Professional Documents
Culture Documents
net/publication/235760318
CITATIONS READS
92 273
4 authors, including:
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:
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.
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
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]).
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
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
wavelength in the horizontal direction and 25 times the
minimum wavelength in the vertical direction. In this way
10
4
we guarantee that errors due to grid dispersion will be negli-
gible. The grid model was bounded laterally and at the
10
5
bottom with transparent, Reynolds’ type [10], non-reflecting
boundaries. However, to verify that no unwanted artificial
10
6
reflections were introduced by the finiteness of our grid, we
0.1 1.0 10.0 computed results for two similar models, different only in
Frequency [Hz]
the extension of the grid both laterally and at the bottom.
Fig. 3. The Gabor pulse used as incident signal in the 2D finite difference The comparison of the results showed clearly that no signif-
simulations. (a) Signal in the time domain. (b) Fourier amplitude spectrum icant artificial reflections were generated at the transparent
of the time signal. boundaries of the model. This was expected, as that most of
the seismic energy will be trapped within the soft sediments
has large variations in thickness across the valley. Forma- given the large velocity contrasts involved. Attenuation is
tion C consists of marly silts and silty sands. Formation D is taken into account through three relaxation mechanisms, at
an intermediate sublayer in which marly, sandy clays and frequencies chosen to insure constant-Q in the frequency
clayey silts are the predominant soil materials. Formations E range of validity of the computations (from 0.1 to 4 Hz),
and F are composed of thick alternating sublayers of clayey as thoroughly described in Moczo [7] and Moczo and Bard
silt–sand and sandy clay with small stones and gravels. [8].
Layer G p corresponds to weathered rock, with some inser- The excitation to the model was given by vertical inci-
tions of clayey sand and sandy silt. Finally, the deepest dence of a plane SH wave. Its time variation was given by a
formation G is the gneiss basement. The wave propagation Gabor pulse by the equation:
values of Table 1 are the average values, determined from
s
t e2a cosvp
t 2 ts 1 c
all the measurements. The four faults (F1–F4) shown in Fig.
2 divide the sediments within the valley in three main blocks where
and affect significantly the thickness of the different layers
along the cross section. From the four faults, only F1 is an vp
t 2 ts 2
a
active seismic fault, while the other three are accretionary g
faults. Wave attenuation in each formation is given in Table The parameters we used are fp 2pvp 0:23; g 0:15;
1 in terms of quality factors for S waves (QS). QS can be c 0:0; and ts 0:25: Fig. 3 shows the time variation of
correlated with the material damping of soils, j; through a this pulse, together with its amplitude spectrum. We observe
simple empirical formula, QS 1=2j: that the pulse has non-negligible frequency content up to
The model of Fig. 2 shows a very irregular structure. For 6 Hz. Thus, we will have to filter the synthetic seismograms
this reason, we have used a finite difference method, which to eliminate contribution of frequencies that cannot be
allows a detailed representation of all the irregularities of correctly propagated by our grid.
the geologic structure of the valley. We have used the 2D,
SH-wave, finite difference method of Moczo [7], refined in 2.2. Analysis of results
Moczo and Bard [8] and Moczo et al. [9]. The accuracy of
this method has been validated for a wide range of wave 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 27
0 .0
5 .0
1 0 .0
T im e (sec )
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. 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 for the vertical profile of the model at the center of the
receivers distributed along the free surface. They have been valley, at the location of TST accelerograph. We can iden-
low-pass filtered with a Butterworth filter having a corner tify the Love waves observed in the synthetics as the funda-
frequency of 3.5 Hz. Duration of ground motion is over- mental and first higher mode of Love waves corresponding
estimated by the lack of attenuation in the model, but this to this stratigraphy. The frequency content of each of those
allows to identify the different waves that compose the two modes of surface waves coincides with the minimum of
synthetics. The receivers at the center of the valley show the corresponding group velocity curves at about 0.8 and
clearly the 1D resonance of the sediments, but the largest 2 Hz. The phase velocities change rapidly at these frequencies,
amplitudes are not related to vertical propagation. The as was to be expected, in agreement with the variations of
synthetic seismograms are very clearly dominated by the phase velocity observed in the synthetics.
