for high velocity zones (red and yellow areas in
Fig. 4d), which are dense sand and limestone.

Line 2 – 28A–28K
Line 2 was conducted using a linear array of 24 4.5-Hz vertical geophones at a spacing of 1.2 m
for a total receiver spread of 27.6 m (station 0.6 m to 28.2 m). Twenty-five shots at 1.2-m
spacing were recorded for a total shot spread of 28.8 m (station 0.0 m to 28.8m). Unlike line 1,
the geophone spacing of 1.2 m (instead of 1.5 m) was used in an attempt to characterize smaller
chimneys.

Similar to line 1, the S-wave velocity was determined from spectral analysis of dispersion data
with a range from 200 to 550 m/s. A linear increasing S-wave velocity (Fig. 6a, top) from 200
m/s at the surface to 550 m/s to a depth of 14.4 m (half of the test length) over a length of 28.8 m
was selected. The initial P-wave velocity (Fig. 6a, bottom) for the domain was calculated from
the S-wave velocities assuming that the initial Poisson’s ratio throughout the domain was 0.25.
Three inversion runs were performed for frequency ranges with central frequencies of 10, 15,
and 20 Hz, beginning from the lowest frequency range. The medium of 14.4 m u 28.8 m was
divided into 1152 cells of 0.6 m u 0.6 m. During the inversion, S-wave and P-wave velocities of
cells were updated independently, and each run was stopped after 20 iterations when the
observed waveform data and the estimated waveform data were similar.

The inverted results of the three runs are shown in Figs. 6b, c and d for analysis of the data at 10,
15 and 20 Hz, respectively. Locations of chimneys 2 and 3 are also shown in Fig. 6d for
comparison. The final inverted S-wave profile (Fig. 6d, top) shows a low-velocity zone at
distance 20 m near chimney 2, along with high lateral and vertical variations in limestone
boundaries (S > 800 m/s) at the bottom of profile. A valley of low-velocity area was found at
distance 8 m near chimney 3. The inverted P-wave profile (Fig. 6d, bottom) was consistent with
the estimated S-wave profile. The Poisson’s ratio profile was calculated from the independently
determined S-wave and P-wave velocities (Fig. 6d) and shown in Fig. 7. Again, the computed
Poisson’s ratio is consistent with soil/rock profiles. High values (0.30 to 0.50) were found for
shallow materials of silt and silty sands, and low velocity zones (blue area in Fig. 6d), which may
be voids filled with raveled soil and water. Low values (0.1 to 0.2) were found for high velocity
zones (red and yellow areas in Fig. 6d), which are dense sand and limestone.

For further verification of the inverted profiles, S-wave velocity profiles from two different
perpendicular lines that intersected are shown in Fig. 8. The intersection was at distance 22 m of
line 1 and distance 18 m of line 2. The similarity of two independent S-wave profiles suggested
consistency and credibility of the FWI.

6
TRB 2014 Annual Meeting Paper revised from original submittal.
 Conclusions
An application of Gauss-Newton inversion of full seismic elastic waveforms is presented for a
highly variable (horizontal and vertical) site. The full waveform inversion successfully identified
complex subsurface profiles including low-velocity embedded zones and highly variable
limestone surfaces at the bottom of profiles. The inverted results are consistent to known open
chimneys observed from the ground surface. The independent inverted S-wave velocity profiles
at the intersection of two perpendicular test lines are similar, suggesting consistency and
credibility of the full waveform inversion technique. Poisson’s ratio determined from the
inverted P-wave and S-wave velocities is consistent with soil types; high values of 0.30 to 0. 5
for silt and clay and low values of 0.1 to 0.2 for limestone. For the cases presented, full
waveform inversion is computationally practical, as the results obtained were all achieved in
about three hours of computer time on a standard laptop computer.

