Prediction Equations for Significant

Duration of Earthquake Ground Motions

Considering Site and Near-Source Effects
Justin J. Kemptona… and Jonathan P. Stewart,b… M.EERI

For engineering systems having a potential for degradation under cyclic

loading !e.g., landslides, soil profiles subject to liquefaction, some structural
systems", the characterization of seismic demand should include the amplitude
and duration of strong shaking within the system. This article is concerned
with significant-duration parameters, which are defined as the time interval
across which a specified amount of energy is dissipated !as measured by the
integral of the square of the ground acceleration or velocity". We develop
ground-motion prediction equations for significant-duration parameters as a
function of magnitude, closest site-source distance, site parameters that reflect
shallow geologic conditions as well as deep basin structure, and near-source
parameters. The relations are developed using a modern database and a
random-effects regression procedure. We find significant duration to increase
with magnitude and site-source distance !effects that had been identified
previously", but also to decrease with increasing shear-wave velocity of near-
surface sediments and to increase with increasing basin depth. Parameters that
principally measure the duration of body waves were also found to decrease in
near-fault areas subject to forward rupture directivity, although such effects
were not apparent for other duration parameters that tend to reflect the
combined duration of body and surface waves. #DOI: 10.1193/1.2358175$

INTRODUCTION
For structural or geotechnical systems whose performance is measured by damage
that accumulates during shaking, duration has been shown to be a meaningful predictor
of performance, along with amplitude and frequency content parameters. Example ap-
plications of duration for seismic performance characterization include the following:
• Seismic displacements of landslide masses have been related to the spatially av-
eraged peak amplitude of shaking within the slide mass as well as the duration of
shaking !e.g., Bray and Rathje 1998". Slope displacements were found to in-
crease with duration.
• Pore pressure generation in liquefiable soils !e.g., Seed and Lee 1966" and volu-
metric strain accumulation in unsaturated soils !e.g., Silver and Seed 1971" both

a"
Senior Engineer, Kleinfelder, Inc., 1370 Valley Vista Drive, Suite 150, Diamond Bar, CA 91765
b"
Associate Professor and Vice Chair, Civil and Environmental Engineering Dept., University of California,
Los Angeles, CA 90095-1593

985
Earthquake Spectra, Volume 22, No. 4, pages 985–1013, November 2006; © 2006, Earthquake Engineering Research Institute
986 J. J. KEMPTON AND J. P. STEWART

Figure 1. Evaluation of bracketed duration !acceleration threshold= 0.05 g" for ground motions
recorded at the North Palm Springs Airport site during M = 6.0 and 7.3 earthquakes.

increase with the amplitude of shaking experienced by a soil element and with
the number of cycles of shaking, N, at that amplitude. The N parameter, in turn,
is correlated to the duration of shaking, although the degree of correlation de-
pends on the specific duration definition and the method for counting cycles
!Hancock and Bommer 2005, Bommer et al. 2006".
• Lateral spread displacements resulting from soil liquefaction have been related to
amplitude and duration parameters of earthquake ground motions !Rauch and
Martin 2000", with displacements found to increase with duration.
• Many types of structural components and systems can be subject to cyclic deg-
radation, and hence would be expected to exhibit sensitivity to duration. None-
theless, whether duration is a statistically significant predictor of structural dam-
age remains an open question, with some research indicating no effect !Cornell
1997, Shome et al. 1998, Iervolino et al. 2006" and other research indicating a
possible correlation !Reinoso et al. 2000, Hancock and Bommer 2004, Bommer
et al. 2004". In support of the latter argument, inelastic response spectra !and its
reduction from elastic spectra" have been correlated with duration by Chai et al.
!1998" and Tiwari and Gupta !2000".
As described by Bommer and Martinez-Pereira !1999" and the references contained
therein, many types of duration parameters have been proposed in the literature, but the
most commonly used duration parameters are bracketed duration and significant dura-
tion. Example calculations of these parameters are shown in Figures 1 and 2. Figure 1
illustrates bracketed duration, which is defined as the time elapsed between the first and
last excursions beyond a specified threshold acceleration !typically 0.05 g or 0.1 g".
Bracketed duration parameters can be sensitive to the threshold accelerations and to
small subevents occurring towards the end of a recording. For these and other reasons,
other definitions of duration are often preferred.
Significant-duration parameters are defined as the time interval across which a speci-
fied amount of energy is dissipated. In this context, energy is represented by the integral
of the square of the ground acceleration or velocity. The integral of ground acceleration
is related to the Arias intensity !IA" !Arias 1970", which is defined using the integral,
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 987

Figure 2. Evaluation of significant-duration parameters Da5-75 and Da5-95 for ground motions
recorded at the North Palm Springs Airport site.

IA =
!
2g
% a !t"dt
T

0
2
!1"

where a!t" is the acceleration time history, g is the acceleration of gravity, and T repre-
sents the complete duration of recording a!t". Husid plots, used to track the build up of
energy in time !Husid 1969", are shown in Figure 2. Two common measures of signifi-
cant duration are time intervals between 5-75% and 5-95% of IA !denoted as Da5-75 and
Da5-95, respectively", which are indicated in Figure 2. The integral of ground velocity,
introduced by Sarma !1970" and referred to as the energy integral by Anderson !2004",
can also be used to define duration:

IE = % v !t"dt
"

0
2
!2"

where v!t" is the velocity time history. Significant-duration parameters evaluated from
the energy integral are denoted as Dv5-75 and Dv5-95.
The focus of the present work is on significant duration measured from the Arias and
energy integrals. We select significant duration because it is relatively stable with respect
to the definitions of beginning and end thresholds !Bommer and Martinez-Pereira 1999".
Moreover, significant duration from the Arias integral has arguably seen the most use in
recent engineering practice. It is recognized that nonzero significant durations can be
measured for low amplitude records that are of little engineering significance; however,
this problem is overcome when amplitude is coupled with duration for damage assess-
ment, as is usually the case.
Ground motion prediction equations have been developed previously for bracketed
duration, significant duration, and other duration parameters. All existing relationships
include magnitude as a predictive parameter; some also consider site-source distance
and site condition. Previous relations for significant duration are summarized in Table 1.
Most previous relations were developed for Da5-95. Table 1 indicates that the available
prediction equations use different magnitude and distance definitions from one another
and that site condition has most often been parameterized as rock or soil.
988 J. J. KEMPTON AND J. P. STEWART

Table 1. Papers presenting prediction equations for significant duration defined from broadband
accelerograms

Duration Magnitude Distance

Author Parameter1 Type2 Type3 Site Parameter4

Trifunac and Brady Da5-95, Dv5-95, n/a repi, h soft alluvium, intermediate
1975 and Dd5-95 rock, and hard rock
Dobry et al. 1978 Da5-95 n/a n/a rock
McGuire and DB, DF, Da5-95 n/a r or repi soil, rock
Barnhard 1979
Kamiyama 1984 Da5-95 MJ repi, h f!VS , di"
Abrahamson and Silva Da5-75 and Da5-95 M r soil, rock
1996
1
Duration Parameters:
DB = bracketed duration
DF = fractional duration !time interval over which Arias intensity is uniformly distributed at an average power
#Vanmarcke and Lai 1980$"
Da5-95 = significant duration as function of the acceleration record
Dv5-95 = significant duration as function of the velocity record
Dd5-95 = significant duration as function of the displacement record
2
Magnitude Parameters:
MJ = Japan Meteorological Agency magnitude
M = moment magnitude
3
Distance Parameters:
r = site to source distance
repi = epicentral distance
h = hypocentral depth
4
Site parameters:
VS = shear-wave velocity
di = layer thicknesses

