Anda di halaman 1dari 22

Bulletin of the Seismological Society of America, Vol. 78, No. 1, pp.

42-63, February 1988

THE EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND


MOTION: A REVIEW AND NEW RESULTS
BY LOUIS GELI*, PIERRE-YVES BARD, AND BI~ATRICE JULLIEN

ABSTRACT
A brief review of experimental and theoretical results about the effect of
topography on seismic motion shows that they are consistent only on a qualitative
basis. Amplification at mountain tops for wavelengths comparable with mountain
widths is predicted and observed, but the numerical simulations often underestimate the actual observed effects. We propose that this disagreement is because
current models are not complex enough.
We therefore computed the response to incident SH waves of a set of different
complex configurations that include two-dimensional surface topography, with or
without periodic ridges and subsurface layering. The results show that: (1) the
topographic effect in itself is diffcult to isolate from other effects, like surface
layering, and therefore the amplification on top of geomorphologically complex
sites cannot be predicted by a priori estimations based solely on topography;
and (2) the high crest/base amplification ratio observed in the field cannot,
usually, be matched even with complex two-dimensional structures with incident
plane SH waves, which suggests that more complex models are needed to
incorporate more complex wave fields (e.g., SV, surface) and three-dimensional
geologic configurations.
INTRODUCTION

It is often reported after destructive earthquakes in hilly areas that buildings at


the top of massive crests suffer more intensive damage than those located at the
base; examples of such observations may be found in Angot (1909) (Lambesc,
France, 1909 earthquake), Brambati et al. (1980) (Friuli, Italy, 1976 earthquake),
Siro (1982) (Irpinia, Italy, 1980 earthquake), and ~elebi and Hanks (1986) (Chile
earthquake of 1985). Instrumental data also suggest that surface topography plays
an important role in ground motion (Davis and West, 1973; Rogers et al., 1974;
Berardi et al., 1978; Griffiths and Bollinger, 1979; Tucker et al., 1984).
In spite of these observations, the amplification effect of topography is not clearly
understood, except for qualitative trends. More precisely, all of the presently
available numerical and laboratory studies fail to match the high amplifications
observed during field experiments. As these numerical or laboratory models are
generally very simple, such as isolated two-dimensional ridges at the surface of a
homogeneous half-space, the purpose of this paper is to show how far the gap
between theory and observation may be narrowed when a more complex subsurface
geological structure is taken into account in numerical models. With that aim, we
first review the most important experimental results, trying in each case to point
out whether the disagrement between theory and observation may be caused by
such a complex geological structure. Then, we present the numerical results obtained
with an extension of the Aki-Larner (1970) method for a variety of complex twodimensional hilly structures with vertically incident S H waves and compare them
with the average quantitative trends inferred from experimental observations.

* Present address: IFREMER, Centre de Brest, BP 337, 29273, Brest Cedex, France.
42

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION

43

COMPARING THEORY AND OBSERVATIONS: A REVIEW

Theoretical results. Many numerical methods are now available for the computation of the seismic response of two-dimensional topographic irregularities above
a homogeneous half-space, ranging from finite difference (Boore, 1972; Zahradnik
and Urban, 1984), finite elements (Smith, 1975), integral equation method (Sills,
1978), boundary methods (Sfinchez-Sesma et al., 1982), and discrete wavenumber
methods (Bouchon, 1973; Bard, 1982). All of them consider the response of an
isolated two-dimensional ridge at the surface of an homogeneous half-space to
incident plane waves and provide consistent results, some of which are summarized
in Figure i for time-domain crest/base amplifications, Figure 2 for frequency domain
crest/base amplifications, and Figure 3 for the corresponding frequencies.
All of the theoretical studies show an amplification patternat the top and the
generation on the topographic irregularity of outward propagating waves. The time
domain crest/base amplification ratios (= ratio of peak crest motion to peak base
motion), shown in Figure 1, are generally increasing with increasing shape ratio (=
ratio of mountain height, h to mountain half-width, 1), but remain below 2 for
smooth isolated ridges.
The frequency domain studies in Figure 2 show that this crest/base amplification
corresponds to a combination of two effects
1. At mountain top, a broadband amplification for wavelengths comparable to
mountain width.
2. At mountain base, an alternation of amplification and deamplification, beginning at low frequencies (i.e., for wavelengths larger than mountain width) with
HORIZONTAL COMPONENT

gB7g

0#73

a TKHN84
.

GBT

VERTICAL COMPONENT

D#73

TKHNS4

RKB74

RKB74

+ Boo72

Boo72

x Smi75

Sm475

Ba82

S~82

S{78

S178

ZU84

ZU84
o

ZBYSO

GSJS6

to

ZBY80

GBJS6

ol

:11

so

D~a

io

0~

'

ol2

'

o!,

'

0'6

0!, >

Shape Batio

I
02

i
Shape

I
04
Ratio

'

&

J, >

Shape ratio : h / !

FIG. 1. Peak-to-peak time-domain crest/base amplification as reported in the literature and plotted
as a function of the apparent shape ratio (i.e., ratio between crest and base recording site altitudes to
their horizontal distance). Open symbols (squares, circles, etc.) indicate experimental field studies, while
others (crosses, dots, etc.) indicate results of theoretical studies. (a) Horizontal motion. (b) Vertical
motion. Each symbol corresponds to a paper as follows: GB79 = Griffiths and Bollinger, 1979; R K B 7 4
= Rogers e t ol., 1974; DW73 = Davis and West, 1973; Ba82 = Bard, 1982; Smi75 = S mi t h, 1975; Si78 =
Sills, 1978; T K H N 8 4 = Tucker e t al., 1984; ZBYS0 = Zhenpeng et al., 1980; Boo72 = Boore, 1972; and
ZU84 = Zahradnik and Urban, 1984.

