Topography Effects in the 1999 Athens Earthquake:

Engineering Issues in Seismology

D. Assimaki1, E. Kausel2, G. Gazetas3

1
Department of Civil & Environmental Engineering, Massachusetts Institute of Technology, U.S.A.
2
Department of Civil & Environmental Engineering, Massachusetts Institute of Technology, U.S.A.
3
Department of Civil Engineering, National Technical University of Athens, Greece

Abstract—During the 1999 Athens Earthquake, the town amplification patterns on hill slopes. The problem of
of Adàmes, located on the eastern side of the Kifissos river scattering and diffraction of seismic waves by topographical
canyon, experienced unexpectedly heavy damage. Despite irregularities has been also studied by many authors. The
the significant amplification potential of the slope geometry, majority of these studies focus on two-dimensional
topography effects cannot alone explain the uneven damage simulations in which the topographic asperities are treated as
distribution within a 300m zone behind the crest, chara- isolated ridges or depressions, usually on the surface of
cterized by a rather uniform structural quality. This paper homogeneous elastic media. Comparison between
illustrates the important role of soil stratigraphy, material instrumental and theoretical results reveals that there is
heterogeneity and soil-structure interaction on the formu-
indeed qualitative agreement between theory and
lation of surface ground motion. For this purpose, we first
observations on topography effects. Nevertheless, from a
perform elastic two-dimensional wave propagation analyses
based on available local geotechnical and seismological data, quantitative viewpoint, there still exists clear discrepancy in
and validate our results by comparison with aftershock numerous cases, where the observed amplifications are
recordings. Next, we conduct inelastic time-domain simula- significantly larger than the theoretical predictions.
tions that include spatial variability of soil properties and Furthermore, results from instrumental studies on weak
soil-structure interaction effects, to reveal their additive motion data or ambient noise may not be applicable to
contribution in the topographic motion aggravation. describe topography effects for strong ground shaking, which
is usually associated with inelastic soil response. Indeed,
Keywords—Site effects, topography, stratigraphy, 2D there exist very few –if any– well documented case studies
wave propagation, inelastic soil behavior, random where topography effects are illustrated for strong ground
properties, soil-structure interaction motion.
This paper uses a case-study from the Athens 1999
earthquake to illustrate the decisive role of local stratigraphy,
INTRODUCTION material heterogeneity and soil-structure interaction in
altering the energy focusing mechanism at the vertex of
It has been long recognized that topography can convex topographies. The effects of local soil conditions are
significantly affect the amplitude and frequency characte- validated by comparison with weak motion data. The effects
ristics of ground motion during seismic events. In the recent of inelastic soil behavior and nonlinear soil-structure
past, documented observations from destructive seismic interaction are then illustrated for the strong motion
events show that buildings located at the tops of hills, ridges recordings. Our conclusions can be used as guideline for
and canyons, suffer more intensive damage than those more rigorous analyses to be performed, accounting for the
located at the base: the Lambesc Earthquake [France 1909], additive contribution of engineering issues in the extensively
the San Fernando Earthquake [1971], the Friuli Earthquake, studied seismological problem of topographic motion
[Italy 1976], the Irpinia Earthquake [Italy, 1980], the Chile amplification.
Earthquake [1985], the Whittier Narrows Earthquake [1987],
the “Eje-Cafetero” Earthquake [Colombia, 1998] and recent THE 7-SEPT-99 EARTHQUAKE
earthquakes in Greece [Kozani, 1995 and Athens, 1999] and
Turkey [Bingöl, 2003] are only some examples of The Ms 5.9 event that shook Athens just three weeks
catastrophic events, during which severe structural damage after the Ms 7.4 Kocaeli Earthquake has been characterized
has been reported on hilltops or close to steep slopes. as the worst natural disaster in the modern history of Greece.
Still nowadays, topographic amplification is poorly This moderate event had a major socio-economical impact,
understood and the insufficient number of documented resulting in the loss of 150 lives, the collapse of 200
evidence prevents these effects from being incorporated in residential and industrial buildings and the severe damage of
most seismic code provisions and microzonation studies, another 13,000. Despite the acquired knowledge after two-
despite their undisputable significance in engineering and-a-half millennia of history, no earthquake has ever been
practice. Instrumental studies that have been performed in assigned to or near the 1999 seismogenic fault. The location
recent years verify the macroseismic observations, by of the ruptured fault and the geography of the heavily
predicting systematic amplification of seismic motion over damaged region, are schematically illustated in Fig. 1. Also
convex topographies such as hills and ridges, de- shown on the same Figure are the locations of the four
amplification over concave topographic features such as accelerograph stations, which recorded the strongest
canyons and hill toes, and complex amplification and de- motions: KEDE, MNSA, SGMA and SPLB.

