See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/235760318

Site effects at Euroseistest—II. Results from 2D numerical modeling and

comparison with observations

Article  in  Soil Dynamics and Earthquake Engineering · March 2000

DOI: 10.1016/S0267-7261(99)00026-3

CITATIONS READS

92 273

4 authors, including:

Dimitrios Raptakis Konstantia Makra

Aristotle University of Thessaloniki Institute of Engineering Seismology and Earthquake Engineering
80 PUBLICATIONS   1,545 CITATIONS    75 PUBLICATIONS   864 CITATIONS   

SEE PROFILE SEE PROFILE

Kyriazis Pitilakis
Aristotle University of Thessaloniki
343 PUBLICATIONS   4,318 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Seismic deformation patterns of rectangular tunnels View project

PERPETUATE: Performance-based Approach to Earthquake Protection of Cultural Heritage in European and Mediterranean Countries (EC-FP7, Ref: 244229) View project

All content following this page was uploaded by Dimitrios Raptakis on 12 February 2018.

The user has requested enhancement of the downloaded file.

 SDEE 2411

Soil Dynamics and Earthquake Engineering 19 (2000) 23–39

www.elsevier.com/locate/soildyn

Site effects at Euroseistest—II. Results from 2D numerical modeling and

comparison with observations
F.J. Chávez-Garcı́a a, D. Raptakis b,*, K. Makra b, K. Pitilakis b
a
Instituto de Ingenierı́a, UNAM, Apdo. Postal 70-472, Coyoacán, 04510 Mexico, D.F. Mexico
b
Department of Civil Engineering, Aristotle University of Thessaloniki, P.O. Box 450, GR-54006 Thessaloniki, Greece
Received 18 November 1998; received in revised form 14 May 1999; accepted 15 August 1999

Abstract
This paper presents results of numerical modeling of site response for Euroseistest. Ground motion across a very detailed model of the
subsoil of this valley has been simulated for vertically incident SH waves. The predominance of locally generated surface waves is very clear
in the synthetic seismograms. These results are then compared with published studies of observed site effects at this basin and with a detailed
analysis of two events in the time domain. It is discussed in which sense it is possible to obtain a good fit between observations and 1D
models, even though the real behavior involves locally generated Love waves. For this reason, it can be misleading to rely on an incomplete
observation such as empirical transfer functions. Finally, it is stressed that in order to predict ground motion in alluvial valleys the
information contained in the phase cannot be neglected. q 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Euroseistest; Test site; 2D model; Finite difference; Site response; Locally generated surface waves

1. Introduction during 1992 as part of a project financed by the European

The importance of local geology on destructive earth-
quake ground motion is largely recognized in earthquake
engineering. Local site amplification accounts for large
variations in the intensity of earthquake ground motion
over relatively short distances. However, there is not a wide-
spread agreement as regards what could be the best way to
estimate the amplification caused by site effects using obser-
vational data. There are also very different approaches to
model and account for site effects in seismic risk studies.
One example of these disagreements is the current ongoing
discussion as to the relative merits of the H/V ratio techni-
que. Another example is the very different approach to
model site effects, which range from 1D to 3D models,
using linear or non-linear material behavior. One of the
reasons for this is that very few site effect studies have
been done where both a detailed study of subsurface struc-
ture and numerous high-quality observations of earthquake
ground motion have been available.
These problems prompted the development of a test site
in Europe, aimed at improving our understanding of site
effects and our capability of accounting for them. Set up
* Corresponding author. Tel.: 130-31-995808; fax: 130-31-995619.
E-mail address: raptakis@evripos.civil.auth.gr (D. Raptakis).
0267-7261/00/$ - see front matter q 2000 Elsevier Science Ltd. All rights reserved.
PII: S0267-726 1(99)00026-3
Economic Community, Euroseistest has concentrated a
large amount of European earthquake engineering research
during the past few years. Additionally, its continuing
operation is making possible the development of a very
large database that will be useful during many years to
come. The location of Euroseistest in Northern Greece is
shown in Fig. 1. General information and description of
Euroseistest may be found in several publications (e.g.
Ref. [1]), but a better reference is the official web site
(http://euroseis.civil.auth.gr/). Euroseistest is an ambitious
project in that a large number of open problems in earth-
quake engineering are addressed. These include effects of
surface geology on ground motion, non-linear soil response,
structure response to ground motion, and soil–structure
interaction.
The purpose of this paper is to take advantage of the high-
quality data gathered at Euroseistest to improve our under-
standing of site effects in shallow alluvial valleys. To this
end, we take advantage of a recent study that has improved
significantly our knowledge of the structure and mechanical
properties of the materials filling the valley of Euroseistest
(Ref. [2] part I of this study). This study has proposed a 2D
benchmark model of Euroseistest. We have computed site
response for this model using a finite difference method. In a
second time, we review the previous site effects studies at
24 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39

Fig. 1. (Top) Map of the Thessaloniki area showing the location of the Volvi basin between the lakes of Volvi and Lagada. The cross-section selected for
Euroseistest is the line joining the villages of Profitis and Stivos, shown with solid squares. (Bottom) Geological map of the Volvi basin (after Jongmans et al.
[5]).

Euroseistest concerning observations of site response. We 2. Numerical modeling

complement these studies by picking two events and look-
ing at the records in time domain. Our purpose is to take a
fresh look at these observations, no longer with the purpose
of measuring site effects, but rather to bring together obser-
vations and results of modeling. In this way, we show how
we can understand together observations and synthetics.
Our results allow to shed some light on the controversies
pointed above and improve our understanding of the physics
of site effects at Euroseistest. Our main point is that it is not
enough to obtain successful comparisons between observa-
tions and models in the frequency domain. In order to make
successful predictions of ground motion, we should
consider site effects also in time domain.
Modeling of Euroseistest basin is a subject where
research is currently engaged by several authors (e.g.
Marrara and Suhadolc [3]; Tolis et al. [4]). Indeed, an
important part of Euroseistest project was to investigate
with as much detail as possible the geologic structure of
the valley and the mechanical properties of the sediments
that fill it. To this end, a very detailed geophysical and
geotechnical exploration program was carried out. A first
2D model of Euroseistest was presented in Jongmans et al.
[5]. However, their paper deals more with the determination
of the 2D structure than with modeling its site response.
Moreover, due to limitations of their method, site response
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39 25