locally generated Love waves. The asymmetrical structure The results for finite Q are shown in Fig. 6, again low-
of the valley generates very strong Love waves at fault F4 pass filtered at 3.5 Hz cutoff. The QS values used for each
(northern edge) with group velocity about 210 m/s and formation are given in Table 1. Locally generated Love
phase velocity about 700 m/s. The southern edge of the waves are significantly affected by the low Q values of the
valley, fault F1, generated similar Love waves (group and topmost layers. However, they are still the more important
phase velocities of 200 and 1100 m/s, respectively), but features of the synthetic seismograms. The asymmetry in
with smaller amplitude. The Love waves generated by faults the amplitude of these surface waves is again evident,
F1 and F4 are not significantly affected by the presence of with energetic wavetrains generated at the northern edge
faults F2 and F3. This indicates that the topmost layers of the valley, in contrast with lower amplitude waves gener-
guide those Love waves, where lateral contrast at faults ated at the southern edge. At the center of the valley, ground
F2 and F3 is quite small. However, contrary to fault F2, motion shows large amplitudes for more than 12 s, in large
fault F3 does generate an additional Love wavetrain, propa- contrast with ground motion duration outside the valley. We
gating faster (group velocity of 300 m/s, phase velocity of may also observe that topography around Profitis strong
1250 m/s) than those generated by faults F1 and F4. Thus, motion station does not have a significant effect in the
clearly the Love waves generated by fault F3 are guided by frequency band included in the synthetics.
the deeper layers, which have higher propagation velocities We have used the seismograms of Fig. 6 to compute
and present a significant discontinuity at this fault. transfer functions relative to input ground motion. The
Fig. 5 shows the Love wave dispersion curves computed 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
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
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
Fig. 7. Theoretical transfer function for vertical incidence of SH waves on During 1994, an ambitious recording campaign using
the model shown in Fig. 2. This result was obtained from the synthetics high quality seismographs was carried out. Thirty-nine digi-
shown in Fig. 6. The positions of the surface accelerographs of the perma- tal seismographs were installed in the valley for more than
nent network have been indicated for reference. We have corrected for the 2 months. The instruments were disposed in three networks
free field factor, thus the transfer function value in absence of the valley
(Figs. 9 and 12), identified by the type of instruments, which
would be a factor of one.
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
20.0
BED
CHU
ONI
FIE
YEL
TES
DEP
MAI
SCU
GRE
TRE
MUR
BAN
THA
BUT
ROC
PRO
TOB
KOR
FAR
15.0
Time [s]
10.0
5.0
0.0
0.0 1.0 2.0 3.0 4.0 5.0 6.0
Distance [km]
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 permanent network of digital accelerogaphs across the
the direction N59.5E (SH motion). Also in Fig. 13(b) are valley. Seven surface instruments plus two borehole instru-
shown the phase velocity dispersion curves expected for the ments were installed in early 1994, and have already
fundamental and first higher mode from the velocity profile recorded a significant number of earthquakes (more than
at the center of CIES array. In spite of aliasing problems 50). There have been two site effect studies of Euroseistest
(evident for those points showing a phase velocity increas- using the data from the accelerograph array, Raptakis et al.
ing linearly with frequency in some frequency bands) we [15] and Dimitriu et al. [16].
observe a good agreement between observed and theoretical The results from both papers Raptakis et al. [15] and
dispersion curves for the fundamental mode in the Dimitriu et al. [16] are very similar. Both of them deal
frequency band 1–3 Hz. Our analysis suggests that the S- with the evaluation of empirical transfer functions in the
wave window of CIES records consists of Love waves frequency domain. The techniques used are the same in
between 1 and 2 Hz, while the later coda of S-waves both cases: standard spectral ratios relative to a reference
consists of Love wave in the frequency band between 1 site, and horizontal-to-vertical spectral ratios. Thus, the
and 3 Hz. The phase velocity of these surface waves is results of their analysis are very similar to those obtained
comprised between 300 and 400 m/s in the frequency from the Reftek array. The differences between Raptakis et
band between 1 and 3 Hz, in good agreement with the al. [15] and Dimitriu et al. [16] lay on their respective
results from finite difference modeling. Thus, wavelengths emphasis. For Raptakis et al. [15] the main point is the
of these Love waves should be between 100 and 400, which comparison between the two techniques to estimate the
explains why individual pulses of these Love waves could amplitude of the transfer functions across the valley. Dimi-
not be correlated among the stations of the Reftek array. triu et al. [16] are more interested in determining site
We observe then that locally generated surface waves, response for station TST, at the center of the valley. Then,
predicted by numerical modeling, are indeed present in Dimitriu et al. [16] proceed to model the observed spectrum
actual earthquake records. These surface waves cannot be at TST during five significant events by multiplying the
identified in the frequency domain transfer functions, spectrum observed at PRO (the reference station on rock)
because they do not appear as separate peaks of amplifica- for each of those events with a 1D, S-wave propagation
tion. Rather, they contribute to the main “resonance peak” transfer function. The stratigraphy they used for TST was
of the empirical transfer functions. taken from the well-constrained measurements of Pitilakis
