Académique Documents
Professionnel Documents
Culture Documents
1785/0120150194
Introduction
In earthquake-prone regions, design of new or evalu- number of smaller magnitude (5.05.7) earthquakes from
ation of existing structures relies on prediction of ground Turkey, California, and other shallow crustal continental
shaking. The level of ground shaking is frequently defined regions.
using ground-motion prediction equations (GMPEs) that are a The new GraizerKalkan GMPE (GK15) is composed of
function of earthquake magnitude, style of faulting, site-to- two predictive equations. The first equation predicts PGA
source distance, and site parameters. GMPEs are often, in (Graizer and Kalkan, 2007), and the second equation con-
empirical form, guided by earthquake physics and/or simu- structs the spectral shape (Graizer and Kalkan, 2009). The
lations, and their estimator coefficients are typically com- term spectral shape refers to the PSA response spectrum nor-
puted by single or multistage regression on ground-motion malized by PGA. The PSA response spectrum is obtained by
amplitudes recorded from previous earthquakes. anchoring the spectral shape to the PGA. In this model, the
An initial GMPE for peak ground acceleration (PGA) and PSA response spectrum is a continuous function of the spec-
5% damped pseudospectral acceleration (PSA) response or- tral period (T). Typically GMPEs (e.g., NGA-West 2 models
dinates of the horizontal component of randomly oriented of Abrahamson et al., 2014; Boore et al., 2014; Campbell
ground motions was developed by Graizer and Kalkan and Bozorgnia, 2014; Chio and Youngs, 2014) use a discrete
(2007, 2009) using the Next Generation Attenuation functional form for predicting the PSA response ordinates.
(NGA)-West1 database (see Data and Resources; Chiou et al., The concept of a continuous function assumes cross corre-
2008) along with additional records from major California lation of spectral ordinates at different periods (Baker, 2011),
earthquakes, including the 2004 Parkfield (M 6.0, M, moment and de facto eliminates the difference between period inter-
magnitude) and 2003 San Simeon (M 6.5) earthquakes, and a vals by making period intervals infinitesimally short. As a
687
688 V. Graizer and E. Kalkan
Table 2
Earthquakes Used for Updating the GraizerKalkan (GK15) GMPE
Epicenter Coordinates
Style of Moment Depth Number Distance
No Event Date Faulting Magnitude (km) Latitude () Longitude () of Data Range (km)
These events were also used in the development of the GK07-09 model (Graizer and Kalkan, 2007, 2009). Data sources are provided in Graizer
and Kalkan (2015).
Among 2583 ground-motion recordings, 1450 are from earthquake faults from events in the 4.97.9 magnitude
reverse-fault events, 1120 are from strike-slip fault events, range. The data used in the analysis represent main-
and 13 are from normal-fault events. The distributions of data shocks only.
with respect to M and V S30 against the closest distance to Approximately half of the stations in our dataset have
fault rupture plane (R) are shown in Figure 1. The current measured V S30 values, and the rest are inferred using surface
dataset includes data recorded within 0.2250 km of the geology (e.g., Wills et al., 2000). The V S30 values range
690 V. Graizer and E. Kalkan
7
1000
(m/s)
M
600
200
0.1 1 10 50 150 0.1 1 10 50 150
R (km) R (km)
Figure 1. Earthquake data distribution with respect to (a) moment magnitude (M), and (b) shear-wave velocity in the upper 30 m of the
geological profile (V S30 ) against closest distance to fault rupture plane (R). Diamond, circle, and plus markers show National Earthquake
Hazard Reduction Program (NEHRP) site categories SB , SC , and SD , respectively. The color version of this figure is available only in the
electronic edition.
Updated GMPE
8 2
0.1
PGA (g)
M
0.01
0.001
0.001 0.01 0.1 1 2 0.1 1 10 100 200
PGA (g) R (km)
Figure 3. Horizontal peak ground acceleration (PGA) distribution with respect to moment magnitude (M), and closest distance to fault
rupture plane (R).
amplification owing to shallow site conditions, and G5 is and c4:9 are estimator coefficients. In equation (2), lnG3 ,
a basin scaling function. Equation (1) can be expressed in lnG4 , and lnG5 are
natural logarithmic space as
lnG3 c10 R=Q0 ; 5
lnPGA lnG1 lnG2 lnG3 lnG4
EQ-TARGET;temp:intralink-;df5;313;277
EQ-TARGET;temp:intralink-;df2;55;266
q
EQ-TARGET;temp:intralink-;df7b;55;94 ABdepth c11 1= f1 c12 =Bdepth 0:12 g2 4 c213 1:5=Bdepth 0:12 ; 7b
692 V. Graizer and E. Kalkan
PSAnorm T I exp 0:5
anchored to PGA S
2
controlled by and Tsp,0 T T 0:5
1 4 Dsp 2 ;
T sp;0 T sp;0
Period 9a
Figure 4. Generic spectral shape (PSAnorm ) model, and its in which T is the spectral period, and Dsp is the estimator
controlling parameters: I defines the peak spectral intensity, coefficient. As illustrated in Figure 4, I defines the peak
and T sp;0 define the predominant period of the spectrum, S defines
spectral intensity, and T sp;0 define the predominant period
the widenessarea under the spectral shapeand controls the
decay of the spectrum at long periods depending upon basin depth. of the spectrum, S defines the spectral wideness (area under
The color version of this figure is available only in the electronic the spectral shape), and controls the decay of the spectrum
edition. at long periods depending upon basin depth (Bdepth ). These
functions are
q
1= f1 c14 =R 0:12 g2 4 c213 40=R 0:12 ; EQ-TARGET;temp:intralink-;df9c;313;453 I a1 M a2 expa3 R; 9c
7c
S s1 R s2 M s3 ; 9d
in which c10:14 are the estimator coefficients. The values of
EQ-TARGET;temp:intralink-;df9d;313;418
0:3
T sp;0 max ; 9e
jt1 R t2 M t3 V S30 t4 j
GMPE for Spectral Acceleration
Similar to our previous model, the 5% damped PSA re-
sponse ordinates are obtained by anchoring the spectral EQ-TARGET;temp:intralink-;df9f;313;328 1:763 0:25 arctan 1:4 Bdepth 1; 9f
shape to PGA. The updated spectral shape model in GK15
has 15 coefficients and six independent predictor parameters. in which m1:4 , a1:3 ; t1:4 , and s1:3 are the estimator
The independent parameters are M, R, V S30 , F, Q0 , and coefficients, and the values of which are presented in
Bdepth . Updates on the spectral shape model include the Table 3.
