Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

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

net/publication/250074950

Effect of Spatial Correlation on Estimated Ground-Motion Prediction Equations

Article  in  Bulletin of the Seismological Society of America · April 2009


DOI: 10.1785/0120080172

CITATIONS READS
53 313

3 authors, including:

Katsuichiro Goda
The University of Western Ontario
249 PUBLICATIONS   4,789 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

PREPARE: Enhancing PREParedness for East African Countries through Seismic Resilience Engineering View project

INSPIRE: Integrated Disaster Cycle Approach for Catastrophic Earthquake Impact Reduction View project

All content following this page was uploaded by Katsuichiro Goda on 18 January 2019.

The user has requested enhancement of the downloaded file.


Bulletin of the Seismological Society of America, Vol. 99, No. 2A, pp. 928–934, April 2009, doi: 10.1785/0120080172

Short Note
Effect of Spatial Correlation on Estimated Ground-Motion
Prediction Equations
by H. P. Hong, Y. Zhang, and K. Goda

Abstract The consideration of spatially correlated seismic hazard could be of im-


portance for seismic risk assessment. The estimation of this correlation for the peak
ground acceleration and the pseudospectral acceleration has been reported in the lit-
erature, although it is not presented as a necessary ingredient for developing ground-
motion prediction equations (GMPEs). In the present study, we show that spatial
correlation can be incorporated in the existing regression algorithms given by Joyner
and Boore for assessing a GMPE. The modified algorithms can be used to estimate
both GMPE coefficients and spatial correlation model parameters simultaneously. In
particular, they are used to investigate the influence of spatial correlation on GMPEs
and to assess parameters of an empirical spatial correlation model by considering a set
of 592 California records. Analysis results indicate that the effects of incorporating
spatial correlation on the estimated GMPEs are insignificant, and that spatial correla-
tion parameters obtained using the modified algorithm are similar to those estimated
based on statistical analysis of regression residuals.

Introduction
Infrastructures in a seismic region are subjected to si- It is noted that the analyses carried out by Kawakami
multaneous excitations caused by large earthquakes. Such and Mogi (2003) and Boore et al. (2003) were based on the
correlated seismic effects can be important in assessing seis- ratios of the PGA at two recording stations, which require
mic risk for spatially distributed structures and can signifi- adequate corrections of the PGA for different soil conditions
cantly affect the probability distribution of seismic damage and for different source-to-site distances using an adopted
costs in the tail regions (Goda and Hong, 2008b). The mo- GMPE; those by Wang and Takada (2005) and Goda and
tions are often measured in terms of the peak ground accel- Hong (2008a) were based on statistical analysis of residuals
eration (PGA) and the pseudospectral acceleration (PSA). of a GMPE. The methods used by Wang and Takada (2005)
Ground-motion prediction equations (GMPEs) for these mea- and Goda and Hong (2008a) are computationally efficient,
sures are developed and used for seismic hazard assessment. but strictly speaking, they invoke an inconsistent argument
The GMPEs contain both interevent variability and in- because the residuals are calculated from GMPEs, which are
traevent variability (Abrahamson and Youngs, 1992; Joyner obtained by regression analysis considering that the residuals
and Boore, 1993), where the latter is spatially correlated. Ka- are uncorrelated. The methods used by Kawakami and Mogi
wakami and Sharma (1999) studied the spatial intraevent (2003) and Boore et al. (2003) may suffer a similar problem
variability of the PSA, whereas Kawakami and Mogi (2003), if the correction to the PGA (or any other ground-motion
Boore et al. (2003), and Wang and Takada (2005) investi- measures) is carried out by using a GMPE, which is obtained
gated the spatial intraevent correlation of the PGA. More by considering the same independent residual assumption.
recently Goda and Hong (2008a) estimated the overall cor- However, the magnitude of errors introduced by using the
relation of the PSA considering interevent variability and (uncorrelated) residuals to assess the spatial correlation is un-
spatial intraevent variability and provided an empirical equa- known, and the effects of considering spatial correlation on
tion to predict the correlation, which depends on the natural the developed GMPE have not been investigated.
vibration period. The mentioned studies indicate that the de- The main objectives of this study are to investigate the
gree of correlation of the ground-motion parameters de- influence of including spatial correlation in regression analy-
creases with increasing distance between two sites, although sis on GMPEs and to develop a spatial correlation model di-
the rate of decrease differs for different studies. rectly from regression analysis. The validity of this direct

928
Short Note 929

approach is assessed by comparing spatial correlation model tion coefficient when Δ equals zero (i.e., ρT 0; T n1 ; T n2 ),
parameters, which are directly obtained from regression which is given by Baker and Cornell (2006); T max 
analysis, with those derived from statistical analysis of re- maxT n1 ; T n2 ; and the intraevent spatial correlation coeffi-
gression residuals. To achieve these objectives, the algo- cient ρε Δ; T n  is given by
rithms given by Joyner and Boore (1993) for regression
analysis are modified to consider correlated residuals. More ρε Δ; T n   expαΔβ ; (3)
specifically, an empirical parametric spatial correlation
model is incorporated in the covariance matrix of residuals, in which α and β are the model parameters , which could be
which is associated with both interevent and intraevent varia- approximated by α  0:16 lnT n   0:62 and β  0:50
bility. Details of the modification are given in the following for a set of California records. In particular if only a single
section. The modified algorithms are illustrated by consider- vibration period is of interest, equation (2) is simplified to
ing a set of California records.
ρT Δ; T n ; T n   ρη T n   ρε Δ; T n 1  ρη T n ; (4)

Incorporation of a Parametric Spatial Correlation where ρη T n f ση T n =σT T n 2 g represents the inter-
Model in Regression Analysis event correlation coefficient that was considered by Wesson
and Perkins (2001).
A simple GMPE for ground-motion measures, such as To incorporate the spatial correlation in regression
the PGA and PSA, is expressed as (Abrahamson and Youngs, analysis we use Joyner and Boore’s (1993) one-stage and
1992; Joyner and Boore, 1993) two-stage algorithms and consider that the parametric corre-
lation model shown in equations (2–4) is adequate. To sim-
ln Y ij T n   fMj ; Rij ; λij ; T n   ηj T n   εij T n ; (1)
plify the notation, the functional dependence on T n is not
shown explicitly in the following; thus, the spatial correlation
where Y ij T n  is the ground-motion parameter at a vibration
coefficient ρε Δ; T n  shown in equation (3) is simply de-
period T n for the i-th site due to the j-th seismic event, and
noted by ρε Δ.
fMj ; Rij ; λij ; T n  is a function of the earthquake magnitude
The algorithms maximize the likelihood functions,
Mj , the site-to-source distance Rij , and a set of other ex-
which are established based on an adopted GMPE by turning
planatory variables λij . The interevent variability ηj T n  is
a nonlinear regression problem into a linear regression prob-
assumed to be normally distributed with zero mean and stan-
lem using the Taylor series expansions, where the solution is
dard deviation ση T n , and the intraevent variability εij T n 
obtained iteratively (Joyner and Boore, 1993). In particular,
is assumed to be normally distributed with zero mean and
if the one-stage algorithm is considered for a single vibration
standard deviation σε T n . Note that if the PGA is consid-
period, the problem is expressed as
ered, equation (1) is no longer a function of T n .
Given a set of records, by considering that ηj T n  is Y  XB  e; (5)
an independent identically distributed normal variate, and
εij T n  is an independent identically distributed normal vari- where Y is an N × 1 vector containing ln Y ij (and some other
ate, model parameters of the GMPE shown in equation (1) terms such as the soil amplification factor, which can be
can be determined based on the algorithms given by Joyner treated as constant in each iteration), in which N is the total
and Boore (1993). If the spatial correlation of intraevent number of data points; X is a matrix that depends on Mj and
variability is considered the algorithms can be modified to Rij ; B is a vector of unknown model parameters to be solved;
accommodate an adopted empirical parametric spatial corre- and e is an N × 1 vector containing ηj  εij , which
lation model, which is described in the following. is normally distributed with zero mean and covariance ma-
An empirical model for the spatial correlation of trix V. V is given by
ground-motion measures was proposed by Goda and
Hong (2008a): V  σ2T v; (6)
ρT Δ; T n1 ; T n2  and v is formed by zeros except for its block-diagonal sub-
 ρ0 T n1 ; T n2  matrix vk for the k-th earthquake and is given by
ση T n1 ση T n2   ρε Δ; T max σε T n1 σε T n2  2 3
× ; (2) 1 ρη  ρη
σT T n1 σT T n2  6 ρη 1  ρη 7
6 7
vk  6 . .. .. .. 7; k  1; …; N e ; (7)
where ρT Δ; T n1 ; T n2  represents the correlation coefficient 4 .. . . . 5
between ηj T n1   εij T n1  and ηj T n2   εkj T n2  for a ρη ρη  1
randomly selected seismic event and at two recording
stations separated by a distance Δkm; σT T n 2  where vk is an N rk × N rk matrix, and its diagonal elements
ση T n 2  σε T n 2 ; ρ0 T n1 ; T n2  represents the correla- are unity. Its off-diagonal elements are ρη ; N rk represents the
930 Short Note

number of records for the k-th earthquake, and N e is the However, if the spatial correlation expressed in equation (4)
number of earthquakes. The solution is the one that maxi- ^ 1 shown in equation (11) is replaced by
is incorporated, B
mizes the following log-likelihood function L:
^ 1  XT V1 X1 1 XT V1 Y1 ;
B (12)
1 1 1 1
N N 1
L   ln2π  ln σ2T  ln jvj
2 2 2
where
1 T 1
 2 Y  XB v Y  XB; (8)
2σT 2 3
v11 0  0
6 0 v12  0 7
where j • j represents the determinant of a matrix. The maxi- 6 7
v1  σ2ε v1  σ2ε 6 .. .. .. .. 7 and (13)
mization requires iteration or use of a search algorithm in 4 . . . . 5
dealing with nonlinear terms. Joyner and Boore (1993) pro- 0 0    v1N e
vided an efficient algorithm to find the model parameters
(i.e., model parameters for a GMPE, ση , and σε ).
If the incorporation of the empirical spatial correlation 2 3
model expressed in equation (4) for assessing a GMPE is con- 1 ρε Δ1;2     ρε Δ1;N rk 
sidered, vk becomes 6 ρε Δ2;1  1    ρε Δ2;N rk  7
6 7
v1k 6 .. .. .. .. 7;
2 3 4 . . . . 5 (14)
1 ρ1;2    ρ1;N rk
ρε ΔN rk ;1  ρε ΔN rk ;2   1
6 ρ2;1 1    ρ2;N rk 7
6 7
vk  6 . .. .. .. 7; k  1; …; N e ; k  1; …; N e
4 .. . . . 5
ρNrk ;1 ρN rk ;2  1
is an N rk × N rk matrix for the k-th earthquake. In such a
(9) case, estimates of the parameters controlling distance depen-
dence, σε , α, and β are obtained such that the following log-
where ρi;j  ρj;i  ρη  ρε Δi;j 1  ρη , Δi;j is the dis-
likelihood function L1 is maximized:
tance between recording stations for the i-th and j-th records
due to the k-th seismic event, and ρε Δi;j  is given in equa-
N N 1
tion (3). Therefore, the matrix v contains two additional L1   ln2π  ln σ2ε  ln jv1 j
model parameters α and β as compared to equation (7), 2 2 2
1
and in general, an analytical inverse matrix is not available.  2 Y1 X1 B1 v1
1 Y1 X1 B1 : (15)
Because all remaining equations and matrices are the same as 2σε
those given by Joyner and Boore (1993), their algorithm is
still applicable except that the inverse of v is evaluated nu- In the second stage, the magnitude dependence is character-
merically, and the maximization of L is carried out for two ized as
additional parameters α and β.
In the two-stage algorithm, model parameters Y2  X2 B2  e2 ; (16)
controlling distance dependence and magnitude dependence
are determined in the first and second stages, respectively. where Y2 equals the estimated P, P, ^ obtained in the first
More specifically, in the first stage the distance depen- stage, X2 is the matrix relating Y2 to the vector B2, which
dence is expressed as (see equation (18) in Joyner and contains the parameters controlling magnitude dependence,
Boore (1993)): and e2 is a residual vector that is composed of P–P^ and ηj . If
the spatial correlation is not considered, the covariance ma-
Y1  X1 B1  e1 ; (10) trix associated with e2 , V2 , which is needed for the remaining
part of the two-stage algorithm, is given by
where Y1 is an N × 1 vector containing ln Y ij (and some
other terms, which can be treated as constant in each itera- 2 3
v21 0  0
tion); B1 contains the model parameters controlling distance 6 0 v22  0 7
dependence as well as the auxiliary N e × 1 vector P, which is 6 7
V2  6 .. .. .. .. 7; (17)
linearly related to the parameters controlling magnitude de- 4 . . . . 5
pendence and ηj , X1 is the matrix relating Y1 to B1 , and e1 is 0 0  v2Ne
an N × 1 vector containing εij . If the spatial correlation
among εij is ignored, the estimate of B1 , B ^ 1 , is given by where

^ 1  XT X1 1 XT Y1 :
B (11) v2k  σ2η  σ2ε =N rk ; k  1; …; N e : (18)
1 1
Short Note 931

Table 1
Ground-Motion Prediction Equation Based on the Geometric Mean for a Set of 592 California Records

Ground-Motion Prediction Equation


ln Y  b1  b2 M  7  b3 M  72  b4  b5 M  4:5 lnr2jb  h2 0:5   AFs , where bi , (i  1; …; 5) are the model parameters listed in the next
section of Table 1; M is the moment magnitude; rjb km is the closest horizontal distance from the station to a point on the Earth’s surface that lies
directly above the rupture and is taken equal to the Joyner and Boore distance or the epicentral distance when the former is not available; hkm
represents a fictitious depth; and AFs represents the amplification factor due to linear and nonlinear soil behavior (see Hong and Goda [2007] and Boore
and Atkinson [2008]).
*
Regression Coefficients
Y (g) b1 b2 b3 b4 b5 h ση σε σT
PGA 1.096 0.444 0.0 1:047 0.038 5.7 0.190 0.433 0.473
1.143 0.398 0.0 1:125 0.064 5.6 0.150 0.438 0.463
PSA at 0.3 sec 1.473 0.571 0.0 0:842 0:007 4.9 0.188 0.486 0.521
1.554 0.458 0:032 0:946 0.026 5.0 0.157 0.489 0.514
PSA at 1.0 sec 0.481 0.644 0:229 0:843 0:002 4.9 0.326 0.559 0.647
0.806 0.834 0:150 0:914 0:009 5.8 0.251 0.570 0.623
PSA at 3.0 sec 1:203 1.255 0:235 0:511 0:100 2.5 0.408 0.557 0.691
0:969 1.631 0:123 0:428 0:159 3.7 0.285 0.569 0.636
*
The first entry is obtained by ignoring the spatial correlation (Hong and Goda, 2007), and the second entry is obtained by incorporating the spatial
correlation in regression analysis.

If the spatial correlation is considered, it can be shown that Impact of Spatial Correlation
v2k shown in equation (18) is replaced by on Regression Parameters
For the assessment of the influence of spatial correlation
X
N rk X
N rk  on a GMPE, we use a set of 592 California records (from
v2k  σ2η  σ2ε = wk;ij ; k  1; …; N e ; (19) 39 earthquakes), which was extracted from the Next Genera-
i1 j1
tion Attenuation (NGA) database of the Pacific Earthquake
Engineering Research (PEER) Center. The selected records
where wk;ij denotes the elements of the inverse matrix of v1k are for earthquakes with M ≥ 5:0 (with magnitude dependent
shown in equation (14). As expected, if ρε Δi;j  for i ≠ j is distance cutoff limits) and are for the National Earthquake
ignored in equation (14), v1k , as well
P asrk its
Pinverse, becomes Hazards Reduction Program NEHRP site class B, C, or D
the N rk × N rk identity matrix,  N i1
N rk
j1 wk;ij   N rk ,
with the geomatrix’s classification I, A, or B (see PEER Cen-
and equation (19) reduces to equation (18). This indicates ter for details). These records are band-pass-filtered with the
that the only change required to incorporate the spatial cor- high-pass corner frequency less than 0.5 Hz (for the PSA at
relation in the second stage is the use of equation (19) instead T n greater than 2.0 sec, the high-pass corner frequency less
of equation (18). than 0.33 Hz is considered for T n from 2.0 to 3.0 sec, result-

Figure 1. Ground-motion prediction equations obtained by considering and ignoring spatial correlation: (a) PGA and (b) PSA at 1.0 sec.
932 Short Note

Figure 2. Comparison of the spatial correlation model parameters for the geometric mean: (a) α and (b) β.

ing in 484 records from 34 earthquakes). The detailed criteria same functional form of the GMPE shown in Table 1 and the
as well as a complete list of the selected records are available same set of records. The obtained GMPE parameters based on
in Hong and Goda (2007). Because the model parameters the geometric mean are shown in Table 1 for the PGA and for
given by Hong and Goda (2007, see Table 1) were calculated the PSA at T n equal to 0.3 sec, 1.0 sec, and 3.0 sec. Compar-
based on the geometric mean of two orthogonal horizontal ison of the results shown in Table 1 suggests that there are
components by using the two-stage algorithm, the numerical differences in the estimated model parameters by exclud-
results presented in this section are also carried out by using ing or including the spatial correlation. By considering the
the modified two-stage algorithm. However, a verification spatial correlation, the variability of interevent residuals de-
analysis by using the one-stage algorithm for many cases creases, the variability of intraevent residuals increases, and
was carried out as well. the variability of overall residuals decreases. To visualize the
We carried out regression analysis with the spatial cor- impact of the spatial correlation on the estimated GMPEs, a
relation structure discussed in the previous section for the plot of the median GMPEs for the PGA and PSA at T n at

Table 2
Results of Regression Analysis for a Randomly Oriented Horizontal Component (Based on 50 Runs)
*
Regression Coefficients

Y (g) b1 b2 b3 b4 b5 h (km) ση σε σT
PGA 1.059 0.383 0:006 1:083 0.056 5.7 0.187 0.463 0.500
(0.074) (0.095) (0.014) (0.068) (0.028) (0.40) (0.014) (0.008) (0.008)
1.087 0.337 0:011 1:144 0.077 5.6 0.151 0.467 0.491
(0.072) (0.096) (0.018) (0.069) (0.027) (0.38) (0.015) (0.008) (0.008)
PSA at 0.3 sec 1.460 0.555 0:011 0:838 0:006 4.8 0.186 0.528 0.560
(0.083) (0.123) (0.019) (0.089) (0.037) (0.35) (0.019) (0.009) (0.009)
1.510 0.470 0:034 0:914 0.019 4.8 0.161 0.530 0.554
(0.093) (0.137) (0.031) (0.087) (0.036) (0.34) (0.019) (0.008) (0.009)
PSA at 1.0 sec 0.494 0.547 0:249 0:905 0.021 5.1 0.320 0.611 0.690
(0.121) (0.196) (0.049) (0.118) (0.049) (0.69) (0.017) (0.010) (0.012)
0.728 0.703 0:189 0:943 0.010 5.6 0.263 0.617 0.671
(0.140) (0.211) (0.049) (0.127) (0.053) (0.67) (0.019) (0.010) (0.011)
PSA at 3.0 sec 1:184 1.150 0:262 0:582 0:073 2.7 0.402 0.611 0.732
(0.112) (0.207) (0.059) (0.108) (0.050) (0.45) (0.026) (0.012) (0.015)
0:980 1.461 0:161 0:527 0:118 3.8 0.296 0.621 0.689
(0.157) (0.249) (0.063) (0.134) (0.061) (0.73) (0.038) (0.012) (0.015)
*
The first and second sets of entries are obtained by ignoring and considering the spatial correlation in regression
analysis, respectively. The first value in a set denotes the mean and the value inside the parentheses denotes the
standard deviation.
Short Note 933

Table 3
Spatial Correlation Parameters α and β Based on Regression Analysis (This Study) and Based on
Statistical Analysis of Regression Residuals (Goda and Hong, 2008a)
Ground-Motion Parameter Regression Analysis (This Study) Statistical Analysis of Residuals (Goda and Hong, 2008a)

α β α β
PGA 0.85 0.41 0.77 0.52
PSA at 0.1 sec 1.19 0.33 1.0 0.47
PSA at 0.2 sec 1.20 0.37 0.96 0.49
PSA at 0.3 sec 1.09 0.42 0.81 0.53
PSA at 0.5 sec 0.60 0.47 0.52 0.52
PSA at 1.0 sec 0.43 0.56 0.41 0.58
PSA at 1.5 sec 0.39 0.59 0.37 0.64
PSA at 2.0 sec 0.32 0.59 0.28 0.61
PSA at 3.0 sec 0.32 0.56 0.20 0.72

1.0 sec is illustrated in Figure 1. The results indicate that the stable. To overcome this problem, one could repeat such an
influence of considering the spatial correlation on the esti- analysis many times, each with a new randomly selected set
mated GMPEs is negligible. of record components. The obtained results for 50 runs by
It should be noted that as a by-product from regression ignoring and considering the spatial correlation model are
analysis, one obtains the parameters α and β, which are de- shown in Table 2 and Figure 4 for the PGA and for the
picted in Figure 2, listed in Table 3, and compared with those PSA at T n equal to 0.3 sec, 1.0 sec, and 3.0 sec. These results
given by Goda and Hong (2008a). Note that the values of α are obtained by considering all 592 records. Comparison of
and β given by Goda and Hong (2008a) were evaluated by the results shown in Tables 1 and 2 indicates that the GMPE
considering six (out of 39) earthquakes (i.e., 1979 Imperial parameters for the two cases are in good agreement. This
Valley, 1983 Coalinga, 1987 Whittier–Narrows, 1989 Loma observation is true whether the spatial correlation is ignored
Prieta, 1994 Northridge, and 1999 Hector Mine), each with or incorporated in regression analysis. Because the standard
more than 25 available records. The figure indicates that the deviations associated with the GMPE parameters shown in
former is similar to the latter. This suggests that the use of Table 2 are significant, it is important to consider sufficient
uncorrelated residuals for assessing the spatial correlation runs when randomly selected components are employed for
provides valid estimates of α and β. More importantly, it is regression analysis. The results shown in Figure 4 and com-
beneficial to incorporate the empirical parametric spatial cor- pared with those obtained based on the geometric mean in-
relation model in regression analysis because the analysis dicate that the spatial correlation for a randomly oriented
results provide not only the parameters for the GMPE but horizontal component is smaller than that for the geometric
also the parameters for the spatial correlation model, which
is valuable for risk analysis of a portfolio of spatially distrib-
uted structures (Goda and Hong, 2008b).
To investigate the influence of the considered records
on the estimated spatial correlation, regression analysis was
repeated by only considering the records for the six earth-
quakes previously mentioned rather than all selected earth-
quakes. The obtained spatial correlation parameters are
depicted in Figure 3 and compared with those obtained by
considering all records. The comparison indicates that use
of one or the other set of records does affect the estimated
(intraevent) spatial correlation and that differences in the es-
timated spatial correlation are less than 0.1.
It must be emphasized that the results shown in Table 1
and Figures 2 and 3 are based on the geometric mean of two
orthogonal horizontal components rather than a randomly
oriented horizontal component. To investigate the spatial cor-
relation for the latter, we randomly selected a component
from each record and used these selected components to de-
Figure 3. Comparison of the spatial correlation based on all
velop the GMPEs by ignoring and considering the spatial cor- events with that based on major six events (for comparison purpose,
relation. However, the analysis results based on a single run the samples given by Goda and Hong [2008a] are also included in
of the randomly selected components may not be sufficiently the figure).
934 Short Note

base (http://peer.berkeley.edu/nga/index.html) sponsored by


Pacific Earthquake Engineering Research (PEER) Center.

Acknowledgments
The financial support of the Natural Science and Engineering Re-
search Council of Canada and the University of Western Ontario are grate-
fully acknowledged.

References
Abrahamson, N. A., and R. R. Youngs (1992). A stable algorithm for re-
gression analysis using the random effects model, Bull. Seismol.
Soc. Am. 82, 505–510.
Baker, J. W., and C. A. Cornell (2006). Correlation of response spec-
tral values for multicomponent ground motions, Bull. Seismol. Soc.
Am. 96, 215–227.
Boore, D. M., and G. M. Atkinson (2008). Ground-motion prediction
equations for the average horizontal component of PGA, PGV, and
Figure 4. Comparison of the spatial correlation for a randomly 5%-damped PSA at spectral periods between 0.01 s and 10.0 s, Earthq.
oriented horizontal component (based on the mean of 50 runs) Spectra 24, 99–138.
with that based on the geometric mean of two orthogonal horizontal Boore, D. M., J. F. Gibbs, W. B. Joyner, J. C. Tinsley, and D. J. Ponti (2003).
components. Estimated ground motion from the 1994 Northridge, California, earth-
quake at the site of the Interstate 10 and La Cienega boulevard bridge
collapse, west Los Angeles, California, Bull. Seismol. Soc. Am. 93,
mean, which is consistent with the results reported by Goda 2737–2751.
Goda, K., and H. P. Hong (2008a). Spatial correlation of peak ground mo-
and Hong (2008a).
tions and response spectra, Bull. Seismol. Soc. Am. 98, 354–365.
Goda, K., and H. P. Hong (2008b). Estimation of seismic loss for spatially
Conclusions distributed buildings, Earthq. Spectra 24, 889–910.
Hong, H. P., and K. Goda (2007). Orientation-dependent ground-motion
It is shown that the intraevent spatial correlation of measure for seismic hazard assessment, Bull. Seismol. Soc. Am. 97,
ground-motion parameters can be incorporated in the popu- 1525–1538.
lar algorithms given by Joyner and Boore (1993) for devel- Joyner, W. B., and D. M. Boore (1993). Methods for regression analysis of
strong-motion data, Bull. Seismol. Soc. Am. 83, 469–487.
oping ground-motion prediction equations (GMPE). This is
Kawakami, H., and H. Mogi (2003). Analyzing spatial intraevent variability
advantageous because the use of the modified algorithms of peak ground accelerations as a function of separation distance, Bull.
provides not only GMPE parameters but also parameters for Seismol. Soc. Am. 93, 1079–1090.
the spatial correlation model, which is valuable for seismic Kawakami, H., and S. Sharma (1999). Statistical study of spatial variation of
risk analysis of a portfolio of spatially distributed structures. response spectrum using free field records of dense strong ground mo-
tion arrays, Earthq. Eng. Struct. Dyn. 28, 1273–1294.
The modified algorithms are applied to a set of 592 Cal-
Wang, M., and T. Takada (2005). Macrospatial correlation model of seismic
ifornia records. The results indicate that the differences in ground motions, Earthq. Spectra 21, 1137–1156.
the estimated GMPE parameters by excluding or including Wesson, R. L., and D. M. Perkins (2001). Spatial correlation of probabilis-
the spatial correlation are insignificant and that the obtained tic earthquake ground motion and loss, Bull. Seismol. Soc. Am. 91,
parameters for an empirical spatial correlation model are 1498–1515.
similar to those based on statistical analysis of regression
residuals, which were reported previously for the same set
Department of Civil and Environmental Engineering
of records.
University of Western Ontario
Canada N6A 5B9
Data and Resources hongh@eng.uwo.ca

We acknowledge that strong ground motion data are


obtained from the Next Generation Attenuation (NGA) data- Manuscript received 4 July 2008

View publication stats

You might also like