et al. [6]. Dimitriu et al. [16] use the incidence angle of the
3.2. Results using accelerographic data plane shear waves as free parameter to obtain the best fit.
Of the two papers, only Raptakis et al. [15] mentions
In addition to the temporary installation of seismograph briefly the possible influence of 2D site effects in the
arrays, Euroseistest project included deployment of a 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 event, which are useful in this analysis. They were recorded
between observed average transfer function at TST site in borehole instruments, at depths of 17 and 72 m, under
and a numerical transfer function obtained from a 1D, verti- TST surface station. We have analyzed these records to
cal propagation of S-waves model. Now, it is clear from Fig. determine if they support our claim of the importance of
6 that the site response at the center of the valley is largely surface waves at TST. First, we aligned the arrival time of
influenced by locally generated surface waves. To verify P wave in the vertical components, to insure an almost
this point, we have again chosen a representative event, common time base for the three records. We incur in an
well recorded by the accelerograph array, the event of uncertainty as large as the vertical travel time of P wave
Arnea. This event occurred on 4 May 1995, had a local between each couple of stations. According to the P-wave
magnitude of 5.3 and was located 32 km to the South– velocities determined in the model, this travel time is of
East of TST. Fig. 14 shows the seismic section obtained 0.03 s between 72 and 17 m depth stations, and 0.02 s
for the records of this event at the accelerograph array. between 17 m depth and the surface. Given our sampling
We plot the motion in the direction N59.5E, i.e. SH motion. rate, these times correspond to seven and four samples
The acceleration records shown in this figure were differences, which are quite small. Second, our analysis
processed in the same way used for the weak motion records suggested that the instrument at 72 m depth has a different
shown previously (low-pass filtered with a 3.5 Hz cutoff). orientation from the other two accelerographs (at 17 m
The result shown in Fig. 14 is very similar to what we depth and at the surface). This is shown in Fig. 16, where
obtained from the weak motion records. At low frequencies, we plot the horizontal components recorded at the three
the most striking differences between records outside the accelerographs, low-pass filtered with a 0.5 Hz frequency
valley and those on the sediments are not in amplitude but cutoff. At such low frequencies, the traces should be very
in the duration of shaking. At stations within the valley, late similar, as Fig. 16 shows. However, the traces for the record
arrivals with significant amplitude are readily identified, and at 72 m depth suggest that the deepest instrument has rotated
very likely are locally generated Love waves. Fig. 15 shows 2908. In what follows, we have corrected for this rotation.
the evolutionary spectra for stations PRO, on rock, and TST, The results of numerical modeling suggest that two
at the center of the valley. We observe that the largest peaks modes of Love waves (fundamental and first higher mode)
in the low frequency range at station TST occur between 20 are important in Euroseistest, at 0.8 and 2 Hz. For this
and 25 s time, while S-wave arrival is at 14 s. It is clear that reason we filtered the observed borehole records around
the late arrivals play a significant role in the “resonant” peak these two frequencies. Fig. 17 shows the transverse compo-
identified at 0.8 Hz by Raptakis et al. [15] and Dimitriu et al. nent (parallel to the valley’s axis) of the three surface and
[16] at TST station. downhole records band-pass filtered between 0.4 and 1 Hz.
There are two additional acceleration records of Arnea As expected, the traces are similar. The dashed lines in
0.02
0.0
-0.02
6.77E-05
4.0
Frequency [Hz]
3.0
2.0
1.0 0.0
0.0
5.0 10.0 15.0 20.0 25.0
Time [s]
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.