We note that a number of additional prediction equations !not shown in Table 1" have
been developed for frequency-dependent duration parameters evaluated from band-
passed accelerograms !e.g., Bolt 1973, Trifunac and Westermo 1977, Mohraz and Peng
1989, Novikova and Trifunac 1994". Novikova and Trifunac !1994" argue that an advan-
tage of modeling duration using frequency bands is that the arrival time and duration of
each separate strong-motion pulse can be evaluated, which can provide insight into
source effects and energy dissipation between strong-motion pulses. However, engineer-
ing applications of duration consistently use duration parameters derived from full
broadband records. This practice is followed here because seismic excitation of struc-
tures is broadband, and intensity measures should characterize that broadband excitation
!an analogous situation is the use of spectral acceleration, defined from broadband ac-
celerograms, to characterize the amplitude of shaking in structures". Accordingly, we fo-
cus on duration parameters for broadband accelerograms.
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 989

There are several shortcomings to the existing ground-motion prediction equations

for significant duration, which we seek to address in the present study:
1. Most existing relations utilize a functional form that lacks a physically based
representation of magnitude effects on source duration.
2. The distance measure used previously has typically been epicentral distance, not
closest site-source distance !accounting for the source dimension".
3. Relations derived from the energy integral are not available; such relations are
of interest because they would be expected to capture the duration of long-
period components of ground motions more effectively than duration param-
eters derived from accelerograms.
4. The effects of site condition on duration have not been satisfactorily investi-
gated.
5. The only existing model for near-fault rupture directivity effects on duration
!Somerville et al. 1997" utilizes a database that is now dated, especially in light
of significant new data inventories compiled as part of the PEER Next Genera-
tion Attenuation !NGA" project !http://peer.berkeley.edu/nga/".
6. Existing relationships were derived using least-squares regression. This method
of analysis gives equal weight to each recording, and hence does not properly
account for correlations among the data recorded during a given event !which
might cause the durations for a particular earthquake to be unusually large or
small".
The objective of the work described in this article is to develop prediction equations
for significant duration that !1" rationally account for the effects of magnitude, distance,
site condition, and near-fault rupture directivity; and !2" are based on a mixed-effects
regression procedure that accounts for inter- and intra-event ground-motion variability.

STRONG-MOTION DATABASE
We used a ground-motion database consisting of 1,829 recordings from 149 earth-
quakes. These recordings are based on worldwide shallow crustal earthquakes near ac-
tive plate margins. Subduction and intra-plate events are excluded. Due to recordings
with unknown or poor estimates of magnitude, distance, or site condition, the data set
was reduced to 1,559 recordings from 73 events. Table 2 lists the reduced data set, which
provides a reasonable amount of data over a magnitude range of M & 5 – 7.6 and closest
site-source distance range r & 0 – 200 km. Figure 3 shows the M and r distribution rep-
resented in the database. Magnitude !M" is taken as moment magnitude where available,
and is otherwise taken as surface wave magnitude for M # 6 and local magnitude for
M $ 6. The data were obtained from the strong-motion database maintained by the Pa-
cific Earthquake Engineering Research Center !http://peer.berkeley.edu/smcat/". This
data set has been uniformly processed, which is important to avoid artificial record-to-
record variations in duration that might result from variable processing.
Included in the strong-motion database are near-fault rupture directivity parameters
as defined by Somerville et al. !1997". Rupture directivity parameters were originally
compiled by Somerville et al. !1997", and were recently updated by Somerville and co-
990 J. J. KEMPTON AND J. P. STEWART

Table 2. Summary of strong-motion database used in this study

Distance Number of
Event Year Mo-day Time Magnitude Range !km" Recordings

Imperial Valley 1940 519 437 7.0 8.3 1

Kern County 1952 721 1153 7.4 41–127 5
San Francisco 1957 322 1944 5.3 9.5 1
Parkfield 1966 628 426 6.1 0.1–60 5
Borrego Mtn 1968 409 230 6.8 46–217.4 5
Lytle Creek 1970 912 1430 5.4 15.4–107.8 10
San Fernando 1971 209 1400 6.6 2.8–223 55
Point Mugu 1973 221 1445 5.8 25 1
Hollister 1974 1128 2301 5.2 11.1–12.3 3
Oroville 1975 801 2020 6.0 9.5 1
Oroville 1975 802 2022 5.0 12.7–14.6 2
Oroville 1975 802 2059 4.4 11.1–15 3
Oroville 1975 808 700 4.7 6.5–13.3 9
Friuli, Italy 1976 506 2000 6.5 34.6–97.1 5
Gazli, USSR 1976 517 6.8 3 1
Friuli, Italy 1976 915 315 6.1 10.8–36.1 4
Santa Barbara 1978 813 6.0 14–36.6 2
Tabas, Iran 1978 916 7.4 3–199.1 7
Coyote Lake 1979 806 1705 5.7 3.1–31.2 10
Imperial Valley 1979 1015 2316 6.5 0.5–54.1 35
Imperial Valley 1979 1015 2319 5.2 12.2–52.1 16
Imperial Valley 1979 1016 658 5.5 11.2 1
Livermore 1980 124 1900 5.8 12.9–37.3 7
Livermore 1980 127 233 5.4 3.6–31 8
Anza !Horse Cany" 1980 225 1047 4.9 5.8–40.6 5
Mammoth Lakes 1980 527 1901 4.9 1.8–9.9 6
Mammoth Lakes 1980 531 1516 4.9 7.3–11.8 7
Victoria, Mexico 1980 609 328 6.4 34.8–62.6 5
Mammoth Lakes 1980 611 441 5.0 7.6–14.2 9
Taiwan SMART1!5" 1981 129 5.7 21 7
Westmorland 1981 426 1209 5.8 10.1–26.5 6
Coalinga 1983 502 2342 6.4 8.5–55.2 49
Coalinga 1983 509 249 5.0 12.1–20.3 21
Coalinga 1983 611 309 5.3 9.7–10.5 3
Coalinga 1983 709 740 5.2 10–17 11
Coalinga 1983 722 239 5.8 8.2–17.4 11
Coalinga 1983 722 343 4.9 12.1–13.7 2
Coalinga 1983 725 2231 5.2 12.7–14.7 2
Coalinga 1983 909 916 5.3 13.7–18.4 2
Morgan Hill 1984 424 2115 6.2 0.1–71.2 31
Bishop !Rnd Val" 1984 1123 1912 5.8 19 1
Nahanni, Canada 1985 1223 6.8 6–16 3
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 991

Table 2. !cont."

Distance Number of
Event Year Mo-day Time Magnitude Range !km" Recordings

Hollister 1986 126 1920 5.4 14.9–16.9 4

Taiwan 1986 520 6.4 64 8
SMART1!40"
N. Palm Springs 1986 708 920 6.0 7.3–82.8 32
Chalfant Valley 1986 720 1429 5.9 11–27 5
Chalfant Valley 1986 721 1442 6.2 9.2–50.8 11
Chalfant Valley 1986 721 1451 5.6 14–20 3
Chalfant Valley 1986 731 722 5.8 13–21 2
Whittier Narrows 1987 1001 1442 6.0 9–105 121
Whittier Narrows 1987 1004 1059 5.3 12.6–42.7 11
Superstitn Hills!A" 1987 1124 514 6.3 24.7 1
Superstitn Hills!B" 1987 1124 1316 6.7 0.7–28.3 11
Spitak, Armenia 1988 1207 6.8 30 1
Loma Prieta 1989 1018 5 6.9 5.1–99.2 65
Big Bear 1992 628 805 6.4 8.6–151 39
Cape Mendocino 1992 425 1806 7.1 8.5–44.6 6
Landers 1992 628 1158 7.3 0–194.1 69
Northridge 1994 117 1231 6.7 2.6–146.5 159
Northridge 1994 117 431 5.9 8.9–85 26
Aftershock
Northridge 1994 117 440 5.2 11–46.3 8
Aftershock
Northridge 1994 117 1533 5.6 19.9–58.5 9
Aftershock
Northridge 1994 117 1643 5.2 22.6–61.8 8
Aftershock
Northridge 1994 320 1320 5.2 13.4–92.2 29
Aftershock
Double Springs 1994 912 1223 6.1 17.7 1
Kobe, Japan 1995 116 2046 6.9 0.2–157.2 25
Kozani, Greece 1995 513 847 6.6 18.9–92.8 6
Dinar, Turkey 1995 1001 1557 6.2 26.3–243.3 5
Aqaba, Jordan 1995 1122 616 7.1 93.8 1
Izmit, Turkey 1999 817 0302 7.4 5–183.3 21
Chi-Chi, Taiwan 1999 921 0147 7.6 0.23–185.5 398
Duzce, Turkey 1999 1012 1657 7.2 0.9–188.4 21
Hector Mine 1999 1016 946 7.1 48.6–245 75

workers as part of the NGA project !Graves 2004, pers. comm.". Near-fault parameters
from the NGA project were used in the present study. There are a total of 306 recordings
from 27 events for which rupture directivity parameters are available.
992 J. J. KEMPTON AND J. P. STEWART

Figure 3. Distribution of strong-motion data used in this study.