44

LOUIS GELI, PIERRE-YVES BARD, AND BI~,ATRICE JULLIEN


GB79

GB7g

HORIZONTAL COMPONENT

DU73

Boo72

~6

Sm~75
5o82

S{78
o
a

"B x

X.&~

0.5

VERTICAL COMPONENT

TKHN84
w RKB74

~2

"

ZU84
ZBYS0
GBJB6

~
~

I
6

S~i75
E

Bo82

Si78

Zus4

2
o

"~

ZBYSO

GBJ86

Q5
o

B0o72

D#73

o TKHN84
.
RKB74

i
0.4

0.B
Shape R a t i o

Ol

(h/t)

Shape R a t l o

(h/l)

FIG. 2. Maximum crest to base spectral amplification as reported in the literature and plotted as a
function of tbe sbape ratio as in Figure 1. (a) Horizontal motion. (b) Vertical motion. See Figure 1 for
explanation of symbols.
GB79

HORIZONTAL COMPONENT

0u73

VERTICAL COMPONENT

o TKNN84
a

RKB74

80072
x
--

Smi75

Bo82

S178

''
,?~.+

.S

GB79

ou73

o TKHN84
.

RKB74

Boo72

Sm(75

Be82
S178

Shape Ratio ( h / l )

04
08
Shape Raio ( h / l )

FI(~. 3. Dimensionless frequencies corresponding to the maximum crest to base spectral amplification
peak displayed in Figure 2, still plotted as a function of shape ratio. (a) Horizontal motion. (b) Vertical
motion. The symbols and their meaning are exactly the same as in Figures 1 and 2, except for experimental
data corresponding to marked three-dimensional topographies, which are "double." The real frequencies
were multiplied, for each case, by the quantity 2 x 1//3, where 1 is the apparent half-width of the
topography (measured at its base), and fi is the S-wave velocity. These values were estimated on the
basis of the indications given in the quoted papers and are more a guess than a reliable measurement.
In the cases where there were two spectral peaks and a marked three-dimensional topography, we
assumed that the lowest frequency corresponds to the larger base width and the larger one to the smaller
width.

a deamplification. This alteration is closely related to the interaction of the


primary wave with the outward diffracted waves (Rogers et al., 1974; Bard,
1982).
The crest/base spectral ratios thus exhibit a first maximum for frequencies
corresponding to wavelengths slightly longer than the mountain width (Figure 3).
The amplitude of this maximum is slightly increasing with increasing shape ratios,
but remains largely below 30 for realistic shape ratios (h/l < 0.7; Figure 2).
Other conclusions of these theoretical studies are the following
1. The amplification is lower for incident P waves than for incident S waves.
2. It is also slightly larger for inplane horizontal motion (P-SV case) than for
antiplane motion (SH case).
3. The amplitude and phase are rapidly varying along mountain slopes, giving
rise to differential motions.

EFFECT OF T O P O G R A P H Y ON EARTHQUAKE G R O U N D MOTION

45

These theoretical results are, both qualitatively and quantitatively, consistent


with those obtained by Rogers et al. (1974) through a model study experiment.
There are other theoretical results corresponding to more complex models. Bard
and Tucker (1985) investigated the S H response of a set of ridges with irregular
subsurface layering; Zhenpeng et al. (1980) and S~nchez-Sesma (1983) investigated
the response of three-dimensional axisymmetric homogeneous ridges. Although all
of these models do lead to higher crest amplifications, only those of Zhengpeng et
al. (1980) are directly comparable to the crest/base amplification ratios shown in
Figures 1 to 3. The dimensionless peak frequency is between 0.6 and 0.8, and the
crest/base spectral amplification ranges between 2 and 7, which is far larger than
for two-dimensional models.
M a i n experimental results. In this section, we briefly review the main experimental results obtained in the last 15 yr. As most of them disagree with theoretical
results, in each case, we also look for what may be responsible for this disagreement.
Davis and West (1973) recorded several aftershocks of the San Fernando 1971
earthquake at the base and crest of Kagel Mountain and Josephine Peak, California.
They also compared the seismic signals caused by underground cavity collapse and
recorded at the base and crest of Butler Mountain, Nevada. They computed, in each
case, the crest/base ratios of pseudo-relative velocity response spectrum (with 5 per
cent damping) and observed a large, frequency-dependent amplification. In general,
the wavelengths corresponding to these amplification frequencies are comparable
to the mountain width (see Figure 3), which is in good agreement with theoretical
results. Nevertheless, the amount of amplification, either in the time domain (Figure
1) or in the frequency domain (Figure 2), is much larger than the values predicted
by theoretical models, as the spectral amplification reaches 30 in some cases.
As the surface geology was apparently, in each case, the same at both crest and
base (granite in Kagel Mountain and Josephine Peak, dacite for Butler Mountain),
these qualitative disagreements may be due either to the pronounced three-dimensional character of the investigated surface topographies or to the weathering effects
at the mountain top, as emphasized by Rogers et al. (1974).
Rogers et al. (1974) recorded an underground nuclear explosion at several sites
across the NASA Mountain, Nevada. They performed only time-domain comparisons of crest and base motions, and the observed amplifications are about 25 per
cent, which is in good agreement with theoretical predictions. As in the theoretical
and laboratory models, the NASA Mountain has a rather smooth and regular twodimensional shape, and the incoming wave front was parallel to the ridge axis.
Griffiths and Bollinger (1979) conducted a field program in the Appalachian
Mountains, where they recorded 137 events at the base and crest of six topographic
features. Only one event was natural; the other ones were quarry and mine blasts.
The instruments were smoked-paper recorders, and crest/base comparisons could
be made only in the time domain. Once again, the seismic motion was observed to
be systematically larger at the top than at the base. For a given crest/base pair of
stations, the crest/base amplification ratios obtained for different events exhibit a
much larger scatter (see Figure 1), but the average values are ranging between 1.7
and 3.4, which is significantly larger than the theoretical predictions.
Yet, in this experiment, the topographic features are nearly two-dimensional
(Appalachian Mountains), and many of the seismic waves came from an azimuth
nearly perpendicular to the ridge axis. Moreover, the horizontal seismometers were
aligned parallel to the ridge crest, so that the observations could be directly compared

