Bulletin of the Seismological Society of America, Vol. 93, No. 6, pp.

2531–2545, December 2003

Source Characterization for Broadband Ground-Motion Simulation:

Kinematic Heterogeneous Source Model and
Strong Motion Generation Area
by Hiroe Miyake, Tomotaka Iwata, and Kojiro Irikura

Abstract We estimate strong motion generation areas that reproduce near-source

ground motions in a broadband frequency range (0.2–10 Hz) using the empirical
Green’s function technique. The strong motion generation areas are defined as ex-
tended areas with relatively large slip velocities within a total rupture area. Four
M 6 class (the 1997 Kagoshima on March and May, 1997 Yamaguchi, and 1998
Iwate) and several moderate-size earthquakes in Japan were analyzed. We examine
the relationship between the strong motion generation area and the area of asperities,
which is characterized based on heterogeneous slip distributions estimated from low-
frequency (
1 Hz) waveform inversions (Somerville et al., 1999). We performed
waveform fitting in acceleration, velocity, and displacement, then obtained the strong
motion generation area occupying about a quarter of the total rupture area. The size
and position of the strong motion generation area coincide with those of characterized
asperities. We find self-similar scaling of seismic moment to both the size of the
strong motion generation area and the rise time for a magnitude range of analyzed
earthquakes. Based on our results, we propose a characterized source model for the
broadband ground-motion simulation, which consists of strong motion generation
areas with large slip velocities and a background slip area with a small slip velocity.
Waveform modeling of the characterized source model suggests that the strong mo-
tion generation areas play important roles in simulating broadband ground motions
and have the potential to be an about 10-MPa stress-release area on the fault.

Introduction
Estimation of spatiotemporal slip distributions for large [1996], Ide et al. [1996], Sekiguchi et al. [1996, 2000], Wald
earthquakes has provided essential information for under- [1996], and Yoshida et al. [1996]; 1999 Kocaeli, Turkey, by
standing source physics and rupture mechanisms. Waveform Bouchon et al. [2002], Delouis et al. [2002], and Sekiguchi
inversions of near-source ground-motion records are ex- and Iwata [2002]; and 1999 Chi-Chi, Taiwan, by Chi et al.
pected to have higher resolution than teleseismic waveform [2001], Ma et al. [2001], Wu et al. [2001], and Zeng and
inversions, because the former can treat shorter wavelengths. Chen [2001]). The resolution of waveform inversions is af-
Kinematic waveform inversions based on strong motion re- fected by the limit of the usable frequency range when cal-
cords have been carried out since the 1979 Imperial Valley culating realistic theoretical Green’s functions. This limita-
earthquake (e.g., Hartzell and Heaton, 1983). These wave- tion is determined by the accuracy of the subsurface
form inversions require good spatial coverage of the source structure, which can be validated by waveform modeling of
area with observed records and detailed knowledge of the small earthquakes. At present, waveform inversions have
subsurface structures to accurately calculate the theoretical been performed successfully up to about 1 Hz for near-
Green’s function. Meanwhile, slip distributions of many source ground-motion velocities. We call such source mod-
large crustal earthquakes have been estimated, following the els, estimated from ground-motion data with frequencies less
deployment of dense strong motion networks and increased than 1 Hz, low-frequency source models.
computational power (e.g., the 1989 Loma Prieta by Beroza When studying high-frequency (e.g., 2 Hz) radiation
[1991] and Wald et al. [1991]; 1992 Landers by Cohee and processes on the fault plane, the problem arises because of
Beroza [1994], Wald and Heaon [1994], and Cotton and the difficulties of calculating the Green’s functions deter-
Compillo [1995]; 1994 Northridge by Wald et al. [1996]; ministically by means of a lack of underground structure
1995 Hyogo-ken Nanbu [Kobe], Japan, by Horikawa et al. information. To overcome this, Zeng et al. (1993) and Kak-

2531
2532 H. Miyake, T. Iwata, and K. Irikura

ehi and Irikura (1996) proposed an envelope inversion by In this article, we estimate an area we call the strong
fitting the envelopes of observed strong motion records to motion generation area from waveform fitting using the em-
synthetic ones calculated by ray theory in displacement and pirical Green’s function method in the frequency range of
the empirical Green’s function method in acceleration, re- 0.2–10 Hz. The area is defined as extended areas with rela-
spectively. Nakahara et al. (1998) estimated absolute values tively large slip velocities within a total rupture area. We
of energy intensity by fitting the synthetic acceleration en- estimate the size and location of the strong motion genera-
velopes to observed ones by means of scattering theory. We tion area and rise time there without any information of low-
call these source models high-frequency source models. frequency waveform inversions, then compare the strong
In general, slip heterogeneities have been found for motion generation area and asperities identified from wave-
most large earthquakes from waveform inversions based on form inversions. We also examine the scaling of the size of
low-frequency (
1 Hz) ground-motion data. Areas of high- the strong motion generation area and rise time with the
frequency (2 Hz) radiation obtained by envelope inver- seismic moment. For the broadband ground-motion simu-
sions largely correspond to regions surrounding the large slip lation, we propose a characterized source model and calcu-
area, branching portions of the fault plane, the area of rupture late the stress drop for the strong motion generation area,
termination, or the area close to the surface. Both the low- taking into consideration the difference between the asperity
and high-frequency source models can be used to interpret and crack models (e.g., Das and Kostrov, 1986; Boatwright,
the frequency-dependent mechanisms of strong ground mo- 1988).
tion generation.
Green’s functions for low-frequency ground motions Data
are calculated by a deterministic approach based on coherent
summation of rays or wavenumbers. For high-frequency We investigate the source models of 12 crustal earth-
ground motions, Green’s functions are calculated by a sta- quakes (Mw 4.8–6.0) that occurred in Japan from 1996 to
tistical approach based on random summation. It is therefore 1999 (Table 1; Fig. 1) using near-source ground-motion re-
difficult to reproduce appropriate Green’s functions that sat- cords. Since the 1995 Hyogo-ken Nanbu (Kobe) earthquake,
isfy both the amplitude and phase information in the fre- dense strong motion networks such as the K-NET (Kinoshita,
quency range of 0.5–2.0 Hz, which is very important for 1998) have been installed across Japan, and they have been
earthquake engineering applications (e.g., Kawase, 1998). consistently recording intermediate-size to large earthquakes
Note also that the source processes generating 0.5- to on identical instruments and providing good spatial coverage
2.0-Hz ground motions are not adequately resolved from around the source area. We therefore estimated source mod-
either the low- or high-frequency source models. Our objec- els of mainshocks by using records of small events as em-
tive is to fill the gaps of the source processes shown in the pirical Green’s functions, where the events had similar focal
frequency range of 0.5–2.0 Hz. mechanisms and hypocenter distances to their respective
Source modeling by the empirical Green’s function mainshocks.
method (e.g., Irikura, 1986; Irikura and Kamae, 1994) is Figure 2 shows the epicenters of the mainshocks and
capable of characterizing near-source strong ground motions their aftershocks, along with the K-NET stations. We use data
in a broadband frequency range up to 10 Hz and including at stations located within 50 km of the epicenters. The hy-
1 Hz. This technique requires observed records from appro- pocenter locations are obtained from the Faculty of Science,
priate small events as empirical Green’s functions within a Kyushu University (1997), Faculty of Science, Kagoshima
broadband frequency range, and the source model is as- University (1997), and Japan Meteorological Agency
sumed to have a homogeneous slip and rise time with radial (JMA). The focal depths of most earthquakes were less than
rupture propagation from the rupture starting point. High- 20 km. Fault-plane solutions derived by moment tensor in-
quality simulations of strong ground motions over a wide versions (Harvard centroid moment tensor [CMT] catalog by
frequency range have supported the approval of using an Dziewonski et al. [1996, 1997]; Kuge et al. [1997, 2003];
extended area for strong motion generation with this choice National Research Institute for Earth Science and Disaster
of simple parameters (e.g., Kamae and Irikura, 1998a; Mi- Prevention (NIED) seismic moment tensor catalog by Fu-
yake et al., 1999, 2001). However, the assumption of ho- kuyama et al. [2000a,b, 2001]), and aftershock distributions
mogeneous slip contrasts with the fact that most waveform (Fukuoka District Meteorological Observatory, JMA, 1998;
inversions give heterogeneous slip distributions. Further- Miyamachi et al., 1999; Graduate School of Science, To-
more the areas for strong motion generation are generally hoku University, 1999; Umino et al., 1998) were used for
smaller than the total rupture areas from waveform inver- determining the orientation of the mainshock fault plane.
sions. Therefore it is important to clarify the relationship The waveform fitting for earthquake number 12m was ex-
between the source parameters estimated by strong ground amined for both conjugate fault planes of the moment tensor
motion simulations based on the empirical Green’s function solution, because the number of aftershock was quite small
method and heterogeneous slip distributions obtained by and the aftershock distribution did not clearly show the fault
waveform or envelope inversions. plane.
Source Characterization for Broadband Ground-Motion Simulation 2533

Table 1
List of Earthquakes Studied
No. Date (JST) Latitude (deg) Longitude (deg) Depth (km) MJMA Mw (M0 [N m]) Name of Earthquake

1m 1996/08/11 03:12 38.905N 140.637E 8.63 5.9 5.9 (9.51E Ⳮ 17) ‡

1996 Akita-ken Nairiku Nanbu
1a 1996/08/11 05:39 38.884N 140.651E 9.36 4.2
2m 1996/08/11 08:10 38.863N 140.676E 9.77 5.7 5.7 (3.84E Ⳮ 17)‡ 1996 Miyagi-ken Hokubu
2a 1996/08/11 15:01 38.849N 140.689E 9.84 4.4
3m 1997/03/26 17:31 31.970N* 130.380E* 8.2* 6.5 6.0 (1.38E Ⳮ 18)§ 1997 Kagoshima-ken Hokuseibu (March)
3a 1997/03/26 17:39 31.968N* 130.362E* 11.1* 4.7
4m 1997/05/13 14:38 31.952N† 130.343E† 7.7† 6.3 5.9 (8.89E Ⳮ 17)§ 1997 Kagoshima-ken Hokuseibu (May)
4a 1997/05/25 06:10 31.932N† 130.348E† (7.7) 4.2
5m 1997/06/25 18:50 34.438N 131.669E 8.29 6.3 5.8 (5.66E Ⳮ 17)| 1997 Yamaguchi-ken Hokubu
5a 1997/06/25 18:58 34.459N 131.696E 13.31 4.0
6m 1997/09/04 05:15 35.259N 133.379E 8.91 5.1 5.2 (6.83E Ⳮ 16)| 1997 Tottori-ken Seibu
6a 1997/09/02 02:07 35.250N 133.381E 13.04 4.1
7m 1998/02/21 09:55 37.288N 138.771E 20.83 5.0 5.0 (3.15E Ⳮ 16)| 1998 Niigata-ken Chubu
7a 1998/02/22 02:49 37.296N 138.775E 22.05 3.7
8m 1998/04/22 20:32 35.165N 136.570E 10.47 5.4 5.2 (6.74E Ⳮ 16)| 1998 Shiga-Gifu Kenkyo
8a 1998/04/22 20:28 35.162N 136.570E 11.21 4.0
9m 1998/09/03 16:58 39.796N 140.910E 9.63 6.1 5.9 (7.53E Ⳮ 17)| 1998 Iwate-ken Nairiku Hokubu
9a 1998/09/03 17:10 39.745N 140.900E 9.86 3.9
10m 1998/09/15 16:24 38.278N 140.766E 13.25 5.0 5.0 (3.19E Ⳮ 16)| 1998 Miyagi-ken Nanbu
10a 1998/09/15 16:18 38.275N 140.766E 12.99 3.6
11m 1999/02/26 14:18 39.150N 139.850E 19.00 5.1 5.2 (7.19E Ⳮ 16)| 1999 Akita-ken Engan Nanbu
11a 1999/03/08 17:46 39.150N 139.850E 21.00 4.2
12m 1999/03/16 16:43 35.270N 135.935E 12.00 4.9 4.8 (1.83E Ⳮ 16)| 1999 Shiga-ken Hokubu
12a 1999/03/22 05:05 35.257N 135.930E 12.40 3.3

*Hypocenter information from Faculty of Science, Kyushu University (1997).

†
Hypocenter information from Faculty of Science, Kagoshima University (1997).
‡
Seismic moment determined by Dziewonski et al. (1996, 1997) as Harvard CMT catalog.
§
Seismic moment determined by Kuge et al. (1997).
|
Seismic moment determined by Fukuyama et al. (2000a,b, 2001) as NIED seismic moment tensor catalog.
Hypocenter information is provided by the Japan Meteorological Agency unless indicated. The characters “m” and “a” in the first column indicate
mainshock and aftershock, respectively. Instead of aftershocks, foreshocks are used as empirical Green’s functions for events 6a, 8a, and 10a.

Strong Ground Motion Simulation based on the small events with corrections for the difference in the slip
Empirical Green’s Function Method velocity time function between the large and small events
following the scaling laws mentioned earlier. This method
Formulation of the Empirical Green’s Function does not require knowledge of the explicit shape of the slip
Method velocity time function for the small event.
As stated previously, we define the strong motion gen- The numerical equations to sum records of small events
eration area as the area characterized by a large uniform slip are
velocity within the total rupture area, which reproduces near-
source strong ground motions up to 10 Hz. The strong mo- N N
r
tion generation area of each mainshock is estimated by U(t) ⳱ 兺 兺 F(t)* (C • u(t))
i⳱1j⳱1 rij
(1)
waveform fitting based on the empirical Green’s function
method. The low-frequency limit is constrained by the sig-
nal-to-noise level ratio of the small event record used as an F(t) ⳱ d(t ⳮ tij)
empirical Green’s function. (Nⳮ1)n⬘
(k ⳮ 1)T
冤d冦t ⳮ t 冧冥
1
The technique by which waveforms for large events are
synthesized follows the empirical Green’s function method
Ⳮ
n⬘ 兺
k⳱1
ij ⳮ
(N ⳮ 1)n⬘
(2)
proposed by Hartzell (1978). Revisions have been made by
Kanamori (1979), Irikura (1983, 1986), and others. We use
rij ⳮ ro nij
the empirical Green’s function method formulated by Irikura tij ⳱ Ⳮ , (3)
(1986), based on a scaling law of fault parameters for large Vs Vr
and small events (Kanamori and Anderson, 1975) and the
omega-squared source spectra (Aki, 1967). The waveform where U(t) is the simulated waveform for the large event,
for a large event is synthesized by summing the records of u(t) is the observed waveform for the small event, N and C
2534 H. Miyake, T. Iwata, and K. Irikura

50˚N Figure 1. Epicenter locations (solid stars)

and focal mechanisms used. Focal mechanisms
were determined by moment tensor inversions
of broadband seismic waveforms. Black dots
No.5m M W 5.8
indicate the stations of K-NET.

No.9m M W 5.9
No.11m M W 5.2
40˚N
No.12m M W 4.8

No.1m M W 5.9

No.3m M W 6.0
No.2m M W 5.7
No.7m M W 5.0
30˚N

No.4m M W 5.9
No.6m M W 5.2

No.10m M W 5.0

No.8m M W 5.2

km
20˚N 0 500
120˚E 150˚E
130˚E 140˚E

are the ratios of the fault dimensions and stress drops be- The shape of equation (4) is shown in Figure 3c. In Irikura
tween the large and small events, respectively, and the as- (1986), the scaling parameters needed for this technique, N
terisk indicates convolution. F(t) is the filtering function (integer value) and C, can be derived from the constant lev-
(correction function) to adjust the difference in the slip ve- els of the displacement and acceleration amplitude spectra
locity time functions between the large and small events. Vs of the large and small events with the formulas
and Vr are the S-wave velocity near the source area and the
rupture velocity on the fault plane, respectively. T is the rise U0 M
time for the large event, defined as the duration of the fil- ⳱ 0 ⳱ CN 3 (5)
u0 m0
tering function F(t) (in Fig. 3b,c). It corresponds the duration
of the slip velocity time function on the subfault from the
A0
beginning to the time before the tail starts. The variable n⬘ ⳱ CN. (6)
is an appropriate integer to weaken the artificial periodicity a0
of n and to adjust the interval of the tick to be the sampling
rate. The other parameters are given in Figure 3a. Here, U0 and u0 indicate the constant levels of amplitude of
Regarding the filtering function F(t), Irikura et al. the displacement spectra for the large and small events, re-
(1997) proposed a modification to equation (2) in order to spectively. M0 and m0 correspond to the seismic moments
prevent sag at multiples of 1/T (Hz) from appearing in the for the large and small events. A0 and a0 indicate the constant
amplitude spectra. The discretized equation for the modified levels of the amplitude of the acceleration spectra for the
F(t) is large and small events (Fig. 3d,e).
N and C are derived from equations (5) and (6):
(Nⳮ1)n⬘
1
F(t) ⳱ d(t ⳮ tij) Ⳮ 兺 1/2 1/2

冢
n⬘ 1 ⳮ
1
e 冣 k⳱1

(4)
N⳱ 冢Uu 冣 冢Aa 冣
0

0
0

0
(7)

冤 1
(kⳮ1)
e (Nⳮ1)n⬘
冦
d t ⳮ tij ⳮ
(k ⳮ 1)T
(N ⳮ 1)n⬘ 冥
冧 . C⳱ 冢 冣 冢 冣
u0
U0
1/2
A0
a0
3/2
. (8)
Source Characterization for Broadband Ground-Motion Simulation 2535

(a)

subfault (i, j)
r rij

w r0
1~N
l ξ ij W
small event

1~N large event

L
(b) (c)
F(t) F(t)
1 1

1 (N-1)n' 1
(N-1)n'
n' n'(1-1/e)

t t
T T

log disp. amp. spc. (d) (e)

log Acc. amp. spc.

U0
A0
CN3
u0 fcm a0 CN
fcm
fca CN fca
log freq. log freq.

Figure 3. Schematic illustrations of the empirical

Green’s function method. (a) Fault areas of large and
small events are defined to be L ⳯ W and l ⳯ w,
respectively, where L/l ⳱ W/w ⳱ N. (b) Filtering
function F(t) (after Irikura, 1986) to adjust to the dif-
ference in slip velocity function between the large and
small events. This function is expressed as the sum
of a delta and a boxcar function. (c) Modified filtering
function (after Irikura et al., 1997) with an exponen-
tially decaying function instead of a boxcar function.
T is the rise time for the large event. (d) Schematic
displacement amplitude spectra following the omega-
squared source scaling model, assuming a stress drop
ratio C between the large and small events. (e) Ac-
celeration amplitude spectra following the omega-
Figure 2. Map showing epicenters of the main- squared source scaling model.
shocks (solid stars) and their foreshocks or after-
shocks (open stars) and K-NET stations (triangles) for
the earthquakes (numbers 2m, 3m, 12m). Broken
lines show the borders between prefectures.
parameters by fitting the observed source spectral ratio be-
tween the large and small events to the theoretical source
spectral ratio, which obeys the omega-squared source model
Parameter Estimation for the Empirical of Brune (1970, 1971).
Green’s Function Method by the Source Spectral The observed waveform O(t) can be expressed as a con-
volution of the source effect S(t), propagation path effect
Fitting Method
P(t), and site amplification effect G(t) assuming linear sys-
For an objective estimation of parameters N and C, tems:
which are required for the empirical Green’s function
method of Irikura (1986), Miyake et al. (1999) proposed a
source spectral fitting method. This method derives these O(t) ⳱ S(t)*P(t)*G(t). (9)
2536 H. Miyake, T. Iwata, and K. Irikura

In the frequency domain, the observed spectrum O(f ) is ex- 103 103
pressed as a product of these effects: All stations Average
102 102
O( f ) ⳱ S( f )•P( f )•G( f ).

Ratio
(10)
101 101
From equation (10), the observed source amplitude spectral
ratio of the large and small events for one station is
100 100
10-1 100 101 102 10-1 100 101 102

冫
1 ⳮpfR/Qs( f )Vs Frequency (Hz) Frequency (Hz)
O( f ) e
S( f ) O( f )/P( f ) R KGS002 Average
⳱ ⳱ , (11) KGS004 SSRF

冫1r e
s( f ) o( f )/p( f ) ⳮpfr/Qs( f )Vs KGS005
o( f ) KGS007

Figure 4. Source spectral ratios of the mainshock

where capital and lowercase letters indicate factors for large to aftershock records used as the empirical Green’s
and small events, respectively. The propagation path effect function for the 1997 Kagoshima-ken Hokuseibu
earthquake, March (number 3m). Left: Spectral ratios
is given by geometrical spreading for the body waves and for all the sites. Right: Comparison of average values
by a frequency-dependent attenuation factor Qs(f ) for the S of the observed ratios and best-fit source spectral ratio
waves. function (SSRF, broken line). Closed circles represent
The observed source amplitude spectral ratio for each the average, bars the standard deviation of the ob-
station is first calculated using the mainshock and its after- served ratios.
shock. Then the observed source amplitude spectral ratio is
divided into m windows in the frequency domain, in which
the central frequency is fi (i ⳱ 1 to m), and the log averages M0
⳱ CN3 ( f → 0) (15)
S(f i)/s(f i) and the log standard deviations S.D.(f i) for all m0
stations were obtained (Fig. 4).
The equation for the amplitude spectra, based on the 2
omega-squared source model by Brune (1970, 1971), is
given as
冢 冣冢 冣
M0 fcm
m0 fca
⳱ CN ( f → ). (16)

M0 The high-frequency limit (f → ) corresponds to the fre-

S( f ) ⳱ , (12) quency where the acceleration source spectrum is constant
1 Ⳮ ( f/fc)2 at frequencies of more than f ca and less than f max (f max is the
high-frequency cutoff of the constant level of acceleration
where f c is the corner frequency. Using equation (12), the source spectrum). According to equations (15) and (16), N
source spectral ratio function (SSRF) is and C are

M0 1 Ⳮ ( f/fca)2 fca
SSRF( f ) ⳱ • . (13) N⳱ (17)
m0 1 Ⳮ ( f/fcm)2 fcm

M0 /m0 indicates the seismic moment ratio between a large 3

and small event at the lowest frequency, and f cm and f ca
respectively are corner frequencies for the mainshock and
C⳱ 冢Mm 冣冢ff 冣 .
0
0 cm

ca
(18)

aftershock. We searched the minimum value of the weighted

Amplitude spectra for 30 sec of the waveforms, including
least squares (equation 14), applying to the observed source
the whole S waves, were calculated by the source spectral
amplitude spectral ratio, equation (11), to fit the SSRF (equa-
fitting method. The vector summation of the three compo-
tion 13; Fig. 4) in the frequency domain:
nents for the amplitude spectra was used for the analyses. A
m moving average of Ⳳ10% window for each frequency point
SSRF ( fi) ⳮ S( fi)/s( fi) 2

兺冢
i⳱1 S.D.( fi) 冣 ⳱ min. (14) was applied as a smoothing of amplitude spectra. Parameters
for the empirical Green’s function method are shown in Ta-
ble 2, in which the S-wave velocities based on the crustal
This method provides estimates of M0 /m0, f cm, and f ca. The structure models around the source areas are taken from
relationships between these parameters, N, and C are given Hurukawa (1981) and Kakuta et al. (1991). The frequency-
as dependent Qs(f )’s for the S waves were taken from Ogue et
Source Characterization for Broadband Ground-Motion Simulation 2537

Table 2
Source Parameters M0 /m0, f cm, and f ca Obtained by the Source Spectral Fitting Method (Miyake et al., 1999)
f cm f ca Vs Range of Analyses
No. M0 /m0 (Hz) (Hz) N C (km) (Hz) K-NET Stations Name of Earthquake

1m/1a 143.9 0.28 1.31 5 1.41 3.5 0.25–10.0 AKT019, YMT006 1996 Akita-ken Nairiku Nanbu
MYG004, IWT010
2m/2a 10.57 0.63 1.21 2 1.49 3.5 0.20–10.0 AKT019, YMT006 1996 Miyagi-ken Hokubu
MYG004, IWT011
3m/3a 96.2 0.24 1.08 5 1.06 3.1 0.20–10.0 KGS002, KGS004 1997 Kagoshima-ken Hokuseibu (March)
KGS005, KGS007
4m/4a 123.4 0.39 1.43 4 2.50 3.1 0.20–10.0 KGS002, KGS004 1997 Kagoshima-ken Hokuseibu (May)
KGS005, KGS007
5m/5a 416.9 0.20 0.87 4 5.06 3.1 0.30–10.0 YMG001, YMG003 1997 Yamaguchi-ken Hokubu
YMG009, SMN014
6m/6a 16.32 0.72 1.72 2 1.10 3.1 0.40–10.0 TTR007, TTR009 1997 Tottori-ken Seibu
SMN003, SMN015
7m/7a 61.05 0.80 3.60 5 0.67 3.5 0.50–10.0 NIG017, NIG018 1998 Niigata-ken Chubu
NIG020, NIG021
8m/8a 72.69 0.66 2.06 3 2.39 3.5 0.30–10.0 GIF022, SIG008 1998 Shiga-Gifu Kenkyo
AIC003, MIE002
9m/9a 159.9 0.23 1.03 4 1.77 3.5 0.20–10.0 IWT018, IWT021 1998 Iwate-ken Nairiku Nanbu
IWT025, AKT011
10m/10a 74.28 0.72 2.30 3 2.28 3.5 0.20–10.0 MYG012, MYG013 1998 Miyagi-ken Nanbu
MYG015, YMT010
11m/11a 30.21 0.89 2.90 3 0.87 3.5 0.20–10.0 AKT019, AKT020 1999 Akita-ken Engan Nanbu
YMT001, YMT002
12m/12a 238.3 1.12 4.29 4 4.24 3.4 1.00–10.0 SIG004, SIG006 1999 Shiga-ken Hokubu
KYT007, KYT010

N and C are the respective fault size ratios and stress drop ratios of the mainshocks to aftershocks for the empirical Green’s function method of Irikura
(1986). Vs is the S-wave velocity near the source area.

al. (1997) and Satoh et al. (1998). The variations of the cords from several K-NET stations surrounding the source
spectra at the low frequency are mostly cased by the rupture area. We confirmed that more than four near-source stations
directivity effects, as shown in the spectral simulation in are required for source modeling, because there is an azi-
Miyake et al. (2001). They pointed out the importance of muthal effect due to rupture directivity (Miyake et al., 2001).
having good azimuthal coverage of stations for the deter- The frequency range available for simulation was generally
mination of source parameters from the spectra, because rup- from 0.2 to 10 Hz, depending on the noise levels of the
ture directivity effects may distort the spectral shapes ex- aftershock records. The S-wave arrival times of the velocity
pected from the omega-square source model and therefore waveforms for the target and element earthquakes first were
give different corner frequencies and high-frequency decays. set, then five parameters representing the size (length and
The log average values of the observed corner frequencies width) and position (starting point of the rupture in the strike
among several station data were used to weaken the rupture and dip directions) for the strong motion generation area,
directivity effect. and rise time, were estimated to minimize the residuals of
the displacement waveform and acceleration envelope fit-
Near-Source Ground-Motion Simulation to Estimate ting. The fitting was done by the genetic algorithm method
Strong Motion Generation Area (e.g., Holland, 1975) for most of the events and by forward
The modeling of the strong motion generation area us- modeling for the 1997 Kagoshima-ken Hokuseibu earth-
ing the empirical Green’s function method is one of the use- quakes (Table 1, numbers 3m and 4m; Miyake et al. [1999]).
ful means to kinematically simulate both low- and high- The genetic algorithm method for the parameter determi-
frequency wave generations from the source using the nation was operated 10 times for different initial parameters
appropriate filtering function F(t). The filtering function F(t) for each source model in order to check the accuracy and
provides the rupture growth with Kostrov-like slip velocity stability of the solution. We searched all possibilities for the
time functions. rupture starting point, and then we did grid searching of
Using the source parameters obtained in Table 3, we length and width of the subfault by 0.1 km and of rise time
perform ground-motion simulations of the target earth- corresponding to the element earthquake by 0.01 sec. Source
quakes using the empirical Green’s function method in order modeling is more sensitive for the variation of the rupture
to estimate the strong motion generation area. These simu- starting points rather than the size of the strong motion gen-
lations utilized acceleration, velocity, and displacement re- eration area and rise time (e.g., Miyake et al., 1999). This
2538 H. Miyake, T. Iwata, and K. Irikura

Table 3
Parameters for the Strong Motion Generation Area Determined by the Empirical Green’s Function Method
SMGA Length Width Strike Dip Rise Time
No. (km2) (km) (km) (deg.) (deg.) (sec) Starting Point Name of Earthquake

1m 25.5 8.5 3.0 N358E 47 0.40 (3,5) 1996 Akita-ken Nairiku Nanbu
2m 8.64 3.6 2.4 N225E 84 0.28 (1,2) 1996 Miyagi-ken Hokubu
3m 42.0 7.0 6.0 N278E 89 0.50 (1,5) 1997 Kagoshima-ken Hokuseibu (March)
4m 24.0 3.0 and 3.0 4.0 and 4.0 N280E and N010E 84 and 90 0.50 and 0.50 (1,3) and (1,3) 1997 Kagoshima-ken Hokuseibu (May)
5m 14.4 3.6 4.0 N229E 76 0.28 (2,4) 1997 Yamaguchi-ken Hokubu
6m 5.76 2.4 2.4 N325E 89 0.20 (1,2) 1997 Tottori-ken Seibu
7m 4.00 2.0 2.0 N213E 41 0.10 (1,4) 1998 Niigata-ken Chubu
8m 4.50 3.0 1.5 N024E 67 0.15 (2,3) 1998 Shiga-Gifu Kenkyo
9m 16.0 4.0 4.0 N217E 44 0.32 (1,4) 1998 Iwate-ken Nairiku Nanbu
10m 4.41 2.1 2.1 N203E 37 0.15 (2,3) 1998 Miyagi-ken Nanbu
11m 4.86 2.7 1.8 N181E 66 0.18 (1,3) 1999 Akita-ken Engan Nanbu
12m 1.44 1.2 1.2 N017E or N172E 66 or 26 0.12 (1,3) 1999 Shiga-ken Hokubu

SMGA is the size of the strong motion generation area. Rupture starting points are indicated as (starting fault number of N for the strike direction,
starting fault number of N for the dip direction). Earthquake number 12m has similar residuals for both conjugate fault planes as determined by Fukuyama
et al. (2000a).

indicates that the rupture directivity effects shown in the Scaling

waveforms well constrain the rupture starting point in the
source modeling. We assumed that rupture propagated ra- Relationship between Strong Motion Generation Area
dially from the fixed hypocenter at a speed of 90% of the S- and Seismic Moment
wave velocity near the source area, except the 1997 Kago- For low-frequency source models, the heterogeneous
shima-ken Hokuseibu earthquakes. slip distributions of crustal earthquakes have been analyzed
Following this approach, we estimate the strong motion to study the characteristics of the slip heterogeneity (e.g.,
generation areas for the target earthquakes (Mw 4.8–6.0) Somerville et al., 1999; Mai and Beroza, 2000, 2002). Som-
(Fig. 1). Synthetic waveforms for the best source models fit erville et al. (1999) studied the slip characterizations by an-
the observed data well in acceleration and velocity, while alyzing the heterogeneous slip distributions derived from
some displacement synthetics seem to be of smaller ampli- waveform inversions of strong motion and teleseismic re-
tude than the observations (Fig. 5). The estimated sizes of cords in the low-frequency range. They found that the size
the strong motion generation areas, rupture starting points of the asperities (areas of large slip) follows self-similar scal-
in the area, and the rise time are summarized in Table 3. ing with respect to seismic moment.
Most of the source models were composed of a single strong We compare the relationship between the size of the
motion generation area, except for the 13 May 1997 Kago- strong motion generation area (0.2–10 Hz) and seismic mo-
shima-ken Hokuseibu earthquake (Table 1, number 4m). We ment with the scaling relationship characterizing slip distri-
found that this event consisted of two strong motion gener- butions of crustal earthquakes (
1 Hz) proposed by Som-
ation areas, which propagated along both conjugate fault erville et al. (1999) in Figure 6. The strong motion
planes. Horikawa (2001) also obtained two asperities, one generation area is proportional to M02/3, indicating self-
on each fault plane by kinematic waveform inversion. Rup- similar scaling for the range of seismic moments between
ture propagation directions of the strong motion generation 1.83 ⳯ 1016 N m (Mw 4.8) and 1.38 ⳯ 1018 N m (Mw 6.0).
area in the frequency range of 0.2–10 Hz agreed with those As the absolute seismic moment values cannot be estimated
of the asperities obtained from low-frequency (
1 Hz) from this procedure with the empirical Green’s function it-
waveform inversion studies, reported by Okada et al. (1997) self, the seismic moments for the mainshocks are adopted as
(numbers 1m and 2m), Ide (1999) (number 5m), Nakahara values determined by moment tensor inversions in Table 1.
et al. (2002) (number 9m), Miyakoshi et al. (2000) (numbers This scaling of the strong motion generation area is close to
3m, 5m, and 9m), and Horikawa (2001) (numbers 3m and that of the combined area of asperities reported by Somer-
4m). For the 1999 Shiga-ken Hokubu earthquake (number ville et al. (1999). The effective source area generating
12m), estimates of the size of the strong motion generation strong ground motions occupies almost a quarter of the total
area and rise time were stable; however, we could not de- rupture area obtained by low-frequency waveform inver-
termine the fault plane from the strong ground motion simu- sions. There is no clear dependence of this scaling depending
lation. There was no clear difference in the residuals and on rupture propagation direction or focal mechanism. For
shapes of simulated waveforms for the two conjugate fault comparison, we added the combined size of the three strong
planes by moment tensor inversion. motion generation areas for the 1994 Northridge earthquake
Source Characterization for Broadband Ground-Motion Simulation 2539

Figure 5. Comparison of observed and syn-

thetic waveforms of acceleration, velocity, and
displacement for the (a) 1996 Miyagi-ken Ho-
kubu earthquake (number 2m), (b) 1997 Ka-
goshima-ken Hokuseibu earthquake, March
(number 3m), and (c) 1999 Shiga-ken Hokubu
earthquake (number 12m), at the four nearest
K-NET stations. Numbers show the maximum
amplitude values of the observed waveforms.
2540 H. Miyake, T. Iwata, and K. Irikura

Mw where we define rise time as the duration of the filtering

5 6 7 function F(t). The rise time shows self-similar scaling for
104 earthquake moments less than 1.38 ⳯ 1018 N m (Mw 6.0).
This scaling is also close to the empirical relationship be-
strong motion generation area (~10Hz)
strong motion generation area (km2)

tween rise time and seismic moment derived by Somerville

combined area of asperities (<1Hz) et al. (1999). For events larger than Mw 6.5, only the rise
103 time for the largest strong motion generation area of the 1994
Northridge earthquake (Kamae and Irikura, 1998b) follows
the empirical relation, whereas that for all strong motion
generation areas of the 1995 Hyogo-ken Nanbu earthquake
102 (Kamae and Irikura, 1998a) shows half of the value expected
by the empirical relation, regardless of the distance from the
hypocenter to the strong motion generation area. At this mo-
ment we cannot conclude whether the significantly shorter
101 rise times are required as a characteristic feature of the strong
motion generation area for larger events.

100 Discussion
1023 1024 1025 1026 1027
Source Characterization for Broadband Ground-
M0 (dyne-cm = 10-7 Nm) Motion Simulation
Figure 6. Scaling of the strong motion generation Based on this slip characterization from waveform in-
area to seismic moment. Solid and open circles, re- versions by Somerville et al. (1999), Miyakoshi et al. (2000)
spectively, show the strong motion generation areas proposed a characterized source model composed of asper-
studied and the combined areas of asperities reported
by Somerville et al. (1999). Thick and thin solid lines ities and a background slip area surrounding the asperities,
respectively correspond to the combined area of as- both regions having a constant slip value. This slip-charac-
perities and total rupture areas, as a function of the terized source model is expressed quantitatively both by
seismic moment (Somerville et al., 1999). Small solid outer fault parameters, such as seismic moment or total rup-
circles are the areas for the 1994 Nothridge earth- ture area, and by inner fault parameters, such as asperity size
quake (Kamae and Irikura, 1998b) and for the 1995
Hyogo-ken Nanbu earthquake (Kamae and Irikura, inside the source. So far this source model has been applied
1998a). The moment magnitude Mw is calculated ac- only to simulate low-frequency ground motions, because the
cording to Kanamori (1977). slip distributions are obtained from waveform inversion with
frequencies less than 1 Hz.
For the broadband ground-motion simulation, we pro-
(Mw 6.6) (Kamae and Irikura, 1998b) and the 1995 Hyogo-
pose a characterized source model composed of strong mo-
ken Nanbu earthquake (Mw 6.8) (Kamae and Irikura, 1998a).
tion generation areas with large slip velocities and a back-
They estimated the strong motion generation areas that best
ground slip area with a small slip velocity. To show the
fit the velocity and acceleration waveforms by forward mod-
importance of slip velocities in our source model, we sim-
eling. The sizes of their strong motion generation areas are
ulate ground motions for the 26 March 1997 Kagoshima-
also close to that of the combined area of asperities in Som-
ken Hokuseibu earthquake (number 3m) in the frequency
erville et al. (1999).
range of 0.2–10 Hz based on the empirical Green’s function
For several earthquakes, the strong motion generation
method, and we compare differences in synthetic waveforms
areas are located at almost the same positions as the asperity
from the strong motion generation area, background slip
areas, based on the criteria of Somerville et al. (1999). These
area, and both the strong motion generation area and back-
asperity areas are determined from the heterogeneous slip
ground slip area.
distribution results of waveform inversions in the frequency
As stated previously, Miyakoshi et al. (2000) inverted
range of 0.1–0.5 Hz (Miyakoshi et al., 2000) (Fig. 7). Our
waveforms to obtain the slip distribution for the 26 March
analyses indicate that the large slip areas (asperities from
1997 Kagoshima-ken Hokuseibu earthquake (number 3m)
low-frequency waveform inversion) and large slip velocity
and characterized the slip distribution based on the criteria
areas (strong motion generation areas in this study) occupy
of Somerville et al. (1999). The estimated ratio between the
almost the same regions.
size of the asperity and total rupture area is 0.27, and that of
the average slip on the asperity to total rupture area is 1.76.
Relationship between Rise Time and Seismic Moment
This means that 47.5% (0.27 ⳯ 1.76) of the total seismic
Figure 8 shows the relationship between the rise time moment is generated by the asperity and the other 52.5% by
for the strong motion generation area and seismic moment, the background slip area. The derived ratio between the size
Source Characterization for Broadband Ground-Motion Simulation 2541

No.5m No.9m
No.3m

dip
dip
dip
20.0 11.0 9.0 3.0 0.0 0.0 3.0 43.0
27.0 12.0 19.0 11.0 16.0 2.0

5.0 25.0 21.0 1.0 40.0 0.0 0.0 0.0

65.0 44.0 27.0 39.0 12.0 24.0

12.0 34.0 45.0 1.0 71.0 103.0 17.0 0.0

46.0 56.0 82.0 49.0 45.0 21.0

13.0 58.0 34.0 10.0 32.0 63.0 28.0 0.0

39.0 27.0 38.0 6.0 39.0 17.0

36.0 22.0 19.0 20.0 0.0 0.0 2 km 8.0 0.0 29.0 13.0 0.0 0.0 24.0 22.0

0.0 0.0 9.0 89.0 2km

strike 0.0 6.0 15.0 13.0
2km
0.0 13.0 13.0 1.0
2km strike
2km
strike
2km

Figure 7. Comparison of the strong motion generation areas and asperity areas,
which are characterized from the heterogeneous slip distributions obtained by wave-
form inversions of Miyakoshi et al. (2000). From left to right, the 1997 Kagoshima-
ken Hokuseibu earthquake, March (number 3m), 1997 Yamaguchi-ken Hokubu earth-
quake (number 5m), and 1998 Iwate-ken Nariku Hokubu earthquake (number 9m).
Thick solid lines on the total rupture areas show the boundaries of the characterized
asperities according to Somerville et al. (1999). Broken lines show the boundaries of
the strong motion generation areas. Numbers on each subfault are slip, in centimeters,
estimated by waveform inversion.

Mw
5 6 7
101
risetime (~10Hz)
risetime (<1Hz)
risetime (sec)

100

10-1
Figure 9. Schematic shape of the synthetic slip
velocity time functions on the (a) strong motion gen-
eration area and (b) background slip area. The ele-
ment slip velocity time function is assumed to follow
10-2 the function proposed by Nakamura and Miyatake
1023 1024 1025 1026 1027 (2000). The synthetic slip velocity time functions are
M0 (dyne-cm = 10-7 Nm) expressed as the convolution of the element slip ve-
locity time functions and the filtering functions F(t),
Figure 8. Scaling of the rise time to seismic mo- given in Figure 3c.
ment. Solid and open circles, respectively, show the
rise times found in this study and which were reported
by Somerville et al. (1999). Solid line shows the scal- empirical Green’s function method for both the strong mo-
ing between rise time and seismic moment (Somer- tion generation area and background slip area. Aftershock
ville et al., 1999). Small solid circles are the rise times
(number 3a) records were adopted as the empirical Green’s
for the 1994 Nothridge earthquake (Kamae and Iri-
kura, 1998b) and for the 1995 Hyogo-ken Nanbu function. For the strong motion generation area, we used the
earthquake (Kamae and Irikura, 1998a). The moment same F(t) (NSMGA ⳱ 5, CSMGA ⳱ 1.06, TSMGA ⳱ 0.5 sec
magnitude Mw is calculated according to Kanamori in Fig. 9a) assuming that our strong motion generation area
(1977). coincides with the asperity derived by Miyakoshi et al.
(2000). For the background slip area, we set
of the background slip area and total rupture area is 0.73 (⳱
1 ⳮ 0.27); therefore the ratio of the average slip on the NSMGA 5
Ntotal ⳱ ⳱  10, (19)
background slip area to total rupture area is 0.72 (⳱ 0.525/ 冪SMGA/S 冪0.27
0.73).
We calculate the filtering function, F(t), required in the where S is the size of the total rupture area and SMGA that
2542 H. Miyake, T. Iwata, and K. Irikura

of the strong motion generation area. According to Miyak- source model, both in acceleration and velocity (Fig. 10). In
oshi et al. (2000), the ratio of seismic moment between the contrast, the contribution from the background slip area is
background slip area and strong motion generation area is small both in acceleration and velocity, although it is slightly
more significant in displacement. This suggests that near-
M0BG 0.525 source strong ground motions are mainly controlled by the
⳱ . (20)
M0SMGA 0.475 size of the strong motion generation area and its rise time.
Our characterized source model is constructed from the
This ratio is also expresses as viewpoint of slip velocity, and the target frequency is 0.2–
10 Hz. Miyakoshi et al. (2000) characterized the heteroge-
M0BG CBG (Ntotal)2 ⳮ (NSMGA)2 neous source model from the viewpoint of slip itself, and
⳱ •
M0SMGA CSMGA (NSMGA)2 the target frequency is 0.1–0.5 Hz. We showed that the char-
Ntotal 0.20 102 ⳮ 52 10 acterized source model of slip at low frequency is equivalent
• ⳱ • • . (21)
NSMGA 1.06 52 5 to the characterized source model of slip velocity in the
broadband frequency.
Because Ntotal is equal to 10 (⳱ twice NSMGA), the duration
of F(t) for the background slip area becomes 1.0 sec (⳱ Physical Interpretation of the Strong Motion
twice TSMGA). The maximum amplitude of F(t) for the back- Generation Area
ground slip area is calculated as about 20% (CBG ⳱ 0.20)
of that for the strong motion generation area. The F(t) for We showed that the strong motion generation area co-
the background slip area is summarized as Ntotal ⳱ 10, incides with the asperity area of heterogeneous slip distri-
CBG ⳱ 0.20, and Ttotal ⳱ 1.0 sec (Fig. 9b). We calculate the bution derived from low-frequency (
1 Hz) ground-motion
slip velocity time function in each subfault of the target data. We now compare the strong motion generation area
earthquake by convolution between F(t) and the slip velocity with the single-crack and single-asperity model proposed by
time function for a small event. The shape of the slip velocity Das and Kostrov (1986).
time function for a small event is schematically calculated Das and Kostrov (1986) discussed the difference be-
by following the formulation of Nakamura and Miyatake tween the source amplitude spectrum for the single-asperity
(2000). The amplitude of the slip velocity time function for model (Fig. 11a) and that for the single-crack model (Fig.
a subfault is larger for the strong motion generation area than 11b), when the size of the asperity and crack are assumed
for the background slip area, but the duration is shorter to be the same. They showed that the levels of the source
(Fig. 10). amplitude spectra at high frequencies were almost equal for
The ground motions from the strong motion generation both models, although at low frequencies they differed by
area are almost identical to those from the characterized (R/r)2 times due to the slip existence on the stress-free field.

Acc.(cm/s**2) Vel.(cm/s) Disp.(cm) Schematic slip

distribution
511 13.6 3.17
obs.
L
386 11.8 3.34 W
syn.1
SMGA
∆u

396 12.2 3.28

syn.2
characterized
source model
55 3.3 0.73
syn.3
background
slip area
0 3 6 9 12 15 0 3 6 9 12 15 0 3 6 9 12 15
s s s

Figure 10. Comparison of observed and synthetic waveforms (0.2–10 Hz) of the
east–west component at station KGS002 for the 1997 Kagoshima-ken Hokuseibu earth-
quake, March (number 3m). Top to bottom: Traces show observed waveforms and
synthetic waveforms from the strong motion generation area, both of the strong motion
generation area and background slip area in our characterized source model, and back-
ground slip area. Left to right: Acceleration, velocity, and displacement waveforms.
Numbers above the waveforms are maximum amplitude values.
Source Characterization for Broadband Ground-Motion Simulation 2543

(a) (b) (c) (d)

R/r

disp. amplitude spectra

acc. amplitude spectra

stress-free field

stress field stress field

ty

ck
er i

cra
r R

as p

gle
gle

s in
s in
barrier barrier

frequency (Hz) frequency (Hz)

Figure 11. Geometry of (a) single-asperity and (b) single-crack models (after Das
and Kostrov 1986). Gray, white, and black zones indicate stress field, stress-free field,
and barrier, respectively. (c) Far-field displacement amplitude spectra for the single-
asperity (black line) and single-crack (gray line) models at a given magnitude. (d) Far-
field acceleration amplitude spectra for the single-asperity (black line) and single-crack
(gray line) models at a given magnitude.

We already showed that the ground motions from the strong Table 4
motion generation area are almost the same as those from Total Rupture Area, Strong Motion Generation Area, and
the characterized source model both in acceleration and ve- Estimated Stress Drop in the Strong Motion Generation Area
locity, while they differ in displacement (Fig. 11), indicating No. S (km2) SMGA (km2) DrSMGA (MPa) Name of Earthquake
that the source models expressed as strong motion genera-
3m 120 42.0 7.29 1997 Kagoshima-ken
tion areas and the characterized source models correspond Hokuseibu (March)
to the crack and asperity models, respectively. The strong 5m 112 14.4 7.35 1997 Yamaguchi-ken
motion generation areas can be regarded as the asperity in Hokubu
the stress-free field. 9m 100 16.0 11.46 1998 Iwate-ken
Nairiku Nanbu
Stress drop for the strong motion generation area is cal-
culated based on the formulation of Madariaga (1979) and The total rupture areas are those Miyakoshi et al. (2000) estimated by
Boatwright (1988), assuming the single-asperity model. The waveform inversion.
background slip area is assumed to have no stress drop. The
equation for the stress drop in the strong motion generation
area, DrSMGA, is Conclusions

7 M0 We show that the strong motion generation areas ob-

DrSMGA ⳱ • , (22) tained in this study (0.2–10 Hz) coincide with the areas of
16 Rr2 the asperities of heterogeneous slip distributions derived
from low-frequency (
1 Hz) waveform inversions. The self-
where M0 is the seismic moment for the target earthquake, similar scalings for the size of the strong motion generation
R the equivalent radius for the total rupture area S (S ⳱ area and rise time, as a function of seismic moment, are
pR2), and r the equivalent radius for the strong motion gen- found for the range of earthquakes less than Mw 6.0. These
eration area SMGA (SMGA ⳱ pr2). Note that the displace- scalings are compatible with those for larger earthquakes
ment amplitude spectrum of the single asperity is (R/r) times obtained by Somerville et al. (1999).
of that of the single-crack models (Fig. 11c), while Das and Our analyses suggest that source characterization for
Kostrov (1986) pointed out (R/r)2 times difference. This dif- simulating the broadband frequency ground motion requires
ference occurs because Madariaga (1979) and Boatwright strong motion generation areas that produce high-frequency
(1988) used the average slip of the crack in the reciprocal motions with large slip velocities as well as a background
theorem to derive equation (22) assuming the asperity lo- slip area that produces low-frequency motions with a small
cated at an arbitrary position of the stress-free field, while slip velocity. The near-source strong ground motions are
Das and Kostrov (1986) used the maximum slip of the crack controlled mainly by the size of the strong motion generation
assuming a small asperity at the center of the stress-free field. area and rise time there. The characterized source model
The stress drops in the strong motion generation areas presented in this study looks different from the high-fre-
for crustal earthquakes were calculated using equation (21), quency source models from envelope inversions; however,
where we used the total rupture areas S (in Fig. 7) estimated we address the fact that strong motion generation areas with
by Miyakoshi et al. (2000). The values of the stress drop are 10 MPa stress drop can also reproduce a series of high-
estimated to be about 10 MPa (Table 4). frequency waveforms.
2544 H. Miyake, T. Iwata, and K. Irikura

The strong motion generation area contains both low- Faculty of Science, Kagoshima University (1997). The earthquakes with M
and high-frequency information that is scaled to the wave 6.3 (March 26, 1997) and with M 6.2 (May 13, 1997) occurring in
northwestern Kagoshima prefecture, Rep. Coord. Comm. Earthquake
generations and rupture dynamics of small earthquakes. We Pred. 58, 630–637 (in Japanese).
verified that quantified strong motion generation areas and Faculty of Science, Kyushu University (1997). Seismic activity in Kyushu
the idea of the characterized source modeling have enough (November 1996–April 1997), Rep. Coord. Comm. Earthquake Pred.
ability to perform the broadband ground-motion simulation 58, 605–618 (in Japanese).
in acceleration, velocity, and displacement. Fukuoka District Meteorological Observatory, JMA (1998). On an M 6.3
earthquake in the northern Yamaguchi prefecture on June 25, 1997,
Rep. Coord. Comm. Earthquake Pred. 59, 507–510 (in Japanese).
Fukuyama, E., M. Ishida, S. Horiuchi, H. Inoue, S. Hori, S. Sekiguchi, H.
Acknowledgments Kawai, H. Murakami, S. Yamamoto, K. Nonomura, and A. Goto
We are greatly indebted to K-NET for the strong ground motion data. (2000a). NIED seismic moment tensor catalogue January–December,
We would like to thank Keiko Kuge, Harvard CMT catalog, and NIED 1999, Technical Note of the National Research Institute for Earth
seismic moment tensor catalog for the focal mechanism information and Science and Disaster Prevention 199, 1–56.
the Faculty of Science, Kyushu University and JMA, who provided the Fukuyama, E., M. Ishida, S. Horiuchi, H. Inoue, S. Hori, S. Sekiguchi, A.
hypocenter information. We are also grateful to Ken Miyakoshi for giving Kubo, H. Kawai, H. Murakami, and K. Nonomura (2000b). NIED
us the results of the waveform inversion, Jorge Aguirre for his helpful seismic moment tensor catalogue January–December, 1997, Techni-
support in the use of the genetic algorithm, and Jim Mori for improvement cal Note of the National Research Institute for Earth Science and
of this manuscript. We are very grateful to Mariagiovanna Guatteri and Disaster Prevention 205, 1–35.
Martin Mai for reviewing the manuscript and providing valuable comments. Fukuyama, E., M. Ishida, S. Horiuchi, H. Inoue, A. Kubo, H. Kawai, H.
Some figures were drawn by GMT Ver.3.0 (Wessel and Smith, 1995). H.M. Murakami, and K. Nonomura (2001). NIED seismic moment tensor
was supported by Research Fellowships for Young Scientists from the Ja- catalogue January–December, 1998 (revised), Technical Note of the
pan Society for the Promotion of Science. This work was supported in part National Research Institute for Earth Science and Disaster Prevention
by a Grant-in-Aid for Scientific Research (Number 11209201) and the Spe- 218, 1–51.
cial Coordination Funds titled “Study on the master model for strong Graduate School of Science, Tohoku University (1999). On the seismic
ground motion prediction toward earthquake disaster mitigation” from the activity of the M 6.1 earthquake of 3 September 1998 in Shizukuishi,
Ministry of Education, Science, Sports, and Culture of Japan. Iwate prefecture, Rep. Coord. Comm. Earthq. Pred. 60, 49–53 (in
Japanese).
Hartzell, S. H. (1978). Earthquake aftershocks as Green’s functions, Geo-
phys. Res. Lett. 5, 1–4.
References Harzell, S. H., and T. H. Heaton (1983). Inversion of strong ground motion
Aki, K. (1967). Scaling law of seismic spectrum, J. Geophys. Res. 72, and teleseismic waveform data for the fault rupture history of the 1979
1217–1231. Imperial Valley, California, earthquake, Bull. Seism. Soc. Am. 73,
Beroza, G. C. (1991). Near-source modeling of the Loma Prieta earthquake: 1553–1583.
evidence for heterogeneous slip and implications for earthquake haz- Holland, J. H. (1975). Adaptation in Natural and Artificial Systems, Uni-
ard, Bull. Seism. Soc. Am. 81, 1603–1621. versity of Michigan Press, Ann Arbor.
Boatwright, J. (1988). The seismic radiation from composite models of Horikawa, H. (2001). Earthquake doublet in Kagoshima, Japan: rupture of
faulting, Bull. Seism. Soc. Am. 78, 489–508. asperities in a stress shadow, Bull. Seism. Soc. Am. 91, 112–127.
Bouchon, M., M. N. Toksöz, H. Karabulut, and M. -P. Bouin (2002). Space Horikawa, H., K. Hirahara, Y. Umeda, M. Hashimoto, and F. Kusano
and time evolution of rupture and faulting during the 1999 İzmit (Tur- (1996). Simultaneous inversion of geodetic and strong motion data
key) earthquake, Bull. Seism. Soc. Am. 92, 256–266. for the source process of the Hyogo-ken Nanbu, Japan, earthquake,
Brune, J. N. (1970). Tectonic stress and the spectra of seismic shear waves J. Phys. Earth 44, 455–471.
from earthquakes, J. Geophys. Res. 75, 4997–5009. Hurukawa, N. (1981). Normal faulting microearthquakes occurring near
Brune, J. N. (1971). Correction, J. Geophys. Res. 76, 5002. the moho discontinuity in the northeastern Kinki district, Japan, J.
Chi, W. C., D. Dreger, and A. Kaverina (2001). Finite-source modeling of Phys. Earth 29, 519–535.
the 1999 Taiwan (Chi-Chi) earthquake derived from a dense strong- Ide, S., M. Takeo, and Y. Yoshida (1996). Source process of the 1995 Kobe
motion network, Bull. Seism. Soc. Am. 91, 1144–1157. earthquake: determination of spatiotemporal slip distribution by
Cohee, B. P., and G. C. Beroza (1994). Slip distribution of the 1992 Landers Bayesian modeling, Bull. Seism. Soc. Am. 86, 547–566.
earthquake and its implications for earthquake source mechanics, Ide, S. (1999). Source process of the 1997 Yamaguchi, Japan, earthquake
Bull. Seism. Soc. Am. 84, 692–172. analyzed in different frequency bands, Geophys. Res. Lett. 26, 1973–
Cotton, F., and M. Campillo (1995). Frequency domain inversion of strong 1976.
motions: application to the 1992 Landers earthquake, J. Geophys. Res. Irikura, K. (1983). Semi-empirical estimation of strong ground motions
100, 3961–3975. during large earthquakes, Bull. Disast. Prev. Res. Inst. Kyoto Univ.
Das, S., and B. V. Kostrov (1986). Fracture of a single asperity on a finite 33, 63–104.
fault: a model for weak earthquakes? in Earthquake Source Mechan- Irikura, K. (1986). Prediction of strong acceleration motions using empir-
ics, S. Das, J. Boatwright, and C. H. Scholz (Editors), American Geo- ical Green’s function, in Proc. 7th Japan Earthquake Engineering
physical Monograph 37, 91–96. Symp., 151–156, Tokyo, 10–12 December 1986.
Delouis, B., D. Giardini, P. Lundgren, and J. Salichon (2002). Joint inver- Irikura, K., and K. Kamae (1994). Estimation of strong ground motion in
sion of InSAR, GPS, teleseismic, and strong-motion data for the spa- broad-frequency band based on a seismic source scaling model and
tial and temporal distribution of earthquake slip: application to the an empirical Green’s function technique, Ann. Geofis. 37, 1721–1743.
1999 İzmit mainshock, Bull. Seism. Soc. Am. 92, 278–299. Irikura, K., T. Kagawa, and H. Sekiguchi (1997). Revision of the empirical
Dziewonski, A. M., G. Ekstrom, and M. P. Salganik (1996). Centroid- Green’s function method by Irikura (1986) (program and abstracts),
moment tensor solutions for January–March 1995, Phys. Earth Seism. Soc. Japan 2, B25 (in Japanese).
Planet. Interiors 93, 147–157. Kakehi, Y., and K. Irikura (1996). Estimation of high-frequency wave ra-
Dziewonski, A. M., G. Ekstrom, N. N. Maternovskaya, and M. P. Salganik diation areas on the fault plane by the envelope inversion of accel-
(1997). Centroid-moment tensor solutions for July–September 1996, eration seismograms, Geophys. J. Int. 125, 892–900.
Phys. Earth Planet. Interiors 102, 133–143. Kakuta, T., H. Miyamachi, and A. Takagi (1991). Intermediate earthquakes
Source Characterization for Broadband Ground-Motion Simulation 2545

in a northern part of the Kyushu–Ryukyu arc, Zisin 44, 63–74 (in Ogue, Y., Y. Wada, A. Narita, and S. Kinoshita (1997). Deconvolution for
Japanese with English abstract). the K-NET data from earthquake in southern Kyushu (program and
Kamae, K., and K. Irikura (1998a). Source model of the 1995 Hyogo-ken abstracts), Seism. Soc. Japan 2, B23 (in Japanese).
Nanbu earthquake and simulation of near-source ground motion, Bull. Okada, T., N: Umino, A. Hasegawa, and N. Nishide (1997). Source pro-
Seism. Soc. Am. 88, 400–412. cesses of the earthquakes in the border of Akita and Miyagi prefec-
Kamae, K., and K. Irikura (1998b). A source model of the 1994 Northridge tures in August, 1996 (2) (program and abstracts), Seism. Soc. Japan
earthquake (Mw 6.7), in Proc. 10th Japan Earthquake Engineering 2, B69 (in Japanese).
Symp., 643–648, Yokohama, Japan, 25–27 November 1998 (in Jap- Satoh, T., H. Kawase, and S. Matsushima (1998). Source spectra, attenu-
anese with English abstract). ation function, and site amplification factors estimated from the
Kanamori, H. (1977). The energy release in great earthquakes, J. Geophys. K-NET records for the earthquakes in the border of Akita and Miyagi
Res. 82, 2981–2987. prefectures in August, 1996, Zisin 50, 415–429 (in Japanese with
Kanamori, H. (1979). A semi-empirical approach to prediction of long- English abstract).
period ground motions from great earthquakes, Bull. Seism. Soc. Am. Sekiguchi, H., and T. Iwata (2002). Rupture process of the 1999 Kocaeli,
69, 1645–1670. Turkey, earthquake estimated from strong-motion waveforms, Bull.
Kanamori, H., and D. L. Anderson (1975). Theoretical basis of some em- Seism. Soc. Am. 92, 300–311.
pirical relations in seismology, Bull. Seism. Soc. Am. 65, 1073–1095. Sekiguchi, H., K. Irikura, and T. Iwata (2000). Fault geometry at the rupture
Kawase, H. (1998). Metamorphosis of near-field strong motions by under- termination of the 1995 Hyogo-ken Nanbu Earthquake, Bull. Seism.
ground structures and their destructiveness to man-made structures, Soc. Am. 90, 117–133.
learned from the damage belt formation during the Hyogo-ken Nanbu Sekiguchi, H., K. Irikura, T. Iwata, Y. Kakehi, and M. Hoshiba (1996).
earthquake of 1995, in Proc. 10th Japan Earthquake Engineering Minute locating of faulting beneath Kobe and the waveform inversion
Symp., 29–34, Yokohama, Japan, 25–27 November 1998 (in Japanese of the source process during the 1995 Hyogo-ken Nanbu, Japan, earth-
with English abstract). quake using strong ground motion records, J. Phys. Earth 44, 473–
Kinoshita, S. (1998). Kyoshin Net (K-NET), Seism. Res. Lett. 69, 309–332. 487.
Kuge, K. (2003). Source modeling using strong-motion waveforms: toward Somerville, P., K. Irikura, R. Graves, S. Sawada, D. Wald, N. Abrahamson,
automated determination of earthquake fault planes and moment- Y. Iwasaki, T. Kagawa, N. Smith, and A. Kowada (1999). Charac-
release distributions, Bull. Seism. Soc. Am. 93, 639–654. terizing crustal earthquake slip models for the prediction of strong
Kuge, K., T. Iwata, and K. Irikura (1997). Automatic estimation of earth- ground motion, Seism. Res. Lett. 70, 59–80.
quake source parameters using waveform data from the K-NET (pro- Umino, N., T. Matsuzawa, S. Hori, A. Nakamura, A. Yamamoto, A. Has-
gramme and abstracts), Seism. Soc. Japan 2, B16 (in Japanese). egawa, and T. Yoshida (1998). 1996 Onikobe earthquakes and their
Ma, K. F., J. Mori, S. J. Lee, and S. B. Yu (2001). Spatial and temporal relation to crustal structure, Zisin 51, 253–264 (in Japanese with En-
distribution of slip for the 1999 Chi-Chi, Taiwan, earthquake, Bull. glish abstract).
Seism. Soc. Am. 91, 1069–1087. Wald, D. J. (1996). Slip history of the 1995 Kobe, Japan, earthquake de-
Madariaga, R. (1979). On the relation between seismic moment and stress termined from strong motion, teleseismic, and geodetic data, J. Phys.
drop in the presence of stress and strength heterogeneity, J. Geophys. Earth 44, 489–503.
Res. 84, 2243–2250. Wald, D. J., and T. H. Heaton (1994). Spatial and temporal distribution of
Mai, P. M., and G. C. Beroza (2000). Source scaling properties from finite- slip of the 1992 Landers, California, earthquake, Bull. Seism. Soc.
fault rupture models, Bull. Seism. Soc. Am. 90, 604–615. Am. 84, 668–691.
Mai, P. M., and G. C. Beroza (2002). A spatial random-field model to Wald, D. J., T. H. Heaton, and K. W. Hudnut (1996). The slip history of
characterize complexity in earthquake slip, J. Geophys. Res. 107, no. the 1994 Northridge, California, earthquake determined from strong-
B11, 2308, doi 10.1029/2001JB000588. motion, teleseismic, GPS, and leveling data, Bull. Seism. Soc. Am. 86,
Miyake, H., T. Iwata, and K. Irikura (1999). Strong ground motion simu- S49–S70.
lation and source modeling of the Kagoshima-ken Hokuseibu earth- Wald, D. J., D. V. Helmberger, and T. H. Heaton (1991). Rupture model
quakes of March 26 (MJMA 6.5) and May 13 (MJMA 6.3), 1997, using of the 1989 Loma Prieta earthquake from the inversion of strong
empirical Green’s function method, Zisin 51, 431–442 (in Japanese motion and broadband teleseismic data, Bull. Seism. Soc. Am. 81,
with English abstract). 1540–1572.
Miyake, H., T. Iwata, and K. Irikura (2001). Estimation of rupture propa- Wessel, P., and W. H. F. Smith (1995). New version of the Generic Map-
gation direction and strong motion generation area from azimuth and ping Tools released, EOS 76, 329.
distance dependence of source amplitude spectra, Geophys. Res. Lett. Wu, C., M. Takeo, and S. Ide (2001). Source process of the Chi-Chi earth-
28, 2727–2730. quake: a joint inversion of strong motion data and global positioning
Miyakoshi, K., T. Kagawa, H. Sekiguchi, