Nonlinear Site Amplification as Function

of 30 m Shear Wave Velocity

Yoojoong Choi,a) M.EERI, and Jonathan P. Stewart,b) M.EERI

We develop empirical relationships to predict nonlinear (i.e., amplitude-

dependant) amplification factors for 5% damped response spectral
acceleration as a continuous function of average shear wave velocity
in the upper 30 m, Vs-30 . We evaluate amplification factors as residuals
between spectral accelerations from recordings and modified rock attenuation
relationships for active regions. Amplification at low- and mid-periods
is shown to increase with decreasing Vs-30 and to exhibit nonlinearity that
is dependent on Vs-30 . The degree of nonlinearity is large for NEHRP
Category E (Vs-30⬍180 m/s) but decreases rapidly with Vs-30 , and is small
for Vs-30⬎⬃300 m/s. The results can be used as Vs-30-based site factors
with attenuation relationships. The results also provide an independent
check of site factors published in the NEHRP Provisions, and apparent
bias in some of the existing NEHRP factors is identified. Moreover,
the results provide evidence that data dispersion is dependent on Vs-30 .
[DOI: 10.1193/1.1856535]

INTRODUCTION
Most modern U.S. seismic design codes for building structures represent seismic de-
mand in terms of 5%-damped response spectral ordinates. These spectral ordinates are
affected by seismic source, travel path, and site response effects. In the NEHRP Recom-
mended Provisions for Seismic Regulations for New Buildings and Other Structures,
Part 1: Provisions and Part 2: Commentary (BSSC 2001), source and path effects are
accounted for in maps showing the results of probabilistic seismic hazard analyses
(PSHA) for the United States (Frankel et al. 2000) and so-called maximum considered
earthquake (MCE) maps, which are modified from PSHA maps using deterministic seis-
mic hazard analyses (DSHA) in areas of large hazard by consensus judgment (Leyen-
decker et al. 2000). These maps are prepared for a particular site condition referred to as
the reference site condition. In the NEHRP Provisions, site condition is generally pa-
rameterized on the basis of the average shear wave velocity in the upper 30 m of the site
(Vs-30), which is defined as the ratio of 30 m to the vertical shear-wave travel time
through the upper 30 m of the site. The Vs-30-based site categories in the NEHRP Pro-
visions are given in Table 1. An exception to the Vs-30 criteria is made for soft clays (de-
fined as having undrained shear strength ⬍24 kPa, plasticity index ⬎20, and water con-

a)
Staff Engineer, GeoPentech, 601 N. Parkcenter Drive, #210, Santa Ana, CA 92705
b)
Associate Professor, Civil & Environmental Engineering Dept., University of California, Los Angeles,
CA 90095

1
Earthquake Spectra, Volume 21, No. 1, pages 1–30, February 2005; © 2005, Earthquake Engineering Research Institute
2 Y. CHOI AND J. P. STEWART

Table 1. Site categories in NEHRP Provisions (Martin 1994)

NEHRP Mean Shear Wave

Category Description Velocity to 30 m
A Hard rock ⬎1500 m/s
B Firm to hard rock 760–1500 m/s
C Dense soil, soft rock 360–760 m/s
D Stiff soil 180–360 m/s
E Soft clays ⬍180 m/s
F Special study soils, e.g., liquefiable soils, sensitive
clays, organic soils, soft clays ⬎36 m thick

tent ⬎40%), for which category E is assigned if the thickness of soft clay exceeds 3 m
regardless of Vs-30 . The reference site condition for which the PSHA maps are intended
to apply is the B-C boundary, or Vs-30⫽760 m/s.
The effects on spectral ordinates of site conditions that deviate from the reference
velocity are accounted for with site factors that are a function of site category and the
amplitude of shaking for the reference site condition (Dobry et al. 2000). The site fac-
tors given in the NEHRP Provisions are plotted in Figure 1. By definition, site factors
represent the ratio of spectral ordinates for a particular site condition to the value of the
ordinates that would be expected for the reference condition. The specific factors given
in the provisions are Fa , which is defined over a low-period range (T⫽0.1–0.5 s), and
Fv , which is defined over a mid-period range (T⫽0.4–2.0 s). The ground motion pa-
rameters for the reference site condition that are used in conjunction with site factors are
T⫽0.2 s spectral acceleration (Sa) for Fa (denoted Ss) and Sa at T⫽1.0 s for Fv (denoted
S1). When the design ground motions are estimated as the product of the amplification
factors given in Figure 1 and spectral ordinates Ss or S1 derived from PSHA, two implicit
assumptions are being made: (1) the amplification factor defines the ratio of the median
ground motion amplitude on the subject site condition to the median amplitude on the
reference site condition, and (2) the data dispersion within the two site categories is

Figure 1. Site factors Fa and Fv given in the NEHRP Provisions (BSSC 2001).
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 3

identical. The former assumption is correct as long as both distributions are lognormally
distributed, while the accuracy of the second assumption is investigated subsequently in
the paper.
One important element of PSHA or DSHA is the attenuation relationship used to
evaluate the probabilistic distribution of a given spectral ordinate given that an earth-
quake with particular source characteristics (e.g., moment magnitude, focal mechanism)
has occurred at a particular distance from the site. The output of an attenuation model
applies only for a particular site condition (i.e., the average site condition at the strong
motion accelerometers that produced the data used to derive the attenuation relation),
and hence PSHA/DSHA results also apply only for the average site condition in the at-
tenuation model. It follows from the above that since the B-C boundary is the reference
condition that the NEHRP PSHA and MCE maps are intended to apply for, the attenu-
ation relations used in the hazard analyses should also be appropriate for this site
condition. Unfortunately, this is not the case. The attenuation relations used to develop
the PSHA maps for T⬎0 s spectral ordinates (i.e., not peak acceleration) in the 2000
version of the NEHRP Provisions are Boore et al. (1997) and Sadigh et al. (1997). The
Boore et al. relation can be implemented directly for the B-C boundary because site
condition is parameterized by Vs-30 . However, the ground motions used to define
the rock attenuation model by Sadigh et al. were recorded primarily at rock and shallow
(⬍20 m) soil sites in California, most of which have Vs-30 values significantly less
than 760 m/s. In fact, a borehole compilation by Silva et al. (1997) for this particular
‘‘rock’’ site condition found the median value of Vs-30 to be approximately 520 m/s.
A similar compilation by Boore et al. (1997) found an average velocity for rock sites of
about 620 m/s.
Given the above, the fact that the NEHRP PSHA and MCE maps were derived with
these relations (with equal weight given to each) suggests that the actual reference site
condition is not the assumed value of 760 m/s, but actually corresponds to a softer con-
dition. The hazard analyses underlying the 2003 maps were expanded to include the
Abrahamson and Silva (1997) and Campbell and Bozorgnia (2003) attenuation relations,
although these relations were also developed for site categories inconsistent with the
NEHRP B-C boundary.
In this paper, we develop statistical models for site factors that are a function of Vs-30
and the amplitude of shaking on the reference site condition. The models are developed
from statistical analyses of residuals between recorded ground motions in active regions
and reference motion predictions developed using modified rock attenuation relation-
ships. The models are useful as follows:
1. to validate the existing NEHRP site factors (which were developed based on
both observation and analysis, as discussed further below);
2. as site terms for use with attenuation relations;
3. to identify variations in data dispersion with magnitude, distance, and Vs-30 ;
and
4. to develop correction factors that can be used to adjust the predictions of at-
4 Y. CHOI AND J. P. STEWART

tenuation models (i.e., Abrahamson and Silva 1997, Campbell and Bozorgnia
2003, Sadigh et al. 1997) to the NEHRP-assumed reference condition
(Vs-30⫽760 m/s).
The third item above is important because the NEHRP PSHA maps are based on the
dispersion estimated from attenuation relations (which is generally independent of site
condition). Values of dispersion for specific site conditions that depart significantly from
those in the attenuation relations would imply that the mapped PSHA spectral ordinates
are biased for those site conditions. The fourth item above is important for the develop-
ment of PSHA maps applicable to the NEHRP B-C site condition. Application of cor-
rection factors has been discussed in past committee deliberations, but has not yet been
carried out for the Abrahamson and Silva, and Sadigh et al. attenuation functions (the
Campbell and Bozorgnia results were corrected using the linear site factor model of
Boore et al. 1997; K. Campbell 2003 pers. comm.).

