Generation of uniform-hazard earthquake ground motions

Henri P. Gavin, M.ASCE

1
and Bryce W. Dickinson, A.M.ASCE
2
ABSTRACT
This paper presents statistical models for the generation of bi-axial earthquake ground mo-
tion time histories with spectra that match those from samples of ground motion records. The
model parameters dene near-eld characteristics such as pulse velocity and pulse period, far-fault
characteristics such as velocity amplitude and power spectral density, and envelope characteris-
tics. The samples of ground motions used in this study were previously selected and scaled to be
representative of particular hazard levels in particular geographical regions. A companion paper
presents the tting of the model to samples of ground motion waveforms. In this paper, the new
concept of a parameter-response correlation spectrum establishes the period-dependence of the
correlation between the response spectrum and ground motion parameters. Parameters that corre-
late to variability of the response spectra are retained as random variables and are then t to mean
and mean-plus-standard deviation of bi-axial response-spectra of the samples of historical records.
Parameter statistics also include correlations between velocity amplitudes and pulse periods.
INTRODUCTION
Peak seismic response statistics of structures that behave with certain nonlinearities including,
but not limited to, discontinuities (e.g., impact and fracture), may be poorly represented by com-
mon distributions (e.g., lognormal). In such cases the computation of peak response distributions
requires many simulations; the sample size required to characterize the peak response distribution
may exceed the of number of available natural ground motions characteristic of the hazard at a site.
Nevertheless, as demonstrated by Dickinson and Gavin (2010) and others, parameters character-
izing a sample of natural ground motions are well-characterized by lognormal distributions. Such
parametric ground motions models may then be used to generate any number of additional ground
motions representative of the sample.
The suites of ground motion records developed for the SAC Steel Project were intended to
represent ground motions with an exceedance probability (e.g., 2% in 50 year or 10% in 50 year)
for a representative geographical region (e.g., Los Angeles or Seattle) for a NEHRP site category
S
D
(rm soil). In addition, a data set representative of near-fault ground motions near Los Angeles
was developed. Ground motions were selected and scaled based upon an approximate deaggrega-
tion for each region and local soil condition, with some preference toward records from nearby or
representative ruptures (Somerville et al. 1997). Each pair of horizontal components was scaled
by a common scaling factor. The scaling factor minimized the weighted sum of the squares of the
differences between the geometric mean spectrum from the records and a 1995 uniform-hazard
spectrum for a NEHRP S
D
site in the selected region. The spectra were matched at periods of
0.3, 1.0, 2.0 and 4.0 s with associated weights of 0.1, 0.3, 0.3, and 0.3. In this sense each pair of
ground motions was scaled to approximate a uniform-hazard spectrum (Somerville et al. 1997).
Each suite provides a sample of ten bi-axial ground acceleration records, with the exception of
1
Dept. Civil & Environ. Engg, Duke Univ., Box 90287, Durham NC 27708
2
Rutherford & Chekene, 55 Second Street, Suite 600, San Francisco, CA 94105
the near-fault set which contains twenty tri-axial records. Within each suite paired horizontal
ground motion records are rotated 45 degrees from their fault-normal and fault-parallel directions
(Somerville et al. 2009). Well over one hundred published studies have made use of this ground
motion database, e.g., (Chang et al. 2002), (Lee and Foutch 2002), (Wen and Song 2003), (Yun
et al. 2002).
Acompanion study (Dickinson and Gavin 2010) examined the wave formcharacteristics of ve
of these data sets: Los Angeles 10% in 50 yr (la10in50), Los Angeles 2% in 50 yr (la2in50),
Seattle 10% in 50 yr (se10in50), Seattle 2% in 50 yr (se2in50), and the near-fault (nrfault)
data sets. Each of the twenty or forty records in each of the ve data sets were individually and
systematically t via constrained non-linear least squares methods in order to determine estimates
and standard errors for eleven ground motion parameters. The eleven parameters describe the
power spectral density of the stochastic content of the record (central ground frequency, f
g
, and
damping
g
), the envelope of the stochastic part of the record (rise time,
1
, constant time,
2
, and
decay time,
3
), the peak velocity of the stochastic part of the record (V
s
), and the pulse (peak pulse
velocity, V
p
, pulse period, T
p
, number of pulse cycles, N
c
, phase of the pulse, , and the pulse
arrival time, T
pk
). Ground motion parameter estimates and their standard errors result in empirical
cumulative distribution functions and distribution statistics (mean and variance) for each parameter
for each data set (Dickinson and Gavin 2010).
Synthetic ground motion records are generated from model parameters as follows. A stochastic
acceleration record is generated from power spectral density parameters (f
g
,
g
) over a specied
frequency range (f
lo
to f
hi
) by the method of spectral representation (Shinozuka and Jan 1972;
Shinozuka and Deodatis 1991). The stochastic acceleration is then time modulated with an en-
velope function (
1
,
2
,
3
) and integrated (in the frequency-domain) to give a stochastic velocity
record. The stochastic velocity record is scaled to the specied amplitude, V
s
. A velocity pulse
(T
p
, V
p
, N
c
, T
pk
, ) is generated and added to the stochastic velocity. The combined velocity is
then differentiated (in the frequency-domain) to give the nal synthetic ground motion accelera-
tion. Frequency components above f
lo
are used in the frequency-domain integration of acceleration
records. Frequency components below f
hi
are used for the differentiation of velocity records. Ex-
pressions used in the waveform synthesis are given in (Dickinson and Gavin 2010).
The parameter estimates developed in the companion study demonstrate that the variability
of wave form characteristics represented in each data set are well-characterized by lognormal
distributions. While these distributions provide insight into the temporal characteristics of each
data set, the effect of these earthquake ground motions on structures is of primary importance for
performance-based design. This study examines the parametrization of ve of the SAC data sets
in terms of response spectra.
PARAMETER-RESPONSE CORRELATION SPECTRA
The relationships between ground motion parameters and structural response parameters have
been investigated recently through a variety of methods (Cordova et al. 2001), (Baker and Cor-
nell 2006), (Baker and Jayaram 2008), (Baker 2007). Correlation analysis between ground motion
parameters and structural response parameters gives a clear picture of the effect that various earth-
quake ground motion characteristics have on structural response. The new concept of a parameter-
response correlation spectrum is introduced here in order to investigate the association between
response spectra and ground motion parameters. Parameter-response correlation spectra give the
statistical correlation between a ground motion waveform parameter and the peak response of a
2 H.P. Gavin and B.W. Dickinson
structure of a given natural period and damping. These spectra are computed by generating many
(a thousand or more) ground acceleration records using parameters drawn from (uncorrelated) sta-
tistical distributions of ground motion model parameters. For each generated set k of parameters,
p
k
, a corresponding ground motion and its response spectrum, s
k
, are computed and the covari-
ance V
P
i
,S
j
between parameter P
i
and response spectrum value S
j
(at natural period T
j
) is updated
recursively,
V
P
i
,S
j
,k
=
k 2
k 1
V
P
i
,S
j
,k1
+
k
(k 1)
2
(p
i,k
p
i,k
)(s
j,k
s
j,k
) , (1)
where the averages of each parameter and response spectrum ordinate are also updated recursively,
p
i,k
=
k 1
k
p
i,k1
+
1
k
p
i,k
, (2)
s
j,k
=
k 1
k
s
j,k1
+
1
k
s
j,k
. (3)
The Pearson correlation between parameter P
i
and response spectrum value S
j
(at natural period
T
j
) is