250
NNW SSE
STE
200 ST22

STC
150
PR O C
FRM B G*
100
G
C
A L T IT U D E (m )

GRA GRB
C B TST A
50 E C B D
TST17
F C E
G* D
0 G E F
TST72 D G* F1
F
G
-5 0 E
F4
G*
G F
-1 0 0 F2

F3 G*
-1 5 0
G

-2 0 0
-0 .5 0 .0 0 .5 1 .0 1 .5 2 .0 2 .5 3 .0 3 .5 4 .0 4 .5 5 .0 5 .5 6 .0
D IS T A N C E (K m )

Fig. 2. Final cross-section determined by Raptakis et al. [2] for Volvi basin, between the villages of Profitis and Stivos. The solid diamonds show the location of
the permanent strong motion network. Four faults (F1–F4) and eight different strata (layers A–G) are identified. Properties of the material are given in Table 1
(after Raptakis et al. [2]).

was computed for a 2D model, which was a sharp simplifi- they have summarized all the data available to date concern-
cation of their proposed structure for this valley. Finally, ing the properties and geometry of the material filling this
Jongmans et al. [5] show only a qualitative comparison alluvial valley. We have not introduced any simplifications
between the results of their model and observed site in this complicated structure and have included the irregular
response in the frequency domain. Additional results from topography at the surface of the cross-section and anelastic
2D modeling for Euroseistest were presented in Marrara and attenuation. This latter factor is important, since the soft,
Suhadolc [3] and Tolis et al. [4]. However, both of these surficial soils present in this valley cannot be accurately
studies have used previously published versions of the considered as non-dissipative materials.
underground structure at Volvi valley and they only show
a very limited comparison with observations and only in the 2.1. Model and method used
frequency domain.
In this section, we take advantage of the update of the A recent effort to re-interpret the exploration data gath-
underground structure provided by Raptakis et al. [2] where ered at Volvi basin has now been completed [2]. The result
has been an improvement on the model of Euroseistest
Table 1
proposed by Jongmans et al. [5] and Pitilakis et al. [6] by
Properties of the materials in the model incorporating additional information. Additionally, while
Jongmans et al. [5] relied on P-wave refraction profiles to
Layer Description S-wave Density QS define the gross subsoil structure of the valley’s sediments,
A Silty, clayey sand 130 2.05 15 the new proposed model stressed the S-wave measurements,
B Silty sand and sandy clay 200 2.15 25 given the importance of this parameter in site response of
C Marly silt and silty sand 300 2.075 30 horizontal ground motion. The final model is shown in Fig.
D Marly, sandy clay and clay silt 450 2.100 40 2. The properties of the different layers are given in Table 1.
E Alternating sublayers of 650 2.155 60
This model takes into account the vast amount of data gath-
clayey, silty sand and sandy
clay with stones and gravels ered at Euroseistest, plus a non-negligible amount of engi-
F Alternating sublayers of 800 2.20 80 neering judgment. Eight different formations were
clayey, silty sand and sandy identified, labeled A–G in Fig. 2. Formation A consists of
clay with stones and gravels silty–clayey sands, is a shallow recent loose deposit, and is
Gp Weathered schist bedrock 1250 2.50 100
only found at the center of the valley. Formation B is
G Gneiss basement 2600 2.60 200
composed of silty sands and gray–black sandy clays. It
26 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39

propagation problems in these papers. As explained in

1.0
a Moczo [7], this method allows modeling of a non-flat free
0.8 surface if we constrain it to pass through nodes of the grid.
This restriction does not apply to any other irregular inter-
Amplitude

0.6 faces in the model, which allows including in the model the
precise irregular shapes of the subsoil structure. We have
0.4
used a constant grid step of 2.6 m in the horizontal direction
0.2 and 1.3 m in the vertical direction. We chose a maximum
accurate frequency for our computations of 4 Hz, covering
0.0 the main amplification peaks determined in Raptakis et al.
0.0 0.2 0.4 0.6 0.8 1.0 [2]. Given a minimum shear velocity of 130 m/s (layer A of
Time [s] the model), our minimum wavelength is 32.5 m. The irre-
gular interfaces of the model were approximated with poly-
10
1
noms (one-degree for the faults, five-degree for the more
irregular interfaces). The free surface was quantized at
2
b 1.3 m vertical spacing, and digitized with a 2.6 m horizontal
10
spacing. Thus our grid step is 12.5 times the minimum
Fourier amplitude

10
3
10
4
10
5
10
6
0.1 1.0 10.0
Frequency [Hz]
Fig. 3. The Gabor pulse used as incident signal in the 2D finite difference
simulations. (a) Signal in the time domain. (b) Fourier amplitude spectrum
of the time signal.
has large variations in thickness across the valley. Forma-
tion C consists of marly silts and silty sands. Formation D is
an intermediate sublayer in which marly, sandy clays and
clayey silts are the predominant soil materials. Formations E
and F are composed of thick alternating sublayers of clayey
silt–sand and sandy clay with small stones and gravels.
Layer G p corresponds to weathered rock, with some inser-
tions of clayey sand and sandy silt. Finally, the deepest
formation G is the gneiss basement. The wave propagation
values of Table 1 are the average values, determined from
all the measurements. The four faults (F1–F4) shown in Fig.
2 divide the sediments within the valley in three main blocks
and affect significantly the thickness of the different layers
along the cross section. From the four faults, only F1 is an
active seismic fault, while the other three are accretionary
faults. Wave attenuation in each formation is given in Table
1 in terms of quality factors for S waves (QS). QS can be
correlated with the material damping of soils, j; through a
simple empirical formula, QS ˆ 1=2j:
The model of Fig. 2 shows a very irregular structure. For
this reason, we have used a finite difference method, which
allows a detailed representation of all the irregularities of
the geologic structure of the valley. We have used the 2D,
SH-wave, finite difference method of Moczo [7], refined in
Moczo and Bard [8] and Moczo et al. [9]. The accuracy of
this method has been validated for a wide range of wave
wavelength in the horizontal direction and 25 times the
minimum wavelength in the vertical direction. In this way
we guarantee that errors due to grid dispersion will be negli-
gible. The grid model was bounded laterally and at the
bottom with transparent, Reynolds’ type [10], non-reflecting
boundaries. However, to verify that no unwanted artificial
reflections were introduced by the finiteness of our grid, we
computed results for two similar models, different only in
the extension of the grid both laterally and at the bottom.
The comparison of the results showed clearly that no signif-
icant artificial reflections were generated at the transparent
boundaries of the model. This was expected, as that most of
the seismic energy will be trapped within the soft sediments
given the large velocity contrasts involved. Attenuation is
taken into account through three relaxation mechanisms, at
frequencies chosen to insure constant-Q in the frequency
range of validity of the computations (from 0.1 to 4 Hz),
as thoroughly described in Moczo [7] and Moczo and Bard
[8].
The excitation to the model was given by vertical inci-
dence of a plane SH wave. Its time variation was given by a
Gabor pulse by the equation:
s…t† ˆ e2a cos‰vp …t 2 ts † 1 cŠ
where
 
