Sei sulla pagina 1di 13

Bulletin of the Seismological Society of America, Vol. 101, No. 1, pp. 222234, February 2011, doi: 10.

1785/0120090323

A Ground-Motion Transfer Function Matrix between Two Nearby


Rock and Soil Sites: A System Identification Problem
by Rafael Benites and T. Grant Caldwell

Abstract

Standard single-component spectral ratios used to estimate localized site


effects may be generalized by using a 3 3 matrix transfer function between the threecomponent motion observed at a reference (rock) site and that at a nearby soil site.
This complex matrix (G matrix) represents the complete three-dimensional siteresponse within the linear regime. No prior knowledge of the geology at the soil site
is required, and G may be computed directly from the spectral responses of earthquakes recorded at both sites. Earthquakes from different hypocentral locations
may be used but must be selected so that the body-wave incidence at both stations
is nearly vertical and must be time-windowed to avoid surface waves. G is calculated
using a stochastic inverse of the observed response spectra and corresponding noise
covariance for each response. This method of analysis depends on the availability of a
good quality reference (rock) site, as do standard spectral ratio techniques that compare rock and soil site records. We have applied this method to determine site effects to
data recorded by strong ground motion seismographs near Wellington, North Island,
New Zealand. Magnitudes of the diagonal elements of G are similar to standard
single-component spectral ratios, and large values of the off-diagonal matrix elements
are seen at some locations. At these locations the maximum response is better estimated from the principal values of the transfer function matrix. The orientation at
which the maximum amplification (up to 14 in one case) occurs varies widely according to location.

Introduction
Site effects, or the effects of local site conditions on
ground motion, are often estimated from the ratio of seismic
spectra recorded at a particular location, generally on soil,
with that recorded at a reference station, generally on rock.
The spectra contain both the effect of the earthquake source
and any scattering along its propagation path to the recording
site. An implicit condition of this approach is that the distance between the soil site and the reference station must
be much smaller than the hypocentral distance so that the
source and paths effects are essentially the same for both
sites and, hence, removed by the ratio.
Generally, the ratios are computed for the corresponding
components of motion at both sites (e.g., Borcherdt, 1970;
King and Tucker, 1984). Other authors have used horizontalon-soil-to-vertical-on-soil ratios (Lermo and Chvez-Garca,
1993) and horizontal-on-soil-to-vertical-on-rock ratios
(Lachet and Bard, 1994). Generally, the ratios are computed
by taking time windows of the soil and rock records, usually
centered around the maximum S-wave amplitude, computing
Fourier spectra, and dividing their absolute values. Another
approach using only data from a single station has been introduced by Nakamura (1989) that estimates ground-motion

amplification by computing horizontal-to-vertical spectral


ratios of ambient noise at the same location. In practice, it
is customary to smooth the spectra, as well as the resultant
spectral ratio, in order to avoid instabilities that result when
the values of the denominator are close to zero (Konno and
Ohmachi, 1998).
The validity of these approaches has been studied by
several authors, both experimentally and theoretically, in
terms of frequency, geometrical and elastic properties of the
site, and the type of source of seismic waves. Actual applications have been mostly to weak motion (i.e., for shear strains
not larger than 104 ), and extrapolation to strong ground
motion requires a thorough geotechnical understanding of
the soil stiffness (Beresnev and Wen, 1996). We refer the
reader to the detailed review by Coutel and Mora (1998),
who classify these techniques as sediment-to-bedrock spectral ratio (SBSR), horizontal-to-vertical spectral ratio (HVSR),
and horizontal-to-vertical noise spectral ratio (HVNSR)
and compare the resulting amplification factors. In their
study, they use synthetic seismograms produced by seismic
plane waves incident on one-dimensional (1D) and twodimensional (2D) media, as well as actual earthquake
222

A Ground-Motion Transfer Function Matrix between Two Nearby Rock and Soil Sites

Figure 1.