EXISTING NEHRP AMPLIFICATION FACTORS

The NEHRP site factors shown in Figure 1 are based on both empirical data analysis
and the results of ground response analyses (Dobry et al., 2000). The empirical studies
were performed by Borcherdt and Glassmoyer (1994), Borcherdt (1994), and Joyner
et al. (1994) using strong motion data recorded in the San Francisco Bay Area during the
1989 Loma Prieta earthquake, and provide amplification factors (Fa and Fv) that apply
for relatively weak levels of shaking (peak horizontal acceleration for reference [rock]
site condition, PHAr⬇0.1 g). These amplification factors were derived using a so-called
reference site approach, in which the amplification is defined as the ratio of Fourier
spectral ordinates of motions recorded on soil to those recorded on nearby reference
rock sites, with appropriate corrections for variations in site-source distance between the
two accelerometers. The analytical studies consisted of one-dimensional equivalent lin-
ear and nonlinear ground response analyses by Dobry et al. (1994) and Seed et al.
(1994), and were used to extend the Fa and Fv values to PHAr⬇0.4–0.5 g. For both the
empirical and analytical studies, site factors were defined relative to a competent rock
site condition, which in the San Francisco Bay Area corresponds specifically to Fran-
ciscan formation bedrock of Cretaceous and Jurassic age.
Since the adoption of the site factors in Figure 1, a number of studies have
investigated the adequacy of the NEHRP factors by comparing them to alternative
factors derived using non–Loma Prieta strong motion data sets (e.g., from Northridge
recordings [Borcherdt 2002a, b], numerous southern California earthquakes [Harmsen
1997, Field 2000, Steidl 2000], and strong motion databases for active
regions [Joyner and Boore 2000, Stewart et al. 2003]). Work has also been performed
using alternative site categorization schemes (Rodriguez-Marek et al. 2001, Stewart
et al. 2003), although this is not discussed here as the present focus is on Vs-30-based site
categories.
Borcherdt (2002a, b) investigated amplification levels within NEHRP categories us-
ing recordings from the 1994 Northridge earthquake, mostly from stiff soil and soft rock
sites. A reference site approach was used to define amplification factors, with reference
motions taken from local stations with metamorphic rock (e.g., weathered granite,
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 5

gneiss) or sedimentary rock (in which case amplification factors were adjusted so that
the effective reference site condition is relatively firm rock). Average Northridge ampli-
fication factors were found to match very well with the NEHRP amplification factors at
both small periods (Fa) and at longer periods (Fv). The Northridge results also demon-
strated decreasing amplification with increasing reference motion amplitude, an effect
that had not been observed from the Loma Prieta recordings. This effect was not ob-
served in Loma Prieta because most recordings sites are at large distances from the
source, so that PHAr values were small.
The work by Harmsen (1997) involved the evaluation of amplification factors within
NEHRP categories using data from multiple southern California earthquakes normalized
relative to a single reference rock site (Caltech Seismic Lab). A number of researchers
affiliated with the Southern California Earthquake Center (SCEC) evaluated amplifica-
tion factors using a consistent data set consisting only of southern California earth-
quakes (Field 2000, Steidl 2000). Field (2000) evaluated amplification factors as a direct
function of Vs-30 using a non-reference site approach in which amplification factors were
derived as a term within a southern California attenuation relationship. Steidl (2000)
also used a non-reference site approach, evaluating site factors as a function of Vs-30 us-
ing residuals from the Sadigh et al. (1993) attenuation relationship for rock sites (similar
to the Sadigh et al. 1997 relation). The amplification factors from the Harmsen and Field
studies are independent of PHAr . In the Steidl study, amplification factors were devel-
oped for PHAr⬍0.1 g and all PHAr ranges.
Joyner and Boore (2000) developed amplification factors within NEHRP categories
using a procedure similar to that of Field (2000) described above, although the short pe-
riod factors are expressed as a function of reference motion amplitude. Stewart et al.
(2003) developed nonlinear amplification factors within NEHRP categories relative to
the Abrahamson and Silva (1997) attenuation relationship.
Most of the above models provide discrete amplification factors within NEHRP cat-
egories; only the Field and Harmsen studies provide amplification factors as a continu-
ous function of Vs-30 . At present, there are no amplification models that are both
PHAr-dependent and a continuous function of Vs-30 . Amplification factors from the
above studies are compared to each other and to the results of this study subsequently in
this paper (Figure 10).

DATA RESOURCES

STRONG MOTION DATA

The ground motion database used in this study consists of 1828 recordings from 154
earthquakes. These recordings are from worldwide shallow crustal earthquakes near ac-
tive plate margins. Subduction and interplate events are excluded. Event dates range
from the 1933 Long Beach, California, earthquake to the 1999 Duzce, Turkey, earth-
quake. Removed from the data set for this study were recordings from events with
poorly defined magnitude or focal mechanism, recordings for which site-source dis-
tances are poorly constrained, recordings at large distance (⬎100 km), and recordings
for which problems were detected with one or more components. The recordings at large
6 Y. CHOI AND J. P. STEWART

distance were eliminated because the currently available data is too sparse to support the
development of empirical ground motion models at that distance range. These removals
reduced the data set to 919 recordings from 59 events. Data from the 1999 Chi-Chi, Tai-
wan, earthquake were not used in this study due to the preliminary nature of the site
classifications. Additional information on characteristics of the strong motion data set is
provided by Stewart et al. (2001).
The sources of strong motion data for the western United States include the Califor-
nia Strong Motion Instrumentation Program (CSMIP), the U.S. Geological Survey
(USGS), the University of Southern California (USC), the California Division of Mines
and Geology (CDMG), and the Los Angeles Department of Water and Power (LADWP).
Additional data have been obtained for the 1999 Kocaeli and Duzce, Turkey, earthquakes
from the Kandilli Observatory and Earthquake Engineering Research Institute of Boçizi
University (Kandilli), the Earthquake Research Department of the General Directorate
of Disaster Affairs (ERD), and Istanbul Technical University (ITU). Most of the time
histories used in this study can be obtained at the web site of the Pacific Earthquake
Engineering Research Center (www.peer.berkeley.edu).

DATA USED FOR SITE CLASSIFICATIONS

In order to classify strong motion sites according to the Vs-30 parameter, a GIS
database was developed having the locations of both strong motion stations and
boreholes in California. Each strong-motion station location was checked with instru-
ment owners (USGS and CSMIP), or against published reports (Anderson et al., 1981),
to optimize accuracy. Borehole locations were generally obtained from maps in reports.
The borehole database is similar to that of Wills and Silva (1998), but also contains
additional Caltrans boreholes, boreholes from selected consulting geotechnical
engineers, and data recently compiled in the ROSRINE program (http://geoinfo.usc.edu/
rosrine/). These databases were used to match boreholes with strong motion sites
if (1) both locations are on the same surface geology and (2) the separation distance
was ⬍1600 m.
The above databases were used to pair 209 strong motion stations to boreholes with
geophysical measurements. Of these sites, 174 have borehole-accelerograph separation
distances less than 160 m, 13 from 160 to 450 m, and 22 from 450 to 1600 m. The
borehole geophysical data was used to develop Vs-30 values for the paired strong motion
stations. Sites were classified into NEHRP categories on the basis of these Vs-30 values,
although a site was classified as E if Vs-30⭐180 m/s or if the thickness of soft clay (de-
fined above) is greater than 3 m. The distribution of Vs-30 values for sites in our database
is shown in Figure 2 along with the median Vs-30 value in each well-populated category.
It should be noted that shear wave velocities for USC strong motion stations obtained by
Rodriguez-Ordonez (1994) were not used due to apparent biases in such data as docu-
mented by Boore and Brown (1998) and Wills and Silva (1998). A complete inventory of
the classifications is presented in Stewart et al. (2001).
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 7

Figure 2. Histogram of Vs-30 values for strong motion sites used in this study.

DEVELOPMENT OF AMPLIFICATION MODEL

Model development begins with two stages of preliminary analysis. In the first stage,
amplification factors within bins defined on the basis of Vs-30 are investigated to evaluate
the variation of nonlinearity (i.e., dependence of amplification on PHAr) with Vs-30 . In
the second stage, amplification levels near a baseline reference amplitude of
PHAr⫽0.1 g are studied to identify an appropriate model for the variation of amplifica-
tion (at the baseline amplitude) with Vs-30 . The baseline amplitude of 0.1 g was selected
because it represents a mid-range amplitude on a log-scale for motions in most site cat-
egories. Results from these two stages of analysis are used to develop a functional form
for a ‘‘unified’’ model (i.e., a model that combines the effects of Vs-30 and nonlinearity).
The regression parameters for this unified model are then evaluated using a mixed-
effects regression procedure (e.g., Abrahamson and Youngs 1992). The following sub-
sections describe the two stages of preliminary data analysis, the regression analyses
used to develop the unified model, comparisons of model predictions to data, and the
analysis of standard deviation terms.

AMPLIFICATION WITHIN VS-30 CATEGORIES

Amplification factors within Vs-30 categories are compiled to evaluate the degree to
which nonlinearity in amplification factors varies with Vs-30 . These data analysis proce-
dures are similar to those of Stewart et al. (2003), and hence are reviewed only briefly.
The amplification factor for ground motion j within site category i, Fij , is evaluated
in arithmetic units from the geometric mean of 5% damped acceleration response spec-
tra for the two horizontal components of shaking, Sij , and the reference ground motion
for the site, (Sr)ij , as follows:

Fij共T兲⫽Sij /共Sr兲ij , (1)

8 Y. CHOI AND J. P. STEWART

where T⫽spectral period. In Equation 1, Sij and (Sr)ij are computed at the same spectral
period, which is varied from 0.01 to 5.0 s. Amplification factors are not evaluated for
T⬎1/(1.25⫻f hp), where f hp⫽high-pass corner frequency.
For the preliminary analyses discussed in this subsection, reference motion param-
eter (Sr)ij is taken as the median spectral acceleration calculated from the Abrahamson
and Silva (1997) A&S attenuation relationship for rock sites, with modifications for rup-
ture directivity effects and event terms. The A&S rock attenuation relationship provides
ground motion estimates that are appropriate for a soft rock site condition with Vs-30 val-
ues reported to be in the range of 520 to 620 m/s (Silva et al. 1997, Boore et al. 1997).
The rupture directivity correction is made for sites near the seismic source using the em-
pirical model by Somerville et al. (1997), later modified by Abrahamson (2000). For
well-recorded events, the event term represents the period-dependent average residual
between motions from a given event and the general attenuation model (the event terms
used at this stage of the analyses were provided by Abrahamson [1999, pers. comm.]).
These terms are evaluated during the development of attenuation models with a mixed-
effects regression procedure (Abrahamson and Youngs 1992). The use of an event term
in the evaluation of (Sr)ij is intended to remove bias in the attenuation model that might
be present for a particular event.
Amplification factors computed using Equation 1 were sorted into the following
Vs-30 categories for intracategory regression analysis:

E: Vs-30⬍180 m/s⫹soft clay

Dlv : 180⬍Vs-30⬍310 m/s
CD: 310⬍Vs-30⬍520 m/s
Chv : 520⬍Vs-30⬍760 m/s
B: 760⬍Vs-30⬍1500 m/s

These ranges of Vs-30 essentially match the NEHRP categories, except that NEHRP
C and D are subdivided into three bins (Chv , CD, and Dlv) to better capture the variation
of site nonlinearity with Vs-30 . Within category i, regression analyses were performed to
relate amplification factors, Fij , to ground motion amplitude as follows:

ln共Fij 兲⫽ai⫹bi ln共Gij 兲⫹␧ij , (2)

where ai and bi are regression coefficients specific to category i, Gij is a parameter rep-
resenting the amplitude of the reference ground motion for site j, and ␧ij is an error term
(i.e., ␧ij is the residual between data and model). This same regression equation has been
used by Youngs (1993), Bazzuro (1998), and Stewart et al. (2003), with Gij taken as
PHAr . We also take Gij as PHAr .
Using the data within the above velocity ranges, regression analyses were performed
according to Equation 2 using ordinary least-squares procedures in which equal weight
is given to all data points. The least-squares procedure is used because of the inclusion
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 9

Figure 3. Spectral acceleration amplification factors, intracategory regression results, and pre-
dictions of unified model for velocity categories plotted relative to PHA of reference motion
(PHAr).

of event terms in (Sr)ij . For each Vs-30 category, we plot in Figure 3 spectral amplifica-
tion levels for the periods of T⫽0.3 s 关F(0.3)兴 and T⫽1.0 s 关F(1.0)兴. Also plotted are
results of regression analyses performed according to Equation 2 (solid lines), ⫾95%
confidence intervals on the median amplification (dotted lines), and median regression
⫾lognormal standard deviation term (dashed lines). Note that the thick dotted lines in
Figure 3 represent predictions of the unified model that are discussed subsequently in the
paper. Presented in Table 2 are regression coefficients and standard deviation terms for
the ground motion parameters of F(0.3) and F(1.0). The estimation error terms for pa-
rameters ai and bi in Table 2 are the half-widths of the ⫾95% confidence intervals on the
parameters.
Reductions of amplification factors with increasing reference motion amplitude
are taken as evidence of sediment nonlinearity. This nonlinearity is quantified
10 Y. CHOI AND J. P. STEWART

Table 2. Regression coefficients for Sa amplification factors

Rejection
Std. confidence for
Category Period a b Dev. b⫽0 model (%)
B 0.3 ⫺0.23⫾0.88 0.03⫾0.37 0.44 14
1.0 ⫺0.84⫾1.63 ⫺0.21⫾0.68 0.77 48
Chr 0.3 ⫺0.05⫾0.64 0.01⫾0.26 0.67 4
1.0 ⫺0.09⫾0.85 ⫺0.01⫾0.35 0.84 5
CD 0.3 ⫺0.09⫾0.30 ⫺0.09⫾0.13 0.63 83
1.0 0.43⫾0.37 0.07⫾0.16 0.67 58
DIr 0.3 ⫺0.26⫾0.22 ⫺0.23⫾0.10 0.47 100
1.0 0.19⫾0.22 ⫺0.11⫾0.10 0.45 97
E 0.3 ⫺0.76⫾0.69 ⫺0.57⫾0.28 0.44 100
1.0 ⫺0.37⫾0.73 ⫺0.53⫾0.29 0.46 100

by the bi parameter for each category i. The statistical significance of the nonlinearity
is assessed two ways. The first significance test consists of comparing the absolute
value of bi to the estimation error for bi (both indicated in Table 2). When 兩bi兩 exceeds
the estimation error, the nonlinearity is considered significant. Secondly, sample ‘‘t’’
statistics are compiled to test the null hypothesis that bi⫽0 and ai⫽overall data
median. This statistical testing provides a significance level⫽␣ that the null hypothesis
cannot be rejected. For clarity of expression, we tabulate in Table 2 values of 1⫺␣,
which we refer to as a ‘‘rejection confidence for a b⫽0 model.’’ Large rejection
confidence levels (i.e., greater than 95%) suggest significant PHAr-dependence in
amplification factors.
The b parameters compiled from the above analyses are plotted as discrete data
points with error bounds in Figure 4. The results show statistically significant nonlinear-
ity (by the above criteria) at small Vs-30 , corresponding to the E category. Values of b
decrease to a relatively consistent value slightly offset from zero for Vs-30⬎⬃300 m/s.
The nonlinearity at these large Vs-30 values is not statistically significant. Based on the
trend of the discrete points in Figure 4, we postulate the following model to simulate the
variation of b with PHAr :

b⫽b1 Category E (3a)

b1⫺b2
b⫽b2⫹共Vs-30⫺bV兲2 180⬍Vs-30⬍bV 共m/s兲 (3b)
共180⫺bV兲2

b⫽b2 bV⬍Vs-30⬍520 共m/s兲 (3c)

b2
b⫽b2⫺共Vs-30⫺520兲 520⬍Vs-30⬍760 共m/s兲 (3d)
240

b⫽0 Vs-30⬎760 共m/s兲 (3e)

NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 11

Figure 4. Variation of slope parameter b (defined in Equation 2) with Vs-30 . Plotted are discrete
results for Vs-30 data bins and continuous lines showing the model defined by Equation 3, whose
parameters are determined from mixed effects regression analyses.

where the units of Vs-30 are in m/s, and b1 , b2 , and bV are model parameters estimated
from the data. A parabolic fit for 180⬍Vs-30⬍bV was used in lieu of a linear fit because
the parabola predicts lower levels of nonlinearity, which is more consistent with the data.
The decrease of b2 to zero at high Vs-30 is motivated by the statistical insignificance of
nonlinearity for high velocity sites. Values of parameters b1 , b2 , and bV were estimated
from regression analyses described subsequently in the paper, and the continuous lines
in Figure 4 represent the outcome of those analyses.

VARIATION OF AMPLIFICATION WITH Vs-30

In this section, we investigate the variation of amplification factors with Vs-30 , which
is accomplished by compiling data points from each category ‘‘near’’ a reference site
baseline shaking level of PHAr⫽0.1 g. Our use here of only data near this baseline shak-
ing level is intended to isolate the Vs-30 dependence of the amplification factors from the
dependence on PHAr .
We identify these data points as follows. Suppose, for example, that the median value
of F(0.3) from regression (i.e., Equation 2) at the baseline amplitude is Fba(0.3). We
then find the PHAr values along the median regression fit for the category (i.e., the solid
lines in Figure 3) corresponding to an amplification departure (in natural logarithmic
units) from Fba(0.3) of 0.05 (i.e., amplification levels in natural logarithmic units of
ln关Fba(0.3)兴⫾0.05). Data points between these two PHAr levels are selected. When the
regression fit shows no significant nonlinearity, most or all of the data is selected,
whereas significant nonlinearity limits the data range selected (e.g., data was taken only
from PHAr⫽0.09–0.11 g for Fba(0.3) s in NEHRP Category E). The value of ⫾0.05
12 Y. CHOI AND J. P. STEWART

Figure 5. Variation of amplification factors Fba(0.3) and Fba(1.0) with Vs-30 for consistent
ground motion amplitude (data points and power law regression fit), along with intracategory
variation of F(0.3) and F(1.0) with reference motion amplitude (vertical lines).

used in the above process was selected by judgment; it was found to provide a collection
of data points that is sufficiently large that statistically stable amplification values
can be defined while simultaneously maintaining insignificant PHAr—dependence of
amplification.
Data points selected by the above process are shown in Figure 5 along with a regres-
sion fit using the following power law equation:

F 共T兲⫽ ba
ba
冉 冊
Vs-30
Vref
cba
(4)

ba
where Vref and cba are regression coefficients (given in Figure 5) and superscript ‘‘ba’’ on
F(T) and the regression parameters denotes the use of amplification factors selected by
ba
the above process (i.e., near the baseline amplitude). Note that parameter Vref is simply
ba
the value of Vs-30 at which F (T) is unity. Plotted adjacent to the power law fit are the
⫾95% confidence intervals on the median amplification. Also shown for reference are
within-category median F(0.3) and F(1.0) values at PHAr⫽0.1 g (i.e., the ordinates of
the solid lines from Figure 3 at PHAr⫽0.1 g), which are plotted with an ⫻ at the median
Vs-30 value for within-category data. The vertical line drawn through the ⫻ represents the
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 13

range of amplification values that would be expected for PHAr⫽0.01–1.0 g based on

intracategory regression results. The results in Figure 5 show the expected significant
increase of amplification with decreasing Vs-30 , although the variation with reference
motion amplitude is also important (especially for Category E).