F.J. Chávez-Garcı́a et al. / Soil Dynamics and Earthquake Engineering 19 (2000) 23–39 33
cne1
cnn1
a
cnn2
cnn3
cns1
cns2
cns3
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
TST
GRB
GRA
PRO
20.0
15.0
Time [s]
10.0
5.0
0.0
0.0 1.0 2.0 3.0 4.0 5.0
Distance [km]
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 encouraged the continued use of 1D models, because of the
were normalized by the value at the free surface. The thick simplicity of computing a theoretical transfer function to
lines in this figure show the theoretical modal shapes compare with the observations. This despite the fact that it
computed for the fundamental (for the frequency range has been almost 20 years since the demonstration that in
0.4–1 Hz) and first higher mode (for the frequency range shallow alluvial valleys, such as Euroseistest, locally gener-
1.6–2.4 Hz) for the vertical profile at TST station. The ated surface waves should be the predominant phenomenon
symbols correspond to the amplitudes measured along the [18,19].
dashed lines in Figs. 17 and 18. The agreement between In the particular case of Euroseistest, frequency domain
observed and predicted modal shapes is good. Thus, the analysis of weak motion data showed that the sediments
borehole records support our claim that, in the low filling the valley of Euroseistest amplify ground motion
frequency range, ground motion is dominated by locally by factors larger than 10 between 1 and 3 Hz. In the time
generated surface waves. 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
4. Discussion surface waves. If we look at the data in the time domain,
both seismograph and accelerograph records confirm the
Most of the recent observational site effect studies deal theoretical results. We observe that surface waves are
with the evaluation of empirical transfer functions. Such excited at the same frequencies at which we expect the
techniques as standard spectral ratios, generalized inversion, “resonance frequency” to occur. It is not surprising that
or horizontal-to-vertical spectral ratios dominate in a large the predominant frequencies for body and surface waves
percentage of recently published papers on site effects. For coincide because both are two manifestations of the same
example, a large number of papers have been devoted to phenomena.
discussing the advantages and disadvantages of the former The problem appears when we want to interpret the
and the latter techniques. In this sense, the paper by Safak empirically determined transfer functions. For example,
[17] is revealing in its in-depth exploration of different tech- Dimitriu et al. [16] interpreted the transfer function at
niques, with the aim of evaluating which would be the best TST in terms of oblique incidence of SH waves in a 1D
way to define an empirical transfer function between two model. For example, Fig. 20 shows the very good fit
records. In principle, the use of amplitude transfer functions Dimitriu et al. [15] obtained for the observed spectrum at
does not restrict the universe of models. In practice, this has 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, seismogram. This phase information was lost in the deter-
ground motion at TST cannot be modeled using 1D models. mination of the empirical transfer functions. An additional
If we take the “well-determined” 1D transfer function used problem is that amplitude transfer functions tend to make us
in Fig. 20 to model ground motion in the time domain, we forget the physics of site effects. We suggest that the deter-
would find that amplitudes are overestimated, while dura- mination of empirical transfer functions is not enough to
tion of ground motion is grossly underestimated (see for discriminate between 1D and 2D site effects, and that a
example, Refs. [20,2]). The reason is clearly that the 1D successful modeling of a partial aspect of site response
“resonance” peak includes the contribution of locally gener- (its amplitude) cannot guarantee that we can make success-
ated surface waves, which occur at different times in the 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
Fig. 20. Solid lines: observed Fourier amplitude spectra at PRO and TST strong motion stations during the Kozani-Grevena earthquake (13.05.95, MS 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 XII, for Science, Research & Technology (contracts
information of site response. For this reason, they can be #EV.5V-CT.93-0281 and #ENV4-CT.96-0255).
misleading. A good fit of this partial observation using 1D
models cannot rule out the possible importance of 2D site References
effects. If 2D site effects are important, predictions of time
histories of ground motion made on the basis of the 1D good [1] Pitilakis et al. EURO-SEISTEST. A European Test-Site for Engineer-
fit will lead to large errors. ing Seismology, Earthquake Engineering and Seismology, Final
Scientific Report, task 4.3: Laboratory test, 1995.
This study was made possible by the large effort carried out
[2] Raptakis D, Chávez-Garcı́a FJ, Makra K, Pitilakis K. Site effects at
during the Euroseistest project. The high quality database EUROSEISTEST—I: determination of the valley structure and
accumulated until now is quite large. Moreover, the continu- confrontation of observations with 1D analysis. Soil Dyn Earthq
ing operation of this test site, together with deployment of Engng 2000;19(1):1–22.
additional instruments, will make this test site very important [3] Marrara F, Suhadolc P. Estimation of local site effects in the Volvi
basin (Greece): comparison between observations and modeling. In:
in Earthquake Engineering during many years to come.