following: (1) a modified decay term for long periods as a The estimator coefficients in Table 3 were obtained by
function of basin depth, (2) a revised term for controlling the performing regression in a number of steps, starting with a
predominant period of the response spectrum, and (3) up- more limited dataset for constraining parameters of filter
dated coefficients. functions and proceeding to the full range similar to
Table 3
Estimator Coefficients of GK15
GMPE for PGA
c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12 c13 c14 bv VA
0.14 6.25 0.37 2.237 7.542 0.125 1.19 6.15 0.6 0.345 1.077 1.5 0.7 40 0.24 484.5
Spectral Shape (PSA_norm) Model
m1 m2 m3 m4 a1 a2 a3 Dsp t1 t2 t3 t4 s1 s2 s3
0.0012 0.38 0.0006 3.9 0.01686 1.2695 0.0001 0.75 0.001 0.59 0.0005 2.3 0.001 0.077 0.3251
Summary of the GK15 GMPE for Horizontal PGA and 5% Damped PSA 693
Abrahamson et al. (2014). However, the difference between bump phenomenon (also called oversaturation) was recently
our approach and Abrahamson et al. is that we did not apply demonstrated through modeling geometrical spreading and
smoothing on estimator coefficients because our spectral- relative amplitudes of ground motions in eastern North
shape prediction model is a continuous function of indepen- America. The bump was attributed to radiation pattern effects
dent parameters. combined with wave propagation through a 1D layered earth
model (Chapman and Godbee, 2012; Baumann and Dalguer,
Functions of the GK15 GMPE 2014). In the case of earthquakes, this bump can be a result of
one or many factors, including the aforementioned radiation
The relationships between physical aspects of each filter pattern, directivity, and nonlinear behavior of the soil in the
function in equation (2) are described below. near source of an earthquake fault (e.g., low-velocity fault-
zone-guided waves, Li and Vidale, 1996), and measuring dis-
G1 (Magnitude and Style of Faulting) tance as that closest to the rupture plane and not from the
The following scaling function models the ground- seismogenic (most energetic) part of the fault rupture. In
motion scaling owing to the magnitude and style of faulting: equation (11), D0 shows the damping term that designates
the amplitude of the bump; D0 0:7 results in no bump.
EQ-TARGET;temp:intralink-;df10;55;563 G1 c1 arctanM c2 c3 F; 10
1979 M6.5 Imperial Valley Eq. 2004 M6.0 Parkfield Eq. 2014 M6.0 South Napa Eq.
1
PGA (g)
0.1
GK15
GM data
0.01
3
Residual
3
0.1 1 10 100 0.1 1 10 100 0.1 1 10 100
R (km)
Figure 5. Ground-motion attenuation during the 1979 M 6.5 Imperial Valley, 2004 M 6.0 Parkfield and 2014 M 6.0 South Napa earth-
quakes. Plots show amplified PGA as a bump at near field (R < 10 km); this phenomenon is captured well by the GraizerKalkan (GK15)
ground-motion prediction equation (GMPE). Ground-motion data shown are V S30 adjusted to 760 m=s; solid curve is for median; dashed
curves are for 16th and 84th percentile of predictions; the largest PGA (1g) recorded at Carquinez bridge (R 20 km) during the South
Napa earthquake is significantly affected by local site amplification (R, closest fault distance to rupture plane; V S30 , shear-wave velocity in
the upper 30 m of the geological profile). Residual plots show logarithmic difference between the predictions and observations. The color
version of this figure is available only in the electronic edition.
Although Q may be distance dependent, Q0 in equa- (shown in Fig. 6a). The fact that Q0 values determined using
tion (12) is a constant. In Figure 6b, we made a simple Lg or coda waves can be changed to suit the region of interest
assumption that Q increases with distance from a relatively is an improvement over other GMPEs. Predictions of four
low value of 10 in the vicinity of the fault to higher values NGA-West2 models and the GK15 model against the
of typical Lg-type crustal Q at far-source distances strong-motion data of the 2014 M 6.0 South Napa earthquake
(R > 100 km). Figure 6c compares the effects of constant demonstrate that the GK15 model with lower Q0 50 (mo-
Q0 150 with that of the distance-dependent Q. As ex- tivated partially by the range suggested by Ford et al., 2008)
pected, low Q0 in the near-source region produces slightly fits data better for this specific earthquake (Baltay and Boat-
lower ground-motion intensity. However, this decrease does wright, 2015).
not exceed 3% at distances up to 50 km; higher Q0 at far
distances results in slower attenuation relative to the constant G4 (Site Correction)
Q0 . The effect of distance-dependent Q relative to the con-
stant (distance-independent) Q0 is not significant. Consider- Based on published studies (a list of references is given
ing other uncertainties, we concluded that it is reasonable to in Graizer and Kalkan, 2007), a linear site-correction filter
use a constant Q0 typical for a given region (usually that for was adopted in GK07 because of the large variability in non-
Lg or coda waves). linear site-correction models:
In our updated GMPE for PGA, we assume a frequency-
G4 expbv lnV S30 =V A : 13
independent Q0 . In equation (12), c10 0:345, based on
EQ-TARGET;temp:intralink-;df13;313;125
the average value of Q0 150 for California produces similar Equation (13) is an equivalent form of the linear site-
effects to our previous GMPE for distances of up to 200 km correction formula of Boore et al. (1997), in which
Summary of the GK15 GMPE for Horizontal PGA and 5% Damped PSA 695
(a) 1
0.1
GK15
PGA (g)
(Q = 300)
GK07
0.01 Chi-Chi Eq.
GK07 GK15
GK15 (Q 0 = 75) (Q = 150)
GK15
GK15 (Q 0 = 150) (Q = 75)
GK15 (Q 0 = 300)
0.001
0.1 1 10 100 1000
(b) (c)
1000 1
0.1
PGA (g)
150
100
Q0
0.01
Q =150
10<Q<240
10 0.001
0.1 1 10 100 1000 0.1 1 10 100 1000
R (km) R (km)
Figure 6. Model results for anelastic attenuation with constant and variable Q0 . (a) Comparison of GraizerKalkan (GK07) PGA predictions
with GK15 considering the 1999 M 7.6 Chi-Chi earthquake data. (b) Constant and distance-dependent Q0 . (c) Comparison of attenuation curves
with constant and distance-dependent Q0 for an M 7.0 event (V S30 400 m=s) (R = closest fault distance to rupture plane). The color version of
this figure is available only in the electronic edition.