Two types of site parameters were compiled for the strong-motion sites in the data-
base. The first is intended to represent the properties of near-surface sediments. The pa-
rameter Vs-30 was used for this purpose, which is the ratio of 30 m to the vertical shear-
wave travel time through the upper 30 m of the site. Parameter Vs-30 has been found to
be an effective predictor of site effects on response spectral acceleration !e.g., Borcherdt
1994, Choi and Stewart 2005", and forms the basis of site classifications used in building
codes !e.g., Dobry et al. 2000". A database of Vs-30 parameters was compiled for strong-
motion sites by Stewart et al. !2001" and has been continuously updated over time as
additional data has become available !Stewart et al. 2005". The database used here
matches that of Stewart et al. !2005". The database includes sites where Vs-30 values were
measured based on boreholes near the station, as well as sites where Vs-30 values were
estimated as part of the NGA project. Velocity estimates were based on classified surface
geology by Borcherdt and coworkers !e.g., Borcherdt and Glassmoyer 1994, Borcherdt
2002", geologic classifications by Christopher Wills !developed as part of the NGA
project", or Geomatrix site classifications by Walter Silva. Figure 4 presents a histogram
of Vs-30 values for the strong ground-motion sites used in this study. There are a total of
557 sites for which Vs-30 values are available. Those sites have produced 968 recordings
from 48 events.
The second type of site parameter is intended to represent the effects of relatively
deep basin structure. The parameterization of basin effects involves the use of a depth
term and an index to differentiate sites located in basins overlying the seismic source
from those in basins located to the side of the seismic source. Depth is parameterized by
distance from the ground surface to the 1.5 km/ s shear-wave isosurface, which is de-
noted z1.5. Depth terms are evaluated for sites in southern California and the San Fran-
cisco Bay Area using the basin models of Magistrale et al. !2000" and Hole et al. !2000",
respectively. The index related to source/basin location is defined as follows:
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 993

Figure 4. Histogram of Vs-30 values for sites used in this study !measured+ estimated
velocities".

• Coincident source and site basin locations !CBL"

• Distinct source and site basin locations !DBL"
Detailed protocols for establishment of the CBL and DBL conditions for a given
site/earthquake pair are given in Choi et al. !2005". There are 563 recordings from 22
events for which basin depth parameters are available; a subset consisting of 284 record-
ings also has DBL/CBL designations.

REGRESSION MODEL
The form of the regression model described here is similar to that used previously by
Abrahamson and Silva !1996". Theoretical seismic source models formulated in the fre-
quency domain !e.g., Hanks and McGuire 1981, Boore 1983" suggest that source dura-
tion !Dsource" is inversely related to a corner frequency fc in the Fourier amplitude spec-
trum !when Fourier amplitude spectra are plotted in log-log space, fc denotes the
frequency that separates a relatively flat portion of the spectrum at mid-frequencies from
a decaying region at low frequencies".

1
Dsource = !3"
fc
Brune !1970, 1971" has related fc to the seismic moment and stress drop as follows:

fc = 4.9 · 106%!&'/M0"1/3 !4"

where % is the shear-wave velocity at the source !taken as 3.2 km/ s", &' is the stress
drop !in bars", and M0 is the seismic moment !in dyne-cm". We refer to &' as “stress
drop index” in lieu of “stress drop” to emphasize that it is not the true stress drop of the
earthquake event !Atkinson and Beresnev 1997", but simply an estimate of stress drop
based on analysis of duration parameters within the context of the Brune source model.
994 J. J. KEMPTON AND J. P. STEWART

Increases in significant duration associated with wave propagation and site effects
are taken as additive to the source term, i.e.,

1
SD = + f !r" + f2!S" !5"
fc!M0,&'" 1
where SD represents significant duration, f1!r" represents the distance dependence. and
f2!S" represents the site dependence. The rationale for the additive !as opposed to mul-
tiplicative" distance term is that the spreading of seismic waves with distance would be
expected to be similar for small- and large-magnitude earthquakes since the wave propa-
gation velocities are similar. Likewise, extensions of duration due to site effects such as
resonance within sedimentary layers or basin trapping would be expected to extend du-
rations a fixed amount of time !which would be a function of the sediment stratigraphy
and seismic velocities", regardless of the duration of shaking on the reference !rock"
condition.
The spreading in time of seismic waves with distance from the source is related to
their propagation velocities. Accordingly, the distance dependence is taken as linear:

f1!r" = c2r !6"

where c2 is a regression parameter. The site dependence of significant duration is evalu-
ated in several ways, and is addressed subsequently. Therefore, the base model for the
regression analysis is written as follows:

ln!SD"ij = ln ' ( &'

)
101.5Mi+16.05
4.9 · 106%
−1/3

+ c2rij + f2!Sij" * + (i + )ij !7"

where (i is the event term for earthquake event i !explained below" and )ij represents the
residual for recording j in event i. In Equation 7, seismic moment is converted to mag-
nitude !M" as M0 = 101.5M+16.05 !Hanks and Kanamori 1979". The natural logarithm of
significant duration is used because duration is assumed to be lognormally distributed
!details given below". By using the natural log of the significant duration, )ij is normally
distributed.
Regression analyses utilizing Equation 7 are performed using mixed-effects proce-
dures similar to Abrahamson and Youngs !1992" as implemented in the program R !Pin-
heiro and Bates 2000". Mixed-effects regression is a maximum-likelihood method that
accounts for correlations in the data recorded by a single earthquake !i.e., the data for a
given event may have unusually low or high durations, as represented by event term (i".
Mixed-effects regressions produce two types of error terms: inter-event terms and intra-
event terms. The total standard deviation 'total of the data combines the inter- and intra-
event variability,
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 995

Table 3. Regression parameters for preliminary models with stress drop index independent of
magnitude

&' !bars" c1 c2 % * ' 'total

Da5-75 411± 164 0.52± 0.23 0.07± 0.01 3.2 0.37 0.44 0.57
Da5-95 24± 9 1.91± 0.50 0.15± 0.02 3.2 0.34 0.38 0.51
Dv5-75 236± 114 1.17± 0.38 0.10± 0.02 3.2 0.44 0.52 0.68
Dv5-95 7.7± 3.2 2.47± 0.42 0.15± 0.03 3.2 0.47 0.39 0.61

'total = +'2 + *2 !8"

where ' = intra-event standard deviation !i.e., standard deviation of )ij" and
* = inter-event standard deviation !i.e., standard deviation of (i".

