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

net/publication/245294027

Site Response at Treasure and Yerba Buena Islands, California

Article  in  Journal of Geotechnical and Geoenvironmental Engineering · January 2003

DOI: 10.1061/(ASCE)1090-0241(2003)129:5(415)

CITATION READS
1 202

1 author:

Laurie G. Baise
Tufts University
74 PUBLICATIONS   1,163 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Laurie G. Baise on 18 September 2014.

The user has requested enhancement of the downloaded file.

 University of California

Peer Reviewed

Title:
Site response at Treasure and Yerba Buena Islands, California
Author:
Baise, L G
Glaser, Steven D, Berkeley
Dreger, D
Publication Date:
05-01-2003
Publication Info:
Postprints, Multi-Campus
Permalink:
http://escholarship.org/uc/item/39h9s89j
Additional Info:
Laurie G. Baise, Steven D. Glaser, Douglas Dreger, Site Response at Treasure and Yerba Buena
Islands, California, Journal of Geotechnical and Geoenvironmental Engineering © ASCE / MAY
2003 / 415.
Keywords:
surface waves, finite difference, California, seismic response, soils, rocks, comparative studies
Abstract:
A variety of methods are utilized to reinvestigate the physical relationship between the seismic
response of Treasure Island (TI) and Yerba Buena Island (YBI) in California. These islands are
a soil (TI) and rock (YBI) site pair separated by 2 kin. The site pair has been used previously
by researchers to identify soil response to earthquake shaking. Linear regime ground motions
(M(W)4.0-M(W)4.6 and PGA: 0.014-0.017 g) recorded in the TI vertical array indicate a coherent
wavefield in the sediments and an incoherence between the rock and sediments. Our analyses
show that the greatest change in the wavefield occurred between the rock and soil layers,
corresponding to a significant impedance contrast. The waveforms change very little as they
propagate through the sediments, indicating that the site response is a cumulative effect of the
entire soil structure and not a result of wave propagation within individual soil layers. In order
to highlight the complexity of the site response, correlation analysis was used to demonstrate
that the rock and soil ground motions were not highly coherent between the two sites. YBI was,
therefore, shown to be an inappropriate reference site for TI. One-dimensional (1D) vertical wave
propagation and inverse techniques were used to differentiate between 1D site response and
more complex site behavior. Both 1D methods (vertical wave propagation and inverse transfer
functions) proved incapable of capturing the site response at TI beyond the initial four seconds of
motion. Finite difference waveform modeling, based on a two-dimensional velocity structure of the
northern San Francisco Bay was needed to explain the linear site response at TI as horizontally
propagating surface waves trapped in the bay sediments. A simplified velocity structure for the
San Francisco Bay including a single 100 m basin layer (constant shear-wave velocity of 400 m/
s) over a 1.5 km/s layer of Franciscan bedrock was able to trap energy in the basin and produce

eScholarship provides open access, scholarly publishing

services to the University of California and delivers a dynamic
research platform to scholars worldwide.
surface waveform ringing similar to that observed in the TI data. Due to surface waves propagating
in the San Francisco Bay sediments, any.1D model will not fully characterize site response at TI.
All 1D models will fail to produce the late arriving energy observed in the ground motions.

eScholarship provides open access, scholarly publishing

services to the University of California and delivers a dynamic
research platform to scholars worldwide.
 Site Response at Treasure and Yerba Buena Islands,
California
Laurie G. Baise, M.ASCE1; Steven D. Glaser, M.ASCE2; and Douglas Dreger3

Abstract: A variety of methods are utilized to reinvestigate the physical relationship between the seismic response of Treasure Island
共TI兲 and Yerba Buena Island 共YBI兲 in California. These islands are a soil 共TI兲 and rock 共YBI兲 site pair separated by 2 km. The site pair
has been used previously by researchers to identify soil response to earthquake shaking. Linear regime ground motions (M W 4.0– M W 4.6
and PGA: 0.014 –0.017 g兲 recorded in the TI vertical array indicate a coherent wavefield in the sediments and an incoherence between the
rock and sediments. Our analyses show that the greatest change in the wavefield occurred between the rock and soil layers, corresponding
to a significant impedance contrast. The waveforms change very little as they propagate through the sediments, indicating that the site
response is a cumulative effect of the entire soil structure and not a result of wave propagation within individual soil layers. In order to
highlight the complexity of the site response, correlation analysis was used to demonstrate that the rock and soil ground motions were not
highly coherent between the two sites. YBI was, therefore, shown to be an inappropriate reference site for TI. One-dimensional 共1D兲
vertical wave propagation and inverse techniques were used to differentiate between 1D site response and more complex site behavior.
Both 1D methods 共vertical wave propagation and inverse transfer functions兲 proved incapable of capturing the site response at TI beyond
the initial four seconds of motion. Finite difference waveform modeling, based on a two-dimensional velocity structure of the northern
San Francisco Bay was needed to explain the linear site response at TI as horizontally propagating surface waves trapped in the bay
sediments. A simplified velocity structure for the San Francisco Bay including a single 100 m basin layer 共constant shear-wave velocity
of 400 m/s兲 over a 1.5 km/s layer of Franciscan bedrock was able to trap energy in the basin and produce surface waveform ringing similar
to that observed in the TI data. Due to surface waves propagating in the San Francisco Bay sediments, any 1D model will not fully
characterize site response at TI. All 1D models will fail to produce the late arriving energy observed in the ground motions.
DOI: 10.1061/共ASCE兲1090-0241共2003兲129:6共415兲
CE Database subject headings: Surface waves; Finite difference; California; Seismic response; Soils; Rocks; Comparative studies.

Introduction and YBI is complicated by regional geologic conditions in the

shallow San Francisco Bay sediments. Therefore, a variety of
Earthquake site response at soft soil sites is an important issue
techniques were enlisted to reinvestigate the physical relationship
along the margins of coastal urban cities, and Treasure Island
between the seismic response of Treasure and Yerba Buena Is-
共TI兲, California is a prime example 共Borcherdt 1970; Borcherdt
lands, focusing on recorded ground motions in the linear range of
and Gibbs 1976; Idriss 1990; UCB 1990; Seed et al. 1991兲. TI, a
soil behavior.
man-made island in San Francisco Bay between San Francisco
Many previous studies have used rock ground motions at YBI
and Oakland, consists of several meters of fill overlying marine
共as a reference site兲 as input to estimate the earthquake site re-
sediments, a similar soil profile to other filled sites around the San
Francisco Bay 共de Alba et al. 1994兲. TI is attached to Yerba sponse at TI to subsequently draw conclusions on the nonlinear
Buena Island 共YBI兲, and the two islands are a unique pair for site response at TI 共Jarpe et al. 1989; Idriss 1990; Darragh and Shakal
response studies, since ground motions recorded on the outcrop- 1991; Seed et al. 1991; Rollins et al. 1994兲. To augment these
ping rock of YBI might provide an estimate of incoming energy at previous studies we used a more complete set of ground motions
the base of the soil column at TI. The seismic site response at TI recorded in the linear range of soil response for two events
(M W 4.0 and M W 4.6) to determine the relationship between TI
1 and YBI, and ultimately the linear site response at TI. Using
Assistant Professor, Dept. of Civil and Environmental Engineering,
113 Anderson Hall, Tufts Univ., Medford, MA 02155. E-mail:
ground motions recorded at depth in both soil and rock below the
[email protected] TI site, and at the surface and at depth in rock at the YBI site, the
2
Associate Professor, Dept. of Civil and Environmental Engineering, assembled data set for this study provided a more complete view
440 Davis Hall, Univ. of California, Berkeley, CA 94720. of the earthquake site response than others to date. The surface
3
Associate Professor, Dept. of Earth and Planetary Science, 307 recordings at TI were compared with those in bedrock at TI and at
McCone Hall, Univ. of California, Berkeley, CA 94720. YBI to evaluate the YBI reference site validity and the site re-
Note. Discussion open until November 1, 2003. Separate discussions sponse at TI.
must be submitted for individual papers. To extend the closing date by Site response studies often rely on Fourier-based spectral ra-
one month, a written request must be filed with the ASCE Managing
tios to determine site response from a rock and soil site pair of
Editor. The manuscript for this paper was submitted for review and pos-
sible publication on January 17, 2001; approved on July 15, 2002. This ground motions. Because commonly used Fourier-based site re-
paper is part of the Journal of Geotechnical and Geoenvironmental sponse methods are fundamentally linear and rely on an input/
Engineering, Vol. 129, No. 6, June 1, 2003. ©ASCE, ISSN 1090-0241/ output relationship between a reference rock site and a soil site, a
2003/6-415– 426/$18.00. correlated linear relationship between ground motions is required