31
32 11th ICSDEE / 3rd ICEGE Proceedings

Adàmes: Observations, Topography, Stratigraphy ensuing as profiles A, B and C. All profiles comprise
alternating soil layers of silty-gravely sands and sandy-
One of the most heavily damaged areas was the small gravely clays in the top 20-30m. The approximate average
community of Adàmes, located next to the deepest canyon of velocity, Vs,30 of the 30m surface soil layers for the three
Kifissos river, the main river of the Athens metropolitan profiles are: 500 m/sec for profile A, 400 m/sec for profile B
area. The majority of local buildings, constructed in the 70’s and 340 m/sec for profile C, indicative of very stiff (profile
and early 80’s, comprise 2- to 4-storey concrete reinforced A), just stiff (profile B), and moderately stiff (profile C) soil
structures of rather uniform quality. Nevertheless, the MMI formations according to the European Seismic Code (EC8).
in the 1200m long and 300m wide town ranged from VIII to
IX+, despite its 8-10km distance from the projection of the
causative fault.
The location of the town next to the crest of the canyon
in conjunction with the high damage intensity (as opposed to
numerous other towns located at equal or smaller distances
from the source where MMI did not exceed a mere VII)
brought forward topography effects to justify the
macroseismic observations. Behind the crest however,
damage was bilaterally non-uniform, and in the parallel to
the river axis direction, was concentrated in two zones: one Fig 2: Typical cross-section of the topographic relief of Kifissos
next to the crest and one at a distance about 200m-300m river canyon and the region of Adàmes
from it. Some scattered -yet less intense- damage was Shear Wave Velocity (m/sec)
observed at intermediate locations. It seems therefore, that
0 200 400 600 800
focusing of seismic energy at the vertex certainly played a 0
significant role but was not the only phenomenon involved.

-20
Adàmes
MMI VVI+ – IX-
Depth (m)

-40
Projection of
rupture surface

SPLB -60
MNSA
SGMA
KEDE Profile A
Profile B
-80
Accelerograph Profile C
10 km stations
Fig. 3: Shear wave velocity variation with depth of three
characteristic soil profiles in Adàmes

Fig. 1: Sketch of the map of the stricken region showing the surface Strong Motion Records
of the projection fault, the severely damaged town of Adàmes, and
the accelerograph stations that recorded the strong motion time- Fifteen strong-motion accelerograph stations were
histories used in our analyses (filled triangles). The dots show the triggered by the main shock within 25 km from the causative
location of the 30 collapsed buildings with human casualties.
fault, recording peak ground accelerations (PGA) ranging
from about 0.05 g up to 0.50 g. Their location is depicted in
A topographic survey of the canyon produced the cross- Fig. 1. Nevertheless, there were no records in the meizo-
section shown in Fig. 2. The slightly idealized geometry of seismal area. Due to lack of acceleration records in the area
the canyon used in our investigation is also shown in this of interest, these strong motion time-histories were used in
figure. Note the 40m deep and the nearly 2:1 (horizontal to our simulations. To recover the corresponding motion at
vertical) slope of the canyon cliff. rock-outcropping, numerical 1D or 2D deconvolution ana-
Geotechnical investigations of the area comprised the lyses were performed for the recordings obtained at stations
drilling of ten (10) boreholes with Standard Penetration Blow where local soil conditions or adjacent structures have
Count (NSPT) measurements and laboratory testing for the altered the bedrock input motion.
definition of the variation of plasticity index (Ip) with depth. Nevertheless these motions were recorded within a
Eight of these were performed down to a depth of about 35 narrow region located 10 km away from the end of the
m, and two reached almost 80 m. Some indirect evidence for ruptured zone, in a direction perpendicular to it, whilst -by
greater depths was “extrapolated” from two 150-m-deep contrast- the Kifissos river canyon lies in front of the rupture
boreholes drilled for the under-construction Olympic zone. There is, therefore, strong indication that forward
Village, 1.5 km west to northwest of Adàmes. The overall rupture directivity is likely to have affected the ground
picture emerging from this investigation is shown in Fig. 3, motion in the town of Adàmes. Accounting for near fault
where low-strain shear wave velocity profiles are constructed effects implies the selection of appropriate time histories
for three characteristic locations in Adàmes, referred to in the characterized by a relatively simple long period pulse of
 Seismicity, Ground Motions and Site Effects 33