vp …t 2 ts † 2
aˆ
g
The parameters we used are fp ˆ 2pvp ˆ 0:23; g ˆ 0:15;
c ˆ 0:0; and ts ˆ 0:25: Fig. 3 shows the time variation of
this pulse, together with its amplitude spectrum. We observe
that the pulse has non-negligible frequency content up to
6 Hz. Thus, we will have to filter the synthetic seismograms
to eliminate contribution of frequencies that cannot be
correctly propagated by our grid.
2.2. Analysis of results
The time domain seismograms are shown in Fig. 4 for the
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
PRO
0 .0
5 .0
1 0 .0
T im e (sec )
1 5 .0
2 0 .0
2 5 .0
0 .0 1 .0
Fig. 4. Seismic section computed at the surface of the 2D model shown in Fig. 2 for vertical incidence of SH waves. The positions of the surface accelerographs
of the permanent network have been indicated for reference. Traces have been low-pass filtered with a 3.5 Hz frequency cutoff. In this computation anelastic
attenuation was neglected.
case of infinite Q. We show displacement computed for 155
receivers distributed along the free surface. They have been
low-pass filtered with a Butterworth filter having a corner
frequency of 3.5 Hz. Duration of ground motion is over-
estimated by the lack of attenuation in the model, but this
allows to identify the different waves that compose the
synthetics. The receivers at the center of the valley show
clearly the 1D resonance of the sediments, but the largest
amplitudes are not related to vertical propagation. The
synthetic seismograms are very clearly dominated by
locally generated Love waves. The asymmetrical structure
of the valley generates very strong Love waves at fault F4
(northern edge) with group velocity about 210 m/s and
phase velocity about 700 m/s. The southern edge of the
valley, fault F1, generated similar Love waves (group and
phase velocities of 200 and 1100 m/s, respectively), but
with smaller amplitude. The Love waves generated by faults
F1 and F4 are not significantly affected by the presence of
faults F2 and F3. This indicates that the topmost layers
guide those Love waves, where lateral contrast at faults
F2 and F3 is quite small. However, contrary to fault F2,
fault F3 does generate an additional Love wavetrain, propa-
gating faster (group velocity of 300 m/s, phase velocity of
1250 m/s) than those generated by faults F1 and F4. Thus,
clearly the Love waves generated by fault F3 are guided by
the deeper layers, which have higher propagation velocities
and present a significant discontinuity at this fault.
Fig. 5 shows the Love wave dispersion curves computed
27
GRA GRB TST FRM STC STE
2 .0 3 .0 4 .0 5 .0
D istan ce (k m )
for the vertical profile of the model at the center of the
valley, at the location of TST accelerograph. We can iden-
tify the Love waves observed in the synthetics as the funda-
mental and first higher mode of Love waves corresponding
to this stratigraphy. The frequency content of each of those
two modes of surface waves coincides with the minimum of
the corresponding group velocity curves at about 0.8 and
2 Hz. The phase velocities change rapidly at these frequencies,
as was to be expected, in agreement with the variations of
the phase velocity observed in the synthetics.
The results for finite Q are shown in Fig. 6, again low-
pass filtered at 3.5 Hz cutoff. The QS values used for each
formation are given in Table 1. Locally generated Love
waves are significantly affected by the low Q values of the
topmost layers. However, they are still the more important
features of the synthetic seismograms. The asymmetry in
the amplitude of these surface waves is again evident,
with energetic wavetrains generated at the northern edge
of the valley, in contrast with lower amplitude waves gener-
ated at the southern edge. At the center of the valley, ground
motion shows large amplitudes for more than 12 s, in large
contrast with ground motion duration outside the valley. We
may also observe that topography around Profitis strong
motion station does not have a significant effect in the
frequency band included in the synthetics.
We have used the seismograms of Fig. 6 to compute
transfer functions relative to input ground motion. The
result is shown in Fig. 7. The center of the valley shows a
28 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39

2.5

Group/Phase velocity [km/s]

2.0

1.5

1.0

0.5

0.0
0.0 1.0 2.0 3.0 4.0
Frequency [Hz]

Fig. 5. Love wave dispersion curves computed for the stratigraphy of the model of Fig. 2 at the location of TST. The thin lines show group velocity, while the
thick lines give phase velocity. The fundamental mode is shown by solid lines, while the first higher mode is indicated by dashed lines.

first peak of amplification at about 0.85 Hz. We note we cannot identify any obvious amplitude change in the
however, that this peak is not homogeneous across the wavetrains. Thus, this heterogeneity of the transfer function
valley and that it breaks at two points between 3000 and at the resonant frequency must come from the interaction
4000 m distance. If we look at Fig. 6 in this distance range, of surface waves. This is shown in Fig. 8, where the

PRO GRA GRB TST FRM STC STE

0 .0
5 .0
1 0 .0
T im e (se c)
1 5 .0
2 0 .0
2 5 .0

0 .0 1 .0 2 .0 3 .0 4 .0 5 .0
D istan ce (k m )

Fig. 6. Seismic section computed at the surface of the 2D model shown in Fig. 2 for vertical incidence of SH waves. The positions of the surface accelerographs
of the permanent network have been indicated for reference. Traces have been low-pass filtered with a 3.5 Hz frequency cutoff. Anelastic attenuation was
included in the computations.
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39 29

spectrogram of the synthetic seismogram at the position of

TST (Fig. 2) is presented. Fig. 8 shows that the energy that
contributes to the “resonant” peak at 0.8 Hz is distributed all
along the synthetic, including both 1D resonance and
surface waves.

