Вы находитесь на странице: 1из 23

Bull Earthquake Eng (2014) 12:549–571

DOI 10.1007/s10518-013-9533-5

ORIGINAL RESEARCH PAPER

Ground-motion prediction equations for interface


earthquakes of M7 to M9 based on empirical data
from Japan

Hadi Ghofrani · Gail M. Atkinson

Received: 30 April 2013 / Accepted: 30 September 2013 / Published online: 12 October 2013
© Springer Science+Business Media Dordrecht 2013

Abstract Ground motion prediction equations (GMPEs) have a major impact on seismic
hazard estimates, because they control the predicted amplitudes of ground shaking. The pre-
diction of ground-motion amplitudes due to mega-thrust earthquakes in subduction zones has
been hampered by a paucity of empirical ground-motion data for the very large magnitudes
(moment magnitude (M) > 7) of most interest to hazard analysis. Recent data from Tohoku
M9.0 2011 earthquake are important in this regard, as this is the largest well-recorded sub-
duction event, and the only such event with sufficient data to enable a clear separation of the
overall source, path and site effects. In this study, we use strong-ground-motion records from
the M9 Tohoku event to derive an event-specific GMPE. We then extend this M9 GMPE to
represent the shaking from other M > 7 interface events in Japan by adjusting the source term.
We focus on events in Japan to reduce ambiguity that results when combining data in differ-
ent regions having different source, path and site effect attributes. Source levels (adjustment
factors) for other Japanese events are determined as the average residuals of ground-motions
with respect to the Tohoku GMPE, keeping all other coefficients fixed. The mean residuals
(source terms) scale most steeply with magnitude at the lower frequencies; this is in accord
with expectations based on overall source-scaling concepts. Interpolating source terms over
the magnitude range of 7.0–9.0, we produce a GMPE for large interface events of M7–M9, for
NEHRP B/C boundary site conditions (time-averaged shear-wave velocity of 760 m/s over the
top 30 m) in both fore-arc and back-arc regions of Japan. We show how these equations may be
adjusted to account for the deeper soil profiles (for the same value of VS30 ) in western North
America. The proposed GMPE predicts lower motions at very long periods, higher motions
at short periods, and similar motions at intermediate periods, relative to the simulation-based
GMPE model of Atkinson and Macias (2009) for the Cascadia subduction zone.

Keywords Ground motion prediction equations · Tohoku M9.0 mega-thrust earthquake ·


Probabilistic seismic hazard analysis · Source scaling · Epistemic and aleatory uncertainty

H. Ghofrani (B) · G. M. Atkinson


Earth Sciences Department, Western University, London, ON N6A 5B7, Canada
e-mail: hghofran@alumni.uwo.ca
G. M. Atkinson
e-mail: Gmatkinson@aol.com

123
550 Bull Earthquake Eng (2014) 12:549–571

1 Introduction

Mega-thrust earthquakes that occur along subduction-zone plate interfaces are a major global
source of seismic hazard in many regions, including Japan, South America, Alaska, and the
Pacific Northwest. Several studies have attempted to characterize interface ground motions
in the form of ground motion prediction equations (GMPEs), giving expected amplitudes as
functions of magnitude and distance (e.g. Crouse et al. 1988; Young et al. 1988; Crouse 1991;
Youngs et al. 1997; Atkinson and Boore 2003; Zhao et al. 2006; Kanno et al. 2006). These
studies generally suggest that at regional distances from the rupture (>50 km), these earth-
quakes may produce larger ground motions in comparison to shallow crustal earthquakes
in active tectonic regions. However, the characteristics of ground motions generated during
subduction-zone earthquakes is a subject that has received little attention relative to studies of
ground motions generated by shallow crustal earthquakes (e.g. see Power et al. 2008 and ref-
erences therein). Moreover, there are questions regarding the extent to which ground-motion
data from different regions should be combined in developing such GMPEs, in particular
because of the potential of systematic differences in regional site conditions and/or attenua-
tion to cause bias in the GMPEs (e.g. Atkinson and Boore 2003; Atkinson and Casey 2003).
For engineering applications, the events of most interest are those of moment magnitude
(M)> 7.0. Because interface events happen offshore, typically at distances >50 km from
populated areas, only large events pose a significant hazard. It therefore makes sense to focus
GMPEs for interface events on the prediction of motions for events of M > 7. In the last
decade, the available ground-motion database for such events has grown significantly in some
regions, especially Japan. In particular, the dataset of >600 strong-motion recordings from
the March 11, 2011 M9 Tohoku Japan earthquake, at ∼40–1000 km from the fault plane,
provides important new information on ground motions that should be considered for hazard
analysis in the world’s subduction zones. It is, however, important to recognize that these
records carry within them the signature of highly-significant shallow soil response that is
specific to recording sites in Japan (Ghofrani et al. 2013).
In this study, we use regression analysis to characterize the median ground motion ampli-
tudes (peak motions and response spectra) as a function of closest distance from the fault
plane (Rcd ), and shear-wave velocity in the upper 30 m (VS30 ; Borcherdt 1992) for this
important event. We then evaluate the ground motions from other interface events of M >
7 in Japan relative to the Tohoku motions, to derive an empirical GMPE that is specifically
tuned to the prediction of ground motion from major mega-thrust events. By restricting the
focus to Japan we are able to take advantage of the comprehensive information available on
site conditions for the Japanese stations, which allows us to derive GMPEs for a specified
reference condition (B/C, or VS30 = 760 m/s), and also for specified attenuation conditions.
Each record is classified as forearc or backarc based on its position relative to the volcanic
front; this classification is important because the attenuation of motion versus distance is
slower for forearc regions than for backarc regions (Ghofrani and Atkinson 2011). We also
address the modifications that are needed to make this GMPE applicable to other regions
characterized by deeper soil profiles, specifically western North America. To our knowledge,
this is the first such set of GMPEs for large mega-thrust events developed with a uniform
dataset, and focusing on the events of most concern for hazard.

2 Strong ground motion data and record processing

An extensive set of strong ground-motion data was recorded by Japanese accelerograph


stations during the 2011 Tohoku earthquake and other events. The strong-motion data and site

123
Bull Earthquake Eng (2014) 12:549–571 551

46o
K-NET stations
Trench
Volcanic front
44o GSI’s Fault Plane
2011 M9.0 Tohoku
Selected Aftershocks
M7.0+ Interface 04/11/29-03:32
Earthquakes M7.0
o
42 03/09/26-04:50
M8.1
03/09/26-06:08
M7.3

40o
05/08/16-11:46
M7.1

38o

36o

34o
0 200 km

136o 138o 140o 142o 144o 146o


Fig. 1 Map showing fault plane and stations triggered by the Tohoku event (black dots) at closest distance
from the fault plane ranging from 43 to 1,000 km. A graphical representation of the background fault plane
(hatched rectangle) for the mainshock, adopted from GSI’s finite-fault model is also shown. The hypocenter
of the mainshock is indicated with the large star close to the trench. Details of the source model are given
in Table 2. Selected aftershocks (circles) and M7.0+ interface earthquakes (squares) are also shown in this
figure. The sizes of symbols are representative of magnitudes

information used in this study were downloaded from the Kyoshin network (K-NET; http://
www.kik.bosai.go.jp/, last accessed August 2013) of the NIED (National Research Institute
for Earth Science and Disaster Prevention) of Japan. The K-NET consists of more than 1,000
observation stations installed on the ground surface. Figure 1 shows the distribution of the K-
NET stations triggered by the 2011 Tohoku event relative to the earthquake fault plane surface
projection, the volcanic-front, and the trench. The overall fault geometry of the Tohoku event
was adopted based on a modified version of the Geospatial Information Authority of Japan
(GSI 2011) model. The focal mechanism information is provided in Table 1.
Soil profiles for each station are generally available for the top 10 or 20 m. The most
common parameter for the simplified classification of a site in terms of its seismic response is
VS30 , the time-averaged velocity to a depth of 30 m [National Earthquake Hazards Reduction
Program (NEHRP 2000), Eurocode 8 (2004), and the National Building Code of Canada
(NBCC 2005)]. To calculate VS30 for each site, the site velocity profile is extended to 30 m