P
i
,S
j
=
V
P
i
,S
j
_
V
P
i
,P
i
_
V
S
j
,S
j
. (4)
In this study, response spectra and correlation spectra for linear elastic structures with ve
percent damping are evaluated. Correlation spectra for acceleration response and displacement
response are identical. This should not be the case for responses of inelastic structures.
Parameters with correlation values within the range of 10 percent over the entire range of
natural periods do not systematically affect response variability, and are treated as constants in
the development of statistical models consistent with the historical earthquake ground motion data
sets.
The correlation spectra in this study are generated from 1000 ground motion simulations for
each of the ve SAC data sets considered. In this analysis, the eleven ground motion parameters
were taken as uncorrelated lognormal randomvariables with means, variances, and offsets obtained
from the ts of the original SAC ground motion waveforms (Dickinson and Gavin 2010). In
order to capture the transient response of long-period structures, the simulated earthquake ground
motions had a minimum duration of sixty seconds. Padding earthquake ground accelerations with
zeroes for the rst ve seconds reduces errors associated with the frequency-domain integration
method used here (Dickinson 2008), (Boore 2005).
Correlation spectra for each of the eleven parameters are shown in Figure 1. In Figure 1, |R|
max
refers to the spectral displacement from a single-component ground motion record. Results from
each data set for each parameter are plotted together. The ground motion parameters that correlate
most signicantly to peak response statistics (in order of decreasing correlation) are: the peak ve-
locity of the stochastic part of the ground motion, V
s
, the peak pulse velocity, V
p
, the pulse period,
T
p
, the central frequency of the stochastic ground motion, f
g
, and the number of cycles in the pulse,
N
c
. Correlation between the stochastic ground velocity, V
s
, and peak response characteristics are
signicant for structures of all periods (0.1s < T
n
< 10.0s) and for the ve SAC data sets consid-
ered. Correlations between V
s
and peak responses drop off for longer period structures (especially
for the Los Angeles data sets), but remain signicant (> 0.1), nonetheless. The pulse velocity
amplitude, V
p
, is signicantly correlated to the responses of long-period structures (T
n
> 1 s) and
3 H.P. Gavin and B.W. Dickinson
is signicantly correlated to the response of structures of all periods for the nrfault data set. The
pulse period, T
p
, is negatively correlated to peak responses at periods from 1 to 3 seconds and is
positively correlated at periods from 3 to 10 seconds, especially for the nrfault, la2in50, and
la10in50 suites. The central frequency of the stochastic content, f
g
, is somewhat correlated to
the peak responses of short-period structures (T
n
< 1 s), and the number of cycles in the pulse,
N
c
, is only slightly correlated to peak responses of long-period structures (T
n
> 1.5 s) for the
la2in50 and the la10in50 suites.
Correlations of peak responses to the other ground motion parameters are within 10 percent.
The importance of the pulse phase, , has previously been found to be insignicant (Mavroeidis
and Papageorgiou 2003), (Mavroeidis et al. 2004). This result is conrmed here.
Based on these correlation spectra, a ground motion model involving a reduced number of ran-
dom variables is proposed. The ground motion variables that are important to the response of linear
elastic structures (with ve percent critical damping) are V
s
, V
p
, T
p
, and f
g
. The variability in peak
responses due to variability in the remaining ground motion parameters is small in comparison.
CORRELATIONS BETWEEN HORIZONTAL VELOCITY COMPONENTS AND PULSE
PERIOD
The transient response analysis of asymmetric structures (with lateral-torsional modes) requires
the specication of ground motions as paired horizontal components. This study adopts the des-
ignations X and Y for the paired horizontal components. The X and Y components of the
pulse velocity, V
pX
and V
pY
, and the peak stochastic velocity, V
sX
and V
sY
, are taken as identically-
distributed and partially-correlated parameters. For physically-motivated reasons, the pulse period,
T
p
, and the ground frequency, f
g
, for ground motions in the X and Y directions are considered to be
identical. Further, the ground frequency is assumed uncorrelated with the other ve random vari-
ables. The bi-axial ground motion model therefore involves six random variables, ve of which
are correlated.
Correlations among V
pX
, V
pY
, T
p
, V
sX
, and V
sY
, are estimated from previously-identied pa-
rameter values (Dickinson and Gavin 2010). For a data set containing m X records and m Y
records, the m parameter estimates V
pXi
, V
pYi
, T
pXi
, T
pYi
, V
sXi
, and V
sYi
, are arranged in a 2m-by-
5 matrix P,
P =
_