The sites of the strong-motion network in the


Wellington and Lower Hutt areas, depicted on the corresponding
topographic map. The blue color is elevated topography, and the
light brown and yellow zones correspond to alluvium and silty clay,
respectively. The red line depicts the trace of the Wellington fault.

records. Overall, their conclusions seem to favor the SBSR


and HVSR techniques over the HVNSR technique to reveal
resonances and the fundamental resonant period, within 10%
error. In general, all the single-component ratios examined
fail to reveal 2D effects such as focusing and basin edge wave
generation.
Three aspects of the SBSR and the HVSR techniques are
relevant to our approach. The first is the extent to which
localized near-surface effects at the reference site may bias
the amplification factors estimated from the spectral ratios. In

223

other words, how well does the motion recorded at a surface


site on a bedrock outcrop represent the incident wave at the
base of the soil site? Yu and Haines (2003) use four rock
reference stations in their study of ground motion at Parkway, a small basin close to Wellington, New Zealand, and
show that the spectral ratios vary with the choice of reference
station. These authors attribute the observed variability to the
highly weathered nature of the greywacke outcrop. Steidl
et al. (1996) provide a thorough study on how good a reference site can be; they use local and regional earthquakes
recorded on the surface and in boreholes at two hard-rock
sites close to the San Jacinto Mountains, California, to compare surface with downhole spectra. The assumption is that
the downhole record represent the incident wave. They find
that, in general, for earthquakes recorded on the surface and
at 300 m downhole of a rock site, the corresponding spectra
are almost the same for frequencies up to 5 Hz, but the surface spectrum increases in amplitude at higher frequencies,
peaking at about 12 Hz with values four times larger than
those downhole. They also attribute the amplification to a
near-surface weathering layer.
The second aspect is the presence of noise in the earthquake records, which can strongly influence the calculated
spectral ratio (Aki and Richards, 1980). This is particularly
important when the ratios are averaged over many earthquakes to give a single estimate of the site response. The
general practice is to select only records with high signalto-noise ratios and assume that the effect of noise is negligible. We believe, however, that the proper treatment of noise
in spectral ratio is to follow the analytical approach introduced by Pisarenko (1970), who used a maximum likelihood
estimator of the amplitude ratio and its corresponding phase,

Figure 2. Hypocenter distribution of earthquakes occurred in the period from 1999 to 2007, of magnitudes between 3 and 6, and of
depths larger than 20 km. The small rectangle covers the Wellington area shown in Figure 1, and the large rectangle depicts the epicentral
distribution area from which the earthquakes for this study were selected.

224

R. Benites and T. G. Caldwell

Table 1
Description of Soils underneath Each of the Strong-Motion Network Sites in
the WellingtonLower Hutt Area Used in This Study
Soil Type and Thickness (m)

Depth to Bedrock (m)

Station*

V S30 (m=s)

PGMS
LHUS
PHFS
EBPS
INSS
ARKS
LHES
LHRS
POTS
TFSS
WNKS
WEL
FKPS
MISS

10
10

515
10

60
35

20
15

200
100150

295

195
212

10

5
10
<1

40

5
50
2
10
12

10
12
0
230

145

20
9

40

280
10
0
150
50
2
30
61

541
453
130
632
268
628
266

*Locations of stations are shown in Fig. 1. INSS and POTS are reference sites, on rock.

Unit A, gravelly sand with shells, also silty clay (alluvium); B, blue silty clay (marine); C,
gravel; D, Wildford marine; E, basal gravel; F, greywacke bedrock.

The values of V S30 (i.e., the S-wave velocity at 30 m depth) are given for most sites, for
completeness. From John Louie, personal commun., 2007.
From Martha Savage, personal commun., 2007.

obtained from many independent earthquake recordings (Aki


and Richards, 1980).
The final aspect to be addressed is the coupling among
all components of motion of the soil and rock sites. Any
component of motion at the soil site must result from the
action of forces exerted by the three components of motion
of the incident wave at its base. The size of each contribution
is determined by a transfer function between components,
including cross coupling. Hence, assuming that the reference
rock site adequately represents the incident wave, the amplification factors estimated from single-component spectral
ratios will be less accurate because they do not take into
account the total magnitude and phase contributions from
each input component of motion.

The three-dimensional (3D) character of the site response


has been addressed by Tumarkin (1998) and by Paolucci
(1999). Each of these authors propose a 3 3 transfer function matrix whose elements represent the coupling between

Figure 3. The relative arrival time between the two stations


is measured by the time shift t, which is incorporated in the
algorithm as expit, where is circular frequency.

Figure 4. An example of event used in the inversion of G,


depicting typical time windows selected for seismic noise and
signal. INSS is the reference station (on rock).