46

LOUIS GELI, PIERRE-YVES BARD, AND BI~ATRICEJULLIEN

with S H calculations. As a consequence, the disagreement may come from the


following possible discrepancies
1. The actual geological structure was different from a simple homogeneous halfspace: some of the recording sites (in valleys and at mountain crests) were
covered by a thin, low-velocity alluvium or colluvium layer.
2. The investigated ridges are not isolated, but they are one in a series of
subparallel ridges, which therefore present a "quasi-periodic" structure (see
Figures 2 and 3 of this paper).
Finally, Tucker et al. (1984) measured the spectral differences of the seismic
records obtained at two-dimensional hard rock sites. For a small ridge, they observed
that the spectral ratios were nearly the same, independent of the azimuth, distance,
and size of the seismic events, and that the spectral amplification with respect to a
nearby tunnel (not with respect to the ridge base) reached ratios as high as 8.
A discussion of the discrepancy between these observations and simple theoretical
models may be found in Bard and Tucker (1985). As this discrepancy occurs for
any earthquake source, and since Tucker et al. (1984) also report some other
observations of much smaller ridge amplifications, the amplification may be due to
the internal shallow structure of that small ridge and to the presence of nearby
similar ridges. Bard and Tucker conclude that models including both complex
subsurface layering and complex surface topography could probably predict the
observed amplification.
There also exist two recent experimental studies which confirm the observations
of large topographical amplifications. Umeda et al. (1986) recorded some aftershocks
of the "Western Nagana Prefecture Earthquake" of 1984 on a small (100 m high,
500 m wide) topographic feature where thrown-off boulders gave evidence of
anomalously high acceleration during the main shock. They observed crest/base
spectral amplifications larger than 10, although the ridge is nearly two-dimensional.
But here too, the underground structure is not homogeneous. Meneroud (personal
communication, 1986) recorded some weak local or regional earthquakes at the base
and top of a hill with very steep slopes in southeastern France. He also observed
spectral amplifications up to 30 for some events. This case is under study by the
authors.
Conclusions. The conclusions of this review section are, therefore, the following
1. There is a qualitative agreement between theory and observations about the
existence of seismic motion amplification at mountain tops; this amplification
is larger on horizontal components (~S motion) than on the vertical component (~P motion).
2. There is a rather good quantitative agreement as to the approximate frequency
of this amplification, which corresponds to wavelengths comparable with
mountain widths.
3. There is a quantitative disagreement between theory and observation as to the
amount of amplification. Although some cases exist where field studies exhibit
only very weak amplifications at mountain tops (Rogers et al., 1984; Tucker et
al., 1984), the observed amplifications are often much larger than the theoretical predictions with two-dimensional homogeneous models.
The detailed examination of each of these field studies reveal that, in two cases
(Davis and West, 1973), this disagreement may be due to the marked threedimensionality of the topographic features. In at least four other cases (reported by
Griffiths and Bollinger, 1979; Tucker et al., 1984), it is probably related to other
causes, such as irregular subsurface layering and/or neighboring topography. More-

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION

47

over, the only three-dimensional theoretical spectral study (Zhenpeng et al., 1980)
provides amplification ratios below 7, whereas Davis and West observed values
above 30.
The aim of the next section is, therefore, to investigate with the help of numerical
modeling whether more complex two-dimensional topographic features, such as
irregular subsurface layering and neighboring ridges, may give rise to larger crest/
base amplifications, and consequently whether it is possible, by this approach, to
shorten the discrepancy between theory and observation.
COMBINED EFFECTS OF NEAR-SURFACE LAYERING AND SURFACE
TOPOGRAPHY

Including subsurface layering and neighboring ridges makes the models much
more complex. There are so many parameters that it is impossible to define generic
configurations into which every topographical structure in nature may be classified.
We therefore consider here only two examples: the first one, depicted in Figure 4,
allows us to study the effect of various kinds of subsurface layerings for an isolated
ridge. The second one, depicted in Figure 11, allows the study of the combined
effects of subsurface layering and neighboring ridges.
The response of these complex geological structures to an incoming plane S H
wave is computed with an extension of the Aki-Larner technique, as explained in
the Appendix. For sake of simplicity, only vertically incident S H waves are considered here. Slightly larger amplifications could perhaps be found for S V waves (Bard,
1982; Castellani et al., 1982), but these differences are not quantitatively significant
for our purpose.
Effect of subsurface layering for an isolated ridge. The models investigated in this
section are depicted in Figure 4. The surface topography has a sine-wave shape, and
its shape ratio (defined as the ratio of height, h, to half-width, l) is set equal to 0.4.
The three different configurations, respectively noted as a, b, and c, were selected
according to the following features
Case a: Ridges may uniformly be covered with thin layers of alluviums and
also there may exist a near-surface weathering. The effect of near-surface, lowvelocity layers is to be investigated.
Cases b and c: Elevated topographies are often associated with a downward
increase of the sediment cover thickness on the flanks (configurations b and c), in
relation to natural erosion.

We considered four lithological units, numbered 1 to 4, each with a given density,


velocity, and quality factor ($1, /~, and Q1), respectively, and separated by three
nonplanar interfaces (see Figure 4). For each configuration, the surface layer was
given two different velocity values: 0.6~3 and 0.4/~3. We computed the half-amplitude
for the six configurations and for frequencies ranging between 0 and 2[33/I (e.g., if l
= 200 m and/~3 = 1000 m/sec, the frequencies range between 0 and 10 Hz). The
results are discussed in terms of dimensionless parameters, with the dimensionless
frequency being = 21[/f33 and the dimensionless time being r = (~3/21)t, where f and
t are frequency and time.
Before discussing the results for these rather complex models, we briefly recall
the effects associated with a single ridge over a homogeneous half-space (Figure 5)
and with the simple horizontal layering present at the base of each of the hills we

48

LOUIS GELI, PIERRE-YVES BARD, AND B]~ATRICE JULLIEN


1

~,q

~ __ 2

Configuration a

h
h/5
2h/5
layer

h/51~
T

2h/52~:

Configuration b

~~
/

t/
/

~
layer
- 4

~
~
~

laver 1
~
layer z

"

Configuration c

h/5
2h/5

D (9/cm 3 )

layer I

7O

2.7

layer 2

70

2.8

0.8 6~

layer 3

80

2.9

B~

layer 4

10(9

3.1

63 x 1.2

63x0.6

; (~30.4

FIG. 4. Cross-sections of the three configurations (a, b, a n d c) selected for investigating the effect of
ridge subsurface layering.

considered (Figure 6). As found in previous studies, the effect of the simple ridge is
a broadband amplification of 1.6 at mountain top, for dimensionless frequencies
lower than 2.5. On the bases, the ground motion is alternatively amplified and
deamplified, but never more than 30 per cent. On the flanks, the same pattern is
observed, but with a higher amplitude (up to 50 per cent at points numbered 3 and
4) and with variable spacing in frequency.
The resonance effect associated with the horizontal layering, as depicted in Figure
6, leads to broadband amplifications of about 70 per cent for #1 = 0.6#~ and 150 per
cent for #l = 0.4#3. The dimensionless frequency corresponding to the peak is about
1 in the former case and about 2.5 in the latter case.
As mentioned in the review section, topographic effects are often presented in

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION


=2.