MIXED-EFFECTS REGRESSION FOR UNIFIED MODEL

The models for Vs-30- and PHAr-dependence of amplification in Equations 3 and 4
can now be combined to form a unified model for amplification factors. This model is
expressed as follows:

ln共Fij 兲⫽c ln 冉 冊 冉 冊
Vs-30ij
Vref
⫹b ln
PHArij
0.1
⫹␩i⫹␧ij , (5)

where PHAr is expressed in units of g; b is a function of regression parameters as given

in Equation 3; c and Vref are regression parameters; ␩i is a random effect term for earth-
quake event i (should have zero median across all events, standard deviation is denoted
as ␶); and ␧ij represents the intra-event model residual for motion j in event i (should
have median near zero for well-recorded events, standard deviation is denoted ␴). In or-
der to simplify the regression process to produce stable results, parameter b2 in Equation
3 was estimated using all data with Vs-30⬎bV . However, as a practical matter, the data
controlling b2 in the regression are sites with velocities between approximately 300 and
600 m/s. As noted previously, the decrease of b2 to zero at high Vs-30 is a judgment-based
adjustment to the model motivated by the statistical insignificance of nonlinearity for
high velocity sites.
The total standard deviation that is appropriate for use with the median amplification
from Equation 5 is

␴total⫽冑␴2⫹␶2 (6)
We perform regression analyses according to Equation 5 using a mixed-effects proce-
dure similar to Abrahamson and Youngs (1992) as implemented in the program R (Pin-
heiro and Bates 2000). The amplification factors used in these regressions are modified
from those presented above (i.e., Equation 1), in that event terms are not incorporated
into the reference site ground motions, Sr . We omit event terms from the reference mo-
tion at this stage because event terms are estimated as part of the mixed-effects regres-
sion procedure (i.e., term ␩i). In addition, reference motions are now evaluated using
multiple attenuation models. The models and corresponding site conditions used to
evaluate Sr values are as follows:
A1. Abrahamson and Silva (1997) A&S: rock
A2. Sadigh et al. (1997): rock
A3. Campbell and Bozorgnia (2003) C&B: generic rock
We considered using the Boore et al. (1997) attenuation relationship as well.
We chose not to develop site factors relative to this attenuation model in part because
the site factor in that attenuation model is already cast in terms of Vs-30 , with 760 m/s
14 Y. CHOI AND J. P. STEWART

Table 3. Regression parameters (unsmoothed) for unified model for site amplification. Param-
eters without error terms are estimated deterministically (as described in text).

Atten.
Model Parameter b1 b2 bv c Vref (m/s) ␶ ␴ ␴total1
A1 F(0.3) ⫺0.41 ⫺0.11⫾0.05 300 ⫺0.46⫾0.07 532⫾93 0.35 0.54 0.64
F(1.0) ⫺0.39 0.02⫾0.05 300 ⫺0.69⫾0.07 519⫾69 0.41 0.55 0.69
A2 F(0.3) ⫺0.49 ⫺0.21⫾0.04 300 ⫺0.44⫾0.07 601⫾103 0.29 0.55 0.62
F(1.0) ⫺0.48 ⫺0.12⫾0.05 300 ⫺0.66⫾0.08 646⫾90 0.35 0.57 0.67
A3 F(0.3) ⫺0.51 ⫺0.05⫾0.05 300 ⫺0.44⫾0.07 610⫾106 0.29 0.53 0.61
F(1.0) ⫺0.49 ⫺0.04⫾0.06 300 ⫺0.67⫾0.07 709⫾107 0.39 0.56 0.68
1
␴total
2
⫽␶2⫹␴2

taken as the reference value. Moreover, because the strong motion database contains
few sites with high Vs-30 , the Boore et al. attenuation model for the reference site
condition is based largely on data from softer sites, and hence the attenuation results
are strongly influenced by the (linear) site factor. We felt it was inappropriate to
implement an attenuation model that is so dependent on one site factor with a new
(different) site factor.
Several issues complicated the regression process. First, a stable estimate of bV could
not be obtained from the regression, so alternative values of bV were used as fixed values
during the regression of other parameters. Optimal bV values varied somewhat from pe-
riod to period, but generally a value of 300 m/s provides a reasonable fit to the data. A
second complication is that parameter b1 , when estimated by regression, was found to be
relatively small in an absolute sense (i.e., indicating small nonlinearity) and to be poorly
constrained (i.e., large estimation uncertainty). The low values underpredict the nonlin-
earity for Category E, for which the available data is not sufficiently abundant to
strongly affect the regression results. Accordingly, b1 was set at values from intracat-
egory regressions.
Example values of model parameters (and their estimation errors) derived directly
from the regression are presented in Table 3. The parameters are also listed in the Ap-
pendix for T⫽0.01–5.0 s. The results in the Appendix have been smoothed with respect
to period.
The c and Vref parameters in Table 3 for Model A1 are similar to those given in Fig-
ure 5. As shown by the lines in Figure 4, the parameters describing nonlinearity param-
eter b for Model A1 define a curve consistent with the b-values from discrete velocity
bins.
Median amplification factors for Models A1–A3 are compared in Figure 6 for
velocities at the median of the sites within each NEHRP category. The Model A1
and A2 results are generally similar to each other both in terms of the amplification
level and the dependence of amplification on PHAr . At T⫽0.3 s, Model A3 has
less PHAr-dependence for Categories C–D and thus has higher amplification levels
for PHAr⬎⬃0.1 g. At T⫽1.0 s, A3 amplification levels exceed A1–A2, although the
amount of PHAr-dependence is comparable. For all three models (A1–A3), the
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 15

Figure 6. Variation with PHAr of the median amplification factors from Models A1–A3 at mid-
Vs-30 value for each NEHRP category.

results are similar for Category E. These differences in the amplification factors
for Models A1–A3 result from the different formulations of the respective attenuation
models. For example, the relatively linear short-period site terms associated with A3
are a result of different distance-scaling formulations in the attenuation models,
which produces relatively low reference rock motions at close distance for C&B (2003)
as compared to A&S (1997) or Sadigh et al. (1997). These low reference motions in
turn cause the Model A3 amplification factors at close distance (thus high PHAr)
to be large (nearly as high as those at low PHAr), which results in the minimal
nonlinearity.

COMPARISONS OF MODEL PREDICTIONS TO DATA

The sufficiency of the model is investigated by plotting intra-event prediction residu-
als (␧ij in Equation 5) against Vs-30 and PHAr in Figure 7. Results for Model A1 are
shown, although similar results were obtained for Models A2–A3. The results show no
apparent trend in model residuals with Vs-30 or PHAr (Figure 7a), and no significant bias
for data within the previously used Vs-30 bins, as demonstrated by median residuals near
unity (Figure 7b). In Figure 3 we plot with thick, dotted lines the model predictions
against data within Vs-30 bins. The unified model is seen to provide predicted median
amplification levels for each category that are reasonably consistent with the intracat-
egory regression results.

ANALYSIS OF STANDARD DEVIATION TERMS

The dispersion of the amplification factors was investigated as a function of
magnitude (m), site-source distance (r), and Vs-30 . The magnitude dependence of
dispersion was examined using a procedure similar to that of Youngs et al. (1995).
16 Y. CHOI AND J. P. STEWART

Figure 7. (a) Residuals of Model A1 (in arithmetic units) plotted against Vs-30 and PHAr ; and
(b) residuals of Model A1 (in arithmetic units) within velocity categories along with the median
(␮) and median ⫾ one standard deviation (␴) of residuals.

The data is binned into groups of 0.5 magnitude width with an overlap of 0.25,
and mixed-effects regression analyses are performed within each bin using regression
Equation 5, but with the regression coefficients set to the values from the unsmoothed
mixed-effect analysis results obtained previously. This analysis provides inter- and
intra-event standard deviation terms (␶ and ␴, respectively) within each magnitude
bin. The ␶ and ␴ terms and their 95% confidence intervals are plotted in Figure 8a
for Model A1 at periods 0.01 s (PHA), 0.3 s, and 1.0 s. Additional analyses (not
shown) were performed for Models A2–A3 at the above periods as well as 3.0 s.
The confidence intervals on the dispersion reflect the estimation uncertainty, and in
general are wide when the data bin is sparsely populated. Note that the confidence
intervals around the standard deviation estimates are not symmetric. This is a common
feature of variance estimated with a maximum likelihood procedure (Raudenbush and
Bryk 2002, p. 55).
The results in Figure 8a do not indicate a significant magnitude-dependence of either
␶ or ␴. Similar results were obtained for Models A2–A3. These results differ from
magnitude-dependent standard deviation terms identified by Youngs et al. (1995) and in-
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 17

Figure 8. (a) Variation of inter- and intra-event standard deviation (and their estimation error)
with magnitude, Model A1; (b) variation of intra-event standard deviation with Vs-30 , Model
A1; and (c) variation of intra-event standard deviation with distance, Model A1.