References
Brenders A. J. and Pratt R. G. (2007). Full waveform tomography for lithospheric imaging:
Results from a blind test in a realistic crustal model: Geophysical Journal International;
168(1): 133-151.
Cheong S., Pyun S., and Shin C (2006). Two efficient steepest-descent algorithms for source
signature-free waveform inversion: Journal of Seismic Exploration; 14: 335-348.
Choi Y., and Alkhalifah T. (2011). Source-independent time-domain waveform inversion using
convolved wavefields: Application to the encoded multisource waveform inversion:
Geophysics; 76 (5): R125-R134.
Ge´lis C., Virieux J., and Grandjean G. (2007). Two-dimensional elastic waveform inversion
using Born and Rytov formulations in the frequency domain: Geophysical Journal
International; 168: 605-633.
Kamatitsch D. and Martin R. 2007. An unsplit convolutional perfectly matched layer improved
at grazing incidence for the seismic wave equation. Geophysics 72(5), SM155–SM167.
Louie J. N. (2001). Faster, better, shear-wave velocity to 100 meters depth from refraction
microtremor arrays: Bulletin of Seismological Society of America; 91(2): 347-364.
Nasseri-Moghaddama A., Cascante G., Phillips C., and Hutchinson, D. J. (2007). Effects of
underground cavities on Rayleigh waves-Field and numerical experiments: Soil Dynamics
and Earthquake Engineering; 27: 300-313.
Nazarian S. (1984). In situ determination of elastic moduli of soil deposits and pavement systems
by spectral-analysis-of-surface-waves method: Ph.D. Dissertation, The University of Texas
at Austin.
O’Neill A., Dentith M., and List R. (2003). Full-waveform P-SV reflectivity inversion of surface
waves for shallow engineering applications: Exploration Geophysics; 34(3):158-173.
Park C. B., Miller R. D., and Xia J. (1999). Multi-channel analysis of surface wave (MASW):
Geophysics; 64(3): 800-808.
Plessix R. E. (2008). Introduction: Towards a full waveform inversion: Geophysical Prospecting;
56:761-763.
Ravaut C., Operto S., Improta L., Virieux J., Herrero A., and Dell’Aversana P. (2004).
Multiscale imaging of complex structures from multifold wide-aperture seismic data by
frequency-domain full-wavefield tomography: Application to a thrust belt: Geophysical
Journal International; 159: 1032–1056.

7
TRB 2014 Annual Meeting Paper revised from original submittal.
 Richart F. E., Hall J. R., and Woods R. D. (1970) .Vibrations of Soils and Foundations, Prentice-
Hall, Inc., Englewood Cliffs, New Jersey.
Romdhane A., Grandjean G., Brossier R., Rejiba F., Operto S., and Virieux J. (2011). Shallow-
structure characterization by 2D elastic full-waveform inversion: Geophysics; 76 (3): R81–
R93.
Schäfer M., L. Groos, T. Forbriger, and T. Bohlen, 2012, On the Effects of Geometrical
Spreading Corrections for a 2D Full Waveform Inversion of Recorded Shallow Seismic
Surface Waves: 74th EAGE Conference & Exhibition incorporating SPE EUROPEC
2012Copenhagen, Denmark, 4-7 June 2012
Sheen D.H., Tuncay K., Baag C. E. and Ortoleva P. J. (2006). Time domain Gauss–Newton
seismic waveform inversion in elastic media: Geophysical Journal International ;167:
1373–1384.
Sheehan J.R., Doll W.E., and Mandell W.A. (2005). An Evaluation of Methods and Available
Software for Seismic Refraction Tomography Analysis: Journal of Environmental and
Engineering Geophysics; 10:21-34.
Shipp R. M. and Singh S. C. (2002). Two-dimensional full wavefield inversion of wide-aperture
marine seismic streamer data: Geophysical Journal International; 151: 325-344.
Slob E., Sato M., and Olhoeft G. (2010). Surface and borehole ground-penetrating-radar
developments: Geophysics; 75:A103–20.
Tran K. T. and Hiltunen D. R. (2008). A comparison of shear wave velocity profiles from
SASW, MASW, and ReMi techniques: Geotechnical Earthquake Engineering and Soil
Dynamics IV, Geotechnical Special Publication; 181.
Tran K.T. and D. R. Hiltunen, (2011), Inversion of First-Arrival Time Using Simulated
Annealing, Journal of Environmental and Engineering Geophysics, 16, 25-35.
Tran K.T. and D. R. Hiltunen, (2012a), One-dimensional inversion of full waveform using
genetic algorithm: Journal of Environmental and Engineering Geophysics, 17, 197-213.
Tran K.T. and D. R. Hiltunen, (2012b), Two-dimensional inversion of full waveform using
simulated annealing: Journal of Geotechnical and Geoenvironmental Engineering, 138, no.
9, 1075-1090.
Tran K.T. and D. R. Hiltunen, (2012c). Inversion of combined surface and borehole first-arrival
time: Journal of Geotechnical and Geoenvironmental Engineering, 138, no. 3, 272-280.
Tran K. T. and M. McVay, 2012, Site Characterization Using Gauss-Newton Inversion of 2-D
Full Seismic Waveform in Time Domain: Soil Dynamics and Earthquake Engineering, 43,
16-24.
Virieux J. (1986). P–SV wave propagation in heterogeneous media: velocity–stress finite-
difference method. Geophysics; 51(4):889–901.
Vireux J. and Operto S. (2009). An overview of full-waveform inversion in exploration
geophysics: Geophysics; 74(6): WCC1-WCC26.