strong motion having relatively short duration, which cannot approximately with the P-wave velocity. As a result of this,
be described by a relatively long stochastic process, as is they arrive in the vicinity of the crest almost simultaneously
indeed the case in more distant seismic events. Two historic with the direct SV-wave.
time-histories have been identified in the world strong- The significant enhancement of forward scattered
motion database from the 1966 Ms 5.6 Parkfield, CA Rayleigh waves (resembling forward directivity effects),
Earthquake, which encompass such long-period character- along with the presence of surface waves traveling along the
ristics, recorded at the Cholame Shandon No. 8 and Temblor slope with velocity of propagation which approaches the P-
stations. wave velocity can be explained from the combination of the
The response spectra of these six acceleration time slope angle and material Poisson’s ratio corresponding to the
histories, which represent the general strong motion present analysis: for Ȟ = 0.35, critical incidence is calculated
characteristics of the Athens event and possible directivity as:
effects anticipated at the area of interest, are plotted in Fig. 4. θ cr = arcsin (Vs / V p ) = 28.71
≈ i = 30

For further details on the seismo-tectonics of the event,
the geotechnical investigation in Adàmes and the strong SV
R2
motion data-processing, the reader is referred to Ref. [1], [2] 500
and [3].
1.2 400

Surface Receiver Location (m)

Cholame
1.0
(PGA 0.27g)
300
0.8

200
0.6

0.4
100

0.2 MNSA
R1
(PGA 0.22g) 0
0 SP
0 0.2 0.4 0.6 0.8 1.0 SV
1 1.5 2 2.5 3
T (sec) Ti ( ) R2
R2
500
Fig. 4: Acceleration response spectra of strong-motion recordings,
defined at rock-outcropping in our simulations
400
ELASTIC SIMULATIONS
Surface Receiver Location (m)

We here investigate the diffraction potential of the slope 300

geometry, the frequency-dependence of topographic

amplification and the role of soil layering and material 200
heterogeneity by means of elastic finite-element parametric
simulations. Successively, the local site conditions are
modeled and our results are validated by comparison with 100

aftershock recordings.
Geometry: Fig. 5 illustrates the wavefield generated 0
R1

by a cliff with the geometry of the Kifissos canyon, at the SP

surface of a homogeneous halfspace upon the incidence of 1 1.5 2 2.5 3
Time (sec)
vertically propagating SV-waves. We here use a narrow- Fig. 5: Synthetics of horizontal (top) and vertical (bottom)
band input, namely Ricker wavelet with normalized acceleration surface response, for a cliff with 30o slope subjected to
frequency a0 = 2 f 0 h / Vs =1.00 , where f0 is the central vertically propagating SV Ricker waves with a0 = 1.0
frequency of the pulse, h is the height of the cliff and Vs is
the shear wave velocity of the halfspace. The Poisson’s ratio Therefore, vertically propagating waves strike at the free
of the elastic medium is Ȟ = 0.35, typical of stiff clayey soils. surface of the slope with almost critical incidence, resulting
The direct / diffracted wavefield shown in Fig. 5 in the transformation of practically all the incident energy
comprises the following waveforms: (i) Direct SV waves into surface waves that travel along the slope and
(denoted SV), (ii) forward scattered Rayleigh waves constructively interfere with the direct SV waves that arrive
(denoted R1) generated at the boundaries of the shadow / behind the crest. Therefore, the site conditions in Adàmes
illuminated zone at the lower corner of the cliff, propagating satisfy a priori conditions which favor a complicated and
along the cliff and being forced to change direction at the detrimental diffraction potential, simply by considering the
upper corner, (iii) Backward scattered Rayleigh waves elastic response of the canyon cliff.
(denoted R2) generated at the boundaries of the shadow / Note also that despite the horizontally-polarized particle
illuminated zone at the lower corner of the cliff and motion of the incident seismic input (vertically propagating
propagating outwards, and (iv) surface waves (denoted SP) SV waves), the surface response contains a parasitic vertical
that are generated along the cliff and propagate upwards acceleration component as well. This corresponds to the
34 11th ICSDEE / 3rd ICEGE Proceedings

vertical particle motion of surface diffracted waves, and is amplified within a narrower zone in the vicinity of the crest.
shown to carry significant portion of the seismic wave (iii) The lobes of constructive and destructive interference at
energy. For the illustrated case, its normalized amplitude is the surface -controlled by the frequency content of the
of the order of 35% of the peak surface acceleration at the incident waves- result in significant differential motion
far-field. behind the crest and along the slope, where transition occurs
Frequency of incident motion: Fig. 6 illustrates the between the convex and concave part of the topography.
spatial distribution of peak surface acceleration (horizontal Numerical simulations have also been performed for a
and vertical) behind the crest, for vertically incident SV homogeneous layer overlying elastic halfspace, for various
Ricker wavelets with different central frequencies. The impedance contrasts. Our results show that the bedrock-soil
response is normalized by the peak acceleration at the far- impedance ratio that controls the seismic energy trapped in
field, defined here at distance 300m from the vertex, where the surface layer and the corresponding one-dimensional
2D phenomena are shown to be negligible. amplification of the motion, introduces additional
complexity to the problem studied. Resonance of the shallow
1.75 (in front of the toe) or deep (behind the crest) far-field soil
α0 = 0.5 columns not only controls the overall response of the
1.50
α0 = 1.0 configuration, but indeed enhances the topographic
Νormalized Peak Acceleration