_
V
pX1
V
pY1
T
pX1
V
sX1
V
sY1
V
pX1
V
pY1
T
pY1
V
sX1
V
sY1
V
pX2
V
pY2
T
pX2
V
sX2
V
sY2
V
pX2
V
pY2
T
pY2
V
sX2
V
sY2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
V
pXm
V
pYm
T
pXm
V
sXm
V
sYm
V
pXm
V
pYm
T
pYm
V
sXm
V
sYm
_

_
(5)
Values for the X and Y components of the pulse velocity and stochastic velocity parameters are
repeated for each pulse period value. Since offset lognormal distributions represent the variability
of these parameters, their standardized values are given by the 2m-by-5 matrix Z,
Z =
1
p
_
ln [P1p
o
] 1
_
ln p
1
2
ln
_
p
p
+ 1
___
, (6)
where p
o
is row-vector of parameter offsets given in Table 3 of (Dickinson and Gavin 2010), 1
is a 2m-by-1 vector of 1s, p is a row-vector containing the average of the columns of the matrix
4 H.P. Gavin and B.W. Dickinson
TABLE 1. Correlation matrices for the bi-axial model
V
pX
V
pY
T
p
V
sX
V
sY
nrfault
V
pX
1 0.35 -0.51 0.20 0.10
V
pY
0.35 1 -0.08 -0.03 -0.04
T
p
-0.51 -0.08 1 -0.11 -0.09
V
sX
0.20 -0.03 -0.11 1 0.68
V
sY
0.10 -0.04 -0.09 0.68 1
la2in50
V
pX
1 0.75 -0.58 -0.08 -0.25
V
pY
0.75 1 -0.62 -0.65 -0.68
T
p
-0.58 -0.62 1 0.32 0.43
V
sX
-0.08 -0.65 0.32 1 0.84
V
sY
-0.25 -0.68 0.43 0.84 1
la10in50
V
pX
1 0.52 -0.26 -0.17 0.03
V
pY
0.52 1 -0.51 0.01 0.33
T
p
-0.26 -0.51 1 -0.05 -0.39
V
sX
-0.17 0.01 -0.05 1 0.70
V
sY
0.03 0.33 -0.39 0.70 1
se2in50
V
pX
1 0.94 -0.60 -0.43 -0.23
V
pY
0.94 1 -0.62 -0.48 -0.16
T
p
-0.60 -0.62 1 -0.18 -0.18
V
sX
-0.43 -0.48 -0.18 1 0.55
V
sY
-0.23 -0.16 -0.18 0.55 1
se10in50
V
pX
1 0.82 0.08 -0.46 -0.61
V
pY
0.82 1 -0.05 -0.56 -0.38
T
p
0.08 -0.05 1 0.01 -0.27
V
sX
-0.46 -0.56 0.01 1 0.44
V
sY
-0.61 -0.38 -0.27 0.44 1
[P1p
o
], and p is a row-vector containing the unbiased standard deviation estimate of the columns
of the matrix [P1p
o
]. The correlation matrix of Zis the correlation matrix for the ve parameters.
This method captures correlations between the individual measurements of pulse period and each
of the X and Y components of the pulse amplitude and the peak stochastic velocity.
The resulting 5-by-5 correlation matrices for each data set are given in Table 1. As would be ex-
pected, pulse amplitudes (V
pX
and V
pY
) and peak stochastic velocities (V
sX
and V
sY
) are positively
correlated. Standardized pulse periods are negatively correlated with standardized pulse velocities.
The correlations between peak stochastic velocities and pulse parameters depend substantially on
the data set from which they were estimated.
GENERATING SAMPLES OF GROUND MOTION PARAMETERS AND WAVEFORMS
Given means, variances, offsets, and correlations for lognormal waveform parameters corre-
sponding to a particular ground motion suite, a sample of waveform parameters may be generated.
The lower Cholesky factor, L, of the standardized parameter correlation matrix is used to generate
5 H.P. Gavin and B.W. Dickinson
correlated standard normal samples, Lz, from which correlated lognormal samples for V
pX
, V
pY
,
T
p
, V
sX
, and V
sY
are computed.
p = p
o
+ pexp
_
_
Lz