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003 / 415

for reliable site response estimates. The coherency, a frequency
domain measure of the linear relationship between input and out-
put records, may deviate from one as a result of one or more of
the following 共Bendat and Piersol 1993, p. 84兲: excessive mea-
surement noise in the time series; significant resolution bias in the
spectral estimates; the system relating the two time series is non-
linear; or the output is not produced exclusively by the input. Any
of these reasons are valid with ground motion input/output pairs.
Specifically, an inappropriate reference site, or seismic energy
entering the site from multiple directions 共surface waves兲 are po-
tential causes of incoherence and fall under the fourth source of
low correlation. If input/output signals are uncorrelated, reliable
transfer functions estimates are not possible. Field et al. 共1992兲
looked into this issue for site response estimates for weak-motion
and attributed low coherence between rock and soil sites to signal
generated noise. Field et al. 共1992兲 found that the signal gener-
ated noise caused large uncertainties but no bias in the individual
spectral ratios whereas the cross-spectrum estimate was biased as
a result of the low coherence. Spectral ratios can be used to esti- Fig. 1. Map of San Francisco Bay Region. Location of Treasure
mate a transfer function; however, the resulting estimate can Island vertical array and Yerba Buena Island shown as triangles.
smear a complex response across frequencies. For example, if the Focal mechanisms for Bolinas 共8/18/99兲 and Richmond 共12/04/98兲
site response is due to S waves and surface waves that enter the events are shown.
signal at different times, the spectral ratio taken over the entire
signal will smear the effects of each wave arrival across frequen-
cies to give a single site response estimate. In the case of surface
waves, wave propagation remains linear but the rock ground mo- surface and five stations in the soil 共de Alba et al. 1994兲. A second
tion does not adequately represent the input thereby resulting in bedrock station was added at a later date. The stations were
low correlation. Low correlation between input and output pro- placed to record ground motions below the bedrock surface 共122
vides a warning of this or similar cases, alerting us to the fact that and 104 m兲, twice in the Pleistocene bay mud 共44 and 31 m兲, near
output is not a simple linear function of the input. the top of the Holocene bay mud 共16 m兲, and twice in the hydrau-
To evaluate the use of YBI as a reference site for input mo- lic fill 共7 and 0 m兲 共de Alba et al. 1994兲.
tions to the TI site, correlation analysis was used to quantify the
linear relationship between the input/output signals. In addition,
system identification techniques 共Glaser 1996; Baise and
Glaser 2000兲 were implemented to investigate the effect of the
bay sediments on ground motions. For our studies, the soil system
was characterized as a linear filter, with the buried sensor the
input, and the sensors above, the outputs. The efficacy of one-
dimensional 共1D兲 site response at TI was evaluated with vertical
wave propagation 共Ching and Glaser 2001兲 and with system iden-
tification techniques to determine an empirical transfer function
evaluated from the rock input and surface output ground motions.
Evaluation of particle motions at TI provided a better understand-
ing of the physical nature of the wavefield. Finally, the wavefield
was modeled for a M W 4.6 earthquake using a two-dimensional
共2D兲 velocity structure of the northern San Francisco Bay and a
finite difference wave-propagation model 共Baise et al. 2003a兲.

Treasure and Yerba Buena Island Sites

Treasure Island Downhole Array, California

In 1993, a deep instrumentation array through the soil to the bed-
rock was installed at Treasure Island 共TI兲 to gather information on
site response. TI is a hydraulically filled island constructed on an
existing sand spit northwest of Yerba Buena Island 共YBI兲 共Lee
1969兲. The TI site is located 12.8 km from the Hayward fault and
29 km from the San Andreas fault, as shown in Fig. 1. The fill
was dredged from the bay and is composed primarily of fine sand,
ranging from clayey, to gravelly sand 共Lee 1969兲. As a result, the
Fig. 2. Simplified soil profile at Treasure Island vertical array site
TI fill is loosely packed and is susceptible to liquefaction under
with shear and compressional wave-velocity profiles and instrument
cyclic loading. A representative TI soil profile is shown in Fig. 2.
locations 共after Gibbs et al. 1992兲
The downhole array initially had one station below the bedrock

416 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003

Table 1. Earthquakes at Treasure Island and Yerba Buena Island
Depth Distance Azimuth PGA
Date Location MW Latitude Longitude 共km兲 共km兲 共°兲 共g兲
12/4/98 Richmond 4.0 37.920 ⫺122.287 6.9 13.0 217 0.014
8/18/99 Bolinas 4.6 37.907 ⫺122.686 6.9 29.0 108 0.017

Yerba Buena Island Downhole Array, California response calculation when used as a base motion reference site
共Margheriti et al. 1994; Steidl et al. 1996; Boore and Joyner
Yerba Buena Island is a bedrock island also shown in Fig. 1. The
1997; Archuleta et al. 2000兲. High correlation between reference
YBI downhole site 共61 m depth兲 was installed after the 1989
rock and soil sites will indicate a linear relationship, and therefore
Loma Prieta event as part of the Berkeley Digital Seismic Net-
an appropriate system input/output pair. Low correlation between
work and the Hayward Fault Borehole Network. The uphole YBI
sites, which may result from complexities in the wavefield due to
site was a temporary site installed as part of the Bridge network
topography, variations in the subsurface geology, surface waves,
above the permanent station 共Hutchings et al. 1999兲. The YBI site
or other propagation effects, would lead to an inappropriate and
共both uphole and downhole兲, shown in Fig. 1, is located near Pier
inaccurate transfer function estimate no matter what estimator is
E2 of the San Francisco-Oakland Bay Bridge on the eastern slope
used. Low correlation does not mean that wave propagation is
of the island. The rock below the site is composed of the Fran-
necessarily nonlinear. Rather, the low correlation specifically in-
ciscan melange consisting of sandstone interbedded with siltstone
dicates that output⫽ f (input).
over graywacke interbedded with siltstone and shale.