REGRESSION RESULTS

PRELIMINARY MODEL FOR EVALUATION OF SOURCE AND PATH EFFECTS

We begin by using a simple binary site model in which site condition is parameter-
ized as S = 0 for rock and S = 1 for soil. This is done to minimize the number of regres-
sion parameters, which optimizes the estimation of source and path parameters. With the
simple binary site parameter, site term f2!S" has only one regression coefficient:

f 2 = c 1S !9"
where c1 is the regression coefficient. Regression analyses are performed according to
Equation 7 with the above substitution for the site term.
We begin by taking &' as a constant, and hence simultaneously regress on param-
eters c1, c2, and &'. The results for each of the four duration parameters
!Da5-75 , Da5-95 , Dv5-75 , Dv5-95" are listed in Table 3. The estimation errors in Table 3 are
the half-widths of the ±95% confidence intervals. Figure 5a shows an example histo-
gram of the intra-event model residuals. The natural logs of the intra-event residuals for

Figure 5. Histogram of residuals for significant-duration parameter Da5-95.

996 J. J. KEMPTON AND J. P. STEWART

Table 4. Stress drop index !&'" for discrete magnitude bins

Magnitude bin Da5-75 Da5-95 Dv5-75 Dv5-95

4.4 to 5.0 211± 385 2.7± 4.0 37± 120 0.3± 0.5
5.0 to 5.5 1079± 2493 7.5± 7.9 1652± 5950 1.2± 1.1
5.5 to 6.0 287± 316 6.6± 5.6 52± 70 1.8± 1.9
6.0 to 6.5 725± 752 33± 22 238± 250 9.3± 6.3
6.5 to 7.0 513± 318 61± 33 357± 258 30± 16
7.0 to 7.5 524± 299 93± 42 183± 155 45± 31
7.6 !Chi-Chi" 334± 128 12± 15 147± 142 5.2± 6.9

Da5-75 are found to be normally distributed per the chi-square test !Ang and Tang 1975,
p. 274" at a significance level of 95%; the other duration parameters are not normally
distributed at the 95% significance level !e.g., Da5-75 in Figure 5b", although the normal
distribution provides a better fit than other distributions considered such as lognormal.
Figure 5b similarly shows a histogram of inter-event residuals !event terms", which were
generally found to be normally distributed at a significance level of 95%. Based on these
results, we assume a lognormal distribution for the duration data.
No significant variations in bias were found when the data is grouped by rupture
mechanism !reverse, normal, strike-slip". Accordingly, this effect is not considered sub-
sequently. We note that bilateral fault ruptures !such as from the 1989 Loma Prieta, Cali-
fornia, earthquake" have been shown to produce reduced durations relative to unilateral
ruptures !Bommer and Martinez-Pereira 1999". However, this effect is not considered
since the occurrence of bilateral rupture is not predictable, and hence event-to-event
variability in bilateral versus unilateral rupture contributes to the inter-event standard de-
viation *.
As shown in Table 3, the stress drop index, &', varies significantly between the 5-75
duration parameters !for which it is relatively large" and the 5-95 duration parameters
!for which it is small". This is a consequence of 5-95 durations being significantly larger
than 5-75 durations !e.g., Figure 2", which forces a lower &' term per the selected func-
tional form in Equation 7. These differences highlight that fact that &' is a stress drop
index related to the duration definition and the functional form of the regression model,
and is not a true stress drop.
We expect that the stress drop index &' may be magnitude dependent based on
physical considerations !increased fault slip in large-magnitude earthquakes should
cause increased stress drop" and previous experience with similar models #magnitude-
dependent stress drop terms were identified for duration by Abrahamson and Silva
!1996" and for equivalent number of cycles by Liu et al. !2001"$. To investigate the mag-
nitude dependence of &', we sort the data into magnitude bins !i.e., M
= 4.4– 5.0, 5.0– 5.5, etc.", then perform regression analyses according to Equation 7
!with site term from Equation 9" with c1 and c2 fixed at the values in Table 3. Hence &'
is the only free parameter. The results are presented in Table 4, and are plotted for the
Da5-75 and Da5-95 parameters in Figures 6 and 7.
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 997

Figure 6. Estimated stress drop index and model for stress drop independent of magnitude for
the Da5-75 data.

For parameters Da5-75 and Dv5-75, the stress drop has no clear magnitude dependence,
so a constant &', set at the values from Table 3, is adequate !the horizontal line in Fig-
ure 6 is drawn at this value of &'". For parameters Da5-95 and Dv5-95, &' increases with
magnitude !e.g., Figure 7". To capture the trend of magnitude-dependent stress drop, an
exponential model for &' was adopted:

&' = exp#b1 + b2!M − M*"$ !10"

where b1 and b2 are regression coefficients and M* is a reference magnitude taken as 6.

Figure 7. Estimated stress drop index and model for stress drop as a function of magnitude for
the Da5-95 data.
998 J. J. KEMPTON AND J. P. STEWART

Table 5. Regression coefficients for magnitude-dependent stress drop index !&'"

b1 b2 1 −p 1 c 1* c 2* %* * ' 'total
Da5-75 6.02± 0.40 n/a n/a 0.52 0.07 3.2 0.37 0.44 0.57
Da5-95 2.79± 0.36 0.82± 0.38 1.00 1.91 0.15 3.2 0.27 0.38 0.47
Dv5-75 5.46± 0.48 n/a n/a 1.17 0.10 3.2 0.44 0.50 0.66
Dv5-95 1.53± 0.37 1.34± 0.40 1.00 2.47 0.15 3.2 0.34 0.40 0.52
1
rejection confidence for a b2 = 0 model !null hypothesis of zero slope"
*
Identical to values in Table 3

When Equation 10 is used to represent the stress drop index, the regression equation
becomes

ln!SD"ij = ln ' ( exp!b1 + b2!Mi − M*""

101.5Mi+16.05
4.9 · 106%
) −1/3

+ c2rij + f2!S"ij * + (i + )ij !11"

Parameters b1 and b2 were estimated by performing regression analyses with Equation

11 !site term from Equation 9" with c1 and c2 fixed at the values in Table 3. The results
are listed in Table 5, and the resulting models for stress drop index are plotted as lines in
Figures 6 and 7.
As shown in Figures 6 and 7, the Chi Chi earthquake data shows significantly lower
stress drop index than the neighboring bin of M = 7.0– 7.5. This implies higher durations
from Chi Chi than other earthquakes with similar !although slightly lower" magnitude.
This low value of stress drop is consistent with unusually low high-frequency ground-
motion amplitudes recorded during that earthquake !e.g., Boore 2001". Nonetheless, the
Chi Chi event is now generally included in strong-motion databases for shallow crustal
earthquakes in active tectonic regions !e.g., the data is included in the NGA database",
and hence data from this event was included for the regression analyses of &', b1, and
b 2.
To test the statistical significance of the magnitude dependence of stress drop for pa-
rameters Da5-95 and Dv5-95, we compile sample ‘t’ statistics to test the null hypothesis
that b2 = 0. This statistical testing provides a significance level= p that the null hypothesis
cannot be rejected. For clarity of expression, we tabulate in Table 5 values of 1-p, which
we refer to as a “rejection confidence for a b2 = 0 model.” The large rejection confidence
levels !i.e., #95%" shown in Table 5 indicate significant magnitude-dependence in stress
drop index. This is further demonstrated by the b2 values being greater than their stan-
dard error. Note also that by utilizing the magnitude-dependent stress drop, the standard
deviation !'total" drops by 0.04 for Da5-95 and 0.09 for Dv5-95 !seen by comparing Tables
3 and 5".
The model residuals !i.e., residual= data−model in natural logarithmic units" are
plotted as functions of M and r in Figure 8 for parameter Da5-95. The results show no
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 999

Figure 8. Residuals of regression model with magnitude-dependent stress drop for Da5-95 plot-
ted as function of magnitude !M" and distance !r".

clear bias with respect to magnitude, but a clear overprediction bias at small distances
!r $ , 20 km". Residuals for the other duration parameters show similar trends !Kemp-
ton 2004". This bias is addressed subsequently in this article.