_
ln
_
p
p
+ 1
_

1
2
ln
_
p
p
+ 1
_
_
_
(7)
The offset values, p
o
, listed for V
s
and T
p
in Table 2 enforce a non-zero minimum value for these
parameters. Extremely large values of V
p
, T
p
, and V
s
are avoided by specifying that the generated
parameter fall below an allowable limit,
V
p
< (
Vp
) +n(
Vp
) , (8)
T
p
< (
Tp
) +n(
Tp
) , (9)
V
s
< (
Vs
) +n(
Vs
) , (10)
where the number of standard deviations, n, should be between 3 and 4. Additionally, the arrival
time of the pulse is restricted such that its arrival is neither too early nor too late to meet physical
expectations about impulsive behavior of ground motions. The pulse arrival time should be later
than the start of the strong ground motion and within the rst quarter of the duration of the record.
(
1
+T
p
N
c
)/2 < T
pk
< (
1
+
2
+
3
)/4 . (11)
If a set of random variables does not meet these requirements, the set is discarded and another set
of parameters are generated. Given a sample of ground motion parameters, a sample of associated
waveforms may be generated using the method described briey in the Introduction and completely
in (Dickinson and Gavin 2010).
RESPONSE SPECTRA COMPATIBLE MODEL FOR BI-DIRECTIONAL GROUND
MOTION
Some adjustment to the parameter statistics, informed by correlation spectra, is required to t
the ground motion model to response spectra. In the tting of the ground motion model to response
spectra of the original data sets, parameters N
c
, T
pk
, ,
1
,
2
, and
3
were xed to the mean values
shown in Table 2. The means and variances of V
p
, T
p
, V
s
, and f
g
, and deterministic values of
g
,
f
lo
, and f
hi
were adjusted (by trial and error) in order to match the mean and variance of the SAC
response spectra.
The average and standard deviation of bi-axial response spectra (with 5% of critical damping)
were calculated for each of the original SAC data sets using the paired horizontal ground motion
records. In this study the bi-axial spectral response, max |R(T
n
)|, is the peak magnitude of the
response components r
x
(t; T
n
) and r
y
(t; T
n
) (in the X and Y directions),
max |R(T
n
)| = max
t
_
r
2
x
(t, T
n
) +r
2
y
(t; T
n
)
_
1/2
, (12)
where the responses r
x
(t; T
n
) and r
y
(t; T
n
) are the responses in the X and Y directions of a bi-axial
oscillator of natural period T
n
(and 5% critical damping) to ground accelerations applied simulta-
neously in the X and Y directions. Similarly, bi-axial response spectra were calculated for 1000
pairs of synthetic ground motions drawing ground motion parameters from the sample statistics
6 H.P. Gavin and B.W. Dickinson
for each SAC data set. Within each bi-axial time history analysis total acceleration responses,
max |A(T
n
)|, and relative displacement responses, max |D(T
n
)|, were computed independently.
Means and standard deviations of the random ground motion parameters were adjusted to match
the response spectra for each ground motion data set, in terms of its mean and variance of accel-
eration and displacement response. In an effort to realistically model potentially large motions,
priority was given to matching the mean and the mean plus one standard deviation response
spectra. Figure 2 qualitatively describes the sensitivity of acceleration and displacement response
spectra to variation in ground motion parameter statistics. The pulse velocity was adjusted to match
the response at long periods and the peak stochastic velocity was adjusted to match the response at
short periods. Increasing f
g
increases acceleration response and decreases displacement response.
Increasing V
p
increases the responses of structures with natural periods around T
p
. Increasing T
p
increases the range of natural periods affected by the pulse, and increasing V
p
and V
s
increases
the standard deviation of the response spectrum.
Spectral responses of the SAC records at short and long periods indicate that the frequency
limits for the stochastic part of the ground motion, f
lo
and f
hi
, should be included as parameters in
order to match response spectra. To match the Los Angeles data sets, the frequency range should
be 0.1 Hz to 9 Hz and to match the Seattle data sets, the frequency range should be 0.1 Hz to 20 Hz.
The resulting adjusted parameter means and standard deviations for the bi-axial response spectra-
compatible model are shown in Table 2. As compared to parameter values t to waveforms, the
pulse velocity decreased and the peak stochastic velocity increased (with the exception of the nr-
fault data set). On the other hand, the variability in the pulse velocity increased (except for the
nrfault data set). For the nrfault suite, the mean pulse velocity, mean ground damping, and the
variance of the ground frequency t to waveforms are signicantly lower than those t to response
spectra. For the nrfault records, the variance in the pulse velocity t to waveforms is larger than
that t to response spectra. The adjustments to pulse period statistics were minimal and showed
no uniform trends. The mean and variance of the ground frequency increased for the Seattle data
sets. These results indicate that, except for near-fault scenarios, tting ground motion parameters
to wave forms tends to overestimate pulse amplitudes and underestimate the variability in pulse
amplitudes, as compared to parameters t to response spectra.
The bi-axial response spectra for the SAC data sets are compared to spectra from the param-
eterized model in Figures 3 through 7. In most cases, the mean minus one model spectra also
match the mean minus one spectra from the original data sets. Although not shown here, re-
sponse spectra for 2% and 20% damping show that the model is response-spectrum-compatible at
multiple levels of damping (Dickinson 2008).
The sequence of generating synthetic earthquake ground motions for two of the ve ground
motion suites is illustrated in Figures 8 and 9. Figure 8 illustrates the synthesis process for the
nrfault parameters and Figure 9 illustrates the process for the se10in50 parameters: the limiting
cases of the most and least signicant pulses. The power spectral density of the synthetic ground
motion is compared to the target power spectral density as dened by the ground frequency and
ground damping parameters. Note that the signicant variability among power spectra computed
from a set of synthetic records is due to their short and transient nature. In many cases, long period
and high amplitude peaks occur in the stochastic velocity record. Many of these peaks are similar
to peaks in the original data set and would have been identied as a velocity pulse in tting the
model to waveforms. Such occurrences directly support the conclusion that tting ground motion
parameters to waveforms can overestimate pulse amplitudes and underestimate the amplitude of
7 H.P. Gavin and B.W. Dickinson
TABLE 2. Model parameters for 2D response-spectrum-compatible records.
suite parameter
Vp Tp Nc T
pk
Vs
1