~20

49

.3

.3
ro
u

3.
~1.

I'

3'

1'

4~

(n)

=2.

=2,

2'

4'

(~)

3.
~I.

(n)

=2.
.3

=2,
o

"3

(n)

3.
E

~I.

(n)

dimensionless

frequency

(q)

Top

1
I

o
h/1 ffi 0.4

FIG. 5. S H Fourier transfer functions obtained at six surface sites for an isolated, homogeneous ridge.
The shape ratio (0.4) is the same as in Figure 4, and the definition of the dimensionless frequency is the
same as in Figure 3.

experimental studies as crest to base spectral ratios. Figure 7 thus presents the
Fourier transfer functions obtained at crest (site 1 on Figure 4) and base sites (site
6) for each of the six structures of Figure 4, together with the corresponding crest/
base ratios.
These results show very clearly that these crest/base spectral ratios are much
less sensitive to the actual underground geological structure than the individual
crest and base Fourier transfer functions, at least in the low-frequency domain (<1)
(compare, for instance, the configurations al, a2, bl, and b2: the maximum
amplitudes of the Fourier transfer functions at the crest range between 1.9 and 3.4,
while crest/base amplification ratios range between 2.1 and 2.7 only).

50

LOUIS GELI, PIERRE-YVES BARD, AND BI~ATRICE JULLIEN


IO AMPLIFICATION

STRATIFICATION

TRANSFER FUNCTION

Q2=50 pi=2.7

Bl

h/S

Q2:70 P2:2-8

B2=0.8 63

2h/5

.~

Q3:80 O3:2.9

B
3

2h/5

~Q. 2

O=lO0ON:3.1
p is in glcm3

BI:O.4Ba

BI: 0.663

B=B3xI.2

01 ''I'''~
I ' ' I ' ' ' I~ [
dimensionless frequency (n)

I n = f x 21 and h/1 = 0.4 1


63

FIG. 6. Fourier transform functions at the surface of the horizontally layered structure present at the
base of the ridge for vertically incident S waves.

The most remarkable results on the crest/base ratios are


1. A systematic and stable low-frequency amplification for dimensionless frequencies below 1 (the maximum being about 0.8).
2. A rather constant, low-frequency crest amplification, ranging between 1.9 (c2)
and 2.7 (al), regardless of the geological structure and surface velocity values.
3. The higher frequency narrowband amplification peaks are not due to crest
amplification, but to base deamplification. Nevertheless, there exists a general
trend to high-frequency deamplification, mainly due to resonance and subsequent amplification in low-frequency layers at the mountain base.
4. The comparisons of configurations b and c show that the deep underground
structure, which in general does not correspond to very strong discontinuities,
does not play a significant role for this class of models.
Figure 8 shows some synthetic seismograms at different surface points of the
three configurations, when the incident signal is a Ricker wavelet of characteristic
dimensionless frequency equal to 1. We may see in this case that the base motion
is almost the same, regardless of the geological structure of the hill, while the top
motion is more sensitive.
Figure 8 also reveals that the amplification is less important at mid-slope (points
3 and 4) than at the base (at least for the input motion chosen here). Moreover,
although the peak amplitude is in each case comparable at both sites 3 and 4, the
signal shape changes a lot because of the increase of diffracted wave amplitude from
point 3 to point 4. It is therefore of interest to investigate the differential motions
along the slope. This is illustrated in Figure 9, which displays the spectral ratios
between two sites close to one another along the slopes. The spectral ratios exhibit
very narrowband peaks or troughs, the frequency of which depends primarily on
the site position along the slope and not so much on the surface velocity value.

E F F E C T OF T O P O G R A P H Y
~op

ON EARTHQUAKE GROUND MOTION


bo~o

2l

t'I

3
:l

Configuration

toe/bo~e

3;

51

-'

,,-

~"

. . . . .

2~
!

t
0

t
i
'

'

O( . . . . . . . .

r ....

,,

(n)

EOD

bo~o

.l*,

. ". ' ,
i
,
i

'
:..

s
i

%
(n)

Configuration

toD/bo~e

"',

.... i .... i .... ,


(~)

~op

(n)

top/ho~e

ba~e

Configuration

.
~

F"

:; ~,:;

I
,

,'

","

3
dimensionless

(n)

(n)

frequency

(~)

FIG. 7. SH Fourier t r a n s f e r f u n c t i o n s at t h e top (left column) a n d base (middle column) of configuration a (top row), b {middle row), a n d c (bottom row). In each plot, t h e t h i c k "full" line corresponds to
a surface layer velocity of 0.6/~3 a n d t h e dotted line to a surface layer velocity of 0.4/~3 as in Figure 6.
T h e right column corresponds to t h e crest/base spectral ratios.

Thus, the conclusions drawn from this example are that the crest to base spectral
amplifications remain limited, even in the case of low-velocity layers at the mountain
top, and that the differential motion is maximum in the upper half of the flanks.
Effect of neighboring topographicfeatures. The aim of this section is to investigate
the combined influence of neighboring topography and subsurface layering. We
consider here a series of three similar hills, as illustrated in Figure 12, with a
subsurface layering similar to that of configuration a in the previous section. The
surface layer here is very thin, with low-velocity and very strong damping, in an

52

LOUIS GELI, PIERRE-YVES BARD, AND BEATRICE J U L L I E N

Configuration b2

Configuration a2

Configuration c2

.o~

I.O0

,.

-o-

=o~

4.40

l.

mow

3.22

.o.

2,04

.o~

1,85

.ox

3.11

Inci dent

i.

top (Ptl)

.o.

2.12

.a~

2.03

.o=

3.22

Pt3

~o.

2.68

t.