3. Comparison with observed site response

Observation of site effects in Volvi basin has been an

important objective of Euroseistest project. Several field
experiments using temporary seismograph networks have
been carried out; published results are available for the
1994 experiment. Additionally, a permanent accelerograph
array operates since early 1994. In the following lines we
will review briefly published results, all of them in the
frequency domain. We will complement these studies by
looking at the records of two events in the time domain,
in order to make the link between observed site effects
and the results of our numerical modeling. For convenience
we have divided this section according to the type of instru-
ment used, seismographs or accelerographs.

3.1. Results using seismograph data

Fig. 7. Theoretical transfer function for vertical incidence of SH waves on
the model shown in Fig. 2. This result was obtained from the synthetics
shown in Fig. 6. The positions of the surface accelerographs of the perma-
nent network have been indicated for reference. We have corrected for the
free field factor, thus the transfer function value in absence of the valley
would be a factor of one.
During 1994, an ambitious recording campaign using
high quality seismographs was carried out. Thirty-nine digi-
tal seismographs were installed in the valley for more than
2 months. The instruments were disposed in three networks
(Figs. 9 and 12), identified by the type of instruments, which
were different for each one of them. The first was a line of

Fig. 8. Evolutionary spectrum of the synthetic seismogram computed by finite differences for the location of TST in the 2D model of Fig. 2. The corresponding
trace (shown on top of the evolutionary spectrum) corresponds to the computation including attenuation. Before computing the spectrum, the trace was low-
pass filtered with a 3.5 Hz frequency cutoff. Spectral amplitudes are given by the linear gray scale shown at the right of the figure.
30 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
Fig. 9. Location of the seismographs used in the temporary networks of
Reftek instruments (solid circles with three letter names) and Lennartz
instruments (open squares labeled L1–L8). The location of the villages
Stivos and Profitis (in reference to Fig. 1) are shown with large open circles.
The CIES network is located at the same point as TES at this scale.
23 seismographs crossing the valley (called the Reftek
array). The second was a line of eight seismographs
disposed parallel to the axis of the valley (the Lennartz
array). Finally, the third was a small aperture cross-shaped
array at the valley center of eight instruments (the CIES
array). The events recorded were located using an addi-
tional, regional network of 20 instruments (the TAD
array). The detailed description of the layout of the experi-
ment is presented in Riepl [11] and Riepl et al. [12].
The data obtained was used to study the site effects using
several techniques, all of them working in the frequency
domain: spectral ratios relative to a reference site, a general-
ized inversion, horizontal to vertical spectral ratios both
using earthquake and noise records, and the coda wave
method of Phillips and Aki [13]. The results are given in
Riepl [11] and Riepl et al. [12] in terms of empirical transfer
functions. Using the data of the Reftek array, they observed
amplification of ground motion at the center of the valley
amounting up to a factor of 12 relative to a station on rock.
Interestingly, they could not identify specific resonance
modes related to vertically propagating waves. It was also
observed that significant amplification was present in like
manner in the vertical component, with amplitude compar-
able to that observed for horizontal motion. Using the data
of the Lennartz array, Riepl et al. [12] showed clearly that
the response did not change significantly along the axis
perpendicular to the 2D structure. Thus, it is adequate to
study the response of the instrumented section of Euroseist-
est using 2D numerical models. Finally, as regards the
analysis of the CIES array, only a very preliminary analysis
is available [14], where no evidence of laterally propagating
waves is presented.
The analysis of Riepl et al. [12] was thoroughly done in
the frequency domain without any reference to the possible
influence of laterally propagating waves in the computed
empirical transfer functions. As our synthetics stress the
importance of such locally generated surface waves in the
site response, we decided to look into this dataset. However,
we do not attempt to analyze in detail the very large number
of records obtained during the experiment. Rather, we have
selected a well-recorded event to see whether we can iden-
tify evidences of laterally propagating waves in the records.
The event we have selected occurred on 06.25.94, had a
local magnitude of 2.5 and its epicenter was located 25 km
to the SE of Euroseistest. It was recorded by 17 seismo-
graphs crossing the alluvial valley and by seven stations
of the CIES array. We assume that site effects are determi-
nant in the observed records, therefore we rotated horizontal
components in directions parallel and perpendicular to the
valley’s axis. First we examined the array crossing the
valley. We filtered the records in several frequency bands,
trying to identify coherent wave pulses among the traces. An
example of the results is shown in Fig. 10. In this figure a
seismic section across the valley is shown for the horizontal
component of motion in the direction N59.5E (SH motion).
The traces have been low-pass filtered with a cutoff
frequency of 3.5 Hz. We observe very large differences
between the records along the cross-section, however,
these differences concern more the duration than the ampli-
tude of ground motion. We observe clearly large amplitude
late arrivals at all stations in the distance range between 1.0
and 4.5 km. These late arrivals are very likely surface waves
generated at the lateral variations of the basin, however, it
was impossible to identify a single pulse arriving at different
stations in any frequency band. This suggests that wave-
lengths are shorter than average spacing between Reftek
stations (about 400 m). The frequency of these late pulses
lies below 2 Hz, as shown by the evolutionary spectrum of
Fig. 11. A further evidence of these pulses being surface
waves is given by the time between successive peaks of
energy in the evolutionary spectrum. If these peaks were
related to reflected waves at the free surface and the sedi-
ment–bedrock interface, they should appear more closely
spaced in time, given that the time of vertically propagating
waves is very short.
Observation using the Reftek array suggests that locally
generated, short wavelength, Love waves propagate in
Volvi basin. Thus, the CIES small aperture array would
seem the better array to analyze these waves. However,
the distribution of the seven CIES stations that recorded
the 06.25.94 event (Fig. 12) shows an almost linear array.
This was due to malfunction of one station. Thus, resolution
of the array is low, and the small number of stations leads to
strong spatial aliasing. In spite of these problems, we
computed f–k spectra for different choices of time window
and for a sequence of 0.4 Hz wide frequency bands between
0.3 and 3.0 Hz. Fig. 13(a) shows the unfiltered velocity
records analyzed. The traces shown correspond to the hori-
zontal motion in the direction N59.5E (SH motion) The
amplitude scale is arbitrary but common to all the traces.
Fig. 13(b) shows the phase velocities determined for two
different time windows (shown in the upper panel): one
starting with the S-wave arrival and one sampling the
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
20.0
BED
CHU
ONI
15.0
Time [s]
10.0
5.0
0.0
0.0
Fig. 10. Seismic section obtained using the records of the Reftek array for the event of 06.25.94. The traces correspond to horizontal motion in the direction