2

3
fg g f
lo
f
hi
cm/s s - s rad cm/s s s s Hz - Hz Hz
nrfault
mean 115 1.5 1.4 5.0 0 60 5.0 0.5 5.1 0.7 2.0 0.10 8
std.dev. 40 1.2 0 0 0 35 0 0 0 0.2 0 0 0
la10in50
mean 20 1.5 1.3 5.0 0 60 4.7 0.2 5.5 1.0 1.8 0.07 9
std.dev. 35 1.0 0 0 0 25 0 0 0 0.1 0 0 0
la2in50
mean 50 1.7 1.2 4.8 0 120 5.5 0.5 4.8 0.7 1.2 0.07 9
std.dev. 60 1.0 0 0 0 40 0 0 0 0.1 0 0 0
se10in50
mean 10 2.1 1.3 6.1 0 25 13.2 0.2 11.4 1.8 0.9 0.10 20
std.dev. 20 0.8 0 0 0 15 0 0 0 0.2 0 0 0
se2in50
mean 40 1.0 1.7 6.6 0 60 16.7 0.2 11.6 1.7 1.0 0.10 20
std.dev. 12 1.0 0 0 0 35 0 0 0 0.2 0 0 0
offset 0 0.8 0 0 0 10 0 0 0 0 0 0 0
the stochastic velocity content.
The adjusted ground motion parameter statistics provide a bi-axial ground motion model cal-
ibrated to the ground motions of the SAC data sets. This model contains the means, standard
deviations, correlations, and offsets for ground motion waveform parameters. It also species the
frequency range required for generation of stochastic ground motion content. The model captures
the mean and variability of response spectra through six random variables and is shown to be ac-
curate in terms of displacement and acceleration responses for a range of natural periods from 0.1
sec to 10 sec, and for damping ratios from 2 percent to 20 percent. This model is implemented in
Matlab code available at: http://www.duke.edu/