Pt4

.il

Base (Pt6)

o
Time ( l / n )

time ( l / n )

Time ( l / n )

FIG. 8. Time-domain response at surface sites 1, 3, 4, and 6 for configurations a (/eft), b (m/dd/e),
and c (right), with a low-velocity surface layer (B1 = 0.4 #a). The input signal is a Ricker wavelet
(displayed at the top) with a characteristic dimensionless frequency 7, = 1.

attempt to fit the actual velocity structure of the small ridge investigated by Tucker
et al. (1984).
The results are again presented in dimensionless form. The dimensionless frequency is ~ = 2lf/#m, # ~ being the mean "hill velocity": #~ = #i/6 + # J 3 + #3/2.
The dimensionless time is: r = #~/21t.
Before we discuss the spectra and spectral ratios obtained for that complex
structure, we try to determine the relative importance of the different effects due
to: (1) topography; (2) neighboring topography; and (3) subsurface layering.
The one-dimensional resonance effect due to the layering is depicted in Figure
10. It exhibits a maximum amplification of about 3.5 for a dimensionless frequency
around 1.
The effects due to surface topography alone, without any complication from
surface layering, are shown in Figure 11, which displays the response of: (1) an
isolated ridge and (2) three contiguous hills.
The effect of the simple isolated ridge is somewhat larger, but qualitatively similar
to the one shown in Figure 5. The presence of neighboring ridges induces an increase
both in top amplification and base deamplification, for dimensionless frequencies
slightly shorter than 1. This increase is larger at the "central" ridge than on the
lateral ridges. As a consequence, the crest/base spectral ratios (displayed on the left
part of Figure 13) exhibit a sharp peak for dimensionless frequencies around 0.7,
and with an amplitude slightly larger than 6 for the central ridge.
Finally, the response of the complete model is displayed in Figure 12, together
with the response of an isolated ridge with identical subsurface layering (for

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION


Pt l/Pt

Pt 2/Pt 3

Pt 3/Pt 4
s

....

,''"r'-

53
Pt 5/Pt 6

Pt 4/Pt 5
l,';'"

a
2

:3

s
i

c.

....

,''"l'"'l

....

c_
o

-Q

Dimensionless frequency (n)

FIG. 9. Differential motions along the flanks of the ridge, for configurations a (top), b (middle), and
c (bottom), and two values of surface layer velocity (full line: #1 = 0.6 #3; dotted line: fll = 0.4 #3). The
five columns, from left to right, correspond to spectral ratios between sites 1 and 2, 2 and 3, 3 and 4, 4
and 5, and 5 and 6, respectively.

TRA~;SFER FUNCTION

STRATIFICATION

0
h/6

QFIO P

0.2 S~

%=40 p

G.6 B3

~=60 p

B3

h/2

8
~2
;z

Q~lOO

1.5 B3

1
p = 2.8 g/cm3

dimensionless frequency (n)

n- f'~m

and

-[-- 0.6

FIG. 10. Fourier transform functions at the surface of the horizontally layered structure present at
the base of the ridge depicted in Figure 12 for vertically incident S waves. This is the analog of Figure 6,
but for the geological structure displayed in Figure 12.

54

LOUIS GELI, PIERRE-YVES BARD, AND BI~ATRICE JULLIEN

3'

6o

5.

5.

2.

1.

1.
r

O.

0.5

I.

1.5

2.

0.

6,

e.

5.

5.

~4.

4.

0.5

I.

1.5

2.

1.5

2.

(n)

3.
2.

~
I.

0.

e.

0.5

I.

1.5

2.

O.

(n)

0.5

6.

1.

(~)

5.

5,
4,

4.

3.

/Z

O.

0.5

1.

dimensionless

1.5
frequency

123'~S

:23~5S

/~X..

2.
(n)

4-

3.

0.

0.5

I.

1.5

2.

(n)

12J45G

654321

FIG. 11. Effect of neighboring topographies in the case of an homogeneous half-space. The six
transfer curves are the S H Fourier transfer functions at points 1 to 6, with the line symbol corresponding
to the topography depicted at the bottom of the figure. Full thick line = homogeneous multiple ridges;
dotted lines -- homogeneous multiple ridge with: thin dotted lines -- central ridge; thick dotted lines =
lateral ridge, inner slope; and thin dotted (with small dots) lines = lateral ridge, outward slope.

comparison purposes). The amplification values are much higher, at any site, than
in the homogeneous case (Figure 11) and in the one-dimensional case (Figure 10).
This result is a clear indication of a strong interference between subsurface layering
and surface topography.
The largest effects are obtained for the central ridge. A significant but narrowband
increase of amplification occurs at ridge top, while a significant and narrowband
deamplification occurs at the near base, exactly as in the homogeneous case (Figure
11). This results in a crest/base spectral ratio up to 10 for a dimensionless frequency,

EFFECT OF T O P O G R A P H Y ON E A R T H Q U A K E G R O U N D

12.

, :::,'k

3.

O.

t~.
g

0.5

12.

~-

6.

I.

1.5

55

3.
O.

2.

15.

(~)

MOTION

12.

0.5

1.

1.5

2.

(~)

12.
3

9.

,.&
e.

s.

6.

,~

O.

0.5

I.

1.5

2.

15.

O.

(~)

15,

0.5

1.

1.5

2.

1.5

2.

(~)

g
g

t2,

12,
.~

~o

{;.
3,

0.

0,5

dimensionless

I.

1.5
frequency

2.

0,

0.5

I.

(r~)

;23,~6

Z23456

::F:~!!!;:!!!!!!Z~!!i!!,
i!!!!!!!!!!~!!!!i~::.~::~

111:111::/

'" Z::"

"::::/

"-::::::::i

Fro. 12. Effect of neighboring topographies in the case of a layered structure. T h e explanation for
the line symbols are the same as in Figure 11.

slightly smaller than 1. But in the present case, the peak is mainly due to the base
deamplification and not so much to the top amplification. It may probably happen,
in some cases, that the frequency of maximum base deamplification coincides with
the frequency of maximum top amplification. The resulting spectral ratios could
then reach much higher values, up to 20 or 30.
On the other hand, the effects on the outer ridges are very similar to the effects
over an isolated ridge, mainly because there is not any deamplification at the base.
The rapid changes in Fourier transfer functions near the mountain top (compare
points 1, 2, and 3 in Figure 12) strongly suggest the existence of significant

56

LOUIS

GELI,

PIERRE-YVES

BARD,

10.