N59.5E (SH motion). Distance is measured along the cross section of Fig. 2 with an arbitrary origin. Traces have been low-pass filtered with a 3.5 Hz frequency
cutoff. The traces are identified with the name of the corresponding station, shown in Fig. 9. Amplitude scale is arbitrary but common to all the traces.
coda. These results correspond to the horizontal motion in
the direction N59.5E (SH motion). Also in Fig. 13(b) are
shown the phase velocity dispersion curves expected for the
fundamental and first higher mode from the velocity profile
at the center of CIES array. In spite of aliasing problems
(evident for those points showing a phase velocity increas-
ing linearly with frequency in some frequency bands) we
observe a good agreement between observed and theoretical
dispersion curves for the fundamental mode in the
frequency band 1–3 Hz. Our analysis suggests that the S-
wave window of CIES records consists of Love waves
between 1 and 2 Hz, while the later coda of S-waves
consists of Love wave in the frequency band between 1
and 3 Hz. The phase velocity of these surface waves is
comprised between 300 and 400 m/s in the frequency
band between 1 and 3 Hz, in good agreement with the
results from finite difference modeling. Thus, wavelengths
of these Love waves should be between 100 and 400, which
explains why individual pulses of these Love waves could
not be correlated among the stations of the Reftek array.
We observe then that locally generated surface waves,
predicted by numerical modeling, are indeed present in
actual earthquake records. These surface waves cannot be
identified in the frequency domain transfer functions,
because they do not appear as separate peaks of amplifica-
tion. Rather, they contribute to the main “resonance peak”
of the empirical transfer functions.
3.2. Results using accelerographic data
In addition to the temporary installation of seismograph
arrays, Euroseistest project included deployment of a
31
FIE
YEL
TES
DEP
MAI
SCU
GRE
TRE
MUR
BAN
THA
BUT
ROC
PRO
TOB
KOR
FAR
1.0 2.0 3.0 4.0 5.0 6.0
Distance [km]
permanent network of digital accelerogaphs across the
valley. Seven surface instruments plus two borehole instru-
ments were installed in early 1994, and have already
recorded a significant number of earthquakes (more than
50). There have been two site effect studies of Euroseistest
using the data from the accelerograph array, Raptakis et al.
[15] and Dimitriu et al. [16].
The results from both papers Raptakis et al. [15] and
Dimitriu et al. [16] are very similar. Both of them deal
with the evaluation of empirical transfer functions in the
frequency domain. The techniques used are the same in
both cases: standard spectral ratios relative to a reference
site, and horizontal-to-vertical spectral ratios. Thus, the
results of their analysis are very similar to those obtained
from the Reftek array. The differences between Raptakis et
al. [15] and Dimitriu et al. [16] lay on their respective
emphasis. For Raptakis et al. [15] the main point is the
comparison between the two techniques to estimate the
amplitude of the transfer functions across the valley. Dimi-
triu et al. [16] are more interested in determining site
response for station TST, at the center of the valley. Then,
Dimitriu et al. [16] proceed to model the observed spectrum
at TST during five significant events by multiplying the
spectrum observed at PRO (the reference station on rock)
for each of those events with a 1D, S-wave propagation
transfer function. The stratigraphy they used for TST was
taken from the well-constrained measurements of Pitilakis
et al. [6]. Dimitriu et al. [16] use the incidence angle of the
plane shear waves as free parameter to obtain the best fit.
Of the two papers, only Raptakis et al. [15] mentions
briefly the possible influence of 2D site effects in the
observed transfer functions. However, they only include
32 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
them as a possible suggestion to explain the difference
between observed average transfer function at TST site
and a numerical transfer function obtained from a 1D, verti-
cal propagation of S-waves model. Now, it is clear from Fig.
6 that the site response at the center of the valley is largely
influenced by locally generated surface waves. To verify
this point, we have again chosen a representative event,
well recorded by the accelerograph array, the event of
Arnea. This event occurred on 4 May 1995, had a local
magnitude of 5.3 and was located 32 km to the South–
East of TST. Fig. 14 shows the seismic section obtained
for the records of this event at the accelerograph array.
We plot the motion in the direction N59.5E, i.e. SH motion.
The acceleration records shown in this figure were
processed in the same way used for the weak motion records
shown previously (low-pass filtered with a 3.5 Hz cutoff).
The result shown in Fig. 14 is very similar to what we
obtained from the weak motion records. At low frequencies,
the most striking differences between records outside the
valley and those on the sediments are not in amplitude but
in the duration of shaking. At stations within the valley, late
arrivals with significant amplitude are readily identified, and
very likely are locally generated Love waves. Fig. 15 shows
the evolutionary spectra for stations PRO, on rock, and TST,
at the center of the valley. We observe that the largest peaks
in the low frequency range at station TST occur between 20
and 25 s time, while S-wave arrival is at 14 s. It is clear that
the late arrivals play a significant role in the “resonant” peak
identified at 0.8 Hz by Raptakis et al. [15] and Dimitriu et al.
[16] at TST station.
There are two additional acceleration records of Arnea
0.02
0.0
-0.02
4.0
Frequency [Hz]
3.0
2.0
1.0
0.0
5.0
Fig. 11. Evolutionary spectrum of the observed seismogram at station TES, at the center of the valley, for the event of 06.25.94, component of motion in the
direction N59.5E (SH motion). Before computing the spectrum, the input trace (shown on top of the spectrum) was band-pass filtered between 0.1 and 5.0 Hz.
Spectral amplitudes are given by the linear gray scale shown at the right of the figure.
event, which are useful in this analysis. They were recorded
in borehole instruments, at depths of 17 and 72 m, under
TST surface station. We have analyzed these records to
determine if they support our claim of the importance of
surface waves at TST. First, we aligned the arrival time of
P wave in the vertical components, to insure an almost
common time base for the three records. We incur in an
uncertainty as large as the vertical travel time of P wave
between each couple of stations. According to the P-wave
velocities determined in the model, this travel time is of
0.03 s between 72 and 17 m depth stations, and 0.02 s
between 17 m depth and the surface. Given our sampling
rate, these times correspond to seven and four samples
differences, which are quite small. Second, our analysis
suggested that the instrument at 72 m depth has a different
orientation from the other two accelerographs (at 17 m
depth and at the surface). This is shown in Fig. 16, where
we plot the horizontal components recorded at the three
accelerographs, low-pass filtered with a 0.5 Hz frequency
cutoff. At such low frequencies, the traces should be very
similar, as Fig. 16 shows. However, the traces for the record
at 72 m depth suggest that the deepest instrument has rotated
2908. In what follows, we have corrected for this rotation.
The results of numerical modeling suggest that two
modes of Love waves (fundamental and first higher mode)
are important in Euroseistest, at 0.8 and 2 Hz. For this
reason we filtered the observed borehole records around
these two frequencies. Fig. 17 shows the transverse compo-
nent (parallel to the valley’s axis) of the three surface and
downhole records band-pass filtered between 0.4 and 1 Hz.
As expected, the traces are similar. The dashed lines in
6.77E-05
0.0
10.0 15.0 20.0 25.0
Time [s]
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39 33