corporated into most modern attenuation models (e.g., those underlying Models A1–
A3). Note also that the confidence intervals on ␶ are much larger than those on ␴. This
occurs because there are relatively few earthquakes within each magnitude bin to con-
strain the ␶ estimates.
18 Y. CHOI AND J. P. STEWART

The variation of the dispersion of ␧ij (denoted ␴) with distance and Vs-30 is investi-
gated by partitioning the model residuals according to overlapping distance bins and
non-overlapping Vs-30 bins, and then evaluating ␴ within each bin. The results of these
analyses are shown in Figure 8b (Vs-30) and 8c (distance) for Model A1 at T⫽0.01, 0.3,
and 1.0 s. As illustrated in Figure 8b, standard deviation term ␴ generally increases with
Vs-30 , although the amount of increase is strongly period dependent. At small periods
(T⭐⬃0.15 s) the amount of increase of ␴ is small between well-populated Vs-30 bins for
which the results are reliable (generally⬍0.05). However, for T⭓1.0 s, the amount of
increase between these bins ranges from about 0.1 to 0.3, with larger increases occurring
at longer periods. Results for the largest Vs-30 bin (760–1300 m/s) vary erratically from
period to period due to limited data, and are unreliable.
As shown in Figure 8c, preliminary data analyses indicate that standard deviation
terms increase with distance (r) for periods T⭐1.0 s. However, when the r-dependence
of ␴ is investigated within well-populated Vs-30-bins (results not shown for brevity), the
trend shown in Figure 8c is lost. Moreover, when the Vs-30-dependence of ␴ is investi-
gated within well-populated r-bins, the trend shown in Figure 8b is retained. Thus, the
Vs-30-dependence of ␴ appears to be more significant than the r-dependence.
Based on the above findings, a simple Vs-30-dependent model for the intra-event stan-
dard deviation is proposed. The standard deviation calculated by this model is denoted
␴v ; the symbol ␴ is retained for the overall intra-event standard deviation without con-
sideration of Vs-30 . In this model, ␴v is taken as constant at low and high Vs-30 , with
log-linear interpolation for intermediate velocities. The threshold velocities were se-
lected after analysis of many plots similar to those in Figure 8b. The model is cast as
follows:

␴v⫽e1 Vs-30⭐260 m/s (7a)

␴v⫽e1⫹e2• ln共Vs-30/260兲 260⬍Vs-30⭐360 m/s (7b)

where e2⫽(e3⫺e1)/ln(360/260)

␴v⫽e3 Vs-30⬎360 m/s (7c)

An example fit based on Equation 7 is shown by the lines in Figure 8b. Coefficients
e1 and e3 were evaluated at all periods and are listed in the Appendix. The coefficients
were estimated using data from well-populated bins at low and high velocity. The model
in Equation 7 necessarily smoothes true bin-to-bin variation of ␴, but in general the
model is not systematically biased high or low across the suite of periods considered for
any particular velocity bin. An exception is soft soil sites (i.e., NEHRP E), for which the
model tends to overpredict ␴ at most periods (although, coincidentally, the model pro-
vides a good fit for E at T⫽0.3 and 1.0s, as shown in Figure 8b). For these soft soil sites,
standard deviation is better estimated with site-specific ground response analysis
(Baturay and Stewart 2003), although use of the present model in PSHA will be conser-
vative at the long return periods often used in engineering design.
Smoothed values of ␶, ␴, e1 , and e3 are plotted in Figure 9. While both ␶ and ␴
are period-dependent, the period-dependence of ␴v is dependent on site condition.
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 19

Figure 9. Variation of standard deviation terms with period (Model A1) showing strong period
dependence of ␴ for relatively stiff soils, but weak dependence for softer soils.

We find no significant period-dependence for relatively soft soils (i.e., e1 in Figure 9,

Vs-30⬍260 m/s), but strong dependence for stiffer materials (e3 , Vs-30⬎360 m/s).

COMPARISONS TO PREVIOUS STUDIES

VELOCITY-DEPENDENCE OF AMPLIFICATION
The Model A1–A3 regression results from Equation 5 are plotted for PHAr⫽0.1 g
in Figure 10, and are compared to the results of previous studies discussed above.
Parameters c and Vref are also compared to those from previous studies in Table 4. The
slope values c are seen to be comparable to those from previous studies (except Steidl).
However, the Vref values for Models A1–A3 are significantly smaller than those from

Figure 10. Comparison of Vs-30 dependence of F(0.3) and F(1.0) parameters (evaluated at
PHAr⫽0.1 g) from this study to short- and mid-period amplification functions from previous
studies.
20 Y. CHOI AND J. P. STEWART

Table 4. Comparison of power law slope (c) and reference velocity (Vref ) parameters from this
study (unsmoothed) to those from previous work

This Study B&G Harmsen Field Steidl

Parameter A11 A21 A31 (1994)2 (1997)3 (2000)4 (2000)5

c F(0.3) ⫺0.46⫾0.07 ⫺0.44⫾0.07 ⫺0.44⫾0.07 ⫺0.36 ⫺0.56 ⫺0.35 ⫺0.13

F(1.0) ⫺0.69⫾0.07 ⫺0.66⫾0.08 ⫺0.67⫾0.07 ⫺0.64 ⫺0.66 ⫺0.70 ⫺0.39
Vref F(0.3) 532⫾93 601⫾103 610⫾106 997 1370 760 —
(m/s) F(1.0) 519⫾69 646⫾90 709⫾107 1067 1140 760 1054
1
results of present study—before smoothing
2
results for period range T⫽0.1–0.5 s in F(0.3) row, results for T⫽0.4–2.0 s shown in F(1.0) row
3
results for period range T⫽0.17–0.5 s in F(0.3) row, results for T⫽0.7–2.0 s shown in F(1.0) row
4
value of Vref preselected as 760 m/s and other regression parameters adjusted accordingly
5
results for data with PHA⬍0.1g; — = parameter not established

other studies, which reflects the relatively soft reference site condition associated with
the attenuation relationships used here to develop reference motions. However, our Vref
values are generally similar to the Vs-30 values compiled by Silva et al. (1997) (median
⫽520 m/s) and Boore et al. (1997) (average⫽620 m/s) from boreholes at rock sites in
active regions.

AMPLIFICATION LEVELS WITHIN NEHRP CATEGORIES

In this section we seek to compare the amplification factors within NEHRP catego-
ries predicted by Models A1–A3 with those utilized within the NEHRP Provisions
(BSSC 2001) and those identified by previous investigators. It is necessary to first re-
move the bias associated with inconsistent reference site conditions before such com-
parisons can be made.
The regression model in Equation 5 enables insight to be developed into the bias
associated with the use of a rock-average site condition (in active regions) to represent
the intended NEHRP reference condition of Vs-30⫽760 m/s. This bias can be calculated
as follows:

ln B共T兲⫽c• ln 冉 冊
Vref
760
(8)

where B(T) indicates bias at period T. Equation 8 strictly holds only when nonlinearity
parameter b is the same for velocities of Vref and 760 m/s. While that is generally not
exactly true (due to the linear taper in b indicated by Equation 3d), Equation 8 nonethe-
less provides a good approximation of bias because of small nonlinearity at these high
velocities. At T⫽0.3 and 1.0 s, the resulting biases for Models A1–A3 are approximately
1.09–1.17 and 1.05–1.28, respectively.
The B(T) values are combined with the A1–A3 amplification models to enable com-
parisons to the site factors in the NEHRP Provisions for a consistent reference site con-
dition of Vs-30⫽760 m/s. Plotted in Figure 11 are the NEHRP factors along with the av-
erage of bias-adjusted predictions of amplification Models A1–A3 over a range of Vs-30
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 21

Figure 11. Comparison of bias-adjusted average amplification factors (reference site condition
of Vs-30⫽760 m/s) from Models A1–A3 for the indicated velocity ranges to amplification
factors by others, including (1) NEHRP (intended to apply for a reference site condition
of Vs-30⫽760 m/s); (2) Borcherdt (2002b) (reference condition of approximately 850 m/s);
(3) Stewart et al. (2003) factors bias-adjusted (using Equation 8) to reference condition
of Vs-30⫽760 m/s; and (4) Joyner and Boore (2000) for reference condition of Vs-30⫽760 m/s.

appropriate to the respective categories. In the averaging across Models A1–A3, equal
weight was given to each model. The variability between the models in this case is
smaller than that shown in Figure 6 because of the bias removal, which adjusts all of the
models to a common reference velocity of 760 m/s. Also shown in Figure 11 are (1) the
Borcherdt (2002b) amplification factors, which apply for a slightly stiffer reference site
condition of Vs-30⫽850 m/s; (2) the amplification factors for NEHRP categories by
22 Y. CHOI AND J. P. STEWART