BASE MODEL INCLUDING SHEAR-WAVE-VELOCITY-DEPENDENT SITE TERM

Figure 9a shows the residuals of the preliminary model developed above plotted
against Vs-30. The linear trend line shows decreasing residual with increasing Vs-30 !i.e.,
increasing overprediction of duration with increasing Vs-30". Accordingly, a site term that
linearly varies with Vs-30 is selected:

f2 = c4 + c5!Vs-30" !12"
Functional forms for f2 that use the log of Vs-30 were also considered, but the linear Vs-30
term was found to produce a lower standard deviation of residuals. Incorporating the
substitution in Equation 12, the regression equation !Equation 11" is expanded to
1000 J. J. KEMPTON AND J. P. STEWART

Figure 9. Residuals of regression model for Da5-95 plotted as function of Vs-30 for !a" prelimi-
nary model !rock/soil site term" and !b" base model !velocity-dependent site term".

ln!SD"ij = ln ' ( exp!b1 + b2!Mi − M*""

101.5Mi+16.05
4.9 · 106%
) −1/3

+ rijc2 + !c4 + c5!Vs-30"ij" * + (i + )ij

!13"
Mixed-effects regression analyses were performed according to Equation 13, but
with c2, b1, and b2 fixed at the values in Table 5. The database utilized for these analyses
is smaller than that utilized for the preliminary analyses, because not all strong-motion
sites have Vs-30 values !968 recordings from 48 earthquakes in the present analyses ver-
sus 1,557 recordings from 73 earthquakes previously". These regression analyses are
used only to establish regression coefficients c4 and c5 !other parameters are fixed at
their previous values". The results are listed in Table 6. We define the “base model” for

Table 6. Regression coefficients for model that includes Vs-30-based site term

c4 c5 1 −p 1 b 1* b 2* c 2* %* * ' 'total
Da5-75 0.82± 0.34 −0.0013± 0.0004 1.00 6.02 n/a 0.07 3.2 0.32 0.42 0.53
Da5-95 3.00± 0.82 −0.0041± 0.0008 1.00 2.79 0.82 0.15 3.2 0.26 0.36 0.44
Dv5-75 1.40± 0.58 −0.0022± 0.0005 1.00 5.46 n/a 0.10 3.2 0.45 0.51 0.68
Dv5-95 3.99± 1.29 −0.0062± 0.0011 1.00 1.53 1.34 0.15 3.2 0.31 0.39 0.50
1
rejection confidence for a c5 = 0 model !null hypothesis of zero slope"
*
Identical to values in Table 3
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1001

Figure 10. Plots of model vs. data for Da5-95.

significant duration as Equation 13 coupled with the regression coefficients in Table 6.

The hypothesis test results listed in Table 6 indicate that the slope parameter !c5" is
statistically significant for each of the duration parameters. The standard-deviation terms
!'total" are generally lower as a result of utilizing the Vs-30-dependent site term !compar-
ing Tables 5 and 6". Residuals for Da5-95 are plotted as a function of Vs-30 in Figure 9b
along with linear fit lines. The previously noted trend !from Figure 9a" has been effec-
tively removed by the Vs-30-dependent site term. Residuals against magnitude and dis-
tance are not significantly different from what was shown previously in Figure 8 !Kemp-
ton 2004".
The above results for parameters c4 and c5 can be used to estimate the reference site
condition for the duration models, which is the condition for which f2!S" = 0 = −c4 / c5.
This reference velocity is 630– 640 m / s for Da5-75, Dv5-75, and Dv5-95, and 730 m / s for
Da5-95.
In Figure 10, the model for Da5-95 !±one standard deviation" is plotted through the
data in the r = 10– 20 km and r = 20– 40 km distance ranges for sites with Vs-30
= 200– 400 m / s !model is plotted for Vs-30 = 300 m / s" and Vs-30 = 400– 800 m / s !model
1002 J. J. KEMPTON AND J. P. STEWART

is plotted for Vs-30 = 600 m / s". The model predictions appear to be generally compatible
with the data. Similar results are obtained for other duration parameters !Kempton
2004".

DEPENDENCE OF BASE MODEL RESIDUALS ON BASIN PARAMETERS

We evaluate the effects of deep basin structure on significant duration by adding a
linear function of basin depth to the site term, i.e.,

f2 = c4 + c5!Vs-30" + c6 + c7!z1.5" !14"

where c6 and c7 are coefficients determined by regression analysis. Note that the site
term for basin effects is additive, as was the Vs-30 term. Because only a relatively small
subset of the database has depth parameters, regression analyses are performed with
fixed values of c4, c5, source/distance parameters, and event terms from the base model.
We utilize the regression equation:

ln!SD"ij = ln#!SDm"ij + c6 + c7!z1.5"ij$ + (i + )ij !15"

where !SDm"ij refers to the median prediction of the base model !Equation 13" for site j
in event i !i.e., the term in brackets on the right-hand side of Equation 13", (i is the
event term for earthquake i, and )ij is the residual term defined previously. To define a
regression equation in arithmetic units, we set )ij = 0 and take the exponent of both sides
to obtain

!SD"ij)−( − !SDm"ij = c6 + c7!z1.5"ij !16"

Regressions to estimate c6 and c7 were least squares, which gives equal weight to each
data point. This is appropriate because of the inclusion of the event term in Equation 16.
One set of regression analyses utilizes all data for which z1.5 is defined. Results are
given in Table 7 in rows labeled “All,” and example plots of residuals against z1.5 are
given in Figure 11a. Significant-duration parameters other than Da5-75 have statistically
significant depth dependence, as seen by the relatively high rejection confidence !1-p"
values for the zero slope null hypothesis. Parameter Da5-75 tends to be more dominated
by body waves than the other duration parameters considered, and the duration of body
waves would not be expected to be significantly affected by basin geometry. This may
explain the lack of depth dependence for Da5-75. Surface waves can significantly con-
tribute to the other duration parameters, so the depth dependence observed for those pa-
rameters is not surprising. The size of the correction for those parameters ranges from
nearly zero at shallow depth to 3 – 4 s at about 3,500 m depth.
Additional regression analyses were performed for deep basin sites !z1.5 # 500 m"
with the data segregated according to whether the source underlies the basin !coincident
source and site basin locations, denoted as CBL" or is located outside of the basin pe-
rimeter !distinct source and site basin locations, denoted as DBL". These results are also
given in Table 7, with representative plots in Figures 11b and 11c. CBL consistently
shows stronger depth dependence than DBL. The CBL depth dependence is towards de-
creasing duration in deeper basins, whereas DBL shows the opposite. Durations for DBL
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1003

Table 7. Regression coefficient showing dependence of base model re-

siduals on basin parameters

Std Dev.
Group c6 !s" c7 !s/m" !s" 1 – p*
Da5-75 All 0.40± 0.22 3.0E − 05± 1.9E − 04 3.41 0.13
CBL 1.91± 0.75 −7.5E − 04± 4.2E − 04 2.86 0.93
DBL −0.08± 0.62 7.0E − 04± 4.3E − 04 4.21 0.89

Da5-95 All −0.44± 0.44 1.2E − 03± 3.7E − 04 6.73 1.00

CBL 3.25± 1.17 −1.2E − 03± 6.5E − 04 4.44 0.94
DBL 0.09± 1.25 9.3E − 04± 8.6E − 04 8.41 0.72

Dv5-75 All −0.26± 0.37 1.1E − 03± 3.1E − 04 5.60 1.00

CBL 4.43± 1.05 −1.9E − 03± 5.8E − 04 4.01 1.00
DBL 0.43± 1.00 7.8E − 04± 6.9E − 04 6.74 0.74

Dv5-95 All −0.14± 0.55 7.7E − 04± 4.7E − 04 8.47 0.90

CBL 5.52± 1.31 −2.7E − 03± 7.3E − 04 4.97 1.00
DBL 0.78± 1.60 2.5E − 04± 1.1E − 03 10.79 0.18
*
rejection confidence for a c7 = 0 model !null hypothesis of zero slope"

consistently exceed those for CBL at z1.5 # , 1500 m. The larger durations for the DBL
case are expected because basin edge effects can generate surface waves that travel
across the basin !Graves 1993". The CBL case is dominated by body waves entering the
sediments from beneath; we speculate that durations may decrease with depth because of
increasing effects of material damping, which can reduce high-frequency motions at the
beginning of records to the extent that they do not contribute to significant duration. De-
spite the physically meaningful differences between the data trends for CBL and DBL,
the scatter of the limited available data is such that the statistical distinction between the
two data sets is only marginally significant per the F-test !Cook and Weiberg 1999".
Given the marginal CBL/DBL distinction, the preferred approach for most applica-
tions is to use the coefficients labeled “All” in Table 7 when correcting for basin effects
with Equation 14, although no depth correction is needed for Da5-75. The general mag-
nitude of the basin corrections can be as large as the corrections for shallow site condi-
tions !up to 3 – 4 s in both cases". Both site effects should be considered in the prediction
of significant duration for engineering design.