hpgavin/groundmotions/.
SUMMARY AND CONCLUSIONS
Synthetic ground motions are generated via the superposition of a long-period pulse with a
stochastic acceleration record generated from a power spectral density function and modulated
by an envelope function. In this study, parameter statistics have been calibrated to the mean and
variance of response spectra from a number of ground motion data sets, but could be t to other
spectra as well. Synthetic ground motion records generated using this model qualitatively match
the wave form characteristics of the original records as well as their response spectra.
Parameter-response correlation spectra, as introduced in this paper, illustrate correlations be-
tween the parameters of synthetic ground motion models to peak structural responses as a function
of natural period. Such spectra indicate the association between peak structural responses and
parameters of synthetic ground motion models. In this study, a ground motion model having thir-
teen parameters (pulse velocity, pulse period, number of pulse cycles, pulse phase, arrival time
of the pulse, peak stochastic velocity, central frequency of the stochastic content, bandwidth of
the stochastic content, envelope rise time, envelope constant time, envelope decay time, low fre-
quency limit, and high frequency limit) was t to spectra from ve sets of earthquake ground
motion records. The pulse velocity, the pulse period, the peak stochastic velocity, and the ground
frequency correlate signicantly (greater than ten percent) to peak structural responses. These pa-
rameters are random variables in the ground motion model while the remaining parameters take
constant values.
The resulting bi-axial ground-motion model includes correlations between horizontal compo-
nents of the pulse velocity and random velocity amplitude, and the direction-independent pulse
period. The bi-axial model was used to simulate 1000 synthetic accelerations for each data set and
the average response spectra for these ground motions were calculated and t to corresponding
8 H.P. Gavin and B.W. Dickinson
spectra of the ground motion data sets. The mean pulse amplitudes were less than those t to the
waveforms, except for the near-fault data set. The stochastic velocity amplitudes have a coefcient
of variation (c.o.v.) of approximately 50%. The c.o.v. of the pulse velocity amplitude for the Los
Angeles and the Seattle 10% in 50 years sets was approximately 100%. The c.o.v. for the ground
frequency for all data sets is approximately 10%.
The methods described in this paper should be extended to conrm the ground motion model
for inelastic response spectra before application to hysteretic or collapse-sensitive structures. It
is anticipated that duration parameters such as number of cycles of a pulse and envelope param-
eters will have a more signicant impact on the response of these structures and would need to
be included in the set of random variables of the model. The correlation spectra for acceleration
response and displacement response were nearly identical for the 5% damped, elastic case. It is
expected that this would not be the case for inelastic structures, and correlation spectra for accel-
eration and displacement response of inelastic structures would need to be evaluated separately.
ACKNOWLEDGMENTS
This material is based upon work supported by the the Civilian Research and Development
Foundation for the Independent States of the Former Soviet Union (CRDF) under Award No.
MG1-2319-CH-02 and by the National Science Foundation under Grant No. NSF-CMMI-0704959
(NEES Research), and Grant No. NSF-CMS-0402490 (NEES Operations). Any opinions, nd-
ings, and conclusions or recommendations expressed in this material are those of the authors and
do not necessarily reect the views of the National Science Foundatioon.
The authors express sincere thanks to the reviewers, whose perceptive comments were instru-
mental in revising this manuscript.
REFERENCES
Baker, J. (2007). Correlation of ground motion intensity parameters used for predicting struc-
tural and geotechnical response. 10th International Conference on Application of Statistic and
Probability in Civil Engineering (ICASP10), Tokyo, Japan, 17.
Baker, J. and Cornell, C. (2006). Correlation of response spectral values for multicomponent
ground motions. Bulletin of the Seismological Society of America, 96(1), 215227.
Baker, J. and Jayaram, N. (2008). Correlation of spectral acceleration values from nga ground
motion models. Earthquake Spectra (in press).
Boore, D. (2005). On pads and lters: Processing strong-motion data. Bulletin of the Seismolog-
ical Society of America, 95(2), 745750.
Chang, S., Makris, N., Whittaker, A., and Thompson, A. (2002). Experimental and analytical
studies on the performance of hybrid isolation systems. Earthquake Engineering and Structural
Dynamics, 31(2), 421443.
Cordova, P., Deierlein, G., Mehanny, S., and Cornell, C. (2001). Development of a two-parameter
seismic intensity measure and probabilistic assessment procedure. The Second U.S.-Japan
Workshop on Performance-based Earthquake Engineering Methodology for Reinforced Con-
crete Building Structures, Sapporo, Hokkaido, 187206.
Dickinson, B. (2008). A parametric statistical generalization and simulation of uniform hazard
earthquake ground motions. Masters Thesis, Department of Civil and Environmental Engi-
neering, Duke University.
9 H.P. Gavin and B.W. Dickinson
Dickinson, B. and Gavin, H. (2010). A parametric statistical generalization of uniform hazard
earthquake ground motions. Journal of Structural Engineering, In Press.
Lee, K. and Foutch, D. (2002). Seismic performance evaluation of pre-northridge steel frame
buildings with brittle connections. Journal of Structural Engineering, 128(4), 546555.
Mavroeidis, G., Dong, G., and Papageorgiou, A. (2004). Near-fault ground motions, and the
response of elastic and inelastic single-degree-of-freedom (sdof) systems. Earthquake Engi-
neering and Structural Dynamics, 33, 10231049.
Mavroeidis, G. and Papageorgiou, A. (2003). A mathematical representation of near-fault ground
motions. Bulletin of the Seismological Society of America, 93(3), 10991131.
Shinozuka, M. and Deodatis, G. (1991). Stochastic wave models for stationary and homogeneous
seismic ground motion. Structural Safety, 10(1-3), 235246.
Shinozuka, M. and Jan, C.-M. (1972). Digital simulation of random processes and its applica-
tions. J. Sound and Vibration, 25(1), 111128.
Somerville, P., Smith, N., Punyamurthula, S., and Sun, J. (1997). Development of ground motion
time histories for phase 2 of the fema/sac steel project (report no. sac/bd-97/04). FEMA/SAC
Steel Project.
Somerville, P., Smith, N., Punyamurthula, S., and Sun, J. (2009). Suites of earthquake ground
motions for analysis of steel moment frame structures. http://nisee.berkeley.edu/
data/strong_motion/sacsteel/. Last accessed 22 Sep. 2009.
Wen, Y. and Song, S. (2003). Structural reliability/redundancy under earthquakes. Journal of
Structural Engineering, 129(1), 5667.
Yun, S., Hamburger, R., Cornell, C., and Foutch, D. (2002). Seismic performance evaluation for
steel moment frames. Journal of Structural Engineering, 128(4), 534545.
10 H.P. Gavin and B.W. Dickinson
-1
-0.5
0
0.5
1
0.1 1 10
V
p