Correlation Analysis at Yerba Buena Island and

Data from Treasure Island and Yerba Buena Island
Treasure Island
The stations in the TI downhole array have recorded two earth-
Fig. 3 presents a schematic drawing of the YBI/TI site pair with
quakes, summarized in Table 1, with companion data for the YBI
recorded rock motions for the Richmond 共12/4/98兲 event. Evalu-
site: the August 18, 1999 Bolinas event (M W 5.0, M W 4.6) and the
ating the correlation coefficient between the YBI bedrock motions
December 4, 1998 Richmond event (M W 4.1, M W 4.0). The TI and
and those recorded in rock beneath TI quantified the spatial con-
YBI recording sites are two kilometers apart.
sistency of the incoming wavefield. Fig. 4 compares the ground
motions for YBI downhole 共YBI↓兲-TI downhole 共TI↓兲 rock loca-
tion, as well as the YBI uphole 共YBI↑兲-TI downhole 共TI↓兲 rock
Yerba Buena Island as Reference Site
pair, for the Richmond temblor. The low-correlation coefficients
for the two combinations of TI and YBI waveforms—r„YBI`-TI`…
Most previous TI site response studies have relied on surface
⫽0.59,r„YBI_-TI`…⫽0.41—indicated that the rock motions at these
ground motions recorded at an additional Yerba Buena site on the
two nearby locations were only moderately coherent. As a com-
southern end of the island as representative input base motion.
parison, r共YBI↓-YBI↑兲 was 0.93. Therefore, caution is advised when
Idriss 共1990兲, Seed et al. 共1991兲, and Rollins et al. 共1994兲 com-
using any YBI rock motion as an input to the TI site. The low
pleted ground response analyses for the Loma Prieta earthquake
correlation observed between rock at TI and at YBI is an indica-
ground motions at TI using the SHAKE equivalent linear wave-
tion of complexity in the wavefield.
propagation program. In addition, Jarpe et al. 共1989兲 and Darragh
A further analysis, comparing the coherency of the TI surface
and Shakal 共1991兲 conducted TI site response studies with spec-
共output-TI↑兲 ground motion with the potential input rock motions
tral ratios using YBI as the reference site to investigate nonlinear
at TI and YBI provided an estimate of the quality of each refer-
site response during the Loma Prieta Earthquake. For both stud-
ence site as a base motion. When the entire 30 s record was used
ies, the TI and YBI strong motion spectral ratios exhibited lower
to estimate the correlation coefficient, both reference sites proved
amplification than the weak motion spectral ratios, indicating pos-
inappropriate—r„YBI_-TI_…⫽0.2, r„TI`-TI_…⫽0.2. However, if the
sible soil nonlinearity at TI. In the present study, the assumption
record was windowed to the 4 s of initial motion, the TI bedrock
of YBI as a reference site for TI site response studies was evalu-
motion and the TI surface motion were more coherent than the
ated by using a correlation analysis to determine if the outcrop
YBI surface bedrock motion and the TI surface motion—
ground motions recorded at YBI are a linear mapping of those
r„YBI_-TI_…⫽0.3, r„TI`-TI_…⫽0.5. According to these r values, the TI
actually recorded in the rock beneath TI.

Correlation Analysis Methodology

The correlation coefficient 共r兲, the normalized maximum value of
the cross correlation, is a single-valued statistic which varies from
⫺1 to 1 and describes the validity of an assumed linear relation-
ship. The coherency and the correlation coefficient are directly
related; coherency is in the frequency domain whereas the corre-
lation coefficient is measured in the time domain. A correlation
coefficient of 1 identifies a perfect linear relationship and ⫺1 is a
perfectly inverse linear relationship. We proposed that correlation
Fig. 3. Schematic drawing of the YBI and TI surface and downhole
analysis can be used to help choose an appropriate reference rock
rock stations for Richmond event. Cross section is not to scale.
site. Recent work shows that some rock sites can have a local site
Correlation coefficients 共r兲 for each pair are shown.
response of their own and therefore introduce a bias into a site

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003 / 417

 Fig. 4. Comparison of east-west component displacement ground motions recorded during Richmond event at YBI and TI

bedrock motion may provide a better input rock motion for the TI a least squares-based optimization for fitting the ARX model
site response; however, the low-r values especially for the full 30 共Ljung 1987兲 and results in model parameters that are constant
s record indicate that a linear transfer function may not be reliable with time. The ARX input/output model accounts for all noise
for estimating TI site response no matter which reference site is components with an additive white noise term at the output 共ex-
used. ogenous兲. Autoregressive moving average modeling has been
used to model waves propagating in layered systems 共Robinson
and Treital 1978; Hubral et al. 1980兲 and to characterize earth-
Yerba Buena Island Site Response quake ground motions 共Gersch and Kitagawa 1985; Safak 1988;
Ellis and Cakmak 1991兲. Several researchers 共Claerbout 1968;
Why are the rock motions at YBI so different from those beneath Robinson and Treital 1978; Hubral et al. 1980; Dargahi-Noubary
TI when the sites are within 2 km of each other? Although both 1999兲 have shown that seismic wave propagation through a lay-
rock sites are in the Franciscan bedrock, the bedrock is highly ered system 共i.e., stratigraphic column兲 is an autoregressive pro-
variable possibly resulting in site response effects. The YBI data cess.
set provided an opportunity to examine site response with the Given input and output data from borehole instruments, the
input-output data pairs. ETF captures a mapping of particle motion time histories between
two points in the soil profile, much like the spectral ratio com-
System Identification monly used in site response studies 共Borcherdt 1970兲. The result-
ing ETF is interpreted as part of the site response.
Because the recorded data from vertical arrays are in the form of
input/output time series, inverse methods were an obvious choice
Site Response Using Empirical Transfer Functions
for investigating the phenomena of earthquake site response.
at Yerba Buena Island
Using system identification, an empirical transfer function 共ETF兲
was developed for the site’s seismic behavior. As an inverse At YBI, the ground motions at the surface and those measured at
method, no assumptions are required about the material properties depth in the borehole are highly coherent, 0.93 and 0.92 for the
共stiffness and damping兲 to produce an estimate of the site transfer Richmond and Bolinas events, respectively. This high degree of
function. Rather the assumptions lie in the assumed parametric correlation indicates that the ground motions from the surface and
form and estimation procedure. The system identification 共SI兲 the downhole location should be linearly related and yield a valid
framework was used for these inversions to insure rapid and con- ETF.
fident convergence to the best-fit model for the soil system A rock reference site should ideally have a flat spectral re-
through optimization and validation criteria set out by Ljung sponse with an amplitude of 1, in order to not bias the soil site
共1987兲 and Bohlin 共1987兲. Previous applications of SI to geotech- response estimate. An estimated ETF at YBI for the Richmond
nical problems include Udwadia 共1985兲, Safak 共1989兲, Glaser event is shown in Fig. 5 along with the predicted surface ground
共1995, 1996兲, Elgamal et al. 共1996兲, Zeghal et al. 共1996兲, Stewart motions. The ETF plotted in Fig. 5 has a peak at 8.6 Hz. The 95%
and Fenves 共1998兲, Stewart et al. 共1999兲, Baise and Glaser confidence intervals shown in the figure indicate the uncertainty
共2000兲, and Glaser and Baise 共2000兲. in the ETF estimate and in the amplification at 8.6 Hz from 3.5 to
The parametric model used for the empirical transfer function 5.5. The peak at 8.6 Hz may correspond to a spectral hole result-
is a complex-valued rational polynomial, referred to as an autore- ing from the interference of upgoing and downgoing waves at 61
gressive moving-average model with exogenous noise, or ARX m depth. The frequency location of a spectral hole due to inter-
model 共Glaser 1995; Ljung 1987兲. The linear SI algorithm used is ference of upgoing and downgoing waves in homogeneous media