DEPENDENCE OF BASE MODEL RESIDUALS ON NEAR-FAULT PARAMETERS

Somerville et al. !1997" found that near-fault forward directivity effects reduce
strong-motion duration. Their model for rupture directivity modifies the Abrahamson
and Silva !1996" model, which does not consider near-fault effects.
We evaluate near-fault effects on significant duration using residuals defined as fol-
lows:
1004 J. J. KEMPTON AND J. P. STEWART

Figure 11. Residuals of base model for Da5-95 with respect to basin depth for !a" overall data-
base, !b" sites with coincident source and site basin locations !CBL", and !c" sites with distinct
source and site basin locations !DBL". Residuals are defined per left-hand side of Equation 16.

)ij = ln!SD"ij − #ln!SDm"ij + (i$ !17"

where )ij = residual for recording j in event i !same as residual term used in Equation
13", (i = event term for event i, SDij = significant duration calculated from recording j,
and !SDm"ij = significant duration calculated from the base regression model !Equation
13" with )ij = 0.
Regression analyses were performed to relate residuals as defined from Equation 17
to rupture directivity parameters and to distance for M # 6 earthquakes and site-source
distances r $ 20 km. The distance cutoff of 20 km was selected because of the bias at
r $ 20 km in the base model !Figure 8". The magnitude cutoff of 6.0 was selected based
on previous experience !Somerville et al. 1997, Liu et al. 2001". Regressions against
rupture directivity parameters were performed using the equation
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1005

)ij = c8 + c9RDPij + +ij !18"

where c8 and c9 are regression coefficients, +ij is a residual term with zero mean and
standard deviation ', and RDij is a rupture directivity parameter that describes the de-
gree to which the site is subject to forward versus backward directivity !RDP # , 0.5
for forward directivity, smaller values indicate backward or no directivity #Somerville et
al. 1997$". Our results, which are presented in full in Kempton !2004", indicate statisti-
cally significant values of c8 but not c9. This indicates that near-fault effects are signifi-
cant, but they do not correlate strongly with RDP. Accordingly, a second set of regres-
sion analyses were performed to relate residuals to distance:

)ij = c10!r − 20" + +ij:r $ 20 km

)ij = 0:r # 20 km !19"

where r = site-source distance in km, and c10 is a regression coefficient. Both Equations
18 and 19 provide additive corrections !in log units" to base Equation 13; hence they are
multiplicative in arithmetic units. This is appropriate because near-fault forward direc-
tivity compresses the body wave field !the correction would be expected to be less ap-
plicable to surface waves". Regressions using Equations 18 and 19 are least squares,
which gives equal weight to each data point. This is appropriate because of the inclusion
of the event term in Equation 17.
Results obtained using Equation 19 are summarized in Table 8. Example plots of re-
siduals versus distance are given in Figure 12. Positive values of c10 indicate decreased
duration !relative to the base model" in near-fault regions. When interpreting these re-
sults, it is useful to begin with parameter Da5-75, which tends to be more dominated by
body waves than the other parameters considered. A body wave–dominated duration pa-
rameter would be expected to exhibit the clearest trends with respect to near-fault con-
ditions because the wave stacking phenomena associated with rupture directivity occurs
in shear waves and not surface waves. Moreover, near-fault effects tend to be stronger
for strike-slip earthquakes than for dip-slip earthquakes !Somerville et al. 1997", and
hence we initially focus on data for strike-slip earthquakes.
Figure 12 shows the residuals of data from strike-slip earthquakes against the base
model along with the results of regression analyses for this data using Equation 19. Con-
fidence intervals !±95% " are also shown around the fit line. The base model for Da5-75
has no significant near-fault bias for backward directivity conditions in strike-slip earth-
quakes !i.e., c10 values are nearly zero" but a strong bias exists for forward-directivity
conditions in strike-slip earthquakes !i.e., c10 values are positive and significantly non-
zero". This is consistent with previous findings !Somerville et al. 1997" and expected
patterns of behavior for shear waves.
As shown in Table 8, c10 values for the Da5-75 parameter and dip-slip earthquakes are
significantly positive but not significantly different for forward- and backward-
directivity conditions. This suggests that some bias exists in the model formulation in
near-fault regions that can be corrected by Equation 19, but this bias is unlikely to be
1006 J. J. KEMPTON AND J. P. STEWART

Table 8. Regression coefficients showing dependence of base model residuals on near-fault

rupture directivity

Slip
type1 Directivity2 c10 ' p

SS Backward 0.007± 0.005 0.373 0.166

Forward 0.016± 0.007 0.453 0.041
Da5-75 DS Backward 0.022± 0.007 0.290 0.002
Forward 0.016± 0.010 0.480 0.133
SS+ DS Backward 0.010± 0.004 0.342 0.008
Forward 0.016± 0.006 0.149 0.010
SS Backward 0.014± 0.004 0.337 0.003
Forward 0.013± 0.006 0.372 0.040
Da5-95 DS Backward 0.018± 0.006 0.243 0.004
Forward 0.015± 0.007 0.353 0.056
SS+ DS Backward 0.015± 0.003 0.296 0.000
Forward 0.014± 0.005 0.361 0.004
SS Backward 0.020± 0.006 0.446 0.001
Forward 0.023± 0.009 0.554 0.019
Dv5-75 DS Backward 0.030± 0.010 0.449 0.007
Forward 0.023± 0.012 0.616 0.083
SS+ DS Backward 0.023± 0.005 0.446 0.000
Forward 0.023± 0.007 0.569 0.003
SS Backward 0.020± 0.005 0.360 0.000
Forward 0.016± 0.007 0.420 0.024
Dv5-95 DS Backward 0.012± 0.006 0.260 0.052
Forward 0.026± 0.009 0.464 0.014
SS+ DS Backward 0.018± 0.004 0.318 0.000
Forward 0.020± 0.006 0.434 0.001
1
Oblique earthquakes are treated as SS or DS depending on orientation of principal rupture slip
2
RDP# , 0.5 for forward directivity and RDP$ , 0.5 for backward directivity

related to rupture directivity. We speculate that the lack of a clear rupture directivity ef-
fect for Da5-75 in dip-slip earthquakes results from relatively modest rupture directivity
!e.g., Somerville et al. 1997".
As shown by the results in Table 8, the other duration parameters considered !Da5-95,
Dv5-75, and Dv5-95" show no significant differences between the results for strike-slip ver-
sus dip-slip earthquakes nor for forward versus backward directivity. We speculate that
this lack of a clear rupture directivity effect occurs because these duration parameters
are sensitive to surface waves, which are not subject to rupture directivity effects.
Based on these results, we recommend that the base model be corrected for near-
fault conditions using Equation 19 with the c10 coefficients given in Table 9. This cor-
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1007

Figure 12. Residuals of base model for Da5-75 with respect to site-source distance for case of
limited or no directivity !top frame" and forward directivity !bottom frame". Data shown are
from strike-slip earthquakes.

rects for rupture directivity effects when applied to parameter Da5-75 and strike-slip
earthquakes; otherwise it simply corrects for bias in the distance term used in the base
model. The intra-event standard-deviation terms !'" given in Table 8 are similar to those
for the base model !Table 6"; hence no adjustments to dispersion terms are needed when
using the near-fault corrections. The effects of these near-fault corrections on the Da5-75
and Da5-95 median predictions are shown in Figure 13.