80 Fig. 17 follow peaks and troughs that are readily correlated

cnn3 among the traces. These peaks and troughs are in phase, and
do not show any evidence of vertical propagation. Fig. 18
shows a similar result for the frequency band between 1.6
60
and 2.4 Hz. The traces are less similar than they are in the
cnn2
NS coordinate [m]

lower frequency band, however, it is still possible to recog-

cne1 nize common peaks and troughs in the three records (dashed
40 cnn1 lines in the figure). We note that these peaks and troughs are
again in phase. Within a small uncertainty, the dashed lines
that work as marks of simultaneous time coincide with
cns1
peaks or troughs in the three records. Vertical propagation
20
cns2
of S-waves would require a time delay of 0.16 s (32
samples) between 72 and 17 m depth, and 0.1 s (19 samples)
cns3
between 17 and the surface. These delays are significantly
0 larger than the time difference between the dashed lines and
10 30 50 70 90
EW coordinate [m]
the peaks and troughs in this figure (less than 10 samples).
An additional evidence that the peaks and troughs identified
Fig. 12. Spatial distribution of the seven CIES stations that recorded the in Figs. 17 and 18 are indeed surface waves is shown in Fig.
06.25.94 event. Distances are measured in the NS and EW direction from an 19. In this figure we have plotted the amplitudes of the peaks
arbitrary origin. TST station was located at the middle point between cns1
and troughs followed from trace to trace with the dashed
and cnn1 stations.

cne1

cnn1
a
cnn2

cnn3

cns1

cns2

cns3

0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0

Time [s]
0.8
7-15 s
b 15-22 s

0.6
Phase velocioty [km/s]

0.4

0.2

0
0 0.5 1 1.5 2 2.5 3
Frequency [Hz]

Fig. 13. (a) Horizontal motion recorded at the CIES stations for the 06.25.94 event in the direction N59.5E, i.e. SH motion. The amplitude scale is arbitrary, but
common to all the traces. (b) Results of f–k analysis using the data shown in (a). Each symbol corresponds to the maximum of a spectrum in the kx 2 ky diagram
for a sequence of 0.4 Hz wide frequency bands between 0.3 and 3 Hz. Results are given for two different time windows corresponding to the S-wave (7–15 s)
and the coda of S-waves (15–22 s), indicated with dotted lines in the upper panel. The solid lines correspond to the phase velocities shown in Fig. 5.
34 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
25.0
STE
STC
FRM
20.0
15.0
Time [s]
10.0
5.0
0.0
0.0 1.0
Fig. 14. Seismic section obtained using the records of the permanent accelerograph network for the event of 5.03.95 (Arnea event). We show horizontal motion
in the direction N59.5E (SH motion). Distance is measured along the cross section of Fig. 2 with an arbitrary origin. Traces have been low-pass filtered with a
3.5 Hz frequency cutoff. The traces are identified with the name of the corresponding station, shown in Fig. 2. The amplitude scale is arbitrary but common to
all the traces.
lines in Figs. 17 and 18 as a function of depth. Amplitudes
were normalized by the value at the free surface. The thick
lines in this figure show the theoretical modal shapes
computed for the fundamental (for the frequency range
0.4–1 Hz) and first higher mode (for the frequency range
1.6–2.4 Hz) for the vertical profile at TST station. The
symbols correspond to the amplitudes measured along the
dashed lines in Figs. 17 and 18. The agreement between
observed and predicted modal shapes is good. Thus, the
borehole records support our claim that, in the low
frequency range, ground motion is dominated by locally
generated surface waves.
4. Discussion
Most of the recent observational site effect studies deal
with the evaluation of empirical transfer functions. Such
techniques as standard spectral ratios, generalized inversion,
or horizontal-to-vertical spectral ratios dominate in a large
percentage of recently published papers on site effects. For
example, a large number of papers have been devoted to
discussing the advantages and disadvantages of the former
and the latter techniques. In this sense, the paper by Safak
[17] is revealing in its in-depth exploration of different tech-
niques, with the aim of evaluating which would be the best
way to define an empirical transfer function between two
records. In principle, the use of amplitude transfer functions
does not restrict the universe of models. In practice, this has
TST
GRB
GRA
PRO
2.0 3.0 4.0 5.0
Distance [km]
encouraged the continued use of 1D models, because of the
simplicity of computing a theoretical transfer function to
compare with the observations. This despite the fact that it
has been almost 20 years since the demonstration that in
shallow alluvial valleys, such as Euroseistest, locally gener-
ated surface waves should be the predominant phenomenon
[18,19].
In the particular case of Euroseistest, frequency domain
analysis of weak motion data showed that the sediments
filling the valley of Euroseistest amplify ground motion
by factors larger than 10 between 1 and 3 Hz. In the time
domain, numerical modeling of a very detailed cross section
of Volvi valley shows that site response in this valley is
dominated by lateral propagation of locally generated
surface waves. If we look at the data in the time domain,
both seismograph and accelerograph records confirm the
theoretical results. We observe that surface waves are
excited at the same frequencies at which we expect the
“resonance frequency” to occur. It is not surprising that
the predominant frequencies for body and surface waves
coincide because both are two manifestations of the same
phenomena.
The problem appears when we want to interpret the
empirically determined transfer functions. For example,
Dimitriu et al. [16] interpreted the transfer function at
TST in terms of oblique incidence of SH waves in a 1D
model. For example, Fig. 20 shows the very good fit
Dimitriu et al. [15] obtained for the observed spectrum at
TST during the Kozani earthquake. We have shown that, if
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39 35