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(a)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10
T
p

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(b)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10
N
c

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(c)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10
T
p
k

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(d)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(e)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10
V
s

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(f)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10

1

-

|R
|m
a
x


c
o
r
r
e
la
t
io
n
natural period, sec
(g)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10

2

-

|R
|m
a
x


c
o
r
r
e
la
t
io
n
natural period, sec
(h)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10

3

-

|R
|m
a
x


c
o
r
r
e
la
t
io
n
natural period, sec
(i)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10
f
g

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(j)
nrfault
la10in50
la2in50
se10in50
se2in50
-1
-0.5
0
0.5
1
0.1 1 10

g

-

|
R
|
m
a
x


c
o
r
r
e
l
a
t
i
o
n
natural period, sec
(k)
nrfault
la10in50
la2in50
se10in50
se2in50
FIG. 1. Correlation spectra between ground motion parameters and peak structural
response statistics for ve of the SAC ground motion data sets. Ground motion
parameters in order of decreasing correlation are: V
s
, V
p
, T
p
, f
g
, and N
c
.
11 H.P. Gavin and B.W. Dickinson
peak response displacement
p
e
a
k

r
e
s
p
o
n
s
e

a
c
c
e
l
e
r
a
t
i
o
n
a
v
e
r
a
g
e

r
e
s
p
o
n
s
e

s
p
e
c
t
r
u
m
a
v
e
r
a
g
e

+

1

s
t
d
.
d
e
v
.


r
e
s
p
o
n
s
e

s
p
e
c
t
r
u
m
r
e
s
p
o
n
s
e

s
p
e
c
t
r
u
m
a
v
e
r
a
g
e

1

s
t
d
.
d
e
v
.
f
g
2
V
V
T
p
T
p
2
V
p
V
p
s
s
FIG. 2. Response spectra sensitivity to input parameters
12 H.P. Gavin and B.W. Dickinson
0.5
1
1.5
2
2.5
3
3.5
4
0.1 1 10
m
a
x
|
A
|
,

g
natural period, s
50
100
150
200
250
0.1 1 10
m
a
x
|
D
|
,

c
m
natural period, s
0
0.5
1
1.5
2
2.5
3
3.5
4
0 50 100 150 200 250
m
a
x
|
A
|
,

g
max|D|, cm
nrfault
0.20.5s 1.0s
2.0s
5.0s
FIG. 3. nrfault bi-axial linear elastic response spectra, 5% damping, mean,
mean+std.dev, mean-std.dev, =SAC records, =model.
0.5
1
1.5
2
2.5
3
0.1 1 10
m
a
x
|
A
|
,

g
natural period, s
20
40
60
80
100
120
0.1 1 10
m
a
x
|
D
|
,

c
m
natural period, s
0
0.5
1
1.5
2
2.5
3
0 20 40 60 80 100 120
m
a
x
|
A
|
,

g
max|D|, cm
la10in50
0.2 0.5s 1.0s
2.0s
5.0s
FIG. 4. la10in50 bi-axial linear elastic response spectra, 5% damping, mean,
mean+std.dev, mean-std.dev, =SAC records, =model.
13 H.P. Gavin and B.W. Dickinson
0.5
1
1.5
2
2.5
3
3.5
4
0.1 1 10
m
a
x
|
A
|
,

g
natural period, s
20
40
60
80
100
120
140
160
0.1 1 10
m
a
x
|
D
|
,

c
m
natural period, s
0
0.5
1
1.5
2
2.5
3
3.5
4
0 20 40 60 80 100 120 140 160
m
a
x
|
A
|
,