A Ground-Motion Transfer Function Matrix between Two Nearby Rock and Soil Sites
pairs of ground-motion components at each frequency.
Although Tumarkin (1998) states that the matrix elements
can be obtained from the maximum likelihood solution of
an overdetermined system of equations, he does not elaborate
on the inversion scheme and directly presents his results of
applying the method to data from the downhole array in
Garner Valley, southern California. On the other hand,
Paolucci (1999) calculates the elements of the transfer function matrix for two generic cases, a free-surface topography
and a cylindrical valley, each with a prescribed set of elastic
parameters and geometry. In each, he computes the ground
motion due to each of three perpendicular input motions
and the corresponding transfer matrix elements using the
outputs to each of those input motions. In addition, he convolves the synthetic transfer matrix with real accelerograms
recorded on different types of stiff soils and rock and performs
a comparative study with the corresponding standard spectral
ratios. Paolucci (1999) concludes that the cross-coupling
terms cause dispersion of the site response, introducing peak
values that could be interpreted as resonances if calculated
with the standard spectral ratios.
In the present study, we presume no prior knowledge of
the local geology of the site and elaborate an inversion
scheme based on the stochastic inverse (Franklin, 1970;
Aki and Richards, 1980), to compute a 3D transfer function
of ground motion, conveying the coupling of the three com-

225

ponents of motion recorded at a soil site and a nearby rock


site. This frequency domain transfer function, which we call
the G matrix, is a 3 3 complex matrix or, equivalently, two
3 3 real matrices that represents the amplitude and phase
relations between the motion at the soil site and the rock site.
The principal values of these matrices determine the amplitude and direction of maximum amplification at the soil site
as functions of frequency. The choice of the stochastic inverse is based on both the presence of noise in the records
and the fact that each earthquake can be considered an independent instance of polarized motion at both the rock and the
soil sites. Within the linear assumption, increasing the number of earthquakes will increase the number of polarities that
are used in the analysis, thus providing robust and stable estimates of G. In essence, we extend the analysis of Pisarenko
(1970) to the case of three coupled components.

Method
Let s f and r f represent the frequency domain
ground-motion vectors at a soil site and a nearby rock site,
respectively, at frequency f. If Gij , i; j  1; 2; 3 is the
transfer function between the wave motion in the direction
j at the rock site and the wave motion in the direction i
at the soil site, the most general linear relationship among
the three components of the two vectors is:

Figure 5. The elements of the G matrix for station PGMS. The black lines correspond to the solutions with the stochastic inversion
scheme described in the text, and the gray lines correspond to the solutions obtained by a simple least-squares inversion of the simultaneous
system of equation (2). The latter tends to produce larger values for all components of G through most of the frequency range.

226

R. Benites and T. G. Caldwell

s1  G11 r1  G12 r2  G13 r3


s2  G21 r1  G22 r2  G23 r3

(1)

s3  G31 r1  G32 r2  G33 r3 ;


where the frequency dependency has been omitted for clarity.
Equation (1) implicitly assumes linearity of the elastic
media underneath the rock and soil sites. Assuming that
the rock site accurately represents the input motion of the
incident wave, the Gij s account for the effects of the local
near-surface geology at the soil site.
The problem is reduced to finding the nine complex
unknowns Gij in the general case where no symmetry is
present. At least three nonparallel pairs of data vectors
s1 ; s2 ; s3  and r1 ; r2 ; r3  are necessary to solve equation (1),
but, by using many earthquakes recorded at both sites (soil
and rock), we help ensure stability and convergence of G to
values that best represent only the effect of the local site
conditions on the ground motion and independence from
k k
the earthquake location. Let us call sk  sk
1 ; s2 ; s3 
k k k
k
and r  r1 ; r2 ; r3  the data vectors corresponding
to earthquakes k  1; 2; 3N (N is number of earthquakes).
Posing an equation similar to equation (1) for each pair k and
arranging all in matrix form, we obtain:

s1
1

B 1
Bs
@ 2
s1
3
0

s2
1

s3
1

s2
2

s3
2

s2
3

s3
3

G11

B
 @ G21
G31

G12
G22
G32

sN
1

C
C
sN
2 A

sN
3
10 1
r1
G13
CB
B
G23 A@ r1
2
G33
r1
3

r2
1

r3
1

r2
2

r3
2

r2
3

r3
3

rN
1

C
C (2)
rN
2 A
rN
3

or
s  Gr;

(3)

where s and r are 3 N (rectangular) matrices.


Unlike most inversion problems in seismology in which,
given a data vector d, a model vector m is found from an
assumed mathematical model d  Gm, where G is a prescribed operator, here we solve for G assuming that matrices
s and r are known. This is called a system identification
problem (e.g., Aster et al., 2005).
Having N 3, equation (3) can be solved by a simple
least-squares inversion scheme, but this will not incorporate
the variance of signal and noise amplitudes from earthquake

Figure 6. Comparison of the G matrix elements of stations LHUS (solid line) and PHFS (dash-dotted line). The corresponding
values are similar up to about 3 Hz and depart dramatically for higher frequencies, suggesting the effect of the weathered greywacke directly
underneath PHFS.