0.005

0.0 a
-0.005
1.07E-06
4.0

Frequency [Hz]

3.0

2.0

1.0
0.0

0.0
10.0 15.0 20.0 25.0
Time [s]

0.01

0.0 b
-0.01
2.09E-05
4.0
Frequency [Hz]

3.0

2.0

1.0
0.0

0.0
10.0 15.0 20.0 25.0
Time [s]
Fig. 15. (a) Evolutionary spectrum of the accelerogram recorded at station PRO during the event of 5.03.95 (Arnea event), transverse component (SH motion).
(b) Evolutionary spectrum of the accelerogram recorded at station TST during the event of 5.03.95 (Arnea event), transverse component (SH motion). Above
each spectrum is shown the corresponding observed accelerogram. Before computing the spectra, traces were band-pass filtered between 0.1 and 5 Hz. The
amplitude scales for the traces are given in fractions of g.

we want to account for the physics governing site effects,
ground motion at TST cannot be modeled using 1D models.
If we take the “well-determined” 1D transfer function used
in Fig. 20 to model ground motion in the time domain, we
would find that amplitudes are overestimated, while dura-
tion of ground motion is grossly underestimated (see for
example, Refs. [20,2]). The reason is clearly that the 1D
“resonance” peak includes the contribution of locally gener-
ated surface waves, which occur at different times in the
seismogram. This phase information was lost in the deter-
mination of the empirical transfer functions. An additional
problem is that amplitude transfer functions tend to make us
forget the physics of site effects. We suggest that the deter-
mination of empirical transfer functions is not enough to
discriminate between 1D and 2D site effects, and that a
successful modeling of a partial aspect of site response
(its amplitude) cannot guarantee that we can make success-
ful predictions of ground motion at a site.
36 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39

0.00129
NS

0m
-0.00129
0.00093
EW

0m
-0.00097
-0.00116
NS

-17 m
0.00115
0.00098
EW

-17 m
-0.00100
0.00054
NS

-72 m
-0.00058
0.00082
EW
-72 m
-0.00087
0.0 10.0 20.0 30.0 40.0
Time [s]

Fig. 16. Horizontal components of acceleration recorded at the surface and borehole instruments at TST site for the Arnea event. Traces have been low-pass
filtered with a 0.5 Hz frequency cutoff. We observe that, very likely, the instrument located at 72 m depth has rotated on its vertical axis. The amplitude scale
for the traces is given in fractions of g.

0.00621

0m

-0.00639
0.00624

-17 m

-0.00638
0.00311

-72 m

-0.00338
0.0 5.0 10.0 15.0 20.0 25.0 30.0
Time [s]

Fig. 17. Transverse components of acceleration (in the direction N59.5E, parallel to the valley’s axis) recorded at the surface, and 17 and 72 m depth
accelerographs at TST site for Arnea event. Traces have been band-pass filtered between 0.4 and 1 Hz. The dashed lines follow in-phase peaks and troughs
for the three records. The amplitude scale for the traces is given in fractions of g.
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39 37

0.00462

0m

-0.00454
0.00266

-17 m

-0.00260
0.00294

-72 m

-0.00275
0.0 5.0 10.0 15.0 20.0
Time [s]

Fig. 18. Transverse components of acceleration (in the direction N59.5E, parallel to the valley’s axis) recorded at the surface, and 17 and 72 m depth
accelerographs at TST site for Arnea event. Traces have been band-pass filtered between 1.6 and 2.4 Hz. The dashed lines follow in-phase peaks and troughs
for the three records. The amplitude scale for the traces is given in fractions of g.

5. Conclusions

We have presented a study of site effects at Euroseistest.