α0 = 1.5 amplification of motion by altering the diffraction

α0 = 2.0 mechanism.
/ afar-field

1.25
The frequency-dependence of the amplification
mechanism can be summarized as follows: (i) for a constant
2D

1.00 input motion, topographic aggravation of the response

hor
a

increases with increasing height of the topographic feature

0.75 o
(h), and (ii) for a given feature, topographic amplification
slope inclination 30 increases with frequency, yet occurs within a more confined
Vertical SV incidence
zone in the vicinity of the vertex (it is shown that max(ahor)
0.50
100 150 200 250 300 occurs at x = 0.2Ȝ0 and max(avert) at x = 0m from the crest)
0.7 Soil Layering: The effect of soil layering is here
α0 = 0.5 examined by means of a single surface layer, of thickness
0.6 α0 = 1.0 h1/h = 0.25 and variable shear wave velocity (Vs1),
Normalized Peak Acceleration

0.5
α0 = 1.5 overlaying homogeneous halfspace. Results are shown for a
α0 = 2.0 soft-surface layer with Vs1 / Vs = 0.5, where Vs is the shear
a 2D / afar-field

0.4 wave velocity of the halfspace. This is indeed an idealized

model for typical soil profiles in Adàmes, but also for many
0.3 sedimentary soil deposits.
vert

Fig. 7 illustrates the effects of a surface soft layer on the

0.2
normalized surface response, by comparison with the
0.1 homogeneous halfspace ground motion. Results can be
summarized as follows:
0 (i) The incident wave energy is trapped within the surface
100 150 200 250 300
layer, and multiple reflections interact with the surface waves
Surface Receiver Location (m) that originate from the lower corner of the slope and
Fig. 6: Normalized horizontal (top) and vertical (bottom) peak propagate uphill. The scattered wavefield at the surface
surface acceleration behind the crest (x = 300m) as a function of the comprises Rayleigh waves that are generated at the crest and
frequency content of the incident pulse. travel with the Rayleigh wave velocity of the surface layer,
VR1, and reflections of waves that travel along the layer-
The main conclusions drawn from our investigation are
halfspace interface with the Rayleigh wave velocity of the
the following:
later, VR.
(i) The diffraction problem of topographic amplification is
(ii) For high-frequency incident waves, the peak normalized
strongly frequency-dependent. The location of peak
horizontal acceleration of the stratified medium is lower than
horizontal acceleration behind the cliff is controlled by the
the corresponding of the homogeneous halfspace. Note
dominant wavelength (Ȝ0) of the incident motion (here the
however that the absolute motion amplification is very
central frequency of the Ricker wavelet) and is
significant in the soft surface layer case, a fact that illustrates
systematically observed at a distance x = 0.2 Ȝ0 from the
the dominant role of the far-field stratigraphy on the
crest (see also Ref. [4]). The amplitude of peak acceleration
amplification mechanism behind the crest.
at this location is also frequency-dependent, and increases
(iii) The vertical acceleration component is remarkably
almost linearly with frequency.
enhanced. This effect is prominent for incident waves with
(ii) The amplitude of the parasitic acceleration component is
wavelengths short enough to see the surface layer. In this
also frequency-dependent. For the high-frequency input
case, the vertical acceleration is shown to attain amplitudes
pulse in these simulations, the peak vertical response is on
25% larger than the corresponding response at the far-field.
the order of 60% of the corresponding far-field response.
(iv) In the frequency domain, the far-field/2D transfer
Since the location of peak vertical acceleration is also
function is very erratic for wavelengths comparable with the
frequency-controlled, higher frequency components are
thickness of the surface layer, i.e. its resonant frequencies.
 Seismicity, Ground Motions and Site Effects 35

1.8
2.0
lized material heterogeneities significantly increase the dura-
α0 = 0.5
α0 = 1.0
tion of the surface response. Fig. 8 illustrates the Fourier
1.6 1.5 amplitude surface of the response behind the crest, for a typi-
Normalized Peak Response

α0 = 1.5
α0 = 2.0 cal realization of the stochastic field with șz / Ȝ0 = 0.0625 and
șx / Ȝ0 = 0.625. Clearly, the erratic frequency content of the
(a2D / aff )

1.4 1.0
response and the amplification level of high frequency com-
hor

1.2 0.5 ponents cannot be simulated by means of a homogeneous

medium.
1.0

0.8
200 220 240 260 280 300
1.25
α0 = 0.5
α = 1.0
0
1.00 α0 = 1.5
Normalized Peak Response

α0 = 2.0
(a2D / aff )

0.75
2.0
vert

1.5
0.50
1.0

0.25
0.5

0
200 220 240 260 280 300

Surface Receiver Location (m)

Fig. 7: Normalized peak surface acceleration behind the crest for a (a)
soft surface layer with Vs1 / Vs = 0.5, as a function of the
dimensionless frequency Į0.