418 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003

Fig. 5. Estimated empirical transfer function in frequency domain for YBI uphole/downhole pair and the Richmond event 共top兲. Estimated and
recorded surface displacement ground motions at YBI for empirical transfer function model 共bottom兲.

can be calculated as the shear-wave velocity divided by four times pronounced resonance 共near 1 Hz兲 while the bedrock motion is
the depth of the downhole sensor 共Safak 1997兲. Assuming an primarily composed of the initial body waves. This would indi-
average shear-wave velocity of 1.7 km/s for the rock 共Baise et al. cate that the surface reverberations are trapped energy in the San
2003b兲, the expected spectral hole would occur at 7 Hz. The ETF Francisco Bay sediments.
therefore does not provide clear evidence of site response at the It is helpful to examine the waveform evolution over depth
rock site over the 61 m depth interval as the transfer function is 共shown in Fig. 6兲 for qualitative differences. The ground motions
possibly influenced by the interference between upgoing and in the Pleistocene bay mud 共44 to 31 m兲 are very consistent in
downgoing waves. amplitude and shape, indicating that within this soil, the wave
train does not change. Comparison of the 16 m 共Holocene bay
mud兲 to the 31 m 共Pleistocene bay mud兲 recordings shows a no-
Treasure Island Site Response ticeable amplification. The correlation between these two levels
drops below 0.9 for both events, indicating a subtle change in the
The location of recording stations at TI provided an excellent
wavefield. The change in waveforms as the motions move from
opportunity to study how the different sediment layers affect the
the Holocene bay mud to the fill is less significant but still no-
ground motions. Fig. 6 shows the displacement records from the
ticeable. The two signals recorded in fill have a similar shape.
TI vertical array for the Richmond event. The ground motions are
highly correlated through the soil profile with r between 0.85 and Overall, the wavefield was uniform in the sediments with most of
0.99, indicating the ground motions at adjacent levels in the soil the change at the rock/soil boundary. The site response at TI,
can be related by a linear filter model. The correlation coefficients therefore, was a cumulative response of the soils bounded and/or
for each interval pair are summarized for the two earthquakes in controlled by the soil/rock interference.
Table 2. The only interval with low correlation spans the bedrock/ In order to assess the nature of the late arriving energy present
soil interface 共r⫽0.18 –0.39兲. The breakdown in correlation oc- in the soil records, the three-dimensional particle motion was ex-
curs somewhere between 44 and 104 m depth. Based on our un- amined. Fig. 7 shows the particle motion for two time intervals
derstanding of wave propagation, we assume that the change in recorded at the surface of the TI vertical array during the Rich-
ground motion occurred at the bedrock and sediment interface, mond event. The direction of propagation is indicated on each
with ground motions highly amplified by the impedance contrast. graph. The early arriving shear waves are polarized in the hori-
The waveforms differed significantly in character on either side of zontal plane while the later arriving waves have slightly more
this interface at 44 and 104 m depth 共see Fig. 6兲. The ground elliptical orbits in the vertical plane. The particle motions provide
motions in the sediments for all recorded earthquakes display a only weak evidence towards the existence of surface waves.

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003 / 419

 increased r of 0.70. The shorter windowed input/output pair
would therefore lead to a better estimated transfer function.
Fig. 8共a兲 shows the optimal ETF for downhole rock to surface
at TI, estimated using the first two seconds of displacement mo-
tions at the bedrock and surface level for the Bolinas event. The
first two seconds were chosen to represent the initial shear-wave
arrivals. Although only two seconds of record were used, the site
response is constant and applicable for the entire record assuming
that the site response results from vertically propagating shear
waves. A low-model order was required by the optimization and
resulted in a single fundamental mode at 1.6 Hz. The data did not
support a more complex model. Fig. 8共b兲 compares the surface
motions estimated with this filter to the data. The initial pulse of
motion was matched very well in shape and in amplitude. The
ground motion was not well estimated after the initial arrivals as
indicated by the model residuals shown in Fig. 8共c兲. Because the
rock motion was deficient in energy beyond the direct arrivals and
the ETF only mapped the wavefield from the rock to the surface,
it can be concluded that the late arriving wave energy was not
entering the system from the rock below as vertically propagating
shear waves. An ETF estimated from the TI rock input/TI surface
output will only be able to account for the amplification of the
soil profile, as observed in the initial shear wave arrival but not
the surface wave generation. The ETF estimate residuals shown in
Fig. 9共c兲 correspond to the energy trapped in the sediments alone
共i.e., surface waves, trapped waves, and interface waves兲.
To investigate the possibility of site resonance due to the
strong rock/soil impedance contrast as an alternative explanation
to surface waves, a 1D vertical shear-wave propagation response
in the linear regime 共Ching and Glaser 2001兲 was calculated for
Fig. 6. East-west component displacements for Richmond the TI site. If the late arriving energy was due to 1D resonance
earthquake recorded at Treasure Island vertical array caused by the rock-soil impedance contrast rather than more com-
plex wave propagation such as horizontally propagating surface
waves, 1D wave propagation would be able to predict it. The
Treasure Island Empirical Transfer Function shear-wave velocity profile shown in Fig. 2 was simplified to
The TI site response can be estimated using surface soil and three soil layers over bedrock as shown in Fig. 9. Due to the
nearby rock outcrop sites, or by using borehole reference rock low-intensity ground motions, low damping was assumed in the
sites either beneath TI or YBI. Based on the results of the corre- sediments 共␤⫽0.01兲. Reduced damping did increase the trapped
lation analysis of waveforms recorded in rock at YBI and TI, YBI energy but was not able to replicate the late arriving energy in the
was not considered to be a reliable reference site for TI. The observed records. The resulting site response transfer function is
ground motions recorded in rock beneath the TI site were pre- plotted in Fig. 9共a兲, the predicted surface ground motion is plotted
ferred. As observed in Fig. 6 and Table 2, a major change in the against the actual ground motion in Fig. 9共b兲, and the model
wavefield occurred between rock and soil at the TI site. With a residuals are plotted in Fig. 9共c兲. As shown in the figure, the 1D
low r of 0.18 for the full record of the Richmond event, a linear wave-propagation method matches the first arrivals well but fails
filter will not be able to accurately model the intervening transfer to capture the late arriving surface energy. One-dimensional ver-
function. To improve the chance of finding a reliable transfer tical wave propagation may be able to match direct arrivals at TI
function, we used the Bolinas event 共r⫽0.35兲 to estimate an ETF and therefore intensity measures related to peak acceleration val-
from rock to surface at TI. From a visual inspection of the wave- ues but cannot capture duration and frequency content measures
forms, the early body waves appeared to be more coherent be- of intensity.
tween rock and soil. This was confirmed with the calculated cor- Figs. 9共a and c兲 compare the results from the 1D inverse
relation coefficients. We further windowed the ground motions to method with the 1D forward method. The transfer functions for
two seconds directly around the direct S arrival, resulting in an the two methods identify the same resonant peak at 1.4 Hz for the
forward model and 1.6 Hz in the inverse model although the ETF
is significantly lower in amplitude. The residuals are similar in
Table 2. Correlation Coefficients at Treasure Island amplitude for the late arriving energy indicating a similar lack of
fit. Because surface waves would enter the system horizontally
12/4/98 8/18/99
and are not a part of the bedrock ground motions, they cannot be
Richmond Bolinas
causally captured by any input/output filter mapping.
0–7 m Fill 0.97 0.98
7–16 m Fill to Young Bay Mud 0.96 0.97
16 –31 m Young Bay Mud to Old Bay Clay 0.85 0.79 Surface Waves—3D Õ2D Wave Propagation Effects
31– 44 m Old Bay Clay 0.98 0.98 In order to account for surface waves traveling horizontally in the
44 –104 m Old Bay Clay to Rock 0.18 0.35 sediments, multi-dimensional waveform modeling is required.