Stewart et al. (2003), which have been adjusted to a reference site condition of 760 m/s
using the bias adjustment factor in Equation 8; and (3) the Joyner and Boore (2000) am-
plification factors for reference condition Vs-30⫽760 m/s.
The bias-adjusted average amplification factors from this study are often smaller
than those given in the NEHRP Provisions. In the case of Categories B-D, the upper-
bound bias-adjusted factors from this study are similar to the NEHRP factors. The non-
linearity represented by the NEHRP factors for Categories B-E is generally similar to
that for our factors. In the case of Category D, the NEHRP nonlinearity appears to co-
incide with the mid-range nonlinearity from the present study. For Category E, the bias-
adjusted factors from this study are generally comparable to NEHRP at small period, but
are considerably smaller than NEHRP for mid-periods.
The offset between our bias-adjusted factors and the NEHRP factors warrants further
discussion. At issue here is whether the NEHRP factors are conservatively biased. One
possible explanation for the discrepancy is that the NEHRP factors, as presently formu-
lated, apply for a site condition stiffer than the intended target of 760 m/s. Recall that the
empirical basis for the NEHRP factors is observations from the 1989 Loma Prieta earth-
quake (corresponding to PHAr⬇0.1 g). As reported by Borcherdt and Glassmoyer
(1994), the velocity at which the amplification function derived from those data is unity
is approximately 1000 m/s. This velocity is contradicted somewhat by Borcherdt
(2002b), who reports that the average velocity at those sites based on borehole measure-
ments is 795 m/s. Nonetheless, the regressed site amplification model used in the origi-
nal development of the NEHRP site factors is unity near 1000 m/s, so that is the more
relevant velocity to the present discussion. Thus, the existing NEHRP factors are likely
biased for their intended reference site condition of 760 m/s by amounts on the order of
⬃12% for Fa and ⬃20% for Fv (based on Equation 8). Accordingly, it appears that a
significant portion of the discrepancies observed in Figure 11 can be explained by ap-
parent bias in the present NEHRP factors.
The amplification factors from Borcherdt (2002b) are generally larger than the NE-
HRP factors and the results of this study. This may be due in part to the stiffer reference
site condition of Vs-30⫽850 m/s. The amplification factors from Stewart et al. (2003) ei-
ther fall near the middle of the range of velocity-dependent factors from this study (e.g.,
C, E), or are near the middle of the range at low PHAr but have less nonlinearity and,
hence, higher amplification at high PHAr (e.g., D). The amplification factors by Joyner
and Boore are generally consistent with the results of the present study except for long
period amplification for Category D.

STANDARD DEVIATION TERMS

Figure 12 shows the standard deviation terms calculated in this study along with
those proposed in the various attenuation relationships used here. The top frame com-
pares Model A1 standard deviation terms to those from the A&S attenuation relation-
ship. The inter- and intra-event standard deviation terms are plotted separately, and the
intra-event terms are separated by site condition. Note that the A&S terms are magnitude
dependent.The standard deviation terms from this study are generally consistent with
A&S, except that Model A1 ␶ is period dependent, and exceeds the A&S ␶ for
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 23

Figure 12. Comparison of error terms from this study to those from attenuation models.

T⬎0.3 s. The middle frame is based on Model A2 and Sadigh et al. (1997) soil attenu-
ation, and shows only the total standard deviation (␴total). The standard deviation from
Model A2 is generally similar to the Sadigh results for soil. The bottom frame is based
on Model A3 and C&B (Campbell and Bozorgnia 2003) attenuation, and again shows
total standard deviation (␴total). The Model A3 standard deviation terms for stiff soils/
rock are larger than the C&B terms, whereas the Model A3 results for relatively soft
soils is consistent with the C&B results.

EFFECTIVENESS OF VS-30 AS SITE CONDITION METRIC

Stewart et al. (2003) investigated the relative effectiveness of several classification
schemes for use in strong motion prediction by evaluating an intra-event standard devia-
tion term that represents the average prediction dispersion across all categories in each
scheme. Since the standard deviation terms were calculated across all categories, they
were denoted ‘‘intercategory standard deviation (␴R).’’
24 Y. CHOI AND J. P. STEWART

Figure 13. Intercategory standard deviation terms for spectral acceleration, soil categories.

A scheme is considered to be relatively effective at capturing site-to-site variations in

ground motion when ␴R is small, and is less effective when ␴R is large. As shown in
Figure 13, it was found that detailed surface geology based classification schemes are
more effective than NEHRP categories (i.e., Table 1) or a geotechnical scheme
(Rodriguez-Marek et al. 2001). In Figure 13, the intra-event standard deviation from this
study is compared to those found by Stewart et al. (2003). The relatively low standard
deviations from this study indicate that with the model proposed herein, the Vs-30 site
metric is more effective than NEHRP or geotechnical classification schemes at most pe-
riods, and roughly equally effective as detailed surface geology.

CONCLUSIONS AND RECOMMENDATIONS

In this paper, we have developed a model for ground motion amplification that is a
function of Vs-30 and PHAr . The amplification factors are defined relative to ‘‘rock’’ ref-
erence motions from several attenuation relationships for active tectonic regions, includ-
ing those of Abrahamson and Silva (1997), Sadigh et al. (1997), and Campbell and Bo-
zorgnia (2003). Amplification at short- and mid-period ranges was shown to decrease
with increasing velocity in a manner similar to trends identified in previous studies. The
nonlinearity of amplification factors was found to vary with Vs-30 , being significant for
Vs-30⬍180 m/s and relatively small for Vs-30⬎300 m/s. Standard deviation terms were
found to have a significant dependence on Vs-30 . The databases used in model develop-
ment cover the parameter spaces Vs-30⫽130–1300 m/s and PHAr⫽0.02–0.8 g, and the
model is only considered valid across that range of parameters.
The model resulting from this work can be used as a site term in empirical attenu-
ation relations, and could be utilized to parameterize site effects in the future develop-
ment of attenuation relationships. The model is applied by using Equation 5 with Vs-30
defined from site characterization, PHAr defined for reference rock conditions using one
of the attenuation relationships used here, and b defined per Equation 3. Model param-
eters can be taken from the Appendix for the corresponding attenuation models (i.e.,
A1—A&S; A2—Sadigh et al.; A3—C&B). For modeling ground motions during a fu-
ture earthquake, event term ␩ in Equation 5 is generally taken as zero for calculation of
the median. The corresponding error term can be taken as ␴total from the Appendix, or
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 25

for a more accurate assessment, can be evaluated using Equation 6 with ␶ taken from the
Appendix and ␴ calculated using the site-dependent model in Equation 7 (in which case,
␴ is denoted ␴v).
The results of this work provide insight into the accuracy of the site coefficients in
the existing NEHRP Provisions and Commentary (BSSC 2001). Several important im-
plications of this work are as follows:
1. We utilized an entirely different procedure for evaluating amplification factors
than that employed in the development of the current NEHRP recommenda-
tions (described in Dobry et al., 2000). In many cases, there are significant dis-
crepancies between our factors and those in the NEHRP Provisions, with our
site factors often being lower. These new results warrant consideration for fu-
ture versions of the NEHRP Provisions and Commentary.
2. The standard deviation analysis results provide evidence for Vs-30-dependence
of intra-event dispersion (␴). For relatively soft materials, ␴ has no significant
period dependency and is relatively low. For stiffer materials, ␴ is strongly
period-dependent such that the offset from the soft soil values is small at low
periods but reaches values up to 0.3 at long period. This result suggests a po-
tential for bias in the procedure by which spectral ordinates are evaluated in
the NEHRP Provisions. In that procedure, design spectral ordinates are calcu-
lated as the product of PSHA results and amplification factors. The bias would
arise when the dispersion values used in the attenuation relationship for PSHA
are different from what is appropriate for the site category.
3. The development of national hazard maps appropriate for the reference site
condition of Vs-30⫽760 m/s requires the correction of existing attenuation
models, because the databases used in the development of these models do not
share the NEHRP reference site condition. The correction factors for the vari-
ous attenuation models can be evaluated using Equation 8 and the coefficients
tabulated in the Appendix.

ACKNOWLEDGMENTS
Support for this study was provided by the Pacific Earthquake Engineering Research
Center through the Earthquake Engineering Research Centers Program of the National
Science Foundation under Award Number EEC-9701568. Rie Von Eyben, Xiao
Chen, Philip Ender, and Steve Erickson provided valuable consultation regarding the
statistical analyses. We would also like to thank Kenneth Campbell for his assistance in
compiling distance parameters for strong motions stations. We thank Dr. Campbell
and Dr. Ned Field, along with two anonymous reviewers, for providing helpful review
comments.