g
max|D|, cm
la2in50
0.2 0.5s 1.0s
2.0s
5.0s
FIG. 5. la2in50 bi-axial linear elastic response spectra, 5% damping, mean,
mean+std.dev, mean-std.dev, =SAC records, =model.
0.5
1
1.5
2
2.5
0.1 1 10
m
a
x
|
A
|
,

g
natural period, s
5
10
15
20
25
30
35
40
45
0.1 1 10
m
a
x
|
D
|
,

c
m
natural period, s
0
0.5
1
1.5
2
2.5
0 5 10 15 20 25 30 35 40 45
m
a
x
|
A
|
,

g
max|D|, cm
se10in50
0.2 0.5s
1.0s
2.0s
5.0s
FIG. 6. se10in50 bi-axial linear elastic response spectra, 5% damping, mean,
mean+std.dev, mean-std.dev, =SAC records, =model.
14 H.P. Gavin and B.W. Dickinson
1
2
3
4
5
6
0.1 1 10
m
a
x
|
A
|
,

g
natural period, s
10
20
30
40
50
60
70
80
90
100
0.1 1 10
m
a
x
|
D
|
,

c
m
natural period, s
0
1
2
3
4
5
6
0 20 40 60 80 100
m
a
x
|
A
|
,

g
max|D|, cm
se2in50
0.2 0.5s
1.0s
2.0s
5.0s
FIG. 7. se2in50 bi-axial linear elastic response spectra, 5% damping, mean,
mean+std.dev, mean-std.dev, =SAC records, =model.
15 H.P. Gavin and B.W. Dickinson
0 1 2 3 4 5 6 7 8
0
0.2
0.4
0.6
0.8
1
FREQUENCY, Hz
N
O
R
M
A
L
I
Z
E
D

P
O
W
E
R

S
P
E
C
T
R
A
A
0 5 10 15 20 25 30 35 40
10
8
6
4
2
0
2
4
6
8
10
TIME, sec
A
C
C
E
L
,

c
m
/
s
/
s
B
0 5 10 15 20 25 30 35 40
5
4
3
2
1
0
1
2
3
4
5
TIME, sec
A
C
C
E
L
,

c
m
/
s
/
s
C
0 5 10 15 20 25 30 35 40
150
100
50
0
50
100
150
TIME, sec
V
E
L
O
C
,

c
m
/
s
D
0 5 10 15 20 25 30 35 40
200
150
100
50
0
50
100
150
200
TIME, sec
V
E
L
O
C
,

c
m
/
s
E
0 5 10 15 20 25 30 35 40
200
150
100
50
0
50
100
150
200
TIME, sec
V
E
L
O
C
,

c
m
/
s
F
0 5 10 15 20 25 30 35 40
1000
800
600
400
200
0
200
400
600
800
1000
TIME, sec
A
C
C
E
L
,

c
m
/
s
/
s
G
0 5 10 15 20 25 30 35 40
80
60
40
20
0
20
40
60
80
TIME, sec
D
I
S
P
L
,

c
m
H
FIG. 8. nrfault simulation: (a) power spectral density, (b) stochastic acceleration
content, (c) time modulated stochastic acceleration content, (d) normalized veloc-
ity content, (e) velocity pulse, (f) superimposed pulse and stochastic velocity, (g)
synthetic acceleration, (h) synthetic displacement.
16 H.P. Gavin and B.W. Dickinson
0 2 4 6 8 10 12 14 16 18 20
0
0.2
0.4
0.6
0.8
1
FREQUENCY, Hz
N
O
R
M
A
L
I
Z
E
D

P
O
W
E
R

S
P
E
C
T
R
A
A
0 10 20 30 40 50 60
15
10
5
0
5
10
15
TIME, sec
A
C
C
E
L
,

c
m
/
s
/
s
B
0 10 20 30 40 50 60
10
8
6
4
2
0
2
4
6
8
10
TIME, sec
A
C
C
E
L
,

c
m
/
s
/
s
C
0 10 20 30 40 50 60
30
20
10
0
10
20
30
TIME, sec
V
E
L
O
C
,

c
m
/
s
D
0 10 20 30 40 50 60
6
4
2
0
2
4
6
TIME, sec
V
E
L
O
C
,

c
m
/
s
E
0 10 20 30 40 50 60
30
20
10
0
10
20
30
TIME, sec
V
E
L
O
C
,

c
m
/
s
F
0 10 20 30 40 50 60
500
400
300
200
100
0
100
200
300
400
500
TIME, sec
A
C
C
E
L
,

c
m
/
s
/
s
G
0 10 20 30 40 50 60
8
6
4
2
0
2
4
6
8
TIME, sec
D
I
S
P
L
,

c
m
H
FIG. 9. se10in50 simulation: (a) power spectral density, (b) stochastic acceler-
ation content, (c) time modulated stochastic acceleration content, (d) normalized
velocity content, (e) velocity pulse, (f) superimposed pulse and stochastic velocity,
(g) synthetic acceleration, (h) synthetic displacement.
17 H.P. Gavin and B.W. Dickinson