bv 0:371, whereas our estimates yield bv 0:24. 2001). Our new basin scaling function considers combined
Equation (13), with its parameters given in Table 3, is similar effects of amplification in both shear and surface waves
to the equation of Field (2000) in exhibiting less amplifica- owing to basin depth under the site according to Hruby and
tion as V S30 decreases than that of Boore et al. (1997). In our Beresnev (2003) and Day et al. (2008). For simplicity, other
updated GMPE for PGA, there is no change in G4 from its parameters, such as the basin shape and distance to the basin
original version in Graizer and Kalkan (2007). edge (Joyner, 2000; Semblat et al., 2002; Choi et al., 2005),
are not accounted for, and only the basin depth is used as a
predictor parameter in our GMPE.
G5 (Basin Effect) The mechanisms and results of shear and surface-wave
A basin consists of alluvial deposits and sedimentary amplifications in the basin are different. The basin amplifi-
rocks that are geologically younger and have a significantly cation of S waves affects mostly frequencies less than
lower shear-wave velocity structure than the underlying 10 Hz (Hruby and Beresnev, 2003), and basin amplifica-
rocks, which creates a strong interface, and can trap and am- tion of surface waves affects a range of spectral frequencies
plify earthquake-induced body and surface waves (Hanks, from PGA to long spectral periods. During the 1992 M 7.3
1975; Lee et al., 1995; Campbell, 1997; Frankel et al., Landers, 1999 M 7.1 Hector Mine, and 2010 M 7.2 El
696 V. Graizer and E. Kalkan
(a) (b)
(km) (km)
Figure 7. (a) Dependence of amplitude on basin depth (Bdepth ), and (b) dependence of the response spectrum long period decay term ()
on basin depth.
Los Angeles and San Bernardino basins were much higher 1999 M7.1 Hector Mine Eq. (R =~ 80 km)
than those measured at rock sites owing to amplified surface
1
waves (Graizer et al., 2002; Graves and Aagaard, 2011;
Hatayama and Kalkan, 2012).
In our previous GMPE (Graizer and Kalkan, 2009), the
PSA (g)
0.001
0.01 0.1 1 10
ABdepth defines the amplitude of the basin effect depending Period (s)
upon Bdepth . The parameters of ABdist and ABdepth were con-
strained by regression according to the 1999 M 7.1 Hector Figure 8. Comparison of pseudospectral acceleration (PSA) re-
Mine, 1992 M 7.3 Landers, and 1989 M 6.9 Loma Prieta sponse computed using updated GraizerKalkan GMPE (GK15)
for three cases: nonbasin, basin with 1.5 km depth, and basin with
earthquakes using only ground-motion data with known 3 km depth. Background data are from the 1999 M 7.1 Hector Mine
Bdepth values. earthquake at about 80 km closest fault distance to the fault-rupture
As shown in Figure 7a, ABdepth varies from 0 for a non- plane (R). Depth to basin (Bdepth ) varies from 1 to 3 km for this data-
basin site to 1.077 for a site in a deep basin, and it saturates set. The predictions are shown with the same V S30 430 m=saver-
for basins deeper than 3 km. When Bdepth 0, ABdepth be- age of the stations. The deeper basin produces a response spectrum
with higher amplitudes at all periods with slower decay at long peri-
comes negligibly small, and thus G5 1:0 does not have ods; it affects the long periods more than the short periods. The color
any effect on ground-motion prediction. Our approach on version of this figure is available only in the electronic edition.
modeling the basin effect is based on the 3D simulations of
Day et al. (2008) who found that depth to the 1:5 km=s
S-wave velocity isosurface is an appropriate parameter for
use in GMPEs. Similar basin amplification was observed for 1.42. As shown in Figure 7b, the spectral shape decays
the 1994 M 6.7 Northridge and 1987 M 6.0 Whittier Narrows at long periods faster (T 2 ) for nonbasin sites and slower
earthquakes by Hruby and Beresnev (2003), and for the 2010 (T 1:4 ) for deep basin sites. Figure 8 compares the PSAs
M 7.2 El-MayorCucapah earthquake by Hatayama and for two different basin depths, Bdepth 1:5 and 3 km against
Kalkan (2012). Our dependence of period amplification on the case without basin (Bdepth 0). The predictions in
Z1:5 approximates the period dependence in table 3 of Hruby Figure 8 are shown with the same V S30 430 m=saverage
and Beresnev (2003). of the stations. The deeper basin produces a response spec-
The parameter controlling the decay rate of the spectrum trum with higher amplitudes at all periods with slower decay
at long periods ( in equation 9f) varies over the range of at long periods; it affects the long periods more than the short
Summary of the GK15 GMPE for Horizontal PGA and 5% Damped PSA 697
0.4
PGA 0.1 1 5
Period (s)
0.4
0.1
Slope b
0.1
PGA 0.1 1 5 PGA 0.1 1 5
Period (s) Period (s)
Figure 9. (a) Overall mean bias of GK15 and its standard deviation (maximum-likelihood mean C indicates the overall bias between
the observations and predictions). (b) Distance, and (c) magnitude dependence of GK15 residuals. Intercept a and slope b of maximum-
likelihood line fit through residuals as a function of closest fault distance to rupture plane (R).