420 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003

Fig. 7. Surface displacements at Treasure Island for Richmond earthquake showing particle motions for initial shear waves and later arriving
surface waves. Bold arrows show direction of propagation and small arrows indicate sense of particle motion.

Surface waves are generally a result of a larger regional feature form modeling is discussed in detail in Baise 共2000兲 and Baise
that cannot be described by a single site and a 1D vertical wave- et al. 共2003a兲. We began with a simple layered velocity model
propagation assumption. Most studies of surface waves therefore and drew from the literature to improve the regional velocity
use two-dimensional 共2D兲 and three-dimensional 共3D兲 finite dif- model and waveform fits at several stations. In order to constrain
ference calculations of the wavefield 共e.g., Vidale et al. 1985; the velocity model, we tested the sensitivity of waveforms to
Dreger and Helmberger 1990; Graves 1993; Scrivner and Helm- variations from the simple model. As an inverse problem, con-
berger 1994; Graves 1998; Stidham et al. 1999; Olsen et al. straining the model space is necessary to prevent model conver-
2000兲. gence to a physically unrealistic local minima. Matching the ab-
solute timing and amplitude of phases helps to retain realistic
models. We realize that trial and error inverse modeling is not an
Methods of Waveform Modeling
effective methodology for determining velocity structure. We,
We modeled the waveforms at TI and YBI following Vidale et al. therefore, used waveform modeling to understand the sensitivity
共1985兲 and Dreger and Helmberger 共1990兲. The forward wave- of the wavefield to the basin velocity structure and to estimate a

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003 / 421

Fig. 8. 共a兲 Frequency domain representation of optimized empirical
transfer function for TI site from rock to surface with 95% confidence
intervals; 共b兲 Predicted surface motions; and 共c兲 model residuals Fig. 9. 共a兲 Frequency domain representation of forward wave propa-
gation model transfer function for TI site from rock to the surface
共ETF is shown as dotted line for comparison兲; 共b兲 predicted surface
motions; and 共c兲 model residuals 共ETF model residuals are shown as
possible regional response to ground shaking as a result of the dotted line for comparison兲
San Francisco Bay sediments. Our hypothesis was that a 2D
structure in the sediments could explain the late arriving energy in
the observed ground motions. The modeling effort used realistic The FK method also considers wave-energy dissipation
velocity models drawn from the literature, varying parameters through an elastic wave propagation. The amplitude spectra for
within reasonable ranges and was careful when drawing conclu- the 1D model at epicentral distances of 14 and 30 km 共appropriate
sions on the velocity structure of the region. for the study兲 using a Q ␣ ⫽600 and Q ␤ ⫽300 共appropriate for
We modeled the particle displacement time histories at TI and hard rock兲 were compared to amplitude spectra in which the
YBI bandpass filtered using a zero-phase Butterworth filter with upper layer was assigned a Q ␣ ⫽Q ␤ ⫽50 共average value of
corner frequencies at 0.02 to 2.0 Hz. The lower frequencies were Johnson and Silva 1981兲. The results show that consideration of
matched first to constrain the average velocity model. Then higher low Q in the upper 200 m damps the modes associated with
frequencies were matched by varying the smaller scale basin reverberations in the upper layer at frequencies above 2 Hz. In the
structure. The 2 Hz bound was controlled by the grid spacing of 0.02 to 2 Hz passband, the effect of low Q is almost negligible;
the finite difference models 共20 m兲 and is a low resolution for therefore, damping was ignored in the analysis. Of course when
engineering interest; therefore, a future study will be required to higher frequencies and shallower structures are considered in fu-
investigate the TI wavefield at higher frequencies and develop a ture studies, Q will have a profound effect.
more complete view of TI site response and seismic hazard. Once the ‘‘best-fit’’ 1D structure was determined for the re-
The tools used for this methodology are waveform modeling gion, 2D basin structure was added using a FD waveform mod-
using a frequency-wave number integration scheme 共FK兲 for 1D eling procedure described by Vidale et al. 共1985兲. The San Fran-
velocity structures and finite different 共FD兲 waveform modeling cisco Bay was included as a shallow 2D structure. The amplitudes
for 2D velocity structures. A FK computer code by Saikia 共1994兲 of the synthetics were determined by application of the source
was used to generate Green’s functions for different 1D velocity time function and multiplication of the synthetics by the moment
models, source locations, and epicentral distances. The compo- of the earthquake. Because of the difficulty of tracking individual
nent Green’s functions were combined according to the specific phases with this method, it did not account for attenuation. The
earthquake focal mechanism to create synthetic seismograms. The short paths in this study, however, were not particularly sensitive
synthetic seismograms were also convolved with a source time to the damping in the 0.02 to 2 Hz passband as determined using
function to account for the source rise time. Many models were the FK method and 1D structure.
simulated to test the sensitivity of waveforms to layer thicknesses
and velocities. All three components of motion were generated, Waveform Modeling
and the modeling focused on first matching the tangential com- Although the San Francisco Bay sediments are not deep 共91 m at
ponent and secondly the radial component. TI兲 as compared to some of the major sedimentary basins 共i.e.,