Material Heterogeneity: We next investigate the effects

of material heterogeneity on the topographic amplification of
seismic motion. For this purpose, we generate Gaussian
shear wave velocity stochastic fields using a univariate
spectral density function, namely the exponential decaying
SDF (Eq. 1).
ξ   ξ  
2

ρi ( ξ ) = cos  2 tan −1    1 +    (1)
  θi     θi  
where ȡi is the correlation function in spatial direction i
(horizontal or vertical), ȟi is the separation distance and și the
correlation distance in the ith direction. Separate correlation
structures are assigned to the horizontal and vertical
direction, which are functions of the correlation distance, and (b)
the correlation function of the stochastic field is defined as
their product. The random fields are generated in the Fig. 8: Fourier amplitude surface of (a) horizontal and (b) vertical
wavenumber domain and successively denormalized and response, for random field with șz / Ȝ0 = 0.0625 and șx / Ȝ0 = 0.625
mapped on deterministic finite element models. The effects (top) and homogeneous halfspace with the same background
velocity (Į0 = 2.00)
of correlation distance of the simulated random media,
relative to the propagating wavelengths, are then evaluated
by means of Monte Carlo simulations (Ref. [5]). Local Site Conditions and Recorded Field Evidence
Comparison of time and frequency-domain results with
the corresponding response of a homogeneous halfspace with The 2D response of the stratified soil configurations
the same background stiffness, illustrates phenomenological corresponding to profiles A, B and C is next evaluated by
scattering attenuation for long wavelengths and enhancement means of elastic simulations. The numerical model is now
of frequency components whose wavelengths are comparable subjected to the strong motion time-histories described
with the horizontal correlation distance of the random above, and results of our analyses can be summarized as
medium. follows:
In particular, the spatial distribution of peak normalized (i) For the broad-band seismic input, topographic ampli-
surface acceleration is practically the same –or slightly fication occurs within a zone behind the crest, approximately
lower– than the halfspace response, yet individual simula- equal to the width of the topographic irregularity (L = 70m).
tions show significant amplification of the vertical compo- This is found to be in accordance with results of our
nent and enhancement of the high-frequencies of the incident parametric investigation.
motion. In addition, multiple wave reflections at the loca- (ii) Two-dimensional amplification of the horizontal
response is shown to be rather insensitive to soil stratigraphy,
36 11th ICSDEE / 3rd ICEGE Proceedings

yet enhanced in comparison to the homogeneous halfspace our conclusions. Nevertheless, what should be highlighted
case. Peak amplification is of the order of 0.30 aff, where aff herein is that the incorporation of spatial small-strain
is the far-field peak surface acceleration. This is again stiffness variability and correct calibration of Rayleigh
consistent with amplification computed for a0 ≈ 4 h / Vs in damping coefficients has proven to be of great importance
for the representation of site conditions and subsequent
our parametric study, where Vs is the mean shear wave
successful prediction of topographic amplification.
velocity of the cliff profile.
(iii) The magnitude of parasitic acceleration however, shows INELASTIC SIMULATIONS
strong dependence on the soil stratigraphy. This effect is
primarily controlled by stiffness of the surface layer. In One-dimensional analyses: The effect of local soil
particular, results show that the amplitude of the vertical conditions in modifying the intensity and frequency
acceleration range from 0.25aff for the stiffer profile A to characteristics of ground shaking in Adàmes is first
0.70aff for the softer profile C, where aff is the corresponding investigated by means of 1D inelastic wave propagation
far-field peak surface acceleration. analyses. The far-field profiles A, B and C are subjected to
Significant corroboration of our elastic numerical the six strong-motion time histories and the surface response
simulations comes from two sets of ground motions, is computed in the frequency-domain using an iterative
recorded during two aftershocks of the Athens 1999 event. equivalent linear algorithm, and in the time-domain, by
The instruments were installed in the free field, two at a site incremental nonlinear finite element simulations. For the
x § 300m from the crest, and one at x § 10m from the crest. former, a modified solution was adopted in which the strain-
The two major aftershocks have provided the empirical compatible soil properties are frequency-dependent, thus
transfer function spectra that are plotted in Fig. 9. Since the avoiding artificial damping of the high-frequency low-
seismographs were placed at locations with different soil amplitude components of motion (Ref. [6]). For the latter, we
property characteristics (profile B for the first two and C for used the multi-yield plasticity soil model implemented in the
the later), the Fourier spectra evaluated from the aftershock computer code DYNAFLOW (Ref. [7]). For consistency, the
accelerograms have been initially divided by the one- parameters controlling the shear behavior of the constitutive
dimensional transfer function for each profile. For the model were calibrated to yield the modulus degradation
recorded peak accelerations being of the order of 0.015g, the curves (Ref. [8]) used in the equivalent linear solution. The
low-strain dynamic soil properties were used for this surface response computed by means of the two approaches,
purpose. Thus, the variability arising from soil-column is found to be in remarkable agreement.
flexibility effects has been eliminated. From the ensemble of the analyses performed, the
following conclusions are drawn:
2.5
Aftershocks (Site C) (i) Profile A being the stiffest of the three sites, shows an
Mean response (µ) appreciable degree of amplification in the period range of T
2D/Far-field Transfer Function