COMPARISONS TO PREVIOUS STUDIES

Figures 14 and 15 compare significant durations from our base model !with near-
fault corrections" to predictions from previous models. Figure 14 illustrates magnitude
scaling for r = 30 km and rock sites. The trend of our results is similar to that of Trifunac
and Brady !1975", Kamiyama !1984", and Abrahamson and Silva !1996". The slopes of
the duration-magnitude curves increase with magnitude in our model, an effect shown
previously only by Abrahamson and Silva !1996".

Table 9. Recommended values of parameter c10

Da5-75
Da5-95 Dv5-75 Dv5-95
SS DS SS+ DS SS+ DS SS+ DS

0 !bkwd"
C10 0.020 0.015 0.023 0.019
0.016 !frwd"
1008 J. J. KEMPTON AND J. P. STEWART

Figure 13. Median durations !Da5-75 and Da5-95" for base models and base models with near-
fault corrections for M = 7 earthquake at soil site with Vs-30 = 300 m / s. Note that there is no
correction for strike-slip faulting with backward directivity !SS, bkwd".

The distance scaling shown in Figure 15 indicates a similar trend on rock for our
results and those of Trifunac and Brady !1975" and Abrahamson and Silva !1996". That
trend consists of negligible distance dependence at close distance !r $ , 20 km" and in-
creasing slope as r increases. Other models tend to show relatively flat slopes !Dobry et
al. 1978, Kamiyama 1984" or distance invariant slopes on a log scale !McGuire and
Barnhard 1979". These differences result from the use of a distance term that is additive
to the source duration !Trifunac and Brady 1975, Abrahamson and Silva 1996, this
study" versus one that is multiplicative !Dobry et al. 1978, McGuire and Barnhard 1979,
Kamiyama 1984".
We found significant duration to increase with decreasing Vs-30. Shown in Figure 16
is the Vs-30-based site term from this study relative to similarly formatted !i.e., additive"

Figure 14. Comparison of magnitude dependence of base model from this study with results of
previous research for rock sites.
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1009

Figure 15. Comparison of distance dependence of base model from this study with results of
previous research for rock sites !see legend in Figure 14".

site terms from previous investigators !Trifunac and Brady 1975, Abrahamson and Silva
1996". The site terms are similar in the sense that larger significant durations are con-
sistently found on soil than on rock. However, the magnitude of the duration change with
site condition !as measured by the “slope” of the relations" is different among the three
studies, with our factors having a slope intermediate between those of Trifunac/Brady
and Abrahamson/Silva. Other investigators used a multiplicative site term that cannot be
readily compared to those shown in Figure 16 !McGuire and Barnhard 1979, Kamiyama
1984". No previous investigators have developed additive site terms for basin effects on
significant duration, so no comparisons are possible.

Figure 16. Comparison of additive site term from this study with similar site terms from pre-
vious research. Abrahamson and Silva !1996" site categories are soil and rock. Trifunac and
Brady site categories are soft alluvium, intermediate rock, and hard rock. The velocity ranges
indicated are our estimates of the range of conditions that would be present for each category.
1010 J. J. KEMPTON AND J. P. STEWART

SUMMARY AND CONCLUSIONS

We have developed prediction equations for the significant duration of earthquake
ground motion as a function of magnitude, closest site-source distance, near-surface
shear-wave velocity !Vs-30", basin depth, and near-fault parameters. The model was de-
veloped using a worldwide strong-motion database of recordings from active tectonic
regions such as California and portions of Japan, Turkey, and Taiwan.
Significant duration is found to increase strongly with earthquake magnitude and
moderately with distance. The model is considered valid across a magnitude range of
M & 5 to 7.6 and a distance range of r & 0 to 200 km. The base model represented by
Equation 13 and Table 6 includes a site term that is dependent on the near-surface shear-
wave velocity !Vs-30". A statistically significant dependence of duration on Vs-30 was
found, with duration increasing as velocity decreases. Residuals of the base model are
correlated with basin depth for significant-duration parameters other than Da5-75. Dura-
tion tends to decrease with depth when the seismic source is located beneath the basin,
and increase with depth when the source and basin locations are distinct. For current
application, however, a simple model that does not consider source location is likely ad-
equate !Equation 14 with coefficients in Table 7 from rows labeled “All”".
The near-fault correction is an important part of the model because of overprediction
bias for M # 6 and r $ 20 recordings. For parameter Da5-75, distance-dependent near-
fault factors reduce that bias when applied to strike-slip earthquakes and sites subject to
forward directivity. Near-fault corrections for other duration parameters are apparently
unrelated to rupture directivity effects, but the corrections are nonetheless significant.
The near-fault correction to the base model is presented in Equation 19 with coefficients
in Table 9.
The total standard deviation of the duration models varies from a minimum of 0.44
!for Da5-95" to a maximum of 0.68 !for Dv5-75" in natural logarithmic units. These
standard-deviation terms are comparable to what is generally obtained for response
spectral acceleration. The standard-deviation terms have contributions from both intra-
and inter-event variability in the data. The intra-event variability is the larger of the two,
being about 20–40% larger than inter-event variability. Dispersion terms are not signifi-
cantly affected by the near-fault corrections, although significant reductions in standard
deviation were achieved by refining the source and site functions in the development of
the base model.

ACKNOWLEDGMENTS
The work presented in this paper benefited from support 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. Any opinions, findings, and conclusions or recommendations expressed in this
material are those of the authors and do not necessarily reflect those of the National Sci-
ence Foundation. Drs. Yoojoong Choi and Rick Schoenberg are thanked for their assis-
tance with the statistical analyses. Dr. Julian Bommer and two anonymous reviewers are
thanked for their comments, which improved the manuscript.
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1011

REFERENCES
Abrahamson, N. A., and Silva, W. J., 1996. Empirical Ground Motion Models, Report to
Brookhaven National Laboratory.
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., 2004. Quantitative measure of goodness-of-fit of synthetic seismographs, Pro-
ceedings, 13th World Conference on Earthquake Engineering, Vancouver, B.C., Canada, Pa-
per 243.
Ang, A. H.-S., and Tang, W. H., 1975. Probability Concepts in Engineering Planning and De-
sign, Volume I—Basic Principles, John Wiley & Sons, New York.
Arias, A., 1970. A measure of earthquake intensity, in Seismic Design for Nuclear Power
Plants, edited by R. Hansen, MIT Press, Cambridge, MA, pp. 438–483.
Atkinson, G. M., and Beresnev, I. A., 1997. Don’t call it stress drop, Seismol. Res. Lett. 68,
3–4.
Bolt, B. A., 1973. Duration of strong ground motions, Proceedings, 5th World Conference on
Earthquake Engineering, Acapulco, Paper 84.
Bommer, J. J., and Martinez-Pereira, A., 1999. The effective duration of earthquake strong mo-
tion, J. Earthquake Eng. 3, 127–172.
Bommer, J. J., Magenes, G., Hancock, J., and Penazzo, P., 2004. The influence of strong motion
duration on the seismic response of masonry structures, Bull. Earthquake Eng. 2 !1", 1–26.
Bommer, J. J., Hancock, J., and Alarcón, J. E., 2006. Correlations between duration and number
of effective cycles of earthquake ground motion, Soil Dyn. Earthquake Eng. 26 !1", 1–13.
Boore, D. M., 1983. Stochastic simulation of high-frequency ground motions based on seismo-
logical models of the radiated spectra, Bull. Seismol. Soc. Am. 73, 1865–1894.
———, 2001, Comparisons of ground motions from the 1999 Chi-Chi earthquake with empiri-
cal predictions largely based on data from California, Bull. Seismol. Soc. Am. 91, 1212–
1217.
Borcherdt, R. D., 1994. Estimates of site-dependent response spectra for design !methodology
and justification", Earthquake Spectra 10 !4", 617–653.
———, 2002, Empirical evidence for acceleration-dependent amplification factors, Bull. Seis-
mol. Soc. Am. 92, 761–782.
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. Geol. Surv. Prof. Pap. 1551-A, A77–A108.
Chai, Y. H., Fajfar, P., and Romstad, K. M., 1998. Formulation of duration-dependent inelastic
seismic design spectrum, J. Struct. Eng. 124, 913–921.
Choi, Y., and Stewart, J. P., 2005. Nonlinear site amplification as function of 30 m shear wave
velocity, Earthquake Spectra 21 !1", 1–30.
Choi, Y., Stewart, J. P., and Graves, R. W., 2005. Empirical model for basin effects that accounts
for basin depth and source location, Bull. Seismol. Soc. Am. 95, 1412–1427.
1012 J. J. KEMPTON AND J. P. STEWART