A Ground-Motion Transfer Function Matrix between Two Nearby Rock and Soil Sites
to earthquake. For our analysis, we have selected seismograms with a signal-to-noise ratio of at least 4=1. For each
frequency, the variability of amplitudes from earthquake to
earthquake can be measured by the covariance matrix of the
three components of motion. Let and be the covariance
matrices
0
1
11 12 13
(4)
 @ 21 22 23 A
31 32 33
0

11
 @ 21
31

12
22
32

1
13
23 A
33

corresponding to pre-event noise at the soil and rock sites,


respectively, where
 N


1 X
k
k

si  si sj  sj  
ij 

N  1 k1
and
 N


1 X
k
k

ij 
ri  ri rj  rj  :

N  1 k1


denotes complex conjugate, and  is the average value.

227

The stochastic inverse G1 is the matrix that minimizes


the value of
G1 sk  rk G1 sk  rk 

(5)

for each k, where  denotes the matrix adjoint.


Following the corresponding formulation of Aki and Richards (1980), the stochastic inverse can be written as
G 1 . Both and are symmetric and positive definite,
the covariance matrices can be factorized as
 Ss S
and
 Rr R ;
where S and R are orthogonal matrices whose elements are
the eigenvectors of and , respectively, and s and r are
diagonal matrices whose elements are the eigenvalues of
and , respectively. Let us introduce a transformation of variables sk and rk such that
sk0  s1=2 S sk

(6)

rk0  r1=2 R rk ;


where 0 denotes the transformed variables. Inserting these
into G 1 sk  rk , we obtain

Figure 7. Comparison of the G matrix elements at stations LHES (solid line) and LHRS (dash-dotted line), both located at each side of
the Wellington fault on thick, soft soil and weathered rock, respectively.

228

R. Benites and T. G. Caldwell

1=2 k0
Rr R G S1
s  Rr1=2 rk0
s S S

Data

or
G1 0 sk0  rk0 ;

(7)

G1 0  r1=2 R G Ss1=2 :

(8)

where

We can arrange equation (8) into a form similar to equation (2) to obtain the matrix equation
G1 0 s0  r0

(9)

with G1 , sk , and rk are now expressed in the primed
variables.
From equation (8), the stochastic inverse is given by
G1 0  r0 s0  r0 r0 s0 s0  s0 r0 1 :

(10)

We compute G1 by transforming G1 0 back to the original variables using the transformation in equation (7) and
finally find G  G1 1 .

We apply our scheme to broadband acceleration data


recorded in the Wellington Metropolitan Area (WMA),
New Zealand (Fig. 1) between years 2000 and 2007, with
magnitudes in the range 3.96.2, corresponding, basically,
to weak motion (Fig. 2). Stations INSS and POTS are the
reference sites for the Lower Hutt and downtown Wellington
areas, respectively (Fig. 1). These stations are located on hard
rock and are considered good references sites (Sritharan and
McVerry, 1992; Taber and Smith, 1992). The rock and soils
underneath each station are described in Table 1. The data
include earthquakes from the AprilMay 2004 Upper Hutt
earthquake swarm, with hypocenters clustered at 2530 km
depth, directly beneath Upper Hutt (upper right corner of the
small rectangle in Figure 2).
The earthquakes have been selected such that the angles
of incidence in the WMA are vertical. For earthquakes with
slightly inclined incidence, we account for the difference in
arrival time at the reference site and at the interface between
sediments and bedrock at the soil site (Fig. 3) by measuring
the arrival times of the P wave at both sites and computing
t  tsoil  trock from the slowness and the thickness of the
soil, assuming a 1D model. We then multiply the spectrum of
the seismogram at the soil site by expit (i.e., time shift),

Figure 8. Comparison of the G matrix elements at stations TFSS (solid line) and MISS (dash-dotted line), both on reclaimed soil but
showing distinctive site effects.