8
TRB 2014 Annual Meeting Paper revised from original submittal.
 a) Test locations

b) Chimney 1

Figure 1. Site 2: a) Test location diagram; b) Chimney 1 photo; c) Chimney 2 photo; and d)
Chimney 3 photo (continued on next page).

9
TRB 2014 Annual Meeting Paper revised from original submittal.
 c) Chimney 2

d) Chimney 3

Figure 1—continued

10
TRB 2014 Annual Meeting Paper revised from original submittal.
 800 1

0.9
700
Rayleigh Wave Velocity (m/s)

0.8
600
0.7
500
0.6

400 0.5

0.4
300
0.3
200
0.2
100
0.1

5 10 15 20 25 30
Frequency (Hz)

Figure 2. Line 1: normalized power spectrum of the real data from one shot

Observed data Final estimated data Final residual

0.7 0.7 0.7

0.6 0.6 0.6

0.5 0.5 0.5

Time (s)

Time (s)

Time (s)

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
0 10 20 30 0 10 20 30 0 10 20 30
Receiver position (m) Receiver position (m) Receiver position (m)

Figure 3: Line 1: comparison between observed and estimated data for one shot

11
TRB 2014 Annual Meeting Paper revised from original submittal.
 a) Initial model c) Inverted results at 15 Hz
S-W ave
S-W ave
0 400 0 800

Depth (m)
350 5
Depth (m)

5 600

300 10
10 400
250
15 15 200
200
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Distance (m) Distance (m)
P-W ave P-W ave
0 0 1200
600
Depth (m)

1000

Depth (m)
5 5
800
10 500 10
600
15 400 15 400
0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Distance (m) Distance (m)

b) Inverted results at 10 Hz d) Inverted results at 20 Hz

S-W ave Chimney 1 S-W ave
S Chimney 2
0 0
800
600
Depth (m)

5 5
Depth (m)

600

10 400 10
400

15 200 15 200

0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Distance (m)
Distance (m)
P-W ave P-W ave
0 0 1400
1000 1200
Depth (m)

5 5
Depth (m)

800 1000

10 10 800
600
600
15 15
400 400

0 5 10 15 20 25 30 35 0 5 10 15 20 25 30 35
Distance (m)
Distance (m)

Figure 4. Line 1: FWI results of S-wave and P-wave velocities (m/s): a) Initial model; b)
Inverted model at 10 Hz; c) Inverted model at 15 Hz; and d)Inverted model at 20 Hz.

Poisson Ratio
0
0.45
0.4
5
0.35
Depth (m)

0.3
10
0.25
0.2
15 0.15
0.1
0 5 10 15 20 25 30 35
Distance (m)
Figure 5: Line 1: Poisson’s ratio

12
TRB 2014 Annual Meeting Paper revised from original submittal.
 a) Initial model c) Inverted results at 15 Hz

S-Wave S-Wave
0 0
1000
500

Depth (m)
Depth (m)

5 800
5
400 600

10 10 400
300
200
0 5 10 15 20 25 0 5 10 15 20 25
Distance (m) Distance (m)
P-Wave
P-Wave
0
0 1000
1500

Depth (m)
Depth (m)

800 5
5
1000
600 10
10
500
400
0 5 10 15 20 25
0 5 10 15 20 25 Distance (m)
Distance (m)

b) Inverted results at 10 Hz d) Inverted results at 20 Hz

S-Wave Chimney 3 S-Wavee Chimney 2

0 0
800
600
Depth (m)

Depth (m)

5 5 600
400
10 10 400

200 200
0 5 10 15 20 25 0 5 10 15 20 25
Distance (m) Distance (m)
P-Wave P-Wave
0 0
1000 1200
Depth (m)

Depth (m)

5 5 1000
800
800
10 600
10 600
400
0 5 10 15 20 25 400
Distance (m) 0 5 10 15 20 25
Distance (m)

Figure 6. Line 2: FWI results of S-wave and P-wave velocities (m/s): a) Initial model; b)
Inverted model at 10 Hz; c) Inverted model at 15 Hz; and d)Inverted model at 20 Hz.

13
TRB 2014 Annual Meeting Paper revised from original submittal.
 Poisson Ratio
0

0.4
Depth (m)

5
0.3

10 0.2

0.1
0 5 10 15 20 25
Distance (m)

Figure 7: Line 2: Poisson’s ratio

S- wave velocity (m/s) at intersection of 2 lines

0 100 200 300 400 500 600

0

8
Depth (m)

10

12

14
Line 1: G6_G42

16
Line 2: 28A-28K

18

20

Figure 8. Comparison of inverted S-wave velocity at the intersection of two lines (distance 22
m of line 1 and distance 18 m of line 2).

14
TRB 2014 Annual Meeting Paper revised from original submittal.