periods. Based on a 3D modeling of ground motion, a used the maximum-likelihood method to recursively deter-
possible explanation for the distance-dependent pattern is mine the mean of data points having the error structure of
suggested by Olsen (2000). According to Olsen, the ampli- Joyner and Boore (1993), where the residuals correspond to
fication factors are greater for events located farther from the
basin edge. He suggested that the larger-amplitude surface EQ-TARGET;temp:intralink-;df16;313;320 Resij C i ij ; 16
waves generated for the distant events, in part at basin edges,
are more prone to the amplification than are the predominant
in which C is a constant term (maximum-likelihood mean)
body waves impinging onto the basin sediments from nearby
from the mixed-effects analysis, which is a measure of the
earthquakes.
overall bias between the observations and the GMPE. The
constant term (C) should be close to zero for unbiased
Mixed-Effect Residuals Analysis
estimates. In equation (16), i represents the event term
We performed a mixed-effect residuals analysis to con- (between-event residual) for event i, and ij represents the
firm that GK15 is not biased with respect to M, R, V S30 , style intraevent (within-event) residual for recording j in event i.
of faulting (F), and Bdepth by examining trends of residuals Event term i represents the approximate mean offset of the
against these independent predictor parameters. The resid- data for event i from the predictions provided by the median
uals at each spectral period are computed as follows: of the GMPE. Event terms (i ) are used to evaluate the
GMPEs performance relative to source predictor variables.
EQ-TARGET;temp:intralink-;df15;55;156 Resij ln Y ij ij M; R; V S30 ; Bdepth ; 15 Both event and intraevent terms are assumed to be random
Gaussian variables with zero mean. Their standard deviations
in which i is the event and j is the recording index, Resij is are indicated by and , respectively.
the residual of the jth recording of the ith event, and Y ij is the For each spectral period, equation (16) is solved using
IM (PGA or 5% damped PSA ordinates) from the jth record- the maximum-likelihood formalism given in the appendix of
ing of the ith event; ij represents the GK15 median estimate Spudich et al. (1999). In Figure 9a, maximum-likelihood
in natural logarithmic units. To check for overall bias, we mean values are plotted for each spectral period ranging from
698 V. Graizer and E. Kalkan
3
dashed line is maximum likelihood fit (y=a+bx): a = 0.059 and b = 0.00108
PGA
3
3
a = 0.050 and b = 0.00092
Intraevent Residual
PSA at 0.2 s
3
3
a = 0.148 and b = 0.00271
PSA at 1.0 s
3
0 25 50 75 100 125 150
R (km)
Figure 10. Distribution of intraevent residuals in natural logarithmic units for PGA and PSA at 0.2 and 1.0 s with respect to closest
fault distance to rupture plane (R). Solid horizontal lines indicate size of the bins and their mean value; vertical solid lines show their
standard deviation; dashed line denotes a maximum-likelihood fit to residuals; different symbols indicate different events. Note that a
and b are intercept and slope of maximum-likelihood fit, respectively. The color version of this figure is available only in the electronic
edition.
PGA to 5.0 s. As shown, the overall bias of GK15 is small. independent parameters. Figure 9b,c shows distance and
Some discrepancies are plausible because a continuous, magnitude bias in the residuals; again both intercept and
smooth function of spectral period was forced to fit to all slope are near zero for all periods, indicating negligible dis-
PSA data instead of a discrete data fitting at certain periods tance dependence and no systematic magnitude bias in the
as other GMPE developers have done. residuals of GK15. We also demonstrate the same observa-
To separate interevent (between-event) disparities from tions against the predictor parameters V S30 and Bdepth , shown
intraevent (within-event) variations, we performed a mixed- in Graizer and Kalkan (2015).
effects analysis with respect to independent parameters, M,
R, V S30 , and Bdepth , and we fit an intercept a and slope b to Intraevent (Within-Event) Residuals Analysis of Path,
residuals according to the following formulation: Site, and Basin Depth Effects
EQ-TARGET;temp:intralink-;df17;55;128 Resij a bxi i ij ; 17 The intraevent residuals (ij ) are used to test the GK15
with respect to distance and site effects. In Figure 10, the
in which xi is the independent predictor parameter. Both residuals are shown in natural logarithmic units for PGA
slope and intercept computed using equation (17) are plotted and spectral periods at 0.2 and 1.0 s, similar to Chiou and
against period to check for systematic bias with respect to the Youngs (2013); similar results for 3 s are provided in Graizer
Summary of the GK15 GMPE for Horizontal PGA and 5% Damped PSA 699
3
dashed line is maximum likelihood fit (y=a+bx): a = 0.440 and b = 0.070
number of events = 35
PGA
3
3
a = 0.529 and b = 0.084
Event Term (i)
PSA at 0.2 s
3
3
a = 0.612 and b = 0.098
PSA at 1.0 s
3
4.8 5 5.5 6 6.5 7 7.5 8
M
Figure 11. Distribution of event terms i in natural logarithmic units for PGA and PSA at 0.2 and 1.0 s with respect to moment magnitude
(M); in each plot dashed line indicates a maximum-likelihood fit to all event terms; its slope and intercept are provided on top of each plot.
Note that a and b are intercept and slope of maximum-likelihood fit, respectively. The color version of this figure is available only in the
electronic edition.
and Kalkan (2015). We plot the intraevent residuals against R 1400 m=s at 0.2 and 3.0 sthis is attributed to the scarcity
(0150 km) using the full dataset with means and standard of the data within this range.
errors shown within bins. The bin sizes were adjusted so that
each bin has approximately the same number of data points. Analysis of Source Effects Using Interevent
The maximum-likelihood line is dashed, and its slope and (Between-Event) Residuals
intercept are provided on top of each plot. Although data Figures 11 and 12 show event terms (i ) plotted against
is slightly underpredicted at 1.0 s for distances greater than magnitude in the range 4:9 M 7:9 and style of faulting
110 km, the results generally show no perceptible trend parameter (F) considering PGA and PSA at 0.2 and 1.0 s
within the body of a predictor, indicating that the path-scal- using 35 eventsthe list of these events is given in table 3
ing functions in GK15 reasonably represent the data trends. of Graizer and Kalkan (2015). The majority of the events,
In similar plots given in Graizer and Kalkan (2015) for V S30 especially at small magnitudes, are from California. The
and Bdepth , the flatness of the trends indicates that our linear events with fewer than five recordings were excluded; this
site response function (applicable for V S30 > 200 m=s) is a reduced the number of events from 47 to 35. Our magnitude-
reasonable average for shallow crustal continental regions, scaling function (G1 ) captures the trends from various events
and there is little dependence on Bdepth between 1200 and as evident by near-zero intercept and near-zero slope for
700 V. Graizer and E. Kalkan
3
dashed line is maximum likelihood fit (y=a+bx): a = 1.157 and b = 0.56
number of events = 35
PGA
3
3
a = 1.194 and b = 0.61
Event Term (j)
PSA at 0.2 s
3
3
a = 1.177 and b = 0.64
PSA at 1.0 s
3
1 1.14 1.28
Style of Faulting (F)
Figure 12. Distribution of event terms i in natural logarithmic units for PGA and PSA at 0.2 and 1.0 s with respect to style of faulting (F);
in each plot dashed line indicates a maximum-likelihood fit to all event terms; its slope and intercept are provided on top of each plot. Note
that a and b are intercept and slope of maximum-likelihood fit, respectively. The color version of this figure is available only in the electronic
edition.
magnitude, and near-zero slope for style of faulting of and in natural logarithmic units (their values are tabulated
the maximum-likelihood fit, indicating that there is no sig- in Table 4). The variability is larger at long periods. in-
nificant trend with either parameters or a notable offset from creases with period similar to the NGA-West 2 GMPEs (e.g.,
zero. Abrahamson et al., 2014; Boore et al., 2014; Chiou and
Youngs, 2014). is almost constant for short periods (from
Terms of Standard Deviation 0.01 to 0.3 s).
Recall that our GMPE for spectral shape is a continuous
In GMPEs, the total residual is composed of the intrae- function of spectral period. To be consistent with this con-
vent residual and the interevent residual. The standard tinuous form, we model the total aleatory variability ()
deviation of total residuals (), that is, total aleatory variabil- using the method of least squares as a continuous function
ity is defined as of spectral period (T):
q
2 2 ; 18
EQ-TARGET;temp:intralink-;df18;55;117
0:668 0:0047 logT
in which is the standard deviation of the interevent resid- T max : 19
0:8 0:13 logT
EQ-TARGET;temp:intralink-;df19;313;113
1
Table 4
Interevent ()
0.5
0.120 0.431 0.529 0.683 1.900 0.635 0.616 0.885
0.130 0.429 0.530 0.682 2.000 0.635 0.624 0.890
Actual 0.133 0.426 0.531 0.680 2.200 0.639 0.634 0.900
Approximation 0.140 0.423 0.532 0.680 2.400 0.640 0.648 0.911
0
PGA 0.1 1 5 0.150 0.422 0.534 0.681 2.500 0.644 0.671 0.930
Period (s) 0.160 0.420 0.536 0.681 2.600 0.650 0.693 0.950
0.170 0.419 0.536 0.680 2.800 0.656 0.710 0.967
Figure 14. Total observed standard deviations () of GK15 0.180 0.414 0.536 0.678 3.000 0.660 0.718 0.975
GMPE, and its parameterization via equation (19) in natural logarith- 0.190 0.410 0.539 0.677 3.200 0.665 0.719 0.979
mic units. The color version of this figure is available only in the 0.200 0.407 0.541 0.677 3.400 0.673 0.719 0.984
electronic edition. 0.220 0.407 0.544 0.679 3.500 0.680 0.723 0.992
0.240 0.409 0.547 0.683 3.600 0.683 0.722 0.994
0.250 0.408 0.550 0.685 3.800 0.687 0.727 1.000
Figure 14 compares the parameterized of equation (19) 0.260 0.406 0.554 0.687 4.000 0.682 0.721 0.992
with the observed given in Table 4, showing an excellent 0.280 0.407 0.558 0.690 4.200 0.680 0.715 0.986
match at all periods. The variance analysis is conducted to 0.290 0.405 0.561 0.692 4.400 0.676 0.717 0.985
examine the magnitude, distance, and V S30 dependence of 0.300 0.406 0.564 0.694 4.600 0.670 0.718 0.982
0.320 0.410 0.565 0.698 4.800 0.667 0.718 0.980
interevent standard deviations () and intraevent standard
0.340 0.414 0.566 0.701 5.000 0.699 0.745 1.022
deviations (); for brevity, the results of this analysis are
provided in Graizer and Kalkan (2015).
4 2
1 km
R = 1 km, VS30 = 760 m/s R = 1 km, VS30 = 270 m/s 1
1 km
1 30 km
Spectral Acceleration (g)
30 km
M8 0.1
150 km
Acceleration (g)
M8
M7
0.1 150 km
M7
M6 0.01
M6
M5
0.01
M5 0.001
M=8
M=7 R=1
M=6 R=30 PGA PSA at 0.2 s
M=5 R=150
0.001 0.0001
4 2
R = 30 km, VS30 = 760 m/s R = 30 km, VS30 = 270 m/s 1 1 km
1
Spectral Acceleration (g)
30 km 1 km
0.1 30 km
Acceleration (g)
150 km
0.1 M8
M8 150 km
0.01
M7
M7
0.01 M6
0.001
M6
M5
PSA at 1.0 s PSA at 3.0 s
M5
0.001 0.0001
0.01 0.1 1 5 0.01 0.1 1 5 5 6 7 8 5 6 7 8
Period (s) Period (s) Moment Magnitude (M) Moment Magnitude (M)
Figure 15. Comparison of median PSA for strike-slip magni- Figure 17. Comparison of magnitude scaling for strike-slip
tude (M) 5.0, 6.0, 7.0, and 8.0 earthquakes at R 1 and 30 km earthquakes at closest fault distance to rupture plane, R 1, 30,
and V S30 270 and 760 m=s considering Q0 150 and Bdepth and 150 km for median PGA and PSA at 0.2, 1.0, and 3.0 s consid-
0 km (R, closest fault distance to rupture plane; V S30 , shear-wave ering V S30 760 m=s, Q0 150, and Bdepth 0 km (R, closest
velocity in the upper 30 m of the geological profile; Q0 , quality fault distance to rupture plane; V S30 , shear-wave velocity in the
factor; Bdepth , depth to basin). The color version of this figure is upper 30 m of the geological profile; Q0 , quality factor; Bdepth ,
available only in the electronic edition. depth to basin). Note that the weak scaling of the short-period mo-
tion at short distances reflects the saturation with magnitude. The
2
PGA PSA at 0.2 s
color version of this figure is available only in the electronic edition.
1
0.1
M8
(520 km from the fault), GK15 produces higher accelera-
M8
tion values than those at the closer distances, which appear as
M7
0.01 M6
the bump on the attenuation curves (as discussed earlier in
M7
M=8 M6
M5
the G2 (Distance Attenuation) section).
M=7
M=6 M5
The magnitude scaling of the current model is shown in
M=5
0.001
2
Figure 17 for vertical strike-slip earthquakes on soft-rock site
1
PSA at 1.0 s PSA at 3.0 s conditions (V S30 760 m=s) for PGA, T 0:2, 1.0, and
3.0 s at distances of R 1, 30, and 150 km. The break
in the magnitude scaling at M 5.5 is driven by consistency
Acceleration (g)
0.1
M8 in response spectra of recorded data. The weak scaling of the
M8 M7 short-period motion at short distances reflects the saturation
M7
with magnitude. Further detailed model results considering
0.01
M6 other independent parameters (e.g., V S30 , Q0 , and Bdepth ) can
M6
M5
be found in Graizer and Kalkan (2015).
0.001
0.1 1 10 100 250 0.1 1 10
M5
100 250
In Figures 18 and 19, PSA predictions of GK15 are com-
R (km) R (km) pared with those computed from near-field (020 km) and
far-field (4565 km) records of select major earthquakes
Figure 16. Comparison of distance scaling for strike-slip mag- from California, Turkey, and Taiwan. The ground-motion re-
nitude (M) 5.0, 6.0, 7.0, and 8.0 earthquakes at median PGA and
PSA at 0.2, 1.0, and 3.0 s considering V S30 760 m=s, cords either correspond to NEHRP site category C or D. For
Q0 150, and Bdepth 0 km (R, closest fault distance to rupture each earthquake, the number of records satisfying the dis-
plane; V S30 , shear-wave velocity in the upper 30 m of the geological tance and soil condition selection criteria is listed on each
profile; Q0 , quality factor; Bdepth , depth to basin). At intermediate panel where the average spectra of records are shown by
distance range (520 km from the fault), the GK15 produces higher
acceleration values; these high accelerations look like a bump on the
jagged thick curves, and they are compared with the GMPE
attenuation curves. The color version of this figure is available only predictions shown by smooth thick curves. Individual spectra of
in the electronic edition. records shown by jagged thin curves demonstrate the aleatoric
Summary of the GK15 GMPE for Horizontal PGA and 5% Damped PSA 703
1
Spectral Acceleration (g)
0.1
1979 M6.5 Imperial Valley Eq. 1999 M7.6 ChiChi Eq. 1989 M6.9 Loma Prieta Eq.
(SOF = SS; Site = D; Num. of records = 22 ) (SOF = REV; Site = C; Num. of records = 52 ) (SOF = REV; Site = C; Num. of records = 11 )
0.01
Average ith GM GK15 (median) GK15 (median )
4
1
Spectral Acceleration (g)
0.1
1989 M6.9 Loma Prieta Eq. 1994 M6.4 Nortridge Eq. 1994 M6.4 Northridge Eq.
(SOF = REV; Site = D; Num. of records = 5 ) (SOF = REV; Site = C; Num. of records = 15 ) (SOF = REV; Site = D; Num. of records = 12 )
0.01
0.01 0.1 1 5 0.01 0.1 1 5 0.01 0.1 1 5
Period (s)
Figure 18. Comparison of GK15 predictions for PSA with the observations from select earthquakes. Ground-motion data correspond to
randomly oriented horizontal components of near-field records (Rrup 10 10 km) with NEHRP site class C or D. Dashed curves indicate
16th and 84th percentile predictions; SOF, style of faulting; SS, strike-slip fault; REV, reverse fault; number of records from each earthquake
are indicated. The color version of this figure is available only in the electronic edition.
variability. The predictions from the GK15 are for the average of the two NGA-West2 models for the given range of mag-
V S30 of each individual dataset. As shown, the 16th and 84th nitudes and distances. For PGA and PSA at 0.2 s, GK15 pro-
percentile predictions (shown by dashed curves) bound the ma- duces similar or slightly lower ground motions as compared
jority of the PSA data. For all events, the GK15 clearly results in with BSSA14 and ASK14 at close distances (05 km). At
PSA predictions closer to the average of the observations; the intermediate distances (520 km), GK15 produces higher
predicted and observed trends of the peak (period and ampli- values because of the bump.
tude) of the response spectra with magnitude and distance The response spectra from magnitude 6.0, 7.0, and 8.0
match. The width of the predicted response spectra is also com- earthquakes at R 1 and 30 km from a vertically dipping
parable to the observations. strike-slip fault and V S30 760 m=s are shown in Figure 21.
There is similarity among the three models for these cases. The
observed difference between the models increases for the M 8
Comparisons with Select NGA-West2 Models
case, especially at the long spectral periods, because magnitude-
We compare GK15 with two NGA-West2 models of scaling functions and data used to constrain them vary from one
Abrahamson et al. (2014) (hereafter referred to as ASK14) GMPE to another; it is also related to lack of observations at this
and Boore et al. (2014) (hereafter referred to as BSSA14). We magnitude range. Further comparisons with BSSA14 and
do not include other NGA-West2 models (Campbell and Bo- ASK14 models as well as comparisons with the GK07-09
zorgnia, 2014; Chiou and Youngs, 2014) for simplicity and be- model can be found in Graizer and Kalkan (2015).
cause they themselves compare well with BSSA14 and ASK14
(Gregor et al., 2014). Figure 20 compares the median attenu- Range of Applicability
ation of PGA and PSA at 0.2 and 1.0 s for a vertically dipping
strike-slip earthquake as a function of R at V S30 760 m=s. GK15 is applicable to earthquakes of moment magni-
Our predictions are in good agreement with the predictions tude 5.08.0 (except for M > 7:0 normal-slip events), at
704 V. Graizer and E. Kalkan
2
1992 M7.3 Landers Eq. 1999 M7.6 ChiChi Eq. 1989 M6.9 Loma Prieta Eq.
(SOF = SS; Site = D; Num. of records = 2 ) (SOF = REV; Site = D; Num. of records = 47 ) (SOF = REV; Site = C; Num. of records = 6 )
1
Spectral Acceleration (g)
0.1
0.01
Average ith GM GK15 (median) GK15 (median )
2
1989 M6.9 Loma Prieta Eq. 1999 M7.1 Hector Mine Eq. 1994 M6.4 Northridge Eq.
1 (SOF = REV; Site = C; Num. of records = 6 ) (SOF = SS; Site = C; Num. of records = 3 ) (SOF = REV; Site = D; Num. of records = 17 )
Spectral Acceleration (g)
0.1
0.01
0.01 0.1 1 5 0.01 0.1 1 5 0.01 0.1 1 5
Period (s)
Figure 19. Comparison of GK15 predictions for PSAs with the observations from select earthquakes. Ground-motion data correspond to
randomly oriented horizontal components of far-field records (R 55 10 km) with NEHRP site class C or D. Dashed curves indicate 16th
and 84th percentile predictions; SOF, style of faulting; SS, strike-slip fault; REV, reverse fault; number of records from each earthquake are
indicated. The color version of this figure is available only in the electronic edition.
closest distances to fault rupture plane from 0 to 250 km, at comparable standard deviations. The GK15 models PSA as
sites having V S30 in the range from 200 to 1300 m=s, and for a continuous function of spectral period, whereas the NGA-
spectral periods (T) of 0.015 s. Neither hanging-wall effects West2 models use a discrete functional form to compute PSA
nor directivity effects are explicitly modeled, but the average response ordinates at certain periods only (a total of 21 peri-
directivity effect is included through the variability of the data. ods for ASK14, CB14, and CY14, and 107 periods for
BSSA14 from 0.01 to 10 s). The advantages of continuous
Summary and Discussion function are that period-by-period regression analyses are
eliminated, the number of estimator coefficients is mini-
In this article, the GraizerKalkan GMPE for PGA and
mized, and a smooth spectrum is obtained directly from the
5% damped PSA response ordinates is revised to account for
regression analysis. Because this approach is simple and pro-
differences in ground-motion scaling in terms of regional far-
source distance attenuation and basin effects. The new GMPE vides more control, it is easier to model and constrain param-
is controlled by six input predictor parameters, including eters affecting the spectral shape. The GK15s use of the
moment magnitude, closest distance to the fault rupture, style quality factor (determined using Lg or coda waves) that
of faulting, shear-wave velocity in the upper 30 m of site geo- can be changed to suit the region of interest is an improve-
logical formation, regional quality factor (Q0 ), and basin ment over other GMPEs.
depth (defined as the depth to 1:5 km=s shear-wave velocity Comparisons of GK15 with ASK14 and BSSA14 dem-
isosurface). onstrate that GK15 produces similar or slightly smaller
The revised GraizerKalkan GMPE (GK15) offers a ground motions at very close distances to the fault (up to
much simpler functional form than the recent NGA-West2 about 5 km) and at distances of more than 20 km from the
models (e.g., ASK14; BSSA14; Campbell and Bozorgnia, fault for earthquakes with a magnitude larger than 6.0. At
2014 [CB14]; Chiou and Youngs, 2014 [CY14]), and it has intermediate distances (520 km from the fault), GK15
Summary of the GK15 GMPE for Horizontal PGA and 5% Damped PSA 705
1
PGA PSA at 0.2 s PSA at 1.0 s
Acceleration (g)
0.1
0.01
M=8 (GK15)
M=7
M=6
M=8 (BSSA14)
M=7
M=6
0.001
2
1
Acceleration (g)
0.1
0.01
M=8 (GK15)
M=7
M=6
M=8 (ASK14)
M=7
M=6
0.001
0.1 1 10 100 250 0.1 1 10 100 250 0.1 1 10 100 250
R (km)
Figure 20. Comparison of distance scaling of GK15 with Boore et al. (2014) (BSSA14) and Abrahamson et al. (2014) (ASK14) for
strike-slip magnitude (M) 6.0, 7.0, and 8.0 earthquakes at median PGA and PSAs at 0.2 and 1.0 s considering V S30 760 m=s, Q0 150, and
Bdepth 0 km (V S30 , shear-wave velocity in the upper 30 m of the geological profile; Q0 , quality factor; Bdepth , depth to basin). The color
version of this figure is available only in the electronic edition.
produces higher estimates of ground motion than either FORTRAN, and Excel at http://earthquake.usgs.gov/research/
ASK14 or BSSA14 does. software/ (last accessed July 2015). Also provided in this link
Compared with the GK07-09 model (Graizer and Kal- is an example MATLAB routine to generate a site-specific re-
kan, 2007, 2009), GK15 yields very similar PGA values sponse spectrum for given hazard conditions. Readers may
whereas spectral ordinates predicted are larger at T < 0:2 s, find further details about GK15 in the U.S. Geological Survey
and they are smaller at longer periods. This adjustment was (USGS) open-file report available at http://pubs.usgs.gov/of/
done based on reevaluation of the ground-motion data from 2015/1009/ (last accessed July 2015). The 2014 M 6.0 South
California earthquakes such as Landers, Northridge, and Napa earthquake ground-motion data used here are available
Hector Mine. The aleatory variability of GK15 is larger than at http://earthquake.usgs.gov/earthquakes/shakemap/nc/shake/
that obtained in our previous model. 72282711/#download (last accessed July 2015). V S30 values
In conclusion, the GK15 GMPE is a significant improve- at each station are taken from the Next Generation of Attenu-
ment over our previous model (GK0709), and provides a ation (NGA)-West2 database (http://peer.berkeley.edu/
demonstrable, reliable description of ground-motion ampli- ngawest2/databases/, last accessed July 2015) or the grid.xml
tudes recorded from shallow crustal earthquakes in active file also available with the ShakeMap download link provided
tectonic regions over a wide range of magnitudes, distances, previously.
and site conditions.
Acknowledgments
Data and Resources The authors thank Martin Chapman, Jon Ake, and Dogan Seber for
insightful discussions on ground-motion attenuation. Special thanks are ex-
The GraizerKalkan (GK15) ground-motion prediction tended to Stephen Harmsen, Nick Gregor, and Michael Pasyanos for inde-
equation (GMPE) is available in MATLAB (http://www. pendently testing our ground-motion prediction equation (GMPE), and
mathworks.com/products/matlab/, last accessed March 2016), providing useful feedback. David Boore, Art Frankel, Brad Aagaard, Anne-
706 V. Graizer and E. Kalkan
4
R = 1 km, VS30= 760 m/s R = 30 km, VS30= 760 m/s Bozorgnia, Y., N. A. Abrahamson, L. Al Atik, T. D. Ancheta, G. M. Atkin-
1 son, J. W. Baker, A. Baltay, D. M. Boore, K. W. Campbell, B. S.-J.
Spectral Acceleration (g)
Graizer, V., A. Shakal, C. Scrivner, E. Hauksson, J. Polet, and L. Jones Sadigh, K., C. Y. Chang, J. A. Egan, F. Makdisi, and R. R. Youngs (1997).
(2002). TriNet strong-motion data from the M 7.1 Hector Mine, Attenuation relationships for shallow crustal earthquakes based on
California, earthquake of 16 October 1999, Bull. Seismol. Soc. Am. California strong motion data, Seismol. Res. Lett. 68, no. 1, 180189.
92, no. 4, 15251541. Semblat, J. F., P. Dangla, M. Kham, and A. M. Duval (2002). Seismic effect
Graves, R. W., and B. T. Aagaard (2011). Testing long-period ground- for shallow and deep alluvial basins; in-depth motion and focusing
motion simulations of scenario earthquakes using the Mw 7.2 El effect, Soil Dynam. Earthq. Eng. 22, nos. 9/12, 849854.
Mayor-Cucapah mainshock: Evaluation of finite-fault rupture charac- Singh, S., and R. B. Herrmann (1983). Regionalization of crustal coda Q in
terization and 3D seismic velocity models, Bull. Seismol. Soc. Am. the continental United States, J. Geophys. Res. 88, no. B1, 527538.
102, no. 2, 895907. Spudich, P., W. B. Joyner, A. G. Lindh, D. M. Boore, B. M. Margaris, and
Gregor, N., N. A. Abrahamson, G. M. Atkinson, D. M. Boore, Y. Bozorgnia, J. B. Fletcher (1999). SEA99: A revised ground motion prediction re-
K. W. Campbell, B. S.-J. Chiou, I. M. Idriss, R. Kamai, E. Seyhan, lation for use in extensional tectonic regimes, Bull. Seismol. Soc. Am.
et al. (2014). Comparison of NGA-West 2 GMPEs, Earthq. Spectra 89, no. 5, 11561170.
30, no. 3, 11791197. Trifunac, M. D. (1994). Q and high-frequency strong motion spectra, Soil
Hanks, T. C. (1975). Strong ground motion of the San Fernando, California, Dynam. Earthq. Eng. 13, 149161.
earthquake; ground displacements, Bull. Seismol. Soc. Am. 65, no. 1, Wills, C. J, M. Petersen, W. A. Bryant, M. Reichle, G. J. Saucedo, S. Tan, G.
193225. Taylor, and J. Treiman (2000). A site conditions map for California
Hatayama, K., and E. Kalkan (2012). Spatial amplification of long-period (3 based on geology and shear-wave velocity, Bull. Seismol. Soc. Am.
to 16 s) ground-motions in and around the Los Angeles Basin during 90, no. 6B, 187208.
the 2010 M 7.2 El Mayor-Cucapah earthquake, Proc. of the 15th World Zhang, F., and A. S. Papagergiou (2010). Attenuation characteristics of Tai-
Conf. on Earthquake Engineering, Lisbon, Portugal, 2428 September wan: Estimation of coda Q, S-wave Q, scattering Q, intrinsic Q, and
2012 scattering coefficient, Seismol. Res. Lett. 81, 769777.
Hruby, C. E., and I. A. Beresnev (2003). Empirical corrections for basin effect
in stochastic ground-motion prediction, based on the Los Angeles basin
analysis, Bull. Seismol. Soc. Am. 93, no. 4, 16791690.
Joyner, W. B. (2000). Strong motion from surface waves in deep sedimen- U.S. Nuclear Regulatory Commission
tary basins, Bull. Seismol. Soc. Am. 90, S95S112. Mail Stop T-7F3
Joyner, W. B., and D. M. Boore (1993). Methods for regression analysis of Washington, D.C. 20555-0001
strong-motion data, Bull. Seismol. Soc. Am. 83, no. 2, 469487. vladimir.graizer@nrc.gov
Lee, V. W., M. D. Trifunac, M. I. Todorovska, and E. I. Novikova (1995). (V.G.)
Empirical equations describing attenuation of peak of strong ground-
motion, in terms of magnitude, distance, path effects and site condi-
tions, Department of Civil Engineering, University of Southern California,
Rept. No. CE 9502, 269 pp. Earthquake Science Center
U.S. Geological Survey
Li, Y. G., and J. E. Vidale (1996). Low-velocity fault-zone guided waves:
Mail Stop 977
Numerical investigations of trapping efficiency, Bull. Seismol. Soc. 345 Middlefield Rd.
Am. 86, 371378. Menlo Park, California 94025
Mitchell, B. J., and H. J. Hwang (1987). Effect of low Q sediments and ekalkan@usgs.gov
Crustal Q on Lg attenuation in the United States, Bull. Seismol. (E.K.)
Soc. Am. 77, no. 4, 11971210.
Olsen, K. B. (2000). Site amplification in the Los Angeles basin from three-
dimensional modeling of ground-motion, Bull. Seismol. Soc. Am. 90, Manuscript received 29 January 2016;
no. 6B, S77S94. Published Online 5 April 2016