AND

BI~,ATRICE J U L L I E N

,\

9.

O.
7.
i
1
i

6.
5,

iAi

4.
3.
2.
1,

0.

0,5

1.

1.5

2.

DIMENSIONLESS FREQUENCY (n)

I0o

9.

It

8.

11

7.
6.
H
ta_

ii

i-

li

17

5.

! ,~,
,' i "3,

4.

,'/
,v

3.
2.
I.

0.

5.5

1.

[ .5

2.

DIMENSIONLESS F~EQUENCY (q)


FIG. 13. Effect of underground layering and neighboring topography on crest/base spectral ratios. (a)
Homogeneous half-space. (b) Layered half-space. Thick full lines = isolated ridge; thick dotted lines =
multiple ridge, ratio of top of central ridge to lateral base amplification; and thin dotted lines = multiple
ridge, ratio of top of central ridge to the next base amplification.

differential motion. This appears very clearly in the time domain (Figure 14), where
it is shown that the peak and average motions are 4 to 5 times larger at the ridge
top than at some distance apart (points 4 and 8). Nevertheless, the crest/base peak
motion ratio is significantly smaller (around 2), which is well below the values
observed by Griffiths and Bollinger (1979). We think that this persisting disagreement is at least partly due to the simplicity of the incident signal that we considered
in this study.
In conclusion, for this second example, it seems that a very simple-shaped
topography gives way to a very complex amplification pattern when many phenomena are present. Here, for instance, the lateral interferences due to the periodicity
of the structure induce an amplification pattern which is impossible to predict by a
simple combination of the three simple effects that occur in this example: topography, layering; and periodicity.

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION

~.TS

~,t6

4,1S

3.98

~.B2

2.65

2.1t

4,34

4,27

3.7S

3,3~

2.2O

3.SS

3,9t

~,2S

3.33

4.5~

4,2~

rip-

incident ~
slgnal O.

4,

2.S~

57

2.~3

4,5~

I.

4.
I/n

FIG. 14. Effect of neighboring ridge (in the presence of underground layering) on the time-domain
response at 11 surface sites (displayed at the bottom of the figure). (Left) isolated ridge; (middle) multiple
ridges, central ridge; (right) multiple ridges, lateral ridge. The input signal is a Ricker wavelet with
dimensionless characteristic frequency = 1.0.

CONCLUSIONS

This study leads to the same qualitative conclusions as the previous studies. Two
general trends seem to be quite reliable regarding the topographic effect, regardless
of the internal geological structure and its complexity
1. There exists a significant amplification at hill tops with respect to the base,
for frequencies corresponding to wavelengths about equal to the mountain
width.
2. The hill sides undergo complex amplification-deamplification patterns and
also significant differential motions, especially in the upper parts of the hill.
Quantitatively, this study shows that the topographic complexity (presence of
neighboring ridges) may be responsible for large crest/base amplifications, whereas
complex subsurface layering may not (unless there are low-velocity layers at the
mountain top and not at the mountain base, which is not realistic). This is illustrated
in Figure 15, where the theoretical results are shown for two-dimensional homogeneous isolated ridges, two-dimensional layered and isolated ridges, two-dimensional
homogeneous multiple ridges, two-dimensional layered, multiple ridges, and threedimensional homogeneous ridges. Furthermore, it should be pointed out that we
have only considered SH waves, and the amplifications should be slightly larger for
incoming S V waves (Castellani, 1982; Bard, 1982).

58

LOUIS GELI, PIERRE-YVES BARD, AND BI~.ATRICE JULLIEN

OBS.

06S.

30

Th,

2Dh~

TN,

3DR{

+ T~,

2DL{

x TN,

2Dhm

2Dim

Th,

X
X

M
o
M

!
o.

oA

o.2

"0.3

Shape r a d o

0.4

o.~ o.,

o., >

(h/I)

Fro. 15. Maximum crest/base amplifications for transverse motion, including previous literature
results and present results. Open circles indicate observations. Double open circles indicate that the
investigated topographies are three-dimensional. Other symbols are for theoretical results (SH case),
with the following explanations: 2D and 3D = two-dimensional and three-dimensional models; h =
homogeneous structure; l = layered structure; i = isolated topography; and m = multiple topography.
The results discussed in this study correspond to h/l = 0.4 (2D/i) and h/l = 0.6 (2Dhi, 2D/m).

ACKNOWLEDGMENTS
This work was supported by the Bureau de Recherches G6ologique et Mini~res. The computations
were performed at the Centre de Calcul Vectoriel pour la Recherche.
REFERENCES
Aki, K. and K. L. Larner (1970). Surface motion in a layered medium having an irregular interface due
to incident plane SH waves, J. Geophys. Res. 75, 933-954.
Aki, K. and P. Richards (1980). Quantitative Seismology: Theory and Methods, vol. 1, W. H. Freeman,
San Francisco, California, 557 pp.
Angot, A. (1909). Le tremblement de terre de Provence i l l juin 1909), Annales du Bureau Central
M~tJrologique de France, M~moires, 37-93 (in French).
Bard, P.-Y. (1982). Diffracted waves and displacement field over two-dimensional elevated topographies,
Geophys. J. R. Astr. Soc. 71,111-120.
Bard, P.-Y. and B. E. Tucker (1985). Ridge and tunnel effects: comparing observations with theory, Bull.
Seism. Soc. Am. 75, 905-922.
Bard, P.-Y. and J. C. Gariel (1986). The seismic response of two-dimensional sedimentary deposits with
large velocity gradients, Bull. Seism. Soc. Am. 76, 343-366.
Berardi, Capozza, and Zonetti (1978). Analysis of rock motion accelerograms recorded at surface and
underground during the 1976 Friuli seismic period, Proceedings of the 1976 Friuli Earthquake on
the antiseismic design of nuclear installation conference, Rome, Italy, October 1977.
Boore, D. M. (1972). Note on the effect of topography on seismic SH waves, Bull. Seism. Soc. Am. 62,
275-284.
Bouchon, M. (1973). Effect of topography on surface motion, Bull. Seism. Soc. Am. 63,615-632.
Brambati, A., E. Faccioli, E. B. Carulli, F. Culchi, R. Onofri, S. Stefan-;ni, and F. Ulcigrai (1980). Studio
de microzonizzazione sismica dell'area di Tarcento (Friuli), Edito da Regiona Autonoma FriuliVenezia Giulia.
Castellani, A., A. Peano, and L. Sardella (1982). On analytical numerical techniques for seismic analysis
of topographic irregularities, Proceedings of the 7th European Conference on Earthquake Engineering, Athens, Greece, September 20-25, 1982.
~elebi, M. and T. Hanks (1986). Unique site response conditions of two major earthquakes of 1985:
Chile and Mexico, Proceedings of the International Symposium of Engineering Geology Problems
in Seismic Areas, vol. IV, Bari, Italy, April 1986.