We have taken advantage of a recent re-interpretation of
field data that produced a detailed model of the subsoil
structure at this valley. We have computed the response of
this 2D structure to vertically incident SH waves. The
results confirm that the dominant features of site response
are locally generated surface waves.
We have looked for evidence of these surface waves in
earthquake data recorded at this site including both seismo-
graph and accelerograph data. Although our 2D modeling
considered only linear behavior of soil materials, this
comparison is adequate because all records correspond to
small magnitude events. Our observations suggest that
locally generated surface waves contribute significantly to
ground motion, and that they appear at the same
frequencies as resonance of vertically propagating
shear waves. The velocity of these surface waves is
consistent with modeling results and their wavelength
should lie between 100 and 400 m. Thus, local hetero-
geneities will affect them significantly, which explains the
difficulty of identifying individual pulses common among Fig. 19. A computed modal shape observed at two frequency bands. Thick
the different stations. lines: computed modal shapes for the 1D profile at TST, fundamental mode
We have shown that the fact that layering below TST in the left panel and first higher mode in the right panel. The two thick lines
correspond to the limiting values of the frequency shown on top of each
station is almost flat does not make this site a good candidate
panel. Solid circles: amplitude of acceleration measured on the traces
for 1D analysis. At TST, the dominant feature of site shown in Fig. 17 (left panel) and Fig. 18 (right panel), normalized by the
response at frequencies below 4 Hz are surface waves amplitude value at the surface. Each group of three connected solid circles
generated at the edges of the valley. corresponds to one dashed line shown in Figs. 17 and 18.
38 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
Fig. 20. Solid lines: observed Fourier amplitude spectra at PRO and TST strong motion stations during the Kozani-Grevena earthquake (13.05.95, MS ˆ 6:6†:
Dashed line: result of multiplying the amplitude spectrum observed at PRO by the 1D transfer function computed for SH-wave incidence at 50 on the
stratigraphy underlying TST site (after Dimitriu et al. [16]).
Finally, amplitude transfer functions give only partial
information of site response. For this reason, they can be
misleading. A good fit of this partial observation using 1D
models cannot rule out the possible importance of 2D site
effects. If 2D site effects are important, predictions of time
histories of ground motion made on the basis of the 1D good
fit will lead to large errors.
This study was made possible by the large effort carried out
during the Euroseistest project. The high quality database
accumulated until now is quite large. Moreover, the continu-
ing operation of this test site, together with deployment of
additional instruments, will make this test site very important
in Earthquake Engineering during many years to come.
Acknowledgements
This work was possible thanks to the many people who
have participated in the different field experiment of Euro-
seistest project. P. Moczo and J. Kristek kindly gave assis-
tance with finite difference computations, and granted CPU
time in one of their machines to compute results for the
larger model. This work was done while the senior author
was on a sabbatical leave from UNAM at Aristotle Univer-
sity of Thessaloniki. This visit was supported by Dirección
General de Asuntos del Personal Académico, UNAM,
Mexico, and by the European Union—Directorate General
XII, for Science, Research & Technology (contracts
#EV.5V-CT.93-0281 and #ENV4-CT.96-0255).
References
[1] Pitilakis et al. EURO-SEISTEST. A European Test-Site for Engineer-
ing Seismology, Earthquake Engineering and Seismology, Final
Scientific Report, task 4.3: Laboratory test, 1995.
[2] Raptakis D, Chávez-Garcı́a FJ, Makra K, Pitilakis K. Site effects at
EUROSEISTEST—I: determination of the valley structure and
confrontation of observations with 1D analysis. Soil Dyn Earthq
Engng 2000;19(1):1–22.
[3] Marrara F, Suhadolc P. Estimation of local site effects in the Volvi
basin (Greece): comparison between observations and modeling. In:
Irikura K, Kudo K, Okada H, editors. The effects of surface geology
on seismic motion. Recent progress and new horizon on ESG studies,
2. USSR: Balkema, 1998. p. 973–80.
[4] Tolis S, Faccioli E, Pitilakis K. A 2D simulation of Euroseistest
near Thessaloniki, Greece. In: Irikura K, Kudo K, Okada H,
editors. The effects of surface geology on seismic motion. Recent
progress and new horizon on ESG studies, USSR: Balkema, 1998.
p. 255–62.
[5] Jongmans D, Pitilakis K, Demanet D, Raptakis D, Riepl J, Horrent C,
Tsokas G, Lontzetidis K, Bard P-Y. EURO-SEISTEST: determina-
tion of the geological structure of the Volvi basin and validation of the
basin response. Bull Seism Soc Am 1998;88:473–87.
[6] Pitilakis K, Raptakis D, Lontzetidis K, Tika-Vassilikou Th, Jongmans
D. Geotechnical & geophysical description of EURO-SEISTEST,
using field, laboratory tests and moderate strong-motion recordings.
J Earthq Engng 1999;3(3):381–409.
[7] Moczo P. Finite-difference technique for SH waves in 2-D media
 F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39
using irregular grids: application to the seismic response problem.
Geophys J Int 1989;99:321–9.
[8] Moczo P, Bard P-Y. Wave diffraction, amplification and differential
motion near strong lateral discontinuities. Bull Seism Soc Am
1993;83:85–106.
[9] Moczo P, Labák P, Kristek J, Hron F. Amplification and differential
motion due to an antiplane 2D resonance in the sediment valleys
embedded in a layer over the half-space. Bull Seism Soc Am
1996;86:1434–46.
[10] Reynolds AC. Boundary conditions for the numerical solution of
wave propagation problems. Geophysics 1978;43:1099–110.
[11] Riepl J. Effets de site: évaluation expérimentale et modélisations
multidimensionnelles: application au site test EURO-SEISTEST
(Grèce). PhD thesis, Université Joseph Fourier—Grenoble I, 1997.
p. 227.
[12] Riepl J, Bard P-Y, Hatzfeld D, Papaioannou C, Nechstein S. Detailed
evaluation of site response estimation methods across and along the
sedimentary valley of Volvi (EURO-SEISTEST). Bull Seism Soc Am
1998;88:488–502.
[13] Phillips WS, Aki K. Site amplification of coda waves from local
earthquakes in Central California. Bull Seism Soc Am
1986;76:627–48.
View publication stats
39
[14] Meunier C. Variabilité spatiale des mouvements sismiques sur courtes
distances. DEA report, LGIT, Université Joseph Fourier de Grenoble,
1997.
[15] Raptakis D, Theodulidis N, Pitilakis K. Data analysis of the Euro-
seistest strong motion array in Volvi (Greece): standard and
horizontal-to-vertical spectral ratio techniques. Earthq Spectra
1998;14:203–24.
[16] Dimitriu PP, Papaioannou CA, Theodulidis NP. EURO-SEISTEST
strong-motion array near Thessaloniki, Northern Greece: a study of
site effects. Bull Seism Soc Am 1998;88:862–73.
[17] Safak E. Models and methods to characterize site amplification from a
pair of records. Earthq Spectra 1997;13:97–129.
[18] Bard P-Y, Bouchon M. The seismic response of sediment-filled
valleys. Part 1. The case of incident SH waves. Bull Seism Soc Am
1980;70:1263–86.
[19] Bard P-Y, Bouchon M. The seismic response of sediment-filled
valleys. Part 2. The case of incident P and SV waves. Bull Seism
Soc Am 1980;70:1921–41.
[20] Chávez-Garcı́a FJ, Bard P-Y. Site effects in Mexico City eight years
after the September Michoacan earthquakes. Soil Dyn Earthq Engng
1994;13:229–47.