2.0
µ±σ < 0.3sec, where both PGA as well as spectral acceleration
(SA) values increase by an average of about 25% compared
1.5 to the rock-outcrop input motion. However, soil
FT2D / FTff

amplification does not alone suffice to explain the

1.0 observations. Topography and local soil conditions have
equally contributed to the observed damage distribution at
0.5
this site, which was more intense next to the crest. In fact, for
this stiff and relatively homogeneous profile, the moderate
damage intensity can be even justified by means of our
0
2 3 4 5 6 7 8 elastic 1D and 2D analyses (recall a2D/aff = 1.3 for profile A).
(ii) Profile B is softer than profile A, and simulations show
f [Hz]
larger amplification over a wider period range. Computed
Fig. 9: 2D/Far-field empirical transfer function from the records of
two strong aftershocks, and comparison with numerical results of 20 PGA values are in the range of 0.30g – 0.40g, and the
Gaussian stochastic field realizations highest SA reaches 1.50g at T ≈ 0.2sec. Evidently, there is a
pseudo-resonance condition occurring at this period: the
For the class-A prediction shown in Fig. 9, the fundamental period of the soil column (Tsoil ≈ 0.2sec from
stratigraphy of profile C has been used as the background the surface/rock-outcrop transfer function) nearly coincides
medium stiffness of a Gaussian stochastic field with șz = with the dominant excitation period (Tinput ≈ 0.2sec). The
2.5m and șx = 15.0m, extrapolated from the geostatistical dominant role of soil conditions becomes evident for this
data of an adjacent site. For the denormalization of the field, site, where the damage intensity was similar next to the crest
a constant standard deviation ı = 0.15 Vs has been adopted. and in the far-field.
The mean and standard deviation of the numerically (iii) Profile C is the softest of the three sites. The
predicted transfer functions from 20 realizations of the fundamental natural period of the soil deposit at the last step
stochastic field at x = 10m are shown in Fig. 9. For the of the iteration process is estimated Tsoil ≈ 0.72sec, whilst
simulations, a Ricker wavelet with central frequency f0 = most of the rock-outcrop excitations have much smaller
5Hz has been used. dominant periods, Tinput ≈ 0.1-0.2sec. Hence, no increase or
It can readily be seen that the recorded and computed even deamplification is expected in PGA and in SA values
results are in very good agreement, offering strong support to due to local soil conditions for T < 0.25 sec, a fact which was
confirmed by our results. On the other hand, the spectral
 Seismicity, Ground Motions and Site Effects 37

amplification predicted for periods 0.4 - 0.6 sec, could be mately with the so-called topographic frequency (see Ref.
substantial if the input motion were rich in such relatively [5]), determined from the location where maximum ampli-
long-period components. In summary, moderate elastic fication of motion behind the crest occurs (recall that this is h
topographic amplification of 30% and 1D inelastic soil / Ȝ0 = 0.2 for a homogeneous soil profile). For the stratified
deamplification do not justify the observations for this soil configuration, this can be approximated by 5h / Vs .
profile, characteristic of one of the most heavily damaged
Amplification of high-frequencies is more pronounced
regions in the 7-9-99 earthquake.
when material heterogeneity (a-priori randomness) is also
Two-dimensional analyses: We next perform 2D in-
introduced in the simulations. A more realistic random
elastic simulations and investigate the effects of material
small-strain stiffness field is here simulated as a non-
softening on the 2D amplification of surface motion. To
Gaussian stochastic field, where the theoretical correlation
simulate the inelastic soil response, we perform: (i) 2D
structures have been fitted to available geostatistical data
elastic analyses with strain-compatible soil properties
from an adjacent site (Ref. [9]). The spatial distribution of
computed at the last iteration of the 1D iterative solution, and
peak surface response is shown in Fig. 11 for profile C. The
(ii) time-domain 2D nonlinear analyses using the multi-yield
spectrum of TAF at x = 20m from the crest is compared to
plasticity model. The surface response is again normalized
that obtained from the inelastic wave propagation analysis of
by the far-field motion, which was found to be consistent for
the horizontally stratified configuration in Fig. 11.
the two methods. A more useful measure of topographic
amplification in the frequency domain is the response 1.6
spectral ratio of the 2D horizontal acceleration component to Random Inelastic