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.
Acknowledgements [4] Tolis S, Faccioli E, Pitilakis K. A 2D simulation of Euroseistest
near Thessaloniki, Greece. In: Irikura K, Kudo K, Okada H,
This work was possible thanks to the many people who editors. The effects of surface geology on seismic motion. Recent
have participated in the different field experiment of Euro- progress and new horizon on ESG studies, USSR: Balkema, 1998.
p. 255–62.
seistest project. P. Moczo and J. Kristek kindly gave assis- [5] Jongmans D, Pitilakis K, Demanet D, Raptakis D, Riepl J, Horrent C,
tance with finite difference computations, and granted CPU Tsokas G, Lontzetidis K, Bard P-Y. EURO-SEISTEST: determina-
time in one of their machines to compute results for the tion of the geological structure of the Volvi basin and validation of the
larger model. This work was done while the senior author basin response. Bull Seism Soc Am 1998;88:473–87.
was on a sabbatical leave from UNAM at Aristotle Univer- [6] Pitilakis K, Raptakis D, Lontzetidis K, Tika-Vassilikou Th, Jongmans
D. Geotechnical & geophysical description of EURO-SEISTEST,
sity of Thessaloniki. This visit was supported by Dirección using field, laboratory tests and moderate strong-motion recordings.
General de Asuntos del Personal Académico, UNAM, J Earthq Engng 1999;3(3):381–409.
Mexico, and by the European Union—Directorate General [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 39
using irregular grids: application to the seismic response problem. [14] Meunier C. Variabilité spatiale des mouvements sismiques sur courtes
Geophys J Int 1989;99:321–9. distances. DEA report, LGIT, Université Joseph Fourier de Grenoble,
[8] Moczo P, Bard P-Y. Wave diffraction, amplification and differential 1997.
motion near strong lateral discontinuities. Bull Seism Soc Am [15] Raptakis D, Theodulidis N, Pitilakis K. Data analysis of the Euro-
1993;83:85–106. seistest strong motion array in Volvi (Greece): standard and
[9] Moczo P, Labák P, Kristek J, Hron F. Amplification and differential horizontal-to-vertical spectral ratio techniques. Earthq Spectra
motion due to an antiplane 2D resonance in the sediment valleys 1998;14:203–24.
embedded in a layer over the half-space. Bull Seism Soc Am [16] Dimitriu PP, Papaioannou CA, Theodulidis NP. EURO-SEISTEST
1996;86:1434–46. strong-motion array near Thessaloniki, Northern Greece: a study of
[10] Reynolds AC. Boundary conditions for the numerical solution of site effects. Bull Seism Soc Am 1998;88:862–73.
wave propagation problems. Geophysics 1978;43:1099–110. [17] Safak E. Models and methods to characterize site amplification from a
[11] Riepl J. Effets de site: évaluation expérimentale et modélisations pair of records. Earthq Spectra 1997;13:97–129.
multidimensionnelles: application au site test EURO-SEISTEST [18] Bard P-Y, Bouchon M. The seismic response of sediment-filled
(Grèce). PhD thesis, Université Joseph Fourier—Grenoble I, 1997. valleys. Part 1. The case of incident SH waves. Bull Seism Soc Am
p. 227. 1980;70:1263–86.
[12] Riepl J, Bard P-Y, Hatzfeld D, Papaioannou C, Nechstein S. Detailed [19] Bard P-Y, Bouchon M. The seismic response of sediment-filled
evaluation of site response estimation methods across and along the valleys. Part 2. The case of incident P and SV waves. Bull Seism
sedimentary valley of Volvi (EURO-SEISTEST). Bull Seism Soc Am Soc Am 1980;70:1921–41.
1998;88:488–502. [20] Chávez-Garcı́a FJ, Bard P-Y. Site effects in Mexico City eight years
[13] Phillips WS, Aki K. Site amplification of coda waves from local after the September Michoacan earthquakes. Soil Dyn Earthq Engng
earthquakes in Central California. Bull Seism Soc Am 1994;13:229–47.
1986;76:627–48.