APPENDIX
Tables A1–A3 show the model parameters for the corresponding attenuation models:
Model A1 for A&S (Abrahamson and Silva 1997), Model A2 for Sadigh et al. (1997),
and Model A3 for C&B (Campbell and Bozorgnia 2003).
26 Y. CHOI AND J. P. STEWART

Table A1. Smoothed model parameters for Model A1

Period
(sec) b1 Vref (m/sec) c b2 ␶ ␴ e1 e3
0.01 ⫺0.64 418⫾72 ⫺0.36⫾0.06 ⫺0.14⫾0.04 0.27 0.49 0.44 0.50
0.02 ⫺0.63 490⫾101 ⫺0.34⫾0.06 ⫺0.12⫾0.04 0.26 0.50 0.45 0.51
0.03 ⫺0.62 324⫾58 ⫺0.33⫾0.06 ⫺0.11⫾0.04 0.26 0.50 0.46 0.51
0.04 ⫺0.61 233⫾49 ⫺0.31⫾0.06 ⫺0.11⫾0.04 0.26 0.51 0.47 0.51
0.05 ⫺0.64 192⫾48 ⫺0.29⫾0.06 ⫺0.11⫾0.04 0.25 0.51 0.47 0.52
0.06 ⫺0.64 181⫾53 ⫺0.25⫾0.06 ⫺0.11⫾0.04 0.25 0.52 0.48 0.52
0.075 ⫺0.64 196⫾57 ⫺0.23⫾0.06 ⫺0.11⫾0.04 0.24 0.52 0.48 0.52
0.09 ⫺0.64 239⫾64 ⫺0.23⫾0.07 ⫺0.12⫾0.04 0.23 0.52 0.49 0.52
0.10 ⫺0.60 257⫾61 ⫺0.25⫾0.07 ⫺0.13⫾0.04 0.23 0.52 0.49 0.53
0.12 ⫺0.56 299⫾66 ⫺0.26⫾0.07 ⫺0.14⫾0.04 0.24 0.52 0.49 0.53
0.15 ⫺0.53 357⫾83 ⫺0.28⫾0.07 ⫺0.18⫾0.04 0.25 0.53 0.49 0.54
0.17 ⫺0.53 406⫾86 ⫺0.29⫾0.07 ⫺0.19⫾0.04 0.26 0.53 0.48 0.55
0.20 ⫺0.52 453⫾97 ⫺0.31⫾0.07 ⫺0.19⫾0.04 0.27 0.53 0.47 0.56
0.24 ⫺0.52 493⫾91 ⫺0.38⫾0.07 ⫺0.16⫾0.04 0.29 0.53 0.47 0.56
0.30 ⫺0.52 532⫾93 ⫺0.44⫾0.07 ⫺0.14⫾0.05 0.35 0.54 0.46 0.57
0.36 ⫺0.51 535⫾97 ⫺0.48⫾0.07 ⫺0.11⫾0.05 0.38 0.54 0.46 0.57
0.40 ⫺0.51 535⫾104 ⫺0.50⫾0.07 ⫺0.10⫾0.05 0.40 0.54 0.46 0.57
0.46 ⫺0.50 535⫾87 ⫺0.55⫾0.07 ⫺0.08⫾0.05 0.42 0.54 0.45 0.58
0.50 ⫺0.50 535⫾82 ⫺0.60⫾0.07 ⫺0.06⫾0.05 0.42 0.54 0.45 0.59
0.60 ⫺0.49 535⫾73 ⫺0.66⫾0.07 ⫺0.03⫾0.05 0.42 0.55 0.44 0.60
0.75 ⫺0.47 535⫾75 ⫺0.69⫾0.07 0.00⫾0.05 0.42 0.55 0.44 0.63
0.85 ⫺0.46 535⫾73 ⫺0.69⫾0.07 0.00⫾0.05 0.42 0.55 0.44 0.63
1.00 ⫺0.44 535⫾69 ⫺0.70⫾0.07 0.00⫾0.05 0.42 0.56 0.44 0.64
1.50 ⫺0.40 535⫾63 ⫺0.72⫾0.08 0.00⫾0.06 0.42 0.57 0.44 0.67
2.00 ⫺0.38 535⫾61 ⫺0.73⫾0.08 0.00⫾0.06 0.43 0.58 0.44 0.69
3.00 ⫺0.34 535⫾65 ⫺0.74⫾0.09 0.00⫾0.07 0.45 0.61 0.44 0.71
4.00 ⫺0.31 535⫾110 ⫺0.75⫾0.09 0.00⫾0.07 0.47 0.64 0.44 0.73
5.00 ⫺0.30 535⫾166 ⫺0.75⫾0.14 0.00⫾0.11 0.49 0.66 0.44 0.75
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 27

Table A2. Smoothed model parameters for Model A2

Period
(sec) b1 Vref (m/sec) c b2 ␶ ␴ e1 e3
0.01 ⫺0.61 567⫾110 ⫺0.34⫾0.06 ⫺0.20⫾0.04 0.24 0.49 0.45 0.51
0.05 ⫺0.66 521⫾128 ⫺0.26⫾0.06 ⫺0.25⫾0.04 0.24 0.52 0.48 0.52
0.09 ⫺0.62 497⫾147 ⫺0.21⫾0.07 ⫺0.24⫾0.04 0.24 0.52 0.49 0.53
0.10 ⫺0.58 464⫾122 ⫺0.22⫾0.07 ⫺0.24⫾0.04 0.24 0.52 0.50 0.53
0.12 ⫺0.53 444⫾109 ⫺0.24⫾0.06 ⫺0.24⫾0.04 0.24 0.52 0.50 0.53
0.15 ⫺0.51 508⫾134 ⫺0.25⫾0.06 ⫺0.24⫾0.04 0.24 0.52 0.50 0.53
0.17 ⫺0.50 545⫾128 ⫺0.26⫾0.06 ⫺0.24⫾0.04 0.24 0.52 0.49 0.54
0.20 ⫺0.50 580⫾142 ⫺0.29⫾0.06 ⫺0.23⫾0.04 0.24 0.52 0.48 0.55
0.24 ⫺0.50 600⫾119 ⫺0.36⫾0.07 ⫺0.23⫾0.04 0.25 0.52 0.47 0.55
0.30 ⫺0.49 620⫾103 ⫺0.43⫾0.07 ⫺0.22⫾0.04 0.29 0.53 0.47 0.56
0.40 ⫺0.49 640⫾119 ⫺0.50⫾0.07 ⫺0.21⫾0.04 0.34 0.54 0.46 0.58
0.50 ⫺0.49 640⫾99 ⫺0.55⫾0.07 ⫺0.19⫾0.04 0.35 0.55 0.46 0.60
0.75 ⫺0.48 645⫾99 ⫺0.63⫾0.08 ⫺0.15⫾0.05 0.36 0.56 0.46 0.64
1.00 ⫺0.48 646⫾90 ⫺0.67⫾0.08 ⫺0.15⫾0.05 0.36 0.57 0.46 0.66
1.50 ⫺0.47 640⫾85 ⫺0.70⫾0.08 ⫺0.14⫾0.05 0.36 0.58 0.46 0.69
2.00 ⫺0.46 580⫾75 ⫺0.72⫾0.08 ⫺0.14⫾0.05 0.37 0.59 0.46 0.72
3.00 ⫺0.43 545⫾68 ⫺0.72⫾0.10 ⫺0.14⫾0.07 0.38 0.62 0.46 0.75
4.00 ⫺0.40 540⫾121 ⫺0.72⫾0.10 ⫺0.13⫾0.07 0.39 0.65 0.46 0.77
5.00 ⫺0.39 535⫾130 ⫺0.72⫾0.14 ⫺0.13⫾0.11 0.44 0.68 0.46 0.79

Table A3. Smoothed model parameters for Model A3

Period
(sec) b1 Vref (m/sec) c b2 ␶ ␴ e1 e3
0.01 ⫺0.55 501⫾90 ⫺0.34⫾0.06 ⫺0.04⫾0.05 0.23 0.49 0.45 0.50
0.05 ⫺0.57 676⫾179 ⫺0.26⫾0.06 ⫺0.05⫾0.05 0.21 0.51 0.47 0.51
0.075 ⫺0.61 780⫾280 ⫺0.21⫾0.06 ⫺0.11⫾0.05 0.22 0.51 0.48 0.52
0.10 ⫺0.57 643⫾198 ⫺0.22⫾0.07 ⫺0.12⫾0.05 0.22 0.51 0.49 0.52
0.15 ⫺0.52 541⫾142 ⫺0.24⫾0.06 ⫺0.13⫾0.05 0.23 0.52 0.49 0.53
0.20 ⫺0.51 565⫾129 ⫺0.28⫾0.06 ⫺0.07⫾0.05 0.25 0.52 0.48 0.53
0.30 ⫺0.51 610⫾106 ⫺0.41⫾0.07 ⫺0.04⫾0.05 0.29 0.53 0.46 0.55
0.40 ⫺0.50 640⫾134 ⫺0.50⫾0.07 ⫺0.02⫾0.05 0.37 0.54 0.45 0.57
0.50 ⫺0.50 660⫾102 ⫺0.59⫾0.07 ⫺0.02⫾0.05 0.39 0.54 0.45 0.58
0.75 ⫺0.49 703⫾101 ⫺0.65⫾0.07 ⫺0.02⫾0.06 0.39 0.55 0.45 0.62
1.00 ⫺0.49 709⫾107 ⫺0.68⫾0.07 ⫺0.04⫾0.06 0.39 0.56 0.45 0.64
1.50 ⫺0.48 710⫾117 ⫺0.71⫾0.08 ⫺0.12⫾0.06 0.39 0.57 0.45 0.67
2.00 ⫺0.46 710⫾113 ⫺0.72⫾0.08 ⫺0.17⫾0.06 0.39 0.58 0.45 0.69
3.00 ⫺0.42 710⫾87 ⫺0.72⫾0.09 ⫺0.22⫾0.08 0.39 0.61 0.45 0.72
4.00 ⫺0.40 710⫾161 ⫺0.72⫾0.10 ⫺0.25⫾0.08 0.39 0.63 0.45 0.74
28 Y. CHOI AND J. P. STEWART