123
552 Bull Earthquake Eng (2014) 12:549–571

Table 1 Subduction interface events with M > 7.0 used in study

Event ID Event Lat Lon Depth Strike Dip Length* Width* Ref.
(yyyy/mm/dd- name (◦ ) (◦ ) (km) (◦ ) (◦ ) (km) (km)
hh:mm)

2003/09/26-04:50†† Tokachi 41.78 144.08 17.0 250 20 130 170 F-NET


2003/09/26-06:08†† Tokachi 41.70 143.80 53.0 199 21 66 49 F-NET
(aftershock)
2004/11/29-03:32 Hokkaido 42.97 145.44 37.4 212 28 23 24 GSI
Kushiro
2005/08/16-11:46 Miyagi-ken 38.13 142.55 40.3 211 21 32 33 GSI
2011/03/11-14:46 Tohoku 39.00 143.49 10.0 202 18 400 150 GSI†
2011/03/11-15:15 Tohoku 36.10 141.30 35.0 209 31 122 72 F-NET
(aftershock)
∗ Length (L) and width (W) of earthquake faults with no finite-fault solution (F-NET: http://www.fnet.
bosai.go.jp/) are calculated based on Strasser et al. (2010) relations: L = 10(−2.477+0.585∗M) ; W =
10(−0.882+0.351∗M)
† Focal mechanisms are based on the reports by Geographical Survey Institute (GSI), where, “latitude”,
“longitude”, and “depth” correspond to a corner of the fault plane
†† For the Tokachi-oki earthquake, M = 8.12 (Macias et al. 2008). For the mainshock, the fault-plane geometry
is based on Yagi (2004) (strike, 250◦ ; dip, 20◦ ; latitude, 41.78◦ N; longitude, 144.08◦ E; depth, 17 km; length,
130 km; and width 170 km)

Histogram of VS30 values in forearc and backarc regions


150 Median VS30
312.65 (m/s)
Median VS30
265.60 (m/s) Back-arc
Fore-arc
Number of observations

100

50

0
100 200 300 1000 2000
VS30 (m/s)

Fig. 2 Distribution of VS30 values in forearc and backarc regions. Arrows show the location of 50 % percentiles
(i.e. median)

using the model given by Boore et al. (2011). The median VS30 for forearc stations is ∼266 m/s
and for backarc stations it is ∼313 m/s. Based on the histogram of VS30 values in for-arc
and backarc regions (Fig. 2), most of the stations are in the backarc regions; these sites are
slightly stiffer than those in the forearc region.

123
Bull Earthquake Eng (2014) 12:549–571 553

The processing procedure for all records includes windowing, correction for baseline and
constant DC trends (zero frequency), and band-pass filtering. Signals are zero-padded at both
ends to a sufficient duration for reliable processing, considering the corner frequency of the
filter (Converse and Brady 1992). We have applied non-causal, band-pass Butterworth filters
with an order of 4. The selected frequency range of analysis is 0.04–15 Hz. To make the energy
of the signal zero at the beginning and the end, a 5 % cosine taper is applied to both ends.
Detailed data selection and preparation procedures can be found in Ghofrani et al. (2013). For
each record, the instrument-corrected geometric-mean horizontal component peak ground
acceleration (PGA) and Pseudo-spectral acceleration (PSA; critical damping = 5 %) are
calculated. Log(10) amplitudes of the spectra are tabulated at frequencies having a spacing
of 0.1 log frequency units, where the log(10) amplitudes were averaged within each frequency
bin, centered about the tabulated frequency.

3 Developing a GMPE from the empirical data (regressions of the Tohoku M9.0 data)

To develop a GMPE that accurately describes the observed Tohoku motions, a regression
analysis is performed using the following equation:

log Y = c0 − log Re f f + c1 F Rcd + c2 B Rcd + c3 log (VS30 /760)



where Re f f = Rcd 2 + h 2 and R is the minimum distance from the finite fault model.
cd
Prediction equations are derived using separate anelastic attenuation factors for forearc and
backarc regions. For stations in the forearc region, F = 1 and B = 0, otherwise (for backarc
stations) F = 0 and B = 1. We assumed a fixed geometrical spreading with slope of −1
over all distance ranges. A “fictitious depth term” (h = 60 km) is used to force saturation of
ground-motions at close distance to the source (we use the term fictitious depth because it
mimics a depth term in the distance metric, but is not actually related to depth of the event;
it is merely a device to force near-distance saturation). The value of 60 km is adopted for
this parameter after searching for the value which minimizes both the standard deviation of
residuals and the average absolute residual (Fig. 3).
Variability of ground motion due to site amplification is considered as a linear function of
VS30 through the regression equation. Detailed studies have shown that, overall, nonlinear
effects were relatively small for the Tohoku event at most stations (Ghofrani et al. 2013).
Furthermore, the non-linear part of the response is difficult to determine and trades off against
the near-source saturation term. Therefore, after careful evaluation of the potential effects
of including a nonlinear term or correction, we opted to neglect this factor and treat the soil
response as linear on average. Any stations for which the shear-wave velocity profile is not
available are not used in the analysis (30 out of 697 stations).
The source term coefficient (c0 ) provides the overall level of the regression line, and
represents the source term for the Tohoku event. In calculating the regression coefficients,
we first estimate the c0 by using just the stations in the forearc region, at Rcd ≤ 200 km
(almost all of the stations at close distances are in the forearc region). In the second step, we
fixed c0 and calculated the attenuation and site coefficients.
In Table 2, regression coefficients for the 2011 M9.0 Tohoku, Japan event are reported
(Note: The PSA values of this model can be found in the GMPE tables that are posted on the
Engineering Seismology Toolbox at http://www.seismotoolbox.ca under Misc Resources).
Figure 4 shows this event-specific GMPE (black curves) in comparison to other ground-
motion prediction equations for subduction interface earthquakes. Based on inspection of the

123
554 Bull Earthquake Eng (2014) 12:549–571

Comparing statistics of residuals for varying fictitious depths


0.102 0.221

Rcut = 400 km

0.22

The average absolute residual


Standard deviation of residuals

0.1015

0.219

0.101

0.218
60 km

0.1005
0.217

0.1 0.216
0 50 100 150 200
Fictitious depth (km)

Fig. 3 Statistics of the residuals (i.e. the standard deviation [left y-axis] and the average absolute [right y-axis]
residuals) for varying fictitious depth, including data to 400 km in the fit

figure, the recorded ground motions from Tohoku decay more rapidly than predicted from
most of the GMPEs. This is partly because of the backarc attenuation represented by most of
the Tohoku stations. As shown by tomographic and seismic studies (Zhao 2010), there is a
high-attenuation, low-velocity region in the crust and upper mantle of Central Japan related
to the volcanic activity. This structure filters out the high-frequency content of ground-motion
in the backarc region. Recent studies (Dhakal et al. 2008; Boore et al. 2009; Ghofrani and
Atkinson 2011; Abrahamson et al. 2012) have tried to capture the resulting difference in decay
rate by implementing separate anelastic factors for forearc and backarc regions. The use of
separate attenuation factors for forearc and backarc regions results in a significant reduction
of intra-event variability of ground motions, especially at high-frequencies (Ghofrani and
Atkinson 2011). It is noted on Fig. 4 that the event-specific GMPE is in better agreement
with the Tohoku data than are the other GMPEs, which is expected since these were the only
data used in deriving the GMPE.
Table 2 also lists the original aleatory variability for the Tohoku mainshock motions
(addressed by standard deviation of the residuals “σ”; details are provided in Sect. 6). This is
an initial estimate of the intra-event variability. However, it should be noted that this apparent
variability is large at high frequencies, due to the influence of site resonances in shallow soil
sites in Japan, as discussed in a later section. Also included on Table 2 is our estimate of how
the Tohoku GMPE should be modified for use in western North America, where deeper soil
profiles predominate. These modifications are discussed in a later section.
Figure 5 shows the amplitude levels predicted by our GMPE for a M9 event at 100 km. This
is the typical distance range for many major cities relative to an interface subduction zone (e.g.
Vancouver, Seattle). From Fig. 5, it is clear that the B/C spectrum for the M9 Tohoku event
(grey curve) peaks at a very high frequency. The strength of the high-frequency response is
due to the average effects of shallow soil resonances, which are not actually removed by the