422 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003

Los Angeles Basin兲, they appear to produce a surface-wave train.
Based on the velocities and layer thicknesses reported in other
San Francisco Bay studies, the San Francisco basin was modeled
as a single 100 m thick layer across the bay, and a range of
sediment shear wave velocities from 150 to 500 m/s was tested.
The preferred velocity structure was chosen based on the best fit
for the absolute timing of the wave packet arrivals and the appro-
priate phase, shape, duration, and amplitude of the surface waves
at TI. Although the fill and Holocene bay mud in the upper 20 m
were expected to have lower velocities (V s ⫽150 to 250 m/s兲, the
regional surface waves 共0.02–2 Hz兲 were adequately modeled by
a higher-uniform average velocity (V s ⫽400 m/s兲 in a 100 m sedi-
ment layer as compared to a more realistic multiple layered
model. This unexpected result most likely was caused by the
coarse grid spacing of 20 m which could not adequately represent
the more realistic upper portions of the layered model. The aver-
age uniform layer (V s ⫽400 m/s兲 is more heavily influenced by
the deeper faster layers. A weighted shear-wave velocity average
of the upper 100 m at Treasure Island is 320 m/s.
The modeling exercise indicated that the surface waves ob-
served at TI could be produced by the large impedance contrast
between the low-velocity sediments (V s ⫽400 m/s兲 and the
weathered Franciscan bedrock 共1.5 km/s兲 in a shallow basin. Sur-
face waves initially formed at the basin edges along the east coast
of the Marin Peninsula 共Fig. 1兲, resulting in trapped energy at a
resonant frequency near 1 Hz, similar to that observed in the data.
Fig. 10 compares the best-fit, 2D model, tangential component
synthetics, and observed ground motions for several levels in the Fig. 10. Synthetic calculations for tangential component of motion
vertical array at TI. As seen in the figure, the comparisons are during Bolinas earthquake at TI vertical array
consistent for all depths, matching the shape of the first 10 s of
motion well. The amplitude of the synthetics overpredict the pri-
mary S-wave arrival amplitude by 30% which may result from an consistent in amplitude and shape to the corresponding observed
overestimate of the earthquake moment or errors in the model. waveforms. In the model, the bay sediments pinch out at the edge
The synthetics at 40 m depth do not match the data at 44 m depth of YBI, 1 km from the site instruments, and this effectively pre-
as well as at the surface. The simulated primary S wave is a single vented the surface waves from propagating to YBI as observed at
pulse whereas the observed primary S wave is separated into the site. The data and synthetics are simplified at YBI as com-
upgoing and downgoing waves. These differences indicate that pared to TI indicating that the model produces a similar relative
the shear-wave velocity in the model is faster than the actual response. Dreger and Helmberger 共1990兲 have shown that a basin
material in the upper 40 m of sediments, as expected. The upper margin produces this filtering effect where high frequencies are
40 m of sediments at TI include Holocene deposits and fill with preferentially removed as energy is converted to diving body
shear-wave velocities between 150 and 250 m/s instead of the 400 waves.
m/s in the model.
The sensitivity study of the reported model indicated that the
synthetics could be improved with further adjustments to the Discussion
model; however, the modeling effort was not made since such
detail is too fine for the 20 m grid spacing that was used and a Many researchers have used coherency of data to statistically as-
detailed velocity model was not the desired product of this study. sess the linear relationship between different time series. The pri-
Rather, this study set out to provide an explanation for the site mary use in site response analyses has been to evaluate the spatial
response 共observed resonance兲 at TI. The important achievement coherence of a wavefield 共Menke et al. 1990; Hough and Field
of this model was the production of late arriving 1 Hz energy 1996兲. Hough and Field 共1996兲 found that waveform coherence
similar to the data which was not produced in the 1D models. estimates from earthquakes in the San Fernando Valley support
Although the model residuals are still of a similar magnitude to the general conclusion that a site response estimate is an adequate
the 1D models as a result of timing errors and phase shifts, the representation of expected site response over a region several
synthetic ground motions produce the duration and frequency kilometers in diameter if the local geology is consistent. As com-
content better than the 1D methods. The synthetics at 40 and 120 pared to work in rock regions 共Menke et al. 1990兲, Hough and
m depth also match the data 共at 44 and 122 m depth兲 in shape and Field 共1996兲 concluded that sedimentary basins may possess more
amplitude. These fits indicated that the model results capture the coherent wavefields as a result of the resonant behavior of the
wave propagation from the rock to the sediments. Specifically, the sediments. Therefore, analysis of the correlation of waveforms is
surface waves in the model were suppressed below 100 m depth a valuable tool for site response. Specifically, we have used the
共the rock/soil boundary in the model兲. correlation coefficient to assess the reliability of site response
The 2D San Francisco Bay model was also verified for the estimates from input/output rock/soil pairs. Correlation coeffi-
YBI site. The estimated and observed surface displacements at cients can be used to determine appropriate reference rock sites
YBI and TI are shown in Fig. 11. The synthetics for both sites are for site response studies. Our conclusion that YBI is an inappro-

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003 / 423

 rock sites were coherent and of a short duration. The soft soil site
ground motions, on the other hand, were longer in duration and
higher in amplitude, indicating a strong resonance with a period
near 1.5 s. This resonance was not observed at TI during the
Loma Prieta earthquake due to the liquefaction of the fill; how-
ever, prior to signs of liquefaction in the TI record, the S-wave
packet was phase coherent with the recordings at the Oakland
sites 共Hanks and Brady 1991兲. If the surface waves were propa-
gating throughout the sediments, they would be felt during strong
motion at non-liquefiable sites in the region. Therefore, we be-
lieve that the surface waves observed at TI during weak motion
earthquakes and documented in the literature at other sites will be
a seismic hazard at nonliquefiable sites around the margins of the
San Francisco Bay in future large earthquakes.

Conclusion

The site response of TI was extensively studied after the 1989

Loma Prieta earthquake. Most studies used the YBI site as a rock
reference site to estimate the incoming energy below TI. The
Treasure Island site response was evaluated using 1D equivalent
Fig. 11. Tangential displacements at TI and YBI for Bolinas event linear vertical wave propagation to conclude that soil nonlinearity
plotted against synthetics for best-fit 2D model. Waveforms are fil- had occurred. Since liquefaction was known to have occurred,
tered to a passband below 2 Hz. soil nonlinearity was not a surprise. Due to an improved data set
including ground motions recorded at depth beneath TI and YBI,
we have reevaluated the site response at TI focusing on the linear
range. We found the YBI surface reference site to be an inappro-
priate reference site for TI runs counter to many previous studies
priate site input signal because of low correlation between the
which relied on this site pair for site response investigations of the
rock ground motions at YBI and TI even in the linear range of
Long Prieta earthquake. As a comparison we calculated the cor-
ground response. TI site response was therefore reevaluated using
relation coefficient for the YBI/TI pair during the Loma Prieta
at-depth ground motions recorded at the TI vertical array as the
earthquake. For 20 s of displacement motion 共east/west兲, r was
input reference motion.
0.41. When the records were cut to the direct arrivals 共4 s兲 of
The linear site response at TI includes horizontally propagat-
motion, r increased to 0.46. In the Loma Prieta case, the correla-
ing surface waves trapped in the San Francisco Bay sediments
tion was lower than that observed over the direct arrivals 共2 s兲
during the Bolinas event 共r⫽0.70兲 in the TI vertical array. As a that cannot be captured by 1D wave propagation or inverse meth-
result of low correlation indicating system complexity, both linear ods. The surface waves were identified by examination of particle
estimates of site response transfer functions should be considered motions, and subsequently modeled using 2D finite difference
unreliable. At best, the estimated site response will characterize 共FD兲 waveform modeling. The FD model including a 100 m
the direct shear-wave response but not the complex site response. depth sedimentary basin with a constant shear-wave velocity of
During Loma Prieta, the TI site liquefied. During the Bolinas 400 m/s over a 1.5 km/s layer of Franciscan bedrock successfully
event, the TI site experienced upwards of 20 s of surface waves. trapped energy in the basin and produced ringing in the synthetics
Neither of these complexities can be captured by linear transfer similar to that observed in the TI data. This coarse model provides
function site response methods. both an example of the possible velocity structure that could
The seismic behavior of the San Francisco Bay region has cause the observed ground motions and an indication of the seis-
been extensively investigated in the past. The investigations pri- mic hazard of surface waves in the San Francisco Bay area, es-
marily used aftershock and other weak ground motion, and all pecially at sites in and around the margins of the bay.
concluded that bay sediments and the associated locally generated As a result of surface waves at TI, any 1D model will not be
surface waves control the waveforms 共Johnson and Silva 1981; able to fully characterize the site response, and the optimal em-
Boatwright 1991; Graves 1993兲. We looked specifically at the pirical transfer function from rock to the surface could only pre-
Treasure Island site response and identified a surface wave reso- dict the initial two seconds of motion accurately. After two sec-
nance near 1 Hz, similar to the previous studies in the region. onds, the empirical transfer function and the forward model fail to
During their studies of the Marina district, Graves 共1993兲 and produce the resonance observed in the surface ground motions.
Boatwright 共1991兲 also observed amplification at around 1 Hz in The greatest change in the wavefield, outside of the rock-to-soil
the Loma Prieta aftershock waveforms. transition, occurred between the Pleistocene and the Holocene
The weak motion surface recordings from the TI vertical array sediments. Overall, the waveforms change very little as they
reveal surface waves that are not evident at depth in the bedrock. propagate through the sediments, indicating that the site response
The question becomes how does this weak motion observation is a cumulative effect of the entire soil structure and not a result
carry over to strong motion, and therefore to regional earthquake of individual soil layers. This is further supported by the FD
hazard assessment? Using ground motions recorded at two rock waveform modeling results which accurately predicted the sur-
sites in San Francisco, YBI, TI, and three soil sites in Oakland, face wavetrain at TI with a 100 m thick averaged uniform sedi-
Hanks and Brady 共1991兲 found that the ground motions at the mentary basin.