A Ground-Motion Transfer Function Matrix between Two Nearby Rock and Soil Sites
p
where i  1 and  2f. In a few cases where the
P-wave arrival was not clear, we used the delay in S-wave
arrival since tS tP for nearly vertical incidence. In general, we find that t is of the order of 0.01 s for earthquakes
with epicenters in the region between the large and small
squares of Figure 2. For epicenters inside the smaller square,
the delays are negligible and we have not incorporated the
time shift. We emphasize that the time windows chosen for
analysis must exclude surface waves or any nondirect body
waves arriving at the soil and reference stations.
The orientations of the horizontal components at all stations used in this study are referred to the geographical
northeast coordinates, with index 1 for the north component
of motion, index 2 for the east component, and index 3 for
the vertical component (positive up). The sampling time interval is 0.01 s; and, because the seismograms recorded at
each station are, in general, of different duration, we have
either increased their duration (by adding zeros) or truncated
them all to the same duration of 40.96 s, sampled by 4096
time points. The mean has been removed from all seismograms, and we have chosen time windows that include the
P and S wave arrivals but omit surface and coda waves.
Generally, these windows are between 10 and 15 seconds,
as shown in Figure 4 for an earthquake recorded at stations
INSS (rock) and LHUS (soil). We have band-pass filtered the

229

records between 0.1 and 20 Hz; and, although the Fourier


spectra have not been smoothed in the computation of G,
the results shown in subsequent figures have all been
smoothed with a 19-point triangular window (0.46 Hz wide).

Results
Figure 5 shows the absolute values of the nine components of G for station PGMS, using INSS (on rock) as the
reference site (Fig. 1). The calculations are in the frequency
domain for up to 15 Hz, using the stochastic inverse (black
line) and a simple least-squares inversion numerical scheme
(gray line). In the latter case, the presence of noise in the
signals biases the variability of the response (Aki and
Richards, 1980, p. 637), resulting in a grossly overestimation
of the amplitude of the response (the glitches below 1 Hz
also reflect the instability of the algorithm due to the lowfrequency deficiency of the weak motion).
Strictly speaking, the G matrix maps the three-component motion vector at the rock site into the three-component
vector describing the motion at the soil site. The values of
the off-diagonal terms depend not only on the spectral
amplitudes of the corresponding components but also on the
angles of the motion with respect to the measurement (usually geographic) coordinates. They define the symmetry of the

Figure 9. Comparison of the G matrix elements at station WEL (solid line), on an elevated topography of greywacke weathered at the
top, and WNKS (dash-dotted line), on thin soil 520 m thick over rock, surrounded by topographic features.

230

R. Benites and T. G. Caldwell

response and may be larger or smaller than the diagonal


terms. What it is important is not the value of the off-diagonal
terms (e.g., G12 and G21 ) but the value of the skew
(G21  G12 ). If G21  G12  0 (for a symmetric matrix),
the implication is that the soil has a symmetric geometry.
Considering only the stochastic inverse results of
Figure 5, the G matrix for PGMS suggests significant geometric asymmetry of the response, in general. The large
values of G13 and G23 , in particular, reveal significant transfer of the vertical motion into horizontal motions. On the
other hand, the amplitudes of G12 and G21 are comparable
and similar to those of G11 and G22 (in the diagonal), suggesting a near-symmetric response for the horizontal motion.
We must point out that although there have been regional earthquakes of moderate magnitude (5.26) recorded by
our network, with S-P times larger than 10 seconds, we
could not include them in our analysis because the recording
parameters of the stations were set to have short pre-event
memory, which have prevented the recording of full waveform P and S waves. Hence, our results are not be reliable
below 0.1 Hz.
Figures 69 show the matrix transfer functions for a pair
of close stations, showing, in turn, the strong variability of
the site effects in the Wellington metropolitan area. In
Figure 6, LHUS (solid line) is the response of the valley
at a point where the soil thickness is about one third of that
at PGMS, close to the fault. On the other hand, PHFS (dashdotted line) lies on highly weathered greywacke close to the
shore, where the bedrock pinches out. A similar situation
occurs on both sides of the fault, as shown in Figure 7,
for LHES (solid line) and LHRS (dash-dotted line). Figure 8
and Figure 9 show responses for the downtown Wellington
area, and G has been computed using station POTS (on rock)
as the reference. Figure 8 shows the responses from stations
TFSS (solid line) and MISS (dash-dotted line) both located
on reclaimed soil near sea level; and, in Figure 9, the
responses from WEL (solid lines) shows amplification of
factors up to 13 for frequencies as high as 3 Hz. This high
amplification is attributed to topographic effects (Davis and
West, 1973; Bouchon et al., 1996) because WEL is on the
top of a hill of about 180 m height. The station WNKS, close
to WEL, also shows (1) a peak at 3 Hz in the elements
of G corresponding to the horizontal components, and
(2) a broader response for all elements, showing the effect
of the alluvium cover over the weathered rock on which
the station is located, which is less than 50 m thick and
surrounded by hills.