123
Bull Earthquake Eng (2014) 12:549–571 555

Table 2 Regression coefficients of PSA of horizontal ground-motions (K-NET)

Frequency Period c0 c1 c2 c3 σ Recommended


(Hz) (s) cascadia multiplicative
factor (log)

0.11 9.09 2.985 – −0.00019 −0.413 0.175 1.00 (0.000)


0.14 7.14 3.122 – −0.00019 −0.462 0.201 1.00 (0.000)
0.17 5.88 3.232 – −0.00027 −0.491 0.218 1.00 (0.000)
0.22 4.55 3.351 −0.00005 −0.00040 −0.561 0.237 1.03 (0.013)
0.27 3.70 3.502 −0.00023 −0.00057 −0.574 0.252 1.11 (0.044)
0.33 3.03 3.657 −0.00039 −0.00075 −0.554 0.264 1.20 (0.079)
0.41 2.44 3.829 −0.00059 −0.00097 −0.513 0.275 1.35 (0.129)
0.52 1.92 3.990 −0.00086 −0.00125 −0.518 0.285 1.49 (0.174)
0.64 1.56 4.105 −0.00112 −0.00152 −0.578 0.274 1.38 (0.139)
0.80 1.25 4.214 −0.00133 −0.00177 −0.646 0.261 1.23 (0.089)
0.99 1.01 4.336 −0.00158 −0.00205 −0.690 0.249 1.05 (0.021)
1.23 0.81 4.464 −0.00183 −0.00234 −0.705 0.249 1.03 (0.014)
1.53 0.65 4.587 −0.00200 −0.00262 −0.672 0.257 1.03 (0.011)
1.90 0.53 4.706 −0.00213 −0.00290 −0.606 0.276 1.02 (0.007)
2.37 0.42 4.820 −0.00224 −0.00315 −0.520 0.295 1.00 (0.002)
2.94 0.34 4.907 −0.00231 −0.00333 −0.438 0.307 0.90 (−0.046)
3.66 0.27 4.958 −0.00233 −0.00346 −0.386 0.312 0.77 (−0.114)
4.55 0.22 5.015 −0.00235 −0.00355 −0.289 0.310 0.66 (−0.183)
5.66 0.18 5.051 −0.00235 −0.00358 −0.180 0.312 0.58 (−0.237)
7.04 0.14 5.072 −0.00240 −0.00357 −0.082 0.324 0.53 (−0.272)
8.75 0.11 5.082 −0.00245 −0.00356 0.010 0.329 0.48 (−0.319)
10.88 0.09 5.022 −0.00244 −0.00346 0.027 0.326 0.44 (−0.357)
13.53 0.07 4.888 −0.00236 −0.00329 −0.046 0.313 0.44 (−0.357)
PGA 4.646 −0.00219 −0.00298 −0.219 0.284 0.50 (−0.301)
PGV 3.390 −0.00070 −0.00099 −0.331 0.195 1.00 (0.000)
σ is standard deviation of residuals for Tohoku M9 motions for Rcd ≤ 200 km

inclusion of VS30 as a regression variable, because in Japan a stiff site in terms of VS30 is
typically a shallow soil site underlain by a much harder layer (Ghofrani et al. 2013). Thus
the GMPE we obtain for B/C is specific to the types of shallow soil sites that are typical
in Japan. To adjust the GMPE for deeper soil conditions that predominate in western North
America, we would need to apply a multiplicative factor that represents the average difference
in site response, for the same value of VS30 , between Japan and western North America. We
adopted the factors suggested by Atkinson and Adams (2013) for this purpose, which are
listed in Table 2. They were obtained by taking the average of two alternative sets of such
factors available in the literature: (i) from Atkinson and Boore (2003), who obtained such
differences by considering region-based residuals to global GMPEs for subduction events;
and (ii) from Atkinson and Casey (2003), who looked at such factors in more detail by
comparing recorded motions for an event in Japan to those for a comparable event in western
North America. On Fig. 5, we show how applying the Japan-to-Cascadia correction factors
changes the spectral shape—bringing the high-frequency down and the low-frequency up.

123
556 Bull Earthquake Eng (2014) 12:549–571

AM09 Japan
Zhao06 (B/C)
Abrahamson12
This study [forearc]
This study [backarc]

freq = 0.3Hz freq = 1.0Hz 1000 freq = 3.3Hz

100

100

100
PSA (cm/s/s)

PSA (cm/s/s)

PSA (cm/s/s)
10
10
10

1 1 1

1 1 1
Residual

Residual

Residual
0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1 -1 -1
100 200 300 100 200 300 100 200 300

Rcd (km) Rcd (km) Rcd (km)

Fig. 4 Comparing event-specific prediction equation for the observed Tohoku ground-motions (black curves)
with other GMPEs (Zhao06 = Zhao et al. 2006; AM09 = Atkinson and Macias 2009; and Abrahamson12 = the
central BC Hydro GMPE of Abrahamson et al. 2012, for forearc sites). Forearc stations are shown with red
circles and backarc stations are shown with black cirlces. All GMPEs are plotted for B/C boundary conditions
(VS30 = 760 m/s). We corrected the Zhao06 motions from NEHRP class C (i.e. very dense soil and soft rock;
366 < VS ≤ 762 m/s) to B/C using the factors given by Boore and Atkinson (2008) and assuming that an
average C site has VS30 ∼ 450 m/s

We believe that the site-corrected GMPE (i.e. after applying the factors given in Table 2)
are more representative of expected motions in regions such as western North America that
have a deeper soil profile than typical sites in Japan. Note that after application of the site-
correction factors, the GMPE is quite similar to that of Abrahamson et al. (2012), which was
developed from a global database.
The idea of applying a correction for Japanese site conditions clearly has an important
impact on spectral amplitudes, as seen in Fig. 5. We examine this issue further in Fig. 6, which
shows alternative estimates of the amount of amplification that is likely to be present for the
B/C reference condition in Japan, in comparison to the factors that we adopted (as given in
Table 2). Based on the analysis of the Kiban-Kyoshin borehole network (KiK-NET) data by
Ghofrani et al. (2013) (their Table 3), the site amplification curve for the reference velocity of
760 m/s is near unity for downhole recordings (black line in Fig. 6), as would be commonly
expected. However, their analysis also shows that, at the surface, a B/C boundary site has
significant high-frequency amplification—as determined using surface-to-borehole spectral
ratios corrected for the depth effect (magenta curve). Another estimation of amplification for
B/C site condition can be derived by calculating the horizontal-to-vertical component ratio
for sites with VS30 ≥ 760 m/s (purple line in Fig. 6). The inferred surface amplification
by a factor of two or more at high-frequencies for all these estimates is a clear indication
of shallow low-velocity layer over bedrock, which is characteristic of the Japanese stations.