Cook, R. D., and Weiberg, S., 1999. Applied Regression Including Computing and Graphics,
John Wiley & Sons, New York.
Cornell, C. A., 1997. Does duration really matter? Proceedings of the FHWA/NCEER Workshop
on the National Representation of Seismic Ground Motion for New and Existing Highway
Facilities, pp. 125–133.
Dobry, R., Idriss, I. M., and Ng, E., 1978. Duration characteristics of horizontal components of
strong-motion earthquake records, Bull. Seismol. Soc. Am. 68, 1487–1520.
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 sys-
tem used in recent building seismic code provisions, Earthquake Spectra 16 !1", 41–67.
Graves, R. W., 1993. Modeling three-dimensional site response effects in the Marina District,
San Francisco, California, Bull. Seismol. Soc. Am. 83, 1042–1063.
Hancock, J., and Bommer, J. J., 2004. The influence of phase and duration on earthquake dam-
age in degrading structures, Proceedings, 13th World Conference on Earthquake Engineer-
ing, Vancouver, B.C., Canada, Paper 1990.
———, 2005. The effective number of cycles of earthquake ground motion, Earthquake Eng.
Struct. Dyn. 34, 637–664.
Hanks, T. C., and Kanamori, H., 1979. A moment magnitude scale, J. Geophys. Res. 84, 2348–
2350.
Hanks, T. C., and McGuire, R. K., 1981. The character of high frequency strong ground mo-
tion, Bull. Seismol. Soc. Am. 71, 2071–2095.
Hole, J. A., Brocher, T. M., Klemperer, S. L., Parsons, T., Benz, H. M., and Furlong, K. P.,
2000. Three-dimensional seismic velocity structure of the San Francisco Bay area, J. Geo-
phys. Res. 105 !B6", 13859–13874.
Husid, L. R., 1969. Características de terremotos, Análisis general, Revista del IDIEM 8, San-
tiago del Chile, pp. 21–42.
Iervolino, I., Manfredi, G., and Cosenza, E., 2006. Ground motion duration effects on nonlinear
seismic response, Earthquake Eng. Struct. Dyn. 35 !1", 21–38.
Kamiyama, M., 1984. Effects of subsoil conditions and other factors on the duration of earth-
quake ground shaking, Proceedings, 8th World Conference on Earthquake Engineering, San
Francisco, Vol. 2, pp. 793–800.
Kempton, J. J., 2004. Prediction Models for Significant Duration of Earthquake Ground Mo-
tions, M.S. dissertation, University of California, Los Angeles.
Liu, A. H., Stewart, J. P., Abrahamson, N. A., and Moriwaki, Y., 2001. Equivalent number of
uniform stress cycles for soil liquefaction analyses, J. Geotech. Geoenviron. Eng. 127,
1017–1026.
Magistrale, H., Day, S., Clayton, R., and Graves, R., 2000. The SCEC southern California ref-
erence three-dimensional seismic velocity model version 2, Bull. Seismol. Soc. Am. 90, S65–
S76.
McGuire, R. K., and Barnhard, T. P., 1979. The usefulness of ground motion duration in pre-
diction of severity of seismic shaking, Proceedings, 2nd U.S. National Conference on Earth-
quake Engineering, Stanford, Calif., pp. 713–722.
PREDICTION EQUATIONS FOR SIGNIFICANT DURATION CONSIDERING SITE AND NEAR-SOURCE EFFECTS 1013

Mohraz, B., and Peng, M.-H., 1989. The use of low-pass filtering in determining the duration of
strong ground motion, Publication PVP-182, Pressure Vessels and Piping Division, ASME,
pp. 197–200.
Novikova, E. I., and Trifunac, M. D., 1994. Duration of strong ground motion in terms of earth-
quake magnitude, epicentral distance, site conditions and site geometry, Earthquake Eng.
Struct. Dyn. 23, 1023–1043.
Pinheiro, J. C., and Bates, D. M., 2000. Mixed-Effects Models in S and S-PLUS, Springer-
Verlag, New York.
Rauch, A. F., and Martin, J. R., 2000. EPOLLS model for predicting average displacements on
lateral spreads, J. Geotech. Engrg. 126, 360–371.
Reinoso, E., Ordaz, M., and Guerrero, R., 2000. Influence of strong ground-motion duration in
seismic design of structures, Proceedings, 12th World Conference on Earthquake Engineer-
ing, Paper 1151.
Sarma, S. K., 1970. Energy flux of strong earthquakes, Tectonophysics 11, 159–173.
Seed, H. B., and Lee, K. L., 1966. Liquefaction of saturated sands during cyclic loading, J. Soil
Mech. and Found. Div. 92, 105–134.
Shome, N., Cornell, C. A., Bazzurro, P., and Carballo, J. E., 1998. Earthquakes, records, and
nonlinear responses, Earthquake Spectra 14 !3", 469–500.
Silver, M. L., and Seed, H. B., 1971. Volume changes in sands during cyclic loading, J. Soil
Mech. and Found. Div. 97, 1171–1182
Somerville, P. G., Smith, N. F., Graves, R. W., and Abrahamson, N. A., 1997. Modification of
empirical strong ground motion attenuation relations to include the amplitude and duration
effects of rupture directivity, Seismol. Res. Lett. 68, 199–222.
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., Choi, Y., and Graves, R. W., 2005. Empirical Characterization of Site Conditions
on Strong Ground Motion, Report No. PEER-2005/01, Pacific Earthquake Engineering Re-
search Center, University of California, Berkeley.
Tiwari, A. K., and Gupta, V. K., 2000. Scaling of ductility and damage-based strength reduction
factors for horizontal motions, Earthquake Eng. Struct. Dyn. 29 !7", 969–987.
Trifunac, M. D., and Brady, A. G., 1975. A study on duration of strong earthquake ground mo-
tion, Bull. Seismol. Soc. Am. 65, 581–626.
Trifunac, M. D., and Westermo, B. D., 1977. A note on the correlation of frequency-dependent
duration of strong earthquake ground motion with the modified Mercalli intensity and the
geologic conditions at the recording site, Bull. Seismol. Soc. Am. 67, 917–927.
Vanmarcke, E. H., and Lai, S.-S. P., 1980. Strong motion duration and RMS amplitude of earth-
quake records, Bull. Seismol. Soc. Am. 70, 1293–1307.
!Received 18 August 2005; accepted 22 December 2005"