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION

59

Davis, L. L. and L. R. West (1973). Observed effects of topography on ground motion, Bull. Seism. Soc.
Am. 63, 283-298.
Geli, L. {1985). Propagation d'ondes sismiques dans les formations superficielles, Tbhse de l'Univ. Scient.
Med. Grenoble, Grenoble, France (in French).
Griffiths, D. W. and G. A. Bollinger (1979). The effect of Appalachian Mountain topography on seismic
waves, Bull. Seism. Soc. Am. 69, 1081-1105.
Ilan, A., J. L. Bond, and M. Spivack (1979). Interaction of a compressional impulse with a slot normal
to the surface of an elastic half space, Geophys. J. R. Astr. Soc. 57, 463-477.
Rogers, A. M., L. J. Katz, and T. J. Benett (1974). Topographic effect on ground motion for incident P
waves: a model study, Bull. Seism. Soc. Am. 64, 437-456.
SAnchez-Sesma, F. (1983). Diffraction of elastic waves by three-dimensional surface irregularities, Bull.
Seism. Soc. Am. 73, 1621-1636.
S/mchez-Sesma, F., I. Herrera, and J. Aviles (1982). A boundary method for elastic wave diffraction:
application to scattering of S H waves by surface irregularities, Bull. Seism. Soc. Am. 72, 473-490.
Sills, L. (1978). Scattering of horizontally polarised shear waves by surface irregularities, Geophys. J. R.
Astr. Soc. 54, 319-348.
Siro, L. (1982). Southern Italy November 23 1980 Earthquake, Proceedings of the 7th European
Conference on Earthquake Engineering, Athens, Greece, September 20-25, 1982.
Smith, W. D. (1975). The application of finite element analysis to body wave propagation problems,
Geophys. J. 42,747-768.
Tucker, B. E., J. L. King, D. Hatzfeld, and I. L. Nersesov (1984). Observations of hard rock site effects,
Bull. Seism. Soc. Am. 74, 121-136.
Umeda, Y., A. Kuroiso, K. Ito, Y. Ito, and T. Saeki (1986). High accelerations in the epicentral area of
the Western Nagana Prefecture, Japan, Earthquake of 1984, J. Seism. Soc. Japan 39, 217-228.
Virieux, J. (1984). S H wave propagation in heterogeneous media: velocity-stress finite difference method,
Geophysics 49, 1933-1957.
Zahradnik, J. and L. Urban (1984). Effect of a simple mountain range on underground seismic motion,
Geophys. J. R. Astr. Soc. 79, 167-183.
Zhenpeng, L. Y., L. Baipo, and Y. Yifan (1980). Effect of three-dimensional topography on earthquake
ground motion, Proceedings of the 7th World Conference on Earthquake Engineering, vol. 2,
Istanbul, Turkey, September 8-13, 1980.
LCPC
58 BOULEVARD LEFEBVRE
75015 PARIS CEDEX, FRANCE (P.-Y.B.)

LGIT-IRIGM
BP68
3402 ST. MARTIN D'HERES
CEDEX, FRANCE (L.G., P.-Y.B.)

BRGM/ARGES
DOMAINE DE LUMINY
ROUTE LEON LACHAMP
13009 MARSEILLE, FRANCE (L.G., B.J.)
Manuscript received 1 December 1986
APPENDIX

We assume that the principles of the Aki-Larner (1970) method are known. This
section is devoted to an extension of the method to the case of many irregular
interfaces. Additional details about the numerical procedure are given by Geli

(1985).
The two-dimensional structure is shown in Figure A1, in the plane x - z. In the
following, "medium (j, m)" will be referred to as the jth horizontal layer, inside the
mth geological unit [1 < m < M and 1 < j < J(m)]. Let fljm, Qjrn, and ~)m be,
respectively, the shear wave velocity, quality factor, and density of the medium
(j, m). Only the case of homogeneous media is treated here; when a velocity gradient
is present, the method can be extended (see Bard and Gariel, 1986). Let z = r ( x ) be
the equation of the mth interface.

60

LOUIS GELI, PIERRE-YVES BARD, AND BI~ATRICE JULLIEN

!'"U._'

(21
,)_ ~

_ _ _ ~

Z = r

:1,3)

l(x)

/~_

~-~-=

. _ _

r2(x)

z;
Z2j . . . . . . . . . . . . . . . . . .

" z = r3(x }

FIG. A1. An example of a typical cross-section that can be modeled by our computing code. On the

right, the velocity-depth profile is displayed with constant velocity gradients in the horizontal layers.

The structure is impinged by an incident plane S H wave, of displacement Uo,


such that
Uo(X, z, t)

ei(hx-~Z)e i'~t

(1)

where
k O 2 3L /)0 2 =

6O
--

~o2

and

Im(uo) <

where w is the angular frequency, ~o is the shear wave velocity of the underlying
half-space, and ko and go2 are the horizontal and vertical wavenumbers, respectively.
Because of the presence of irregular interfaces, the incident wave field is diffracted
into upgoing and downgoing waves, characterized by their horizontal wavenumber,
k. The displacement field, at point M ( x ) of medium (j, m), is assumed to be a sum
of plane waves
u(x, z, t) = e i~t

f:

[Ajm(k)e i"/"~ + B j m ( k ) e -i"p~] dk

(2)

where
vj

m2

k2

60 2

and

Im(vj m ) < O .

The originality of the Aki-Larner method is that we compute the wave field
diffracted by an infinity of identical structures, of width L, periodically disposed
along the x axis. The original wave field is recovered by introducing a complex
frequency in order to attenuate the arrivals of the waves diffracted by the adjacent
structures. Then, (2) becomes
u(x, z, t) = e i~t ~
.