424 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003

Acknowledgments Gersch, W., and Kitagawa, G. 共1985兲. ‘‘A time varying AR coefficient
model for modeling and simulating earthquake ground motion.’’
The writers would like to thank David Boore and an anonymous Earthquake Eng. Struct. Dyn., 13, 243–254.
reviewer for their helpful and thorough reviews of the manuscript. Gibbs, J. F., Fumal, T. E., Boore, D. M., and Joyner, W. B. 共1992兲.
Their insights helped improve the paper significantly. ‘‘Seismic velocities and geologic logs from borehole measurements at
seven strong-motion stations that recorded the Loma Prieta earth-
quake.’’ U. S. Geological Survey Open-File Rep. No. 92-287, Menlo
References Park, Calif., 139.
Archuleta, R., Steidl, J. H., and Bonilla, L. F. 共2000兲. ‘‘Engineering in- Glaser, S. D. 共1995兲. ‘‘System identification and its applications to esti-
sights from data recorded on vertical arrays.’’ 12th World Conf. in mating soil properties.’’ J. Geotech. Eng., 121共2兲, 553–560.
Earthquake Engineering, Auckland, New Zealand. Glaser, S. D. 共1996兲. ‘‘Insight into liquefaction by system identification.’’
Baise, L. G. 共2000兲. ‘‘Modeling of the northern San Francisco bay veloc- Geotechnique, 46共4兲, 641– 656.
ity structure for the 18 August 1999 Bolinas earthquake.’’ Masters Glaser, S. D., and Baise, L. G. 共2000兲. ‘‘System identification estimation
thesis, Dept. of Geology and Geophysics, Univ. of California, Berke- of soil properties at the Lotung site.’’ Soil Dyn. Earthquake Eng., 19,
ley, Calif. 521–531.
Baise, L. G., Dreger, D. S., and Glaser, S. D. 共2003a兲. ‘‘Modeling of the Graves, R. W. 共1993兲. ‘‘Modeling three-dimensional site response effects
Northern San Francisco Bay velocity structure for the 18 August 1999 in the Marina district basin, San Francisco, California.’’ Bull. Seismol.
Bolinas earthquake.’’ Bull. Seismol. Soc. Am., in press. Soc. Am., 83共4兲, 1042–1063.
Baise, L. G., and Glaser, S. D. 共2000兲. ‘‘Consistency of ground-motion Graves, R. W. 共1998兲. ‘‘Three-dimensional finite-difference modeling of
estimates made using system identification.’’ Bull. Seismol. Soc. Am., the San Andreas Fault; source parameterization and ground-motion
90共4兲, 993–1009. levels.’’ Bull. Seismol. Soc. Am., 88共4兲, 881– 897.
Baise L., Hutchings, L., and Glaser, S. 共2003b兲. ‘‘Site response at Yerba Hanks, T. C., and Brady, A. G. 共1991兲. ‘‘The Loma Prieta earthquake,
Buena Island, San Francisco Bay, California analyzed with weak mo- ground motion, and damage in Oakland, Treasure Island, and San
tion recordings.’’ BOLLETTINO DI GEOFISICA Teorica ed Appli- Francisco.’’ Bull. Seismol. Soc. Am., 81共5兲, 2019–2047.
cata: Special Issue on ‘‘Site response estimation from observed Hough, S. E., and Field, E. H. 共1996兲. ‘‘On the coherence of ground
ground motion data,’’ Instituto Nazionale di Oceanografia e di Geofi- motion in the San Fernando Valley.’’ Bull. Seismol. Soc. Am., 86共6兲,
sica Sperimentale–OGS, Trieste, Italy, 43共3– 4兲, in press. 1724 –1732.
Bendat, J. S., and Piersol, A. G. 共1993兲. Engineering applications of Hubral, P., Treitel, S., and Gutowski, P. R. 共1980兲. ‘‘A sum autoregressive
correlation and spectral analysis, 2nd Ed., Wiley, New York. formula for the reflection response.’’ Geophysics, 45共11兲, 1697–1705.
Boatwright, J. 共1991兲. ‘‘Ground motion amplification in the Marina dis- Hutchings, L., Kasameyer, P., Foxall, W., Hollfelder, J., Turpin, C.,
trict.’’ Bull. Seismol. Soc. Am., 81共5兲, 1980–1997. Hayek, C., McEvilly, T., Clymer, R., Uhrhammer, R., and Jarpe, S.
Bohlin, T. 共1987兲. ‘‘Model validation.’’ Encyc. of sys. and cont., M. 共1999兲. ‘‘Deep borehole instrumentation along San Francisco Bay
Singh, ed. Pergamon, London. Bridges.’’ Rep. No. UCRL-ID-132137, Lawrence Livermore National
Boore, D. M., and Joyner, W. B. 共1997兲. ‘‘Site amplification of generic Laboratory, Livermore, Calif., March 1, 1999.
rock sites.’’ Bull. Seismol. Soc. Am., 87共2兲, 327–341. Idriss, I. M. 共1990兲. ‘‘Response of soft soils during earthquakes.’’ H.
Borcherdt, R. D. 共1970兲. ‘‘Effects of local geology on ground motion near Bolton seed volume 2 memorial symposium proceedings, J. Michael
San Francisco Bay.’’ Bull. Seismol. Soc. Am., 60共1兲, 29– 61. Duncan, ed., Bitech, Vancouver, B.C.
Borcherdt, R. D., and Gibbs, J. F. 共1976兲. ‘‘Effects of local geological Jarpe, S. P., Hutchings, L. J., Hauk, T. F., and Shakal, A. F. 共1989兲.
conditions in the San Francisco Bay region on ground motions and the
‘‘Selected strong- and weak-motion data from the Loma Prieta Se-
intensities of the 1906 earthquake.’’ Bull. Seismol. Soc. Am., 60共2兲,
quence.’’ Seismol. Res. Lett., 60, 167–176.
467–500.
Johnson, L. R., and Silva, W. 共1981兲. ‘‘The effects of unconsolidated
Ching, J. Y., and Glaser, S. D. 共2001兲. ‘‘1D time-domain solution for
sediments upon the ground motion during local earthquakes.’’ Bull.
seismic ground motion prediction.’’ J. Geotech. Geoenviron. Eng.,
Seismol. Soc. Am., 71共1兲, 127–142.
127共1兲, 36 – 47.
Lee, C. H. 共1969兲. ‘‘Treasure Island fill.’’ Bay mud developments case
Claerbout, J. 共1968兲. ‘‘Synthesis of a layered medium from its acoustic
histories, C. Lee and U. Praszker, eds., California Division of Mines
transmission response.’’ Geophysics, 33, 264 –269.
and Geology, 69–72.
Dargahi-Noubary, G. R. 共1999兲. Time series with applications to seismol-
ogy, Nova Science, New York. Ljung, L. 共1987兲. System identification, Prentice-Hall, Englewood Cliffs,
Darragh, R. B., and Shakal, A. F. 共1991兲. ‘‘The site response of two rock N.J.
and soil station pairs to strong and weak ground motion.’’ Bull. Seis- Margheriti, L., Wennerberg, L., and Boatwright, J. 共1994兲. ‘‘A compari-
mol. Soc. Am., 81共5兲, 1885–1899. son of Coda and S wave spectra ratios as estimates of site response in
de Alba, P., Benoit, J., Pass, D. G., Carter, J. J., Youd, T. L., and Shakal, the Southern San Francisco Bay Area.’’ Bull. Seismol. Soc. Am., 84,
A. F. 共1994兲. ‘‘Deep instrumentation array at the Treasure Island 1815–1830.
Naval Station.’’ Loma Prieta earthquake of October 17, 1989: Strong Menke, W., Lerner-Lam, A. L., Dubendorff, B., and Pacheco, J. 共1990兲.
ground motion and ground failure, R. D. Brocherdt, ed., USGS Prof. ‘‘Polarization and coherence of 5 to 30 Hz seismic wave fields at a
Paper P 1551-A, A155-A168, United States Geological Survey, Den- hard-rock site and their relevance to velocity heterogeneities in the
ver. crust.’’ Bull. Seismol. Soc. Am., 80共2兲, 430– 449.
Dreger, D. S., and Helmberger, D. V. 共1990兲. ‘‘Broadband modeling of Olsen, K. B., Nigbor, R., and Konno, T. 共2000兲. ‘‘3D viscoelastic wave
local earthquakes.’’ Bull. Seismol. Soc. Am., 80共5兲, 1162–1179. propagation in the upper Borrego Valley, California, constrained by
Elgamal, A.-W., Zeghal, M., Parra, E., Gunturi, R., Tang, H. T., and borehole and surface data.’’ Bull. Seismol. Soc. Am., 90共1兲, 134 –150.
Stepp, J. C. 共1996兲. ‘‘Identification and modeling of earthquake Robinson, E. A., and Treitel, S. 共1978兲. ‘‘The fine structure of the normal
ground response—I. Site amplification.’’ Soil Dyn. Earthquake Eng., incidence synthetic seismogram.’’ Geophys. J. R. Astron. Soc., 53,
15, 499–522. 289–309.
Ellis, G. W., and Cakmak, A. S. 共1991兲. ‘‘Effect of spatial variability on Rollins, K. M., McHood, M. D., Hryciw, R. D., Homolka, M., and Sh-
ARMA modeling of ground motion.’’ Struct. Safety, 10共N1-3兲, 181– ewbridge, S. E. 共1994兲. ‘‘Ground response on Treasure Island.’’ Loma
191. Prieta earthquake of October 17, 1989: Strong ground motion and
Field, E. H., Jacob, K. H., and Hough, S. E. 共1992兲. ‘‘Earthquake site ground failure, R. D. Borcherdt, ed., USGS Prof. Paper P 1551-A,
response estimation a weak-motion case study.’’ Bull. Seismol. Soc. A155-A168.
Am., 82共6兲, 2283–2307. Safak, E. 共1988兲. ‘‘Analysis of recordings in structural engineering: adap-

JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003 / 425

 tive filtering, prediction and control.’’ U. S. Geological Society Open Stewart, J. P., Fenves, G. L., and Seed, R. B. 共1999兲. ‘‘Seismic soil-
File Rep. No. 88-647, Menlo Park, Calif. structure interaction in buildings. I: Analytical methods.’’ J. Geotech.
Safak, E. 共1989兲. ‘‘Optimal-adaptive filters for modelling spectral shape, Geoenviron. Eng., 125共1兲, 26 –37.
site amplification, and source scaling.’’ Soil Dyn. Earthquake Eng. Stidham, C., Antolik, M., Dreger, D., Larsen, S., and Romanowicz, B.
8共2兲, 75–95. 共1999兲. ‘‘Three-dimensional structure influences on the strong-
Safak, E. 共1997兲. ‘‘Models and methods to characterize site amplification motiono wavefield of the Loma Prieta earthquake.’’ Bull. Seismol.
from a pair of records.’’ Earthquake Spectra, 13共1兲, 97–129. Soc. Am., 89共5兲, 1184 –1202.
Saikia, C. K. 共1994兲. ‘‘Modified frequency-wavenumber algorithm for Udwadia, F. E. 共1985兲. ‘‘Some uniqueness results related to soil and
regional seismograms using Filon’s quadrature: Modeling of Lg building structural identification.’’ SIAM (Soc. Ind. Appl. Math.) J.
waves in eastern North America.’’ Geophys. J. Int., 118, 142–158. Appl. Math., 45共4兲, 674 – 685.
Scrivner, C. W., and Helmberger, D. V. 共1994兲. ‘‘Seismic waveform mod- University of California at Berkeley/Earthquake Engineering Research
eling in the Los Angeles basin.’’ Bull. Seismol. Soc. Am., 84共5兲, 1310– Center 共UCB/EERC-90/05兲. 共1990兲. ‘‘Preliminary report on the prin-
1326. cipal geotechnical aspects of the October 17, 1989 Loma Prieta earth-
Seed, R. B., Dickenson, S. E., and Idriss, I. M. 共1991兲. ‘‘Principal geo- quake.’’ Rep., Earthquake Engineering Research Center, Univ. of
technical aspects of the 1989 Loma Prieta earthquake.’’ Soils Found., California, Berkeley, Calif.
31共1兲, 1 26. Vidale, J. E., Helmberger, D. V., and Clayton, R. W. 共1985兲. ‘‘Finite-
Steidl, J., Tumarkin, A. G., and Archuleta, R. 共1996兲. ‘‘What is a refer- difference seismograms for SH waves.’’ Bull. Seismol. Soc. Am.,
ence site?’’ Bull. Seismol. Soc. Am., 86, 1733–1748. 75共6兲, 1765–1782.
Stewart, J. P., and Fenves, G. L. 共1998兲. ‘‘System identification for evalu- Zeghal, M., Elgamal, A. W., and Parra, E. 共1996兲. ‘‘Identification and
ating soil-structure interaction effects in buildings from strong motion modeling of earthquake ground response — II. Site liquefaction.’’ Soil
recordings.’’ Earthquake Eng. Struct. Dyn., 27, 869– 885. Dyn. Earthquake Eng., 15, 523–547.

426 / JOURNAL OF GEOTECHNICAL AND GEOENVIRONMENTAL ENGINEERING © ASCE / MAY 2003

View publication stats