the corresponding far-field response. In the ensuing, we shall Stratified Inelastic

Normalized Surface Response

refer to this ratio as Topographic Aggravation Factor (TAF). 1.4 Stratified Eq. Linear

The mean TAF at x = 20m from the crest is plotted in Fig. 10 a2D / aff
as a function of period (T), for profile C and the ensemble of 1.2
hor

strong input motions.

2.00
Elastic 1.0
50m
Eq. Linear
Topographic Aggravation Factor (TAF)

1.75
Inelastic
0.8
200 220 240 260 280 300
1.50
1.0
SA2D / SAff

Stratified Random (µ)

1.25 Stratified Random (µ ± σ)
Normalized Surface Response

0.8
Stratified Inelastic

1.00 Stratified Eq. Linear

0.6
a2D / aff
vert

0.75
0.4
Profile C
0.50
0.2 0.4 0.6 0.8 1 0.2
T [sec]
Fig.10: Mean Spectrum of Topographic Aggravation Factor at
x=20m from the crest, for Profile C and six strong-motions 0
200 220 240 260 280 300

As can readily be seen, the elastic and equivalent linear Surface Receiver Location (m)

solution yield very similar spectral amplification values, Fig 11. Mean normalized peak acceleration for Profile C and a
random medium with the same mean stiffness and șz = 2.5m and șx
whereas the inelastic solution shows significant enhancement
= 16.0m, subjected to six strong-motions
of the high frequency components that is not predicted
otherwise. This can be justified when we consider that the The erratic surface response, which is substantially
incremental variation of strain-compatible soil properties amplified and more confined in the vicinity of the crest, is
introduces a-posteriori randomness in the original consistent with the enhancement of high-frequencies, when
horizontally stratified profile that has been shown to favor material heterogeneity and inelastic soil behavior are
amplification of short wavelength components. It is indeed in modeled. Ref. [9] illustrates that for 1D conditions, strong
the high-frequency regime that theoretical elastic models fail motion input introduces material yielding that is prominent
to predict measured amplification levels in the field. Note in the vertical direction and overshadows the small
also that the elastic finite-element solution is shown to be fluctuations of the stochastic small-strain field, yielding the
sensitive in the correct selection of Rayleigh damping horizontally stratified configuration adequate for seismic
coefficients, which when calibrated for the mean frequency response analyses. This is no longer valid for 2D wave
of the input motion artificially attenuate the high-frequency propagation analyses, where the localization of material
components. yielding is controlled by diffracted rather than direct waves
For larger periods (T > 0.2sec), the spectrum of TAF and does not restore the background medium stratigraphy
shows that in the period range 0.2sec < T < 0.4sec, where no when inelastic effects occur.
1D amplification occurs for the particular profile, topography Even more important for the justification of the damage
effects are negligible. They become important again in the distribution in site C is the amplitude of the vertical
period range 0.4sec < T < 0.6sec, which coincides approxi-
38 11th ICSDEE / 3rd ICEGE Proceedings