[Aj,n(k.)eiV.(h.) z + Bj,n(k~)e_i~/.(k.)~]

2~r

(3)

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION

61

kn = ko + n(2rL).

(4)

where

The infinite sum (3) is truncated to order N, and approximated by [the harmonic
time dependence exp(iwt) has been omitted]
2~ n:+N
U(X, Z, 60) = --'~
Z [ajm(k,~)ei"i "(a")~ "4"sjm(kn)e-i",m(kn)z].
n=-N

(5)

Inside the mth geological unit, the unknowns Aj m(k,) and Bj a(k,) are expressed by
means of the propagator matrix of Thomson-Haskell, noted [T] (see Aki and
Richards, 1980, pp. 273-280)

( Ajm(kn)~

Bjm(kn)] =

[Tjm(kn)]

{ etm(kn)~

\Btm(kn) ]

(6)

where

[Tjm(kn)]

--

Thomson-Haskell Matrix.

(7)

There are (2M - 1)(2N + 1) unknowns that are the Al~(k,~) and B ~ ( k , ) . Substituting (6) into (5), we get
2~r n=+N
U(X, z, w) = --~ Z [Atm(k,)yjm(z, kn) + Btm(kn)Zjm(z, kn)]e -iknx

rt=--N

(8)

where

Yjm(z, kn)~
zjm(z, kn)]

[rpml., ,lt[ eivjm(k)z~


\e -i'jm(k.,z]

= taJ r~nll

(9)

where [T] t is shorthand for the transposed matrix of [T]. Now, let M ( x ) be a point
of the mth irregular interface (m > 1) and b the normal vector oriented "outwards."
Point M belongs to medium (j (x), m), medium (/(x), m + 1), and z = r(x). The
boundary conditions at the interface express the continuity of the displacementstress vector. They are, for any point M of the interface

Uj(x)(X,m r(x)) : Ut(x)rn+l.,IX, r(x))


(10)

rn a[ud~x)(x'
ra
r(x))]
~tltx)
r(x))]
m+l Ofum+l(x,
t z(x)
Ob
= #t<.~)
Ob
0
where ~ stands for the partial derivative relative to the normal vector b.
Following Aki and Larner (1970), we take the spatial Fourier transform of

62

LOUIS GELI, PIERRE-YVES BARD, AND BI~ATRICE JULLIEN

equations (10), and

Iol

r ( x ) ) e (2ip~/L~ clx =

tfo

"

ui(~)(x,'~

ut(~m+l'tx,r ( x ) ) e (2ip~x/L) d x

(11)

m O[Uilx ) (x, r(x))] e(2ip,x/n) d x #J(~)


On

m+x t t(x) Ix, r(x))] e(eip~/L) dx.


On

Yo #l(x)

All along the ruth interface, at the bottom of the (m - 1)th geological unit, the
displacement-stress vector, expressed in the spatial Fourier domain, is
[CBm]Atm(k_n),

. . . , Alm(kn), Btm(k_~), . . . , Blm(kn)) t

(12)

where [C] could be called the "Aki-Larner propagator matrix," given by the
(2N + 1) x (2N + 1) following relations
L

m ~
C Bv~

.,9

e(2i./L)(p-n)x Y ~ ) ( r ( x ) ,

kn) d x

for l _ - _ p _ - < 2 N + l
1-<n=2N+1

C'~pn = fO L e (2i~/L)(p-n')x

Zj~(~)(r(x), kn) d x

for p = 2 N + l
2N+l<n-<2(2N+l)
withn' =n-(2N+

1)

0
C~n = fO L e (2i~'/L)(p ' --n)x #j~(x)~n [ Y ~ x ) r ( x ) ' kn)] d x

for

2 N + l < p _ - < 2 ( 2 N + 1)
1-<n<-2N+l

with p ' = p - (2N + 1)

CmBpn

0
f o L e(2ir/L)(p'-n')x #j~(x) -~n [ZJ~(x)(r(x)' kn)] d x

for

2 N + l<p_-< 2(2N+ 1)
2 N + l < n _ - _ 2 ( 2 N + 1)

withn' = np'=p-

( 2 N + 1)
( 2 N + 1).

(13)

EFFECT OF TOPOGRAPHY ON EARTHQUAKE GROUND MOTION

63

The Aki-Larner propagator matrix at the top of the mth geological unit is denoted
[C]. Its expression is given by replacing in (13) the index m + I by m, and the index
j (x) by l(x). Then, if we substitute these matrix notations into equation (11), we
get
[CB ]

slm

-~ [ ~ T

]~slrn+l]

for

1 < m _-- M - 2.

(14)

Along the deepest interface (m = M), the incident terms should be added to the
right-hand side of equation (10), and equation (14) becomes

M MAL
r~
~ +M[ F ] \ ]
t ~B M-I'[AIM-I~
l\slM-1 ] = [CT ][B

(15)

where [F] is the incident vector

F(p) = -~

uotM(x)(X,rM(x))e (2ip~x/L)dx;

1 /'L

F(p) = -

1 < p < 2N + 1

0gM(x)

tz,M) ~

(16)

(x, rM(x))e (2iv',/L) dx;

2N + 1 < p <- 2(2N + 1)

Since there is no upward diffracted wave in the bedrock (Rayleigh Ansatz Error),
A and B are linearly dependent

ArM(k) = )klM(k) BlM(k)


where

)klM(k) is a

(17)

scalar.

Along the free surface, the stress are vanishing. Let [S] be the Aki-Larner
propagator at the top of the first geological unit (free surface), truncated to the
(2N + 1) lines corresponding to the stress. Then,
0

We have, therefore, M matrix equations that are (18), (15), the (M - 2) equations
and (14). The system is solved step by step, such as in the Thomson-Haskell
method. One equation of system (14) is solved, and the solution is "propagated" up
to the next interface, at the top of the next upper geological unit. With the matrix
formulation, the solution for B is directly expressed as the solution of the following
equation

{ )klMBlM'~

[ S ] x [ G ] X [ C T M] x \ B, M ] = - [ S ] x [ G ] x [ F ]

(19)

where
[G] -- [CB1] -1 X [CT2][CB2]-lx . . . x[CTm][cBm]-lx

. . . X[CTM-1][cBM-1] -1.

(20)

Anda mungkin juga menyukai