123
Bull Earthquake Eng (2014) 12:549–571 557

Fig. 5 Response spectrum for Prediction of M9.0 PSA at Rcd = 100 km, forearc
M9 at 100 km. The grey line is 1000
the regression line for Japan,
which may be modified for
western North America by
application of the factors of Table
2 (black line). The GMPEs of
Atkinson and Macias (AM09), 200

M9.0 PSA (cm/s/s) @ 100 km


Abrahamson et al. (2012) and
Zhao et al. (2006) are also shown,
for B/C conditions. The Zhao 100
GMPE is corrected first to B/C
(from C), using factors from
Boore and Atkinson (2008) and
assuming that C sites are 450 m/s;
then the Japan to Cascadia factors
of Table 2 are also applied (so our
20
estimate of the Zhao curve has
been corrected to represent an K-NET vs-based regression
event in western North America) 10 K-NET vs-based regression
corrected with Cascadia/Japan
site factors [Table 2]
AM09 [Cascadia]
Zhao06 (B/C)
Abrahamson 12

0.1 0.2 1 2 10 20
Frequency (Hz)

Vref 760 m/s (bedrock level)


Surface-to-Borehole Ratio for Vs30 = 760 m/s
3 Averaged H/V for Vs30 > _ 760 m/s
Adopted Japan site factors [Table 2]
Amplification

0.1 0.2 1 2 10
Frequency (Hz)

Fig. 6 Alternative estimates of amplification for B/C sites in Japan. Black line shows adopted factors from
Table 2 (inverse of tabulated values). Estimates of surface amplification for B/C by Ghofrani et al. (2013) based
on surface-to-borehole ratios is shown by the magenta line; dashed grey line shows B/C amplification function
inferred at depth (bedrock level at bottom of borehole). Amplification estimate from horizontal/vertical spectral
ratio for sites with VS30 ≥ 760 m/s is also shown (purple line)

These results support the need to apply regional site correction factors before comparing
GMPEs across regions, even for the reference B/C site condition.
Overall, an important feature of the Tohoku motions is that high-frequency components
of the motion were particularly strong. Median peak ground accelerations from the Tohoku
event were near 30 %g at 100 km, for NEHRP C sites. In large part, the large amplitudes at
high frequencies are due to the prevalence of shallow-soil conditions in Japan which amplify
higher frequencies as discussed above. However, source effects from this complex event may
also play a role in the large amplitudes at high frequencies. Another important feature is that

123
Table 3 Source terms (c0 = c0 [Tohoku] + mean residuals) for the M >7.0 interface earthquakes
558

Frequency (Hz) Tohoku (M9.0) Tokachi-oki (M8.1) Tokachi-oki (M7.3) Hokkaido Kushiro (M7.0) Miyagi-ken (M7.1) Tohoku (M7.8)

123
0.11 2.985 2.926 (±0.300) 2.173 (±0.368) 2.281 (±0.253) 2.097 (±0.196) 2.494 (±0.311)
0.14 3.122 3.099 (±0.340) 2.403 (±0.390) 2.487 (±0.275) 2.354 (±0.203) 2.625 (±0.332)
0.17 3.232 3.222 (±0.366) 2.561 (±0.397) 2.652 (±0.275) 2.508 (±0.244) 2.725 (±0.326)
0.22 3.351 3.335 (±0.386) 2.734 (±0.406) 2.821 (±0.287) 2.684 (±0.305) 2.834 (±0.319)
0.27 3.502 3.449 (±0.378) 2.898 (±0.399) 2.969 (±0.284) 2.865 (±0.328) 2.953 (±0.317)
0.33 3.657 3.587 (±0.370) 3.062 (±0.385) 3.141 (±0.277) 3.058 (±0.319) 3.096 (±0.314)
0.41 3.829 3.730 (±0.370) 3.235 (±0.369) 3.339 (±0.286) 3.265 (±0.309) 3.273 (±0.314)
0.52 3.990 3.865 (±0.355) 3.410 (±0.361) 3.503 (±0.297) 3.475 (±0.313) 3.449 (±0.305)
0.64 4.105 3.964 (±0.341) 3.538 (±0.367) 3.608 (±0.308) 3.639 (±0.297) 3.583 (±0.293)
0.8 4.214 4.040 (±0.327) 3.649 (±0.374) 3.710 (±0.314) 3.796 (±0.279) 3.704 (±0.283)
0.99 4.336 4.123 (±0.306) 3.759 (±0.395) 3.811 (±0.320) 3.942 (±0.263) 3.811 (±0.269)
1.23 4.464 4.223 (±0.291) 3.878 (±0.413) 3.916 (±0.329) 4.098 (±0.247) 3.905 (±0.264)
1.53 4.587 4.341 (±0.287) 4.021 (±0.440) 4.021 (±0.331) 4.260 (±0.219) 4.001 (±0.272)
1.9 4.706 4.463 (±0.291) 4.162 (±0.451) 4.142 (±0.319) 4.414 (±0.195) 4.117 (±0.282)
2.37 4.820 4.575 (±0.304) 4.278 (±0.440) 4.275 (±0.295) 4.583 (±0.189) 4.228 (±0.300)
2.94 4.907 4.659 (±0.324) 4.362 (±0.434) 4.379 (±0.296) 4.734 (±0.202) 4.315 (±0.317)
3.66 4.958 4.708 (±0.335) 4.421 (±0.431) 4.445 (±0.320) 4.823 (±0.223) 4.374 (±0.336)
4.55 5.015 4.753 (±0.335) 4.484 (±0.427) 4.510 (±0.343) 4.892 (±0.223) 4.428 (±0.344)
5.66 5.051 4.777 (±0.342) 4.516 (±0.417) 4.575 (±0.352) 4.942 (±0.218) 4.466 (±0.339)
7.04 5.072 4.779 (±0.349) 4.522 (±0.412) 4.599 (±0.348) 4.978 (±0.222) 4.490 (±0.344)
8.75 5.082 4.767 (±0.340) 4.500 (±0.409) 4.590 (±0.352) 4.987 (±0.242) 4.500 (±0.349)
10.88 5.022 4.703 (±0.329) 4.422 (±0.399) 4.526 (±0.353) 4.924 (±0.256) 4.442 (±0.338)
13.53 4.888 4.582 (±0.311) 4.290 (±0.389) 4.384 (±0.334) 4.782 (±0.253) 4.313 (±0.317)
PGA 4.646 4.367 (±0.300) 4.057 (±0.402) 4.142 (±0.317) 4.482 (±0.223) 4.066 (±0.281)
PGV 3.390 3.254 (±0.272) 2.752 (±0.376) 2.878 (±0.296) 2.958 (±0.202) 2.848 (±0.229)
Bull Earthquake Eng (2014) 12:549–571
Bull Earthquake Eng (2014) 12:549–571 559

the long-period motions were not particularly strong considering the size of the event. This
is a topic we return to in the next section, when comparing the Tohoku motions to those from
other events.

4 Determining event terms (c0 ) for other M > 7 interface earthquakes

We can extend the Tohoku GMPE to other major interface events by adjusting the source
term. Specifically, we assume that all events follow the same attenuation shape as the M9.0
Tohoku event, and that therefore we need only adjust the source term, c0 , to provide GMPEs
for other large events. We selected 4 interface mainshocks with M > 7.0, as listed in Table
1 and estimated their event terms (c0 ). We determine c0 for each event by calculating the
mean residual (where residual = log10 (observed PSA) − log10 (predicted PSA)) for records
from that event with respect to the M9 Tohoku GMPE; the average residual for the event
(c0 ) is the additive adjustment factor (in log units) to be applied to obtain its c0 term. The
mean residuals are calculated using only the forearc stations (which are the nearest stations);
furthermore, for each earthquake we exclude records beyond a cut-off distance (Rcut ) where
PGA starts to flatten on PGA-Rcd plots (i.e. due to instrument noise and/or non-triggered
stations). The adjusted GMPEs for the selected earthquakes are shown in Fig. 7.
It is interesting that for some events the Tohoku attenuation rate matches the observations
well, while for other events (not necessarily the smallest) the Tohoku attenuation rate is
flatter. In such cases (i.e. flatter attenuation), the residuals are positive at Rcd < 100 km.
Therefore, we decided to restrict the distance range for finding the mean residual term, to
Rcd ≤ 200 km for all events, to lessen the residuals at shorter distances, at the expense of
greater over-prediction at Rcd > 200 km. This revised the c0 terms for a couple of the events
upwards. The apparent source terms (c0 ) of M > 7.0 interface earthquakes are calculated by
shifting the Tohoku mainshock source term at each frequency according to the values of mean
residuals, as reported in Table 3. We considered incorporating a magnitude-dependence to
the spectral shape, by making the fictitious depth term (h) a function of magnitude, but did
not find the evidence in the data to be sufficiently strong to warrant this refinement.

5 Scaling of source term (c0 ) and GMPEs with magnitude

Figure 8 shows the adjusted terms (c0 [Tohoku] + c0 ) for the selected interface earthquakes
as a function of magnitude at several frequencies (coefficients are from Table 3), along
with a linear trend line to fit the source terms. The upwards trend of the source-term slope
with magnitude is greatest at lower frequencies. This is as we would expect from basic
seismological scaling concepts (e.g. Abrahamson et al. 2008; Boore and Atkinson 2008;
Boore et al. 2013a,b). Table 4 shows the coefficients of the fit of the source terms versus
magnitude. Note that the slope of the line is of marginal significance at higher frequencies
(i.e. p value of Student’s t test is about the same as the preselected cutoff of 0.05).
The complete GMPE resulting from this study, for large interface events of M7 to M9,
can be obtained using the coefficients in Table 3 to determine the source term (c0 ) for the
magnitude of interest, which is then attenuated using the coefficients in Table 2. The Japan-
to-Cascadia site factor of Table 2 should also be applied if the GMPE is to be used for western
North America rather than Japan. On Fig. 9, we plot the scaling of the resulting GMPE with
magnitude, at a distance of 100 km (chosen because many cities in subduction regions are
located at about this distance from interface events), in comparison to the predictions from

123
560 Bull Earthquake Eng (2014) 12:549–571

2003/09/26 - 04:50
Depth = 17 km, M8.1, Rcut = 200 km, median(Vs30) = 288.6 m/s

PSA (cm/s/s) at freq. = 0.52 Hz 1000 1000

PSA (cm/s/s) at freq. = 0.99 Hz


100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)
1 1
Residual

Residual
0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

1000 1000
PSA (cm/s/s) at freq. = 2.37 Hz

PSA (cm/s/s) at freq. = 5.66 Hz

100 100

10 10
GMPE-Tohoku (forearc) GMPE-Tohoku (forearc)
GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
1 1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)

1 1
Residual

Residual

0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)
2003/09/26 - 06:08
Depth = 53 km, M7.3, Rcut = 200 km, median(Vs30) = 297.6 m/s

1000 1000
PSA (cm/s/s) at freq. = 0.52 Hz

PSA (cm/s/s) at freq. = 0.99 Hz

100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)

1 1
Residual

Residual

0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

Fig. 7 Observed PSA values of the selected M > 7 interface earthquakes at ∼0.5, 1.0, 2.0, and 5.0 Hz,
overlaid by the GMPE curves. The curves plot the Tohoku GMPE for forearc and backarc stations (black
lines), and the event GMPE (magenta lines), as determined using the offset obtained from the mean residuals
(i.e. the required shift in source terms). At each frequency, the corresponding residuals are also plotted. The
residuals are binned into 10 classes and the corresponding means (filled squares) and standard deviations
(bars) are calculated for each class. c0 for each event is based on the values reported in Table 3

123
Bull Earthquake Eng (2014) 12:549–571 561

PSA (cm/s/s) at freq. = 2.37 Hz 1000 1000

PSA (cm/s/s) at freq. = 5.66 Hz


100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)
1 1
Residual

Residual
0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

2004/11/29 - 03:32
Depth = 50 km, M7.0, Rcut = 200 km, median(Vs30) = 274.5 m/s

1000 1000
PSA (cm/s/s) at freq. = 0.52 Hz

PSA (cm/s/s) at freq. = 0.99 Hz

100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)

1 1
Residual

Residual

0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

1000 1000
PSA (cm/s/s) at freq. = 2.37 Hz

PSA (cm/s/s) at freq. = 5.66 Hz

100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)

1 1
Residual

Residual

0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

Fig. 7 continued

123
562 Bull Earthquake Eng (2014) 12:549–571

2005/08/16 - 11:46
Depth = 44 km, M7.1, Rcut = 200 km, median(Vs30) = 364.7 m/s

1000 1000
PSA (cm/s/s) at freq. = 0.52 Hz

PSA (cm/s/s) at freq. = 0.99 Hz


100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)
1 1
Residual

Residual
0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

1000 1000
PSA (cm/s/s) at freq. = 2.37 Hz

PSA (cm/s/s) at freq. = 5.66 Hz

100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)

1 1
Residual

Residual

0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

2011/03/11 - 15:15
Depth = 35 km, M7.8, Rcut = 200 km, median(Vs30) = 304.0 m/s

1000 1000
PSA (cm/s/s) at freq. = 0.52 Hz

PSA (cm/s/s) at freq. = 0.99 Hz

100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)

1 1
Residual

Residual

0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

Fig. 7 continued

123
Bull Earthquake Eng (2014) 12:549–571 563

PSA (cm/s/s) at freq. = 2.37 Hz 1000 1000

PSA (cm/s/s) at freq. = 5.66 Hz


100 100

10 10

1 GMPE-Tohoku (forearc) 1 GMPE-Tohoku (forearc)


GMPE-Tohoku (backarc) GMPE-Tohoku (backarc)
Adjusted GMPE (forearc) Adjusted GMPE (forearc)
Adjusted GMPE (backarc) Adjusted GMPE (backarc)
0.1 0.1
log(PSA_obs) - log(PSA_pred)

log(PSA_obs) - log(PSA_pred)
1 1
Residual

Residual
0 0

-1 -1

30 100 200 300 1000 30 100 200 300 1000


Rcd (km) Rcd (km)

Fig. 7 continued
Near-source amplitude (c0), log units

Near-source amplitude (c0), log units

freq. = 0.2 Hz freq. = 0.5 Hz


3.5
4.0

3.0
3.5

0.258 + 0.350M 1.412 + 0.285M


2.5
GA13 empirical GA13 empirical
3.0

7.0 7.5 8.0 8.5 9.0 7.0 7.5 8.0 8.5 9.0
Moment magnitude (M) Tohoku Moment magnitude (M) Tohoku
Near-source amplitude (c0), log units

Near-source amplitude (c0), log units

4.5 freq. = 1.0 Hz freq. = 5.0 Hz

5.0

4.0

4.5

3.5 1.985 + 0.256M 3.446 + 0.165M


GA13 empirical GA13 empirical
4.0
7.0 7.5 8.0 8.5 9.0 7.0 7.5 8.0 8.5 9.0
Moment magnitude (M) Tohoku Moment magnitude (M) Tohoku

Fig. 8 Source terms (c0 from Tohoku earthquake + adjustment factor = source terms) for the selected inter-
face earthquakes at several frequencies (GA13 empirical; orange symbols). Error bars are the standard devi-
ation of residuals at Rcd ≤ Rcut = 200 km. Trend line shows fit of source terms versus magnitude

other recent GMPEs. Our empirical model follows the slope of the Atkinson and Macias
(2009) GMPEs, which were based on stochastic finite-fault simulations, in terms of the
steepness of magnitude scaling. However we note that the Atkinson and Macias GMPE
predicts much higher amplitudes at long periods (5 s). The results of this study tend to predict
amplitudes that are intermediate to those of Zhao et al. (2006) and Abrahamson et al. (2012),
for large events (M>7.5). Figure 9 also shows, for comparison, the magnitude scaling of
the Boore and Atkinson (2008) GMPE for shallow crustal events; for events of M>8, the

123
564 Bull Earthquake Eng (2014) 12:549–571

Table 4 Coefficients and statistics of fit of source terms versus magnitude: (c0 = a + bM):

Frequency (Hz) Parameter Coefficient Standard error t(Coef = 0)∗ P† ±95 % CL

0.11 Intercept (a) −1.0932 0.796 −1.374 0.242 2.2093


Slope (b) 0.4641 0.103 4.524 0.011 0.2848
0.14 Intercept (a) −0.5012 0.761 −0.658 0.546 2.1138
Slope (b) 0.4119 0.098 4.197 0.014 0.2725
0.17 Intercept (a) −0.1469 0.774 −0.190 0.859 2.1499
Slope (b) 0.3835 0.100 3.842 0.018 0.2771
0.22 Intercept (a) 0.2578 0.754 0.342 0.750 2.0923
Slope (b) 0.3497 0.097 3.600 0.023 0.2697
0.27 Intercept (a) 0.5093 0.709 0.718 0.512 1.9682
Slope (b) 0.3361 0.091 3.678 0.021 0.2537
0.33 Intercept (a) 0.7965 0.704 1.131 0.321 1.9556
Slope (b) 0.3197 0.091 3.521 0.024 0.2521
0.41 Intercept (a) 1.1154 0.676 1.650 0.174 1.8774
Slope (b) 0.3015 0.087 3.459 0.026 0.2420
0.52 Intercept (a) 1.4124 0.623 2.268 0.086 1.7288
Slope (b) 0.2851 0.080 3.552 0.024 0.2229
0.64 Intercept (a) 1.6301 0.581 2.806 0.049 1.6128
Slope (b) 0.2730 0.075 3.645 0.022 0.2079
0.80 Intercept (a) 1.8442 0.564 3.269 0.031 1.5664
Slope (b) 0.2599 0.073 3.573 0.023 0.2019
0.99 Intercept (a) 1.9852 0.592 3.351 0.029 1.6449
Slope (b) 0.2561 0.076 3.353 0.029 0.2120
1.23 Intercept (a) 2.1321 0.671 3.176 0.034 1.8639
Slope (b) 0.2522 0.087 2.914 0.044 0.2403
1.53 Intercept (a) 2.3234 0.753 3.085 0.037 2.0908
Slope (b) 0.2435 0.097 2.509 0.066 0.2695
1.90 Intercept (a) 2.5332 0.796 3.181 0.034 2.2107
Slope (b) 0.2331 0.103 2.271 0.086 0.2850
2.37 Intercept (a) 2.7958 0.881 3.174 0.034 2.4456
Slope (b) 0.2154 0.114 1.897 0.131 0.3153
2.94 Intercept (a) 3.0371 0.979 3.103 0.036 2.7175
Slope (b) 0.1970 0.126 1.561 0.194 0.3503
3.66 Intercept (a) 3.2008 1.021 3.134 0.035 2.8353
Slope (b) 0.1839 0.132 1.397 0.235 0.3655
4.55 Intercept (a) 3.3178 1.039 3.194 0.033 2.8839
Slope (b) 0.1763 0.134 1.317 0.258 0.3718
5.66 Intercept (a) 3.4463 1.058 3.259 0.031 2.9363
Slope (b) 0.1650 0.136 1.210 0.293 0.3785
7.04 Intercept (a) 3.5005 1.088 3.217 0.032 3.0210
Slope (b) 0.1604 0.140 1.144 0.317 0.3894
8.75 Intercept (a) 3.4483 1.119 3.083 0.037 3.1059
Slope (b) 0.1669 0.144 1.157 0.312 0.4004

123
Bull Earthquake Eng (2014) 12:549–571 565

Table 4 continued
Frequency (Hz) Parameter Coefficient Standard error t(Coef = 0)∗ P† ±95 % CL

10.88 Intercept (a) 3.3592 1.128 2.979 0.041 3.1311


Slope (b) 0.1700 0.145 1.170 0.307 0.4036
13.53 Intercept (a) 3.1807 1.105 2.878 0.045 3.0685
Slope (b) 0.1759 0.142 1.234 0.285 0.3956
PGA (99) Intercept (a) 2.8193 1.006 2.803 0.049 2.7929
Slope (b) 0.1908 0.130 1.471 0.215 0.3600
PGV (98) Intercept (a) 0.8540 0.707 1.208 0.294 1.9630
Slope (b) 0.2795 0.091 3.066 0.037 0.2530
* Degrees of freedom for two-tailed ttests = 4
† If P ≤ 0.05, the coefficient is significantly different from 0

GMPE scaling with M


Predictions for B/C site condition @ Rcd = 100 km for forearc stations
GA13 [B/C - forearc] AM09 [B/C - Cascadia] Zhao06 [B/C - Cascadia] Abrahamson 2012 [Global] BA08

100 200

freq. = 0.2 Hz freq. = 0.5 Hz


Predicted PSA (cm/s/s)

Predicted PSA (cm/s/s)

100

20

10
30

20

2 10
7.0 7.5 8.0 8.5 9.0 7.0 7.5 8.0 8.5 9.0
Moment magnitude (M) Tohoku Moment magnitude (M) Tohoku

200

freq. = 1.0 Hz freq. = 5.0 Hz


Predicted PSA (cm/s/s)

Predicted PSA (cm/s/s)

300
100 200

100

40

30

20 30
7.0 7.5 8.0 8.5 9.0 7.0 7.5 8.0 8.5 9.0
Moment magnitude (M) Tohoku Moment magnitude (M) Tohoku

Fig. 9 Comparing interface GMPEs at 100 km. GA13 is our empirical GMPE (black line) for B/C, corrected
with Japan/Cascadia site factors (Table 2) and for forearc stations. Magenta line shows simulation-based
GMPE of Atkinson and Macias (2009). Zhao06 is the predicted motions corrected first from C to B/C and
then for Japan/Cascadia site factors. Abrahamson et al. (2012) is the predicted motions for forearc at B/C
(VS30 = 760m/s), and interface event (yellow line). Purple line shows Boore and Atkinson (2008) GMPE for
shallow crustal events

inferred motions for crustal events at 100 km tend to be lower than those for interface events.
These factors should be considered when evaluating how to weight empirical GMPEs whose
behavior at M9 follows the experience during the Tohoku event.
The tendency of shallow sites to produce strong amplifications at a predominant frequency,
as determined by the layer depth (e.g. Ghofrani et al. 2013), leads to relatively large aleatory

123
566 Bull Earthquake Eng (2014) 12:549–571

M8.12 M7.34 M7.0 M7.1 M7.8 M9.0 average

MYG017 (Vs30 = 122.1 m/s) IWT012 (Vs30 = 202.5 m/s) AOM021 (Vs30 = 249.2 m/s)
1.5 1.5 1.5

1 1 1

0.5 0.5 0.5


residuals

0 0 0

-0.5 -0.5 -0.5

-1 -1 -1

-1.5 -1.5 -1.5


0.1 0.2 1 2 10 0.1 0.2 1 2 10 0.1 0.2 1 2 10
Frequency (Hz) Frequency (Hz) Frequency (Hz)

AOM028 (Vs30 = 368.3 m/s) FKS005 (Vs30 = 426.4 m/s) IWT010 (Vs30 = 756.8 m/s)
1.5 1.5 1.5

1 1 1

0.5 0.5 0.5


residuals

0 0 0

-0.5 -0.5 -0.5

-1 -1 -1

-1.5 -1.5 -1.5


0.1 0.2 1 2 10 0.1 0.2 1 2 10 0.1 0.2 1 2 10
Frequency (Hz) Frequency (Hz) Frequency (Hz)

Fig. 10 Site terms calculated from the average residuals for stations that recorded all events, at Rcd ≤ 200 km.
Solid curves are the residuals for each event and squares are the average residual. Bars are ±1 standard deviation

variability in the GMPEs for these events at higher frequencies, as seen in Table 2. To gain
further insight into this component of the aleatory variability, we calculate the mean residual
for each site, at each frequency (including only those sites that recorded at least 5 of the study
events). This provides a reasonable estimate of the “site term” for the 60 stations that recorded
all the events; note that this site term is primarily capturing the effect of different soil depths
at the sites, because the effect of stiffness has already been included through the use of VS30
as a site parameter in the regression. Figure 10 plots this site term versus frequency for several
representative stations within 200 km of the subduction interface. It is interesting that some
sites (e.g. AOM021) have a nearly-flat relative response versus frequency, suggesting they
follow the typical site response shape for sites in Japan. Other sites show a significant trend
towards negative relative response at low frequencies, with positive relative response at high
frequencies. This is indicative of a large site response at high frequencies, due to shallow
soil resonances. A good example of such a site is IWT012, which is a site with 6 m of soil
overlying a much stiffer material. The large variability in site response for sites in Japan, even
with the same value of VS30 , is a significant reason that the intra-event variability of ground
motions is relatively large in Japan (Table 2). This variability may not be representative for
other regions with more homogeneous conditions, nor is it representative of the expected
variability at a single site. It may be better treated as an epistemic uncertainty in the site
response for a given VS30 , as in principle it is reducible given knowledge of the layer depth.
The subject of aleatory and epistemic uncertainty in the GMPE is considered in more detail
in the next section.

123
Bull Earthquake Eng (2014) 12:549–571 567

6 GMPEs for M7.0 to M9.0 interface events and their uncertainty

GMPEs describe the median and the variability of ground-motion amplitudes. In proba-
bilistic seismic hazard analysis, generally two types of uncertainty are distinguished: epis-
temic uncertainty due to lack of data and knowledge (i.e. uncertainty in what is the correct
median GMPE) and aleatory uncertainty (i.e. random variability/scatter of observations about
a median GMPE) (Toro et al. 1997; McGuire 2004; Al Atik et al. 2010). In GMPEs, the stan-
dard deviation of the residuals (i.e. where the residual is the difference between the logarithm
of an observation and the logarithm of an estimated value), denoted by σ, is treated as an
aleatory variability and integrated across in the hazard analysis. The total aleatory vari-
ability is also separated into the earthquake-to-earthquake (interevent) and the site-to-site
(intraevent) variability (Brillinger and Preisler 1984, 1985; Abrahamson and Youngs 1992;
Joyner and Boore 1993). Sigma has a significant influence on the results of seismic hazard
analyses, particularly for low probabilities (Atkinson and Charlwood 1983; Restrepo-Vélez
and Bommer 2003; Bommer and Abrahamson 2006).
The above GMPE equation and corresponding tables of coefficients provide all the infor-
mation needed to calculate predicted median ground motions (for the geometric mean of two
horizontal components) for large interface events (for either Japan or western North America
site conditions). Table 1 provides intraevent variability as determined in the regression. How-
ever, as noted previously, these values are reflective of highly variable shallow-site conditions
(Fig. 10) and we thus believe they are overestimates of the true aleatory variable for typical
hazard applications. The interevent variability can be assessed from the variability of the
source terms. We calculate the standard deviation of the residuals of the source terms with
respect to the prediction line of Fig. 8 (i.e. the second-stage regression for c0 ). We tabulate
interevent, intraevent and total variability, in Table 5. We consider this to be an upper bound
on actual aleatory variability as it includes some epistemic components as well as the effects
of pronounced site variability.
It is useful to evaluate how much epistemic uncertainty there may be in the proposed
GMPEs, in addition to the aleatory uncertainty. We consider the model of Atkinson and
Adams (2013), who proposed that the epistemic uncertainty in GMPEs for interface events
should grow with distance. This is based on the observation that epistemic uncertainty for
shallow crustal events in active tectonic regions, as represented by alternative GMPEs from the
PEER-NGA project (e.g. see Power et al. 2008 and references therein) grows with distance.
Atkinson and Adams argue that the epistemic uncertainty in interface GMPEs exceeds that
for crustal events, due to the paucity of well-recorded events at the very large magnitudes
of interest. Their proposed uncertainty bounds to represent epistemic uncertainty, applied by
constructing lower and upper GMPE curves (relative to the central GMPE) for large interface
events is given by:
 
Delta = min ((0.15 + 0.0007 Rcd ) , 0.35) log10 units

where Delta is an additive factor, in log units. This formulation prescribes a factor of 2.8 in
amplitude scaling from the lower to the upper GMPE (2 times delta), at 100 km; the factor
grows to 5.2 at 300 km.
As noted in Figs. 4 and 5, the Tohoku motions were weaker than expected for a M9.0 event
at long periods, based on previous GMPEs such as Atkinson and Macias (2009) or Zhao et
al. (2006); thus one may wonder if the Tohoku long-period motions are lower than would be
expected for future M9 events. In the context of the Atkinson and Adams uncertainty model,
this suggests that perhaps the uncertainty band about the Tohoku-based GMPEs should be

123
568 Bull Earthquake Eng (2014) 12:549–571

Table 5 Variability of Tohoku


Frequency Period Intraevent Interevent Combined or
ground-motions
(Hz) (s) (σ) (τ) total variability

(σT = σ2 + τ2 )

0.11 9.09 0.175 0.155 0.233


0.14 7.14 0.201 0.148 0.250
0.17 5.88 0.218 0.151 0.265
0.22 4.55 0.237 0.147 0.279
0.27 3.70 0.252 0.138 0.287
0.33 3.03 0.264 0.137 0.298
0.41 2.44 0.275 0.132 0.305
0.52 1.92 0.285 0.121 0.310
0.64 1.56 0.274 0.113 0.296
0.80 1.25 0.261 0.110 0.283
0.99 1.01 0.249 0.115 0.274
1.23 0.81 0.249 0.131 0.281
1.53 0.65 0.257 0.147 0.296
1.90 0.53 0.276 0.155 0.316
2.37 0.42 0.295 0.171 0.341
2.94 0.34 0.307 0.191 0.361
3.66 0.27 0.312 0.199 0.370
4.55 0.22 0.310 0.202 0.370
5.66 0.18 0.312 0.206 0.374
7.04 0.14 0.324 0.212 0.387
8.75 0.11 0.329 0.218 0.394
10.88 0.09 0.326 0.220 0.393
13.53 0.07 0.313 0.215 0.380
PGA 0.284 0.196 0.345
PGV 0.195 0.138 0.238

shifted upwards for long-period motions, such that both the upper and lower limits are higher
(e.g. an asymmetric band about the central GMPE). This could be achieved by using 1.5*delta
to construct the upper bound for long-period motions ( f ≤ 0.2 Hz), and delta/1.5 to construct
the lower bound. Conversely, one might infer that the Tohoku motions may have been higher
than is typical at high frequencies (based on Fig. 4). Thus the uncertainty band may be
asymmetric in the opposite sense at high frequencies ( f ≥ 5 Hz), with an upper curve being
constructed by the addition of delta/1.5, and a lower curve constructed by the subtraction of
delta*1.5. For intermediate frequencies, a transition between the two behaviors is proposed,
with the uncertainty band being symmetric at 1 Hz. Figure 11 plots the proposed epistemic
uncertainty representation for M9.0 for forearc sites following this rationale, in comparison
to the median GMPE of this study and other selected GMPEs. It is acknowledged that this
epistemic uncertainty is a subjective judgment. It represents our estimate of the epistemic
uncertainty in the proposed GMPE for interface events, considering the available data with
which to constrain the estimates and the implications of other GMPEs. Note that it would
also include much of the epistemic uncertainty in individual site responses relative to the
central GMPE.

123
Bull Earthquake Eng (2014) 12:549–571 569

GMPEs for Interface M9 (B/C site conditions)


AM09 Japan Zhao06 Abrahamson 12 [BCH] GA13med GA13low,high

1000 1000
PSA at freq: 5.0 Hz

PSA at freq: 1.0 Hz


100 100

10 10

T=0.2s T=1.0s

1 1
100 200 300 100 200 300
Closese distance [Rcd] (km) Closese distance [Rcd] (km)

1000 1000

100 100
PSA at freq: 0.5 Hz

PSA at freq: 0.2 Hz

10 10

T=2.0s T=5.0s

1 1
100 200 300 100 200 300
Closese distance [Rcd] (km) Closese distance [Rcd] (km)

Fig. 11 Comparison of proposed alternative interface GMPEs based on Tohoku and other recent large events,
after application of the site correction factors (GA13med, low, high; black lines) with GMPEs of Abrahamson
et al. (2012); Zhao et al. (2006) (also corrected for site factors), and Atkinson and Macias (AM09). Solid black
line is our median GMPE, while dashed black lines show suggested epistemic uncertainty

7 Conclusion

In this study, we provide GMPEs for interface events of M7–9 based on regression analysis
of recorded ground motions from large events in Japan, including the M9 Tohoku event
(Tables 1, 3). The proposed GMPEs using data from Tohoku earthquake are very important
for seismic hazard analysis for subduction zones because of the large number of available
ground motions. In addition to the event-specific GMPE, estimates of aleatory variability
and epistemic uncertainty are also provided (Table 5; Fig. 11). We show how the GMPEs for

123
570 Bull Earthquake Eng (2014) 12:549–571

Japan can be modified to represent the deeper site conditions more typical in western North
America. The inferred surface amplification of Tohoku motions by a factor of two or more
at high-frequencies is a clear indication of shallow low-velocity layer over bedrock, which is
characteristic of the Japanese stations. These results support the need to apply regional site
correction factors before comparing GMPEs across regions, even for the reference B/C site
condition.

Acknowledgments Ground-motion data and site information were obtained from the K-NET at http://www.
kyoshin.bosai.go.jp/, the NIED strong-motion seismograph networks. The authors would like to thank the two
anonymous reviewers for their helpful comments.

References

Abrahamson NA, Youngs RR (1992) A stable algorithm for regression analyses using the random effects
model. Bull Seism Soc Am 82:505–510
Abrahamson N, Atkinson GM, Boore DM, Bozorgnia Y, Campbell K, Chiou B, Idriss M, Silva W, Youngs R
(2008) Comparisons of the NGA ground-motion relations. Earthq Spectr 24:45–66
Abrahamson N A, Gregor N, Addo K (2012) BCHydro ground motion prediction equations for subduction
earthquakes. Earthq Spectr (submitted)
Al Atik L, Abrahamson NA, Cotton F, Scherbaum F, Bommer J, Kuehn N (2010) The variability of ground-
motion prediction models and its components. Seismol Res Lett 81:794–801
Atkinson GM, Adams J (2013) Ground motion prediction equations for application to the 2015 national
seismic hazard maps of Canada. Can J Civil Eng (submitted)
Atkinson GM, Boore DM (2003) Empirical ground-motion relations for subduction zone earthquakes and
their application to Cascadia and other regions. Bull Seism Soc Am 93:1703–1729
Atkinson GM, Casey R (2003) A comparison of ground motions from the 2001 M6.8 in-slab earthquakes in
Cascadia and Japan. Bull Seism Soc Am 93:1823–1831
Atkinson GM, Charlwood RG (1983) Uncertainties in probabilistic seismic hazard assessment as a function
of probability level: a case history for Vancouver, British Columbia. Bull Seism Soc Am 73:1225–1241
Atkinson GM, Macias M (2009) Predicted ground motions for great interface earthquakes in the Cascadia
subduction zone. Bull Seism Soc Am 99:1552–1578
Bommer JJ, Abrahamson NA (2006) Why do modern probabilistic seismic-hazard analyses often lead to
increased hazard estimates? Bull Seism Soc Am 96:1967–1977
Boore DM, Atkinson GM (2008) Ground-motion prediction equations for the average horizontal component
of PGA, PGV, and 5 %-damped PSA at spectral periods between 0.01 s and 10.0 s. Earthq Spectr 24:99–138
Boore DM, Stewart JP, Seyhan E (2013) NGA-West 2 equations for predicting response spectral accelerations
for shallow crustal earthquakes, PEER 2013/05. University of California, Berkeley, California, Pacific
Earthquake Engineering Research Center, 135 pp
Boore DM, Skarlatoudis A, Margaris B, Papazachos C, Ventouzi C (2009) Along-arc and backarc attenuation,
site response, and source spectrum for the intermediate-depth 8 January 2006 M6.7 Kythera, Greece,
earthquake. Bull Seism Soc Am 99:2410–2434
Boore DM, Thompson EM, Cadet H (2011) Regional correlations of VS30 and velocities averaged over depths
less than and greater than 30 m. Bull Seism Soc Am 101:3046–3059
Boore DM, Stewart JP, Seyhan E, Atkinson GM (2013) NGA-West 2 Equations for predicting response
spectral accelerations for shallow crustal earthquakes. Draft report for the NGA-West2 Project, Contract
No. 00007327
Borcherdt RD (1992) Simplified site classes and empirical amplification factors for site-dependent code
provisions. In: NCEER, SEAOC, BSSC workshop on site response during earthquakes and seismic code
provisions proceedings. University Southern California, Los Angeles
Brillinger DR, Preisler HK (1984) An exploratory analysis of the Joyner–Boore attenuation data. Bull Seism
Soc Am 74:1441–1450
Brillinger DR, Preisler HK (1985) Further analysis of the Joyner–Boore attenuation data. Bull Seism Soc Am
75:611–614
Converse AM, Brady AG (1992) BAP: basic strong-motion accelerogram processing software, version 1.0,
US Geology Survey Open-File Report 92–296A, 174 pp
Crouse CB, Yogesh KVYAS, Schell BA (1988) Ground motions from subduction-zone earthquakes. Bull
Seism Soc Am 78:1–25

123
Bull Earthquake Eng (2014) 12:549–571 571

Crouse CB (1991) Ground-motion attenuation equations for earthquakes on the Cascadia subduction zone.
Earthq Spectr 7:210–236
Dhakal YP, Takai N, Sasatani T (2008) Path effects on prediction equations of pseudo-response spectra in
northern Japan. In: The 14th World conference on earthquake engineering. Beijing, China
Eurocode 8 (2004) Design of structures for earthquake resistance, part 1: general rules, seismic actions and
rules for buildings, EN 1998–1, European Committee for Standardization (CEN), www.cen.eu/cenorm/
homepage.htm (last accessed April 2013)
Ghofrani H, Atkinson GM (2011) Forearc versus backarc attenuation of earthquake ground motion. Bull Seism
Soc Am 101:3032–3045
Ghofrani H, Atkinson GM, Goda K (2013) Implications of the 2011 M9.0 Tohoku Japan earthquake for the
treatment of site effects in large earthquakes. Bull Earthquake Eng 11:171–203
Geospatial Information Authority (GSI) of Japan (2011) The 2011 off the Pacific coast of Tohoku earthquake:
crustal deformation and fault model. http://www.gsi.go.jp/cais/topic110422-index-e.html (last accessed
April 2013)
Joyner WB, Boore DM (1993) Methods for regression analysis of strong-motion data. Bull Seism Soc Am
83:469–487
Kanno T, Narita A, Morikawa N, Fujiwara H, Fukushima Y (2006) A new attenuation relation for strong
ground motion in Japan based on recorded data. Bull Seism Soc Am 96:879–897
Macias M, Atkinson GM, Motazedian D (2008) Ground-motion attenuation, source, and site effects for the
26 September 2003 M8.1 Tokachi-Oki earthquake sequence. Bull Seism Soc Am 98:1947–1963
McGuire R (2004) Seismic hazard and risk analysis, vol. 6, EERI Monograph MNO-10. Earthq. Eng. Res.
Inst, Oakland
NBCC (2005) National building code of Canada. National Research Council of Canada, Ottawa, NRCC 47,666
NEHRP (2000) NEHRP recommended provisions for seismic regulations for new buildings and other struc-
tures, Part 1, Provisions, FEMA 368. Federal Emergency Management Agency, Washington
Power M, Chiou B, Abrahamson NA, Bozorgnia Y, Shantz T, Roblee C (2008) An overview of the NGA
Project. Earthq Spectr 24:3–21
Restrepo-Vélez LF, Bommer JJ (2003) An exploration of the nature of the scatter in ground motion prediction
equations and the implications for seismic hazard assessment. J Earthq Eng 7:171–199
Strasser FO, Arango MC, Bommer JJ (2010) Scaling of source dimensions for interface and intraslab
subduction-zone earthquakes with moment magnitude. Seismol Res Lett 81:941–950
Toro G, Abrahamson NA, Schneider J (1997) Model of strong ground motions from earthquake in central and
eastern North America: best estimates and uncertainties. Seismol Res Lett 68:41–57
Yagi Y (2004) Source rupture process of the 2003 Tokachi-oki earthquake determined by joint inversion of
teleseismic body wave and strong ground motion data. Earth Planets Space 56:311–316
Young RR, Day SM, Stevens JP (1988) Near field motions on rock for large subduction zone earthquakes, in
earthquake engineering and soil dynamics II-recent advances in ground motion evaluation. ASCE Geotech
Special Publ 20:445–462
Youngs R, Chiou S, Silva W, Humphrey J (1997) Strong ground motion attenuation relationships for subduction
zone earthquakes. Seismol Res Lett 68:58–73
Zhao JX, Zhang J, Asano A, Ohno Y, Oouchi T, Takahashi T, Ogawa H, Irikura K, Thio HK, Somerville
PG, Fukushima Y, Fukushima Y (2006) Attenuation relations of strong ground motion in Japan using site
classification based on predominant period. Bull Seism Soc Am 96:898–913
Zhao JX (2010) Geometric spreading functions and modeling of volcanic zones for strong-motion attenuation
models derived from records in Japan. Bull Seism Soc Am 100:712–732

123

Вам также может понравиться