REFERENCES
Abrahamson, N. A., 2000. Effects of rupture directivity on probabilistic seismic hazard analy-
sis, Proceedings of 6th International Conference on Seismic Zonation, November 12–15,
Palm Springs, CA, Earthquake Engineering Research Institute, Oakland, CA.
Abrahamson, N. A., and Silva, W. J., 1997. Empirical response spectral attenuation relations for
shallow crustal earthquakes, Seismol. Res. Lett. 68, 94–127.
Abrahamson, N. A., and Youngs, R. R., 1992. A stable algorithm for regression analyses using
the random effects model, Bull. Seismol. Soc. Am. 82, 505–510.
Anderson, J. G., Trifunac, M. D., Teng, T.-L., Amini, A., and Moslem, K., 1981. Los Angeles
vicinity strong motion accelerograph network, Report No. CE 81-04, Univ. of Southern Cali-
fornia.
Baturay, M. B., and Stewart, J. P., 2003. Uncertainty and bias in ground motion estimates from
ground response analyses, Bull. Seismol. Soc. Am. 93, 2025–2042.
Bazzuro, P., 1998. Probabilistic Seismic Demand Analysis, Ph.D. dissertation, Civil Engineer-
ing Dept., Stanford University.
Boore, D. M., Joyner, W. B., and Fumal, T. E., 1997. Equations for estimating horizontal re-
sponse spectra and peak acceleration from western North American earthquakes: A sum-
mary of recent work, Seismol. Res. Lett. 68, 128–153.
Boore, D. M., and Brown, L. T., 1998. Comparing shear wave velocity profiles from inversion
of surface wave phase velocities with downhole measurements: Systematic differences be-
tween the CXW method and down hole measurements at six USC strong motion sites, Seis-
mol. Res. Lett. 69, 222–229.
Borcherdt, R. D., 1994. Estimates of site-dependent response spectra for design (methodology
and justification), Earthquake Spectra 10 (4), 617–653.
Borcherdt, R. D., 2002a. Empirical evidence for acceleration-dependent amplification factors,
Bull. Seismol. Soc. Am. 92, 761–782.
Borcherdt, R. D., 2002b. Empirical evidence for site coefficients in building code provisions,
Earthquake Spectra 18 (2), 189–217.
Borcherdt, R. D., and Glassmoyer, G., 1994. Influences of local geology on strong and weak
ground motions recorded in the San Francisco Bay region and their implications for site-
specific building-code provisions, The Loma Prieta, California Earthquake of October 17,
1989—Strong Ground Motion, U.S. Geological Survey Professional Paper 1551-A, A77-
A108
Building Seismic Safety Council (BSSC), 2001. NEHRP Recommended Provisions for Seismic
Regulations for New Buildings and Other Structures, Part 1: Provisions and Part 2: Com-
mentary, Federal Emergency Management Agency, FEMA-368 and FEMA-369, Washington
D.C., February.
Campbell, K. W., and Bozorgnia, Y., 2003. Updated near-source ground-motion (attenuation)
relations for the horizontal and vertical components of peak ground acceleration and accel-
eration response spectra, Bull. Seismol. Soc. Am. 93, 314–331.
Dobry, R., Martin, G. R., Parra, E., and Bhattacharyya, A., 1994. Development of site-
dependent ratios of elastic response spectra (RRS) and site categories for building seismic
codes, Proceedings of 1992 NCEER/SEAOC/BSSC Workshop on Site Response During
Earthquakes and Seismic Code Provisions, edited by G. R. Martin, University of Southern
California, National Center for Earthquake Engineering Research Special Publication
NCEER-94-SP01, Buffalo, NY.
NONLINEAR SITE AMPLIFICATION AS FUNCTION OF 30 M SHEAR WAVE VELOCITY 29

Dobry, R., Borcherdt, R. D., Crouse, C. B., Idriss, I. M., Joyner, W. B., Martin, G. R., Power, M.
S., Rinne, E. E., and Seed, R. B., 2000. New site coefficients and site classification system
used in recent building seismic code provisions, Earthquake Spectra 16 (1), 41–67.
Field, E. H., 2000. A modified ground motion attenuation relationship for southern California
that accounts for detailed site classification and a basin depth effect, Bull. Seismol. Soc. Am.
90, S209–S221.
Frankel, A., Mueller, C., Barnhard, T., Leyendecker, E., Wesson, R., Harmsen, A., Klein, F.,
Perkins, D., Dickman, N., Hanson, S., and Hopper, M., 2000. USGS national seismic hazard
maps, Earthquake Spectra 16 (1), 1–20.
Harmsen, S. C., 1997. Determination of site amplification in the Los Angeles urban area from
inversion of strong motion records, Bull. Seismol. Soc. Am. 87, 866–887.
Joyner, W. B., and Boore, D. M., 2000. Recent developments in earthquake ground motion es-
timation, Proceedings of 6th International Conference on Seismic Zonation, November 12–
15, Palm Springs, CA, Earthquake Engineering Research Institute, Oakland, CA.
Joyner, W. B., Fumal, T. E., and Glassmoyer, G., 1994. Empirical response spectral ratios for
strong motion data from the 1989 Loma Prieta, California, earthquake, Proceedings of 1992
NCEER/SEAOC/BSSC Workshop on Site Response During Earthquakes and Seismic Code
Provisions, edited by G. R. Martin, University of Southern California, National Center for
Earthquake Engineering Research Special Publication NCEER-94-SP01, Buffalo, NY.
Leyendecker, E. V., Hunt, R. J., Frankel, A. D., and Rukstales, K. S., 2000. Development of
maximum considered earthquake ground motion maps, Earthquake Spectra 16 (1), 21–40.
Pinheiro, J. C., and Bates, D. M., 2000. Mixed-Effects Models in S and S-PLUS, Springer, New
York.
Raudenbush, S. W., and Bryk, A. S., 2002. Hierarchical Linear Models: Applications and Data
Analysis Methods, 2nd Edition, Sage Publications, Thousand Oaks, CA.
Rodriguez-Marek, A., Bray, J. D., and Abrahamson, N. A., 2001. An empirical geotechnical
seismic site response procedure, Earthquake Spectra 17 (1), 65–87.
Rodriguez-Ordonez, J. A., 1994. A New Method for Interpretation of Surface Wave Measure-
ments in Soils, Ph.D. dissertation, North Carolina State University, Raleigh.
Sadigh, K., Chang, C.-Y., Abrahamson, N. A., Chiou, S. J., and Power, M. S., 1993. Specifica-
tion of long-period ground motions: Updated attenuation relationships for rock site condi-
tions and adjustment factors for near-fault effects, Proceedings of Seminar on Seismic Iso-
lation, Passive Energy Dissipation, and Active Control, Applied Technology Council
Publication No. 17-1, Vol. 1, pp. 59–70.
Sadigh, K., Chang, C.-Y., Egan, J. A., Makdisi, F., and Youngs, R. R., 1997. Attenuation rela-
tions for shallow crustal earthquakes based on California strong motion data, Seismol. Res.
Lett. 68, 180–189.
Seed, R. B., Dickenson, S. E., Rau, G. A., White, R. K., and Mok, C. M., 1994. Observations
regarding seismic response analyses for soft and deep clay sites, Proceedings of 1992
NCEER/SEAOC/BSSC Workshop on Site Response During Earthquakes and Seismic Code
Provisions, edited by G. R. Martin, University of Southern California, National Center for
Earthquake Engineering Research Special Publication NCEER-94-SP01, Buffalo, NY.
Silva, W. J., Abrahamson, N., Toro, G., and Costantino, C., 1997. Description and validation of
the stochastic ground motion model, report to Brookhaven National Laboratory, Associated
Universities, Inc., Upton, NY, available at http://www.pacificengineering.org
Somerville, P. G., Smith, N. F., Graves, R. W., and Abrahamson, N. A., 1997. Modification of
30 Y. CHOI AND J. P. STEWART

empirical strong ground motion attenuation relations to include the amplitude and duration
effects of rupture directivity, Seismol. Res. Lett. 68, 199–222.
Steidl, J. H., 2000. Site response in southern California for probabilistic seismic hazard analy-
sis, Bull. Seismol. Soc. Am. 90, S149–S169.
Stewart, J. P., Liu, A. H., Choi, Y., and Baturay, M. B., 2001. Amplification factors for spectral
acceleration in active regions, Report No. PEER-2001/10, Pacific Earthquake Engineering
Research Center, University of California, Berkeley.
Stewart, J. P., Liu, A. H., and Choi, Y., 2003. Amplification factors for spectral acceleration in
tectonically active regions, Bull. Seismol. Soc. Am. 93, 332–352.
Wills, C. J., and Silva, W., 1998. Shear wave velocity characteristics of geologic units in Cali-
fornia, Earthquake Spectra 14 (3), 533–556.
Youngs, R. R., 1993. Soil amplification and vertical to horizontal ratios for analysis of strong
motion data from active tectonic regions, Appendix 2C in Guidelines for Determining De-
sign Basis Ground Motions, Vol. 2, Electrical Power Research Institute, Report No. TR-
102293.
Youngs, R. R., Abrahamson, N. A., Makdisi, F. I., and Sadigh, K., 1995. Magnitude-dependent
variance of peak ground acceleration, Bull. Seismol. Soc. Am. 85, 1161–1176.
(Received 22 September 2003; accepted 12 July 2004)