acceleration component, which attains values 0.8-1.0 aff near Results are shown for the stratigraphy of profile C in
the crest. This is consistent with the results of our parametric Fig.12, where the spectrum of TAF at the centerline of the
analyses that show amplification of the parasitic response for structure is compared to the free-field response at the same
a soft surface layer. location. Note that the high-frequency components of the
Nonlinear soil-structure interaction: The response of a response are geometrically filtered, yet for higher periods,
rigid surface structure founded next to the crest is here the frequency content of motion is practically unaffected by
simulated by means of inelastic 2D analyses. A schematic the presence of the structure. This verifies that no significant
illustration of the configuration is shown in Fig. 12. inelastic effects occur as a result of the structural static
For relatively soft soil formations, repeated loading on loading or inertial soil-structure interaction.
structures creates a zone of yielding and inelastic defor-
mation beneath the foundation. This reduces the effective CONCLUSIONS
dynamic impedance (radiation damping) of the semi-infinite
domain and creates resonant frequencies at which the Using a case study from the Athens 1999 earthquake,
structural motion is amplified. Since material yielding is we have shown that: (i) despite the detrimental diffraction
expected to be associated with seismic wave propagation, the potential of the cliff, geometry could not alone predict the
subsequent response of a structure on a soil profile that level of topographic motion amplification, (ii) even stiff soil
exhibits strain softening, is expected to be altered. sites (with average Vs = 400m/sec of the top 30m) can sub-
On the other hand, upon incidence of seismic waves, stantially amplify seismic motions, (iii) soft surface layers
soil deformations impose subsequent dynamic displacements significantly aggravate the amplitude of parasitic
to the foundation and the supported structure (so-called acceleration, which cannot be neglected for design purposes,
kinematic interaction). In turn, this induced motion of the (iv) weak motion data can be successfully used as a valuable
super-structure generates inertia forces that result in dynamic guidance in reconnaissance studies, yet are not adequate to
forces and moments at the base, subsequently transmitted describe topography effects associated with strong motions,
into the supporting soil. Therefore, additional deformations (v) 2D inelastic soil response introduces localized patches of
are imposed to the surrounding soil while additional waves yielded material, which equivalently to a random medium,
emanate from the soil-foundation interface (so-called inertial amplify high-frequency components and further enhance the
interaction). vertical response, (vi) soil-structure interaction on stiff soil
For the stiff soil formations in Adàmes, altering of the deposits filters the high frequencies of the horizontal motion.
response at the location of the structure is shown to be The normalization of the 2D response to the far-field
governed by kinematic interaction phenomena, namely the allows the decoupling of soil and topography effects in the
inability of the structure to follow the strongly differential estimated response. With an adequate number of strong-
surface response. As a result, frequency components of the motion case studies, the proposed Topographic Aggravation
horizontal response whose wavelengths are comparable or Spectrum can potentially be used for the development of a
shorter than the dimensions of the structure are filtered. simple rule for the estimation of topography effects and
Nevertheless, the vertical acceleration is almost unaffected pseudo-static slope stability analyses.
by the presence of the stiff structure and moreover, the
differential surface ground motion imposes additional REFERENCES
rocking loading.
[1] S.B Pavlides, G. Papadopoulos and A. Ganas, “The Fault that Caused
the Athens September 1999 Ms = 5.9 Earthquake: Field
Observations,” Natural Hazards, vol. 27, no. 1-2, pp.61-84, 2002.
[2] G. Gazetas, P.V. Kallou and P.N. Psarropoulos, “Topography and Soil
Effects in the MS 5.9 Parnitha (Athens) Earthquake: The Case of
Adámes,”Natural Hazards, vol. 27, no. 1-2, pp.133-169, 2002.
[3] D. Assimaki and G. Gazetas, “Soil and topographic amplification on
canyon banks and the Athens 1999 earthquake,” Journal of Earth-
quake Engineering, in press.
[4] S.A. Ashford, N. Sitar, J. Lysmer and N. Deng, “Topographic Effects
1.75 on the Seismic Response of Steep Slopes,” Bulletin of the Seismo-
Stratified Inelastic
logical Society of America, vol. 87, no. 3, pp. 701-709, 1997.
R. Popescu, “Stochastic variability of soil properties: data analysis,
Topographic Aggravation Factor

Random Inelastic [5]

1.50 Stratified Inelastic SSI digital simulation, effects on system behavior,” PhD thesis,
Princeton University, Princeton, New Jersey, 1995.
[6] E. Kausel and D. Assimaki, “Simulation of dynamic, inelastic soil
SA2D / SAff

1.25
behavior by means of frequency-dependent shear modulus and
damping,” Journal of Engineering Mechanics, ASCE, vol. 128, no.
1, pp. 34-47, 2001.
[7] J.H. Prevost, “DYNAFLOW: A nonlinear transient finite element
1.00
analysis program,” Technical report, Department of Civil Engi-
neering and Operations Research, Princeton University, Princeton,
New Jersey, 1981.
0.75 [8] M. Vucetic and R. Dobry, “Effect of Soil Plasticity on Cyclic
0 0.2 0.4 0.6 0.8 1.0
Response,” Journal of Geotechnical Engineering, vol. 117, no. 1,
T [sec] pp. 89-107, 1991.
Fig.12: Schematic illustration of numerical model, and mean [9] D. Assimaki, A. Pecker, R. Popescu and J.H. Prevost, “Effects of
spectrum of Topographic Aggravation Factor at x=20m from the Spatial Variability of Soil Properties on Surface Ground Motion,”
Journal of Earthquake Engineering, vol.7, no. 1, pp. 1-44, 2003.
crest, for Profile C and six strong-motions