Principal Direction of Motion: Singular


Value Decomposition
Site effects cannot be properly represented by the absolute value of the calculated response alone. As pointed out
previously in this article, the transfer function matrix G is
complex; that is, it is composed of two real matrices Gre
and Gim , conveying magnitude and phase of the relative

motion between the soil and the reference rock sites, respectively. To estimate the maximum response, we first compute
the principal values and directions (axes) of each Gre and
Gim . Then, we add these principal values in groups of two
(nine vector values) to obtain a single matrix S. The principal
values and axes of S determine an upper bound on amplification and the corresponding direction of the relative motion
between soil and rock. This must be done for each frequency.
Because G is nonsymmetric, in general, we use singular
value decomposition (SVD) to find the principal values and
directions of matrices Gre and Gim . These, in turn, determine
the principal vectors sre j and sim j, j  1; 2; 3 for Gre and
Gim , respectively. The matrix S is constructed by adding
these principal vectors:
Si; j 

q
s2re i  s2im j  2 cosij ;

(11)

where cosij   sre i sim j, with i; j  1; 3.


The principal values of S (Gp , Gq , and Gn [Gp
~ q,
~
Gq Gn > 0]) and associated orthogonal directions p,
and n~ bound the maximum possible value of the siteresponse amplification. Gp , Gq , and Gn comprise the axes
of an ellipsoid representing the ground motion for each
frequency.
Figure 10a shows the principal values Gp (black), Gq
(red), and Gn (green) at PGMS as function of frequency,

(a) Principal values for PGMS: Gp (black), Gq


(red), and Gn (green). The circled numbers denote the three peak
values of Gp in the frequency range from 0 to 10 Hz. (b) Azimuth
angle with respect to the geographical north of the horizontal motion (black), obtained by projection of the 3D motion ellipsoid on
the northeast plane. The gray line is the angle of the ellipsoid with
respect to depth (positive up), normal to the northeast plane.
(c) The orientation of ellipses of horizontal motion corresponding
to the three peaks in (a), whose amplitude values, frequencies, and
orientation angles (in degrees) are listed on the right side of the figure. The Hutt Valley is oriented about 50 with respect to north.

Figure 10.

A Ground-Motion Transfer Function Matrix between Two Nearby Rock and Soil Sites

231

obtained from the SVD of the matrix S. We have computed


the Euler angles of the ellipsoid encompassing the response
with respect to the geographical coordinates and plotted the
azimuth and the dip angle (in black and gray, respectively) as
functions of frequency in Figure 10b. Because the maximum
response is greatest along the p and q axes up to about 12 Hz
and Gn is negligible, we show only the projection of the ellipsoid on the horizontal plane. The bounds of the horizontal
motion, shown by the ellipses in Figure 10b correspond
to the three large peaks in Figure 10a. This is shown in
Figure 10c, with black for 1.98 Hz, blue for 5.40 Hz, and
orange for 8.28 Hz. The peak at 1.98 Hz of amplitude 13.14
is oriented in the direction 31 (northeast, or roughly perpendicular to the Wellington fault (Fig. 1), while the other
peak values are roughly aligned along the strike.
The previous example shows the variability of the amplitude and directions of the ground response with frequency.
Let us look at the variability of the maximum value of the
response in space. In Figure 11, we have plotted the bounding ellipses corresponding to the frequency at which the
maximum amplitude occurs for the stations shown in
Figure 1, along with their corresponding azimuthal angle.
Except for PGMS, all the maximum values are oriented in
the northeast direction, for both the Wellington and Lower
Hutt areas.

The orientation of the maximum response also depends


on frequency. For example, the amplitudes for the stations in
Figure 11 for frequencies 1.5 Hz, 2 Hz, and 3 Hz, along with
their azimuthal angles, are plotted in Figure 12, with black
for 1.5 Hz, blue for 2 Hz, and orange for 3 Hz. It is remarkable the change of ground motion from 2 Hz to 3 Hz, drastically in amplitude and rather mild in direction, at two
stations as close as LHRS and LHES and separated by the
Wellington fault line. The amplitude differences are due to
the impedance contrast between LHRS situated on bedrock
(greywacke) and LHES situated on soft alluvium; while the
difference on direction might be due to the generation of
weak surface waves at the faults edge (Adams et al., 2003).
Finally, for comparison, we calculate the standard single-component spectral ratios R11 , R22 , and R33 for several
stations in the WMA using the same earthquake records as
used to compute G. For each station, we calculated the
geometrical mean and standard
deviation of the horizontal
p
spectral ratios (i.e., SRh  R211  R222 ) for all earthquakes.
The spectra were also smoothed with a 19-point triangular
window. The mean spectral ratios are plotted on the right
column of Figure 13, where the thick line is the mean value
and the dashed lines the corresponding standard deviations.
The left column of Figure 13 shows the corresponding values

Figure 11. Ellipses representing the maximum values of the


response at all stations. Below each ellipse are the values of the
amplitude of the response, corresponding frequency, and direction
(with respect to north).

Figure 12.

Horizontal motion ellipses for prescribed frequencies 1.51 Hz (black), 2 Hz (blue), and 3 Hz (orange) at all stations.
Below are the corresponding amplification factors, written with
colors corresponding to each ellipse.

232

R. Benites and T. G. Caldwell

Figure 13. The left column shows the principal values Gp (black line) and Gq (gray line) for the stations listed at the far right. The right
column shows the mean value of the horizontal standard spectral ratios SRh (solid line) and the corresponding standard deviation (dashed
lines) for the same stations.
of Gp (black line) and Gq (gray line). The comparison shows
that, for stations PGMS, LHUS, and PHFS, Gp exhibits
similar variability to the corresponding mean spectral ratio
with frequency. This is not the case for LHES, ARKS,

WELS, and MISS, where the Gp values are larger than


the corresponding peak mean ratios. In particular, at ARKS,
the peak at about 2.2 Hz is not present in the corresponding
spectral ratio.

A Ground-Motion Transfer Function Matrix between Two Nearby Rock and Soil Sites

Conclusions
In this article, we have introduced a method based on the
stochastic inverse to compute a 3D representation of the
effects of the local site conditions on ground motion. This
representation is a 3 3 complex matrix, or transfer matrix
function G whose elements transform the three components
of the ground motion recorded at a reference rock site into
the three components of motion at a soil site, in the frequency
domain. G is derived from an overdetermined system of
simultaneous equations formed with the unknown elements
of G and the three-component complex spectra of many
different earthquakes recorded at both the soil and rock sites.
We used a total of 25 earthquakes for our analysis (only five
of which were of M 5 or larger). The earthquakes were
selected hypocenters distributed over a wide azimuthal range
but such that the angle of incidence is vertical, or near
vertical, at both the soil and rock reference stations. These
conditions help ensure that a variety of different polarizations
are present, which stabilizes the inversion. Within the linear
regime, the elements of the G matrix converge to constant
values as the number of earthquakes used in the inversion
is increased.
For the cases treated in this work, we used data from 25
earthquakes per station in the inversion. Adding the data
from M 34 earthquakes makes negligible difference to
the G matrix estimates above 2 Hz. The availability of a number of earthquakes with magnitude greater than M 5 would
modify the results in the low-frequency range (< 0:5 Hz) due
to their capability of exciting the low-frequency modes of
the basin.
Using the G matrix, we have determined the local site
effects at the recording sites of the permanent strong ground
motion network in the WMA. The results show that the
maximum amplification factors occur at different frequencies
and in different orientations, depending on the site location.
The overall orientation of the maximum motion in the Wellington area is northeastsouthwest. In general, at any given
frequency, the amplifications and maximum orientations are
highly variable, even where sites are close together. This
suggests a strong dependency of the G matrix on the geometry of the near-surface geological heterogeneity that produces
the amplification.
In some cases the frequency behavior of Gp is similar to
that of the corresponding mean horizontal spectral ratio, SRh ,
while in other the behaviors are different. The differences are
explained by the extent to which the response at the soil site
depends on the polarization of the input motion at the rock
site. Sites with symmetric geometry will produce a
symmetric G matrix that will have only diagonal terms in
a particular coordinate system. In this situation the magnitude of the diagonal terms will be the same as the standard
spectral ratios, and linearly polarized input motion at the rock
site will produce a linearly polarized response at the soil site.
Although we have calculated our results for up to 15 Hz,
their validity may be questionable beyond 5 Hz, in the light

233

of the results from Steidl et al. (1996). Stations POTS and


INSS, both on hard rock, are the best reference stations available for the Lower Hutt valley and the Wellington downtown
area, respectively. However, we consider that the best reference station for the application of our method would be a
downhole seismometer situated at the bottom of a 1-km-deep
drill hole in the basement rock.

Data and Resources


The seismograms used in this study were obtained from
the New Zealand strong motion network: www.geonet.org
.nz/resources/basicdata/strongmotiondata/index.html. The
soil description and thickness at the seismic stations listed in
Table 1 were obtained from the station information available
at magma.geonet.org.nz/delta/.

Acknowledgments
The authors express their gratitude to John Haines for valuable discussions at the start of this project, as well as to Russell Robinson and Terry
Webb for their time dedicated to improve the quality of this manuscript. The
authors are indebted to Francisco Chvez-Garca, Roberto Paolucci, and Stefano Parolai, an associate editor of BSSA, whose critical comments have
been crucial to clarify the concepts involved in this work. Our gratitude
is also extended to the staff of the Lido Cafe (Wellington) who graciously
hosted our endless discussions on the subject.

References
Adams, B. M., N. M. Osborne, and J. J. Taber (2003). The basin-edge effects
from weak ground motions across the fault-bounded edge of the Lower
Hutt valley, New Zealand, Bull. Seismol. Soc. Am. 93, 27032716.
Aki, K., and P. G. Richards (1980). Quantitative Seismology: Theory and
Methods, W. H. Freeman and Co., San Francisco, California,
768 pages.
Aster, R. C., B. Borchers, and C. H. Thurber (2005). Parameter Estimation
and Inverse Problems, Vol 90 in the International Geophysics Series,
Elsevier Academic Press, San Diego, California, 301 pages.
Beresnev, I. A., and K.-L. Wen (1996). Nonlinear soil responseA reality?,
Bull. Seismol. Soc. Am. 86, 19641978.
Borcherdt, R. D. (1970). Effects of local geology on ground motion near San
Francisco Bay, Bull. Seismol. Soc. Am. 60, 2981.
Bouchon, M., C. A. Schultz, and M. N. Toksoz (1996). Effect of threedimensional topography on seismic motion, J. Geophys. Res. 101,
58355846.
Coutel, F., and P. Mora (1998). Simulation-based comparison of four siteresponse estimation techniques, Bull. Seismol. Soc. Am. 88, 3042.
Davis, L. L., and L. R. West (1973). Observed effects of topography on
ground motion, Bull. Seismol. Soc. Am. 63, 283298.
Franklin, J. N. (1970). Well-posed stochastic extensions of ill-posed linear
problems, J. Math. Anal. Appl. 31, 682716.
King, J. L., and B. E. Tucker (1984). Observed variations of earthquake
motions across a sediment-filled valley, Bull. Seismol. Soc. Am. 74,
137151.
Konno, K., and T. Ohmachi (1998). Ground-motion characteristics estimated
from spectral ratio between horizontal and vertical components of
microtremor, Bull. Seismol. Soc. Am. 88, 228241.
Lachet, C., and P.-Y. Bard (1994). Numerical and theoretical investigations
on the possibilities and limitations of Nakamuras technique, J. Phys.
Earth 42, 377397.
Lermo, J., and F. J. Chvez-Garca (1993). Site effects evaluation using
spectral ratios with only one station, Bull. Seismol. Soc. Am. 83,
15741594.

234
Nakamura, Y. (1989). Inference of seismic responses of surface layer based
on microtremor measurement, Q. Rep. Railway Tech. Res. Inst. 4, 18
27 (in Japanese).
Paolucci, R. (1999). Numerical evaluation of the effect of cross-coupling of
different components of ground motion in site response analysis,
Bull. Seismol. Soc. Am. 89, 877887.
Pisarenko, V. F. (1970). Statistical estimates of amplitude and phase corrections, Geophys. J. R. Astr. Soc. 20, 8998.
Sritharan, S., and G. H. McVerry (1992). Microzone effects in the Hutt
Valley in records from a strong-motion accelerograph array, Bull.
New Zeal. Natl. Soc. Earthq. Eng. 25, 246264.
Steidl, J. H., A. G. Tumarkin, and R. J. Archuleta (1996). What is a reference
site?, Bull. Seismol. Soc. Am. 86, 17331748.
Taber, J. J., and E. G. C. Smith (1992). Frequency dependent amplification
of weak ground motions in Porirua and Lower Hutt, New Zealand,
Bull. New Zeal. Natl. Soc. Earthq. Eng. 303331.

R. Benites and T. G. Caldwell


Tumarkin, A. G. (1998). Site response analyisis in 3D, in The Effects
of Surface Geology on Seismic Motion, Vol. 2, K. Irikura, K. Kudo,
H. Okada, and T. Sasatani (Editors), Balkema, Rotterdam,
365370.
Yu, J., and J. Haines (2003). The choice of references sites for seismic
ground amplification analyses: Case study at Parkway, New Zealand,
Bull. Seismol. Soc. Am. 93, 713723.

GNS Science
1 Fairway Drive, Avalon
P.O. Box 30-368
Lower Hutt 5040, New Zealand
Manuscript received 2 November 2009

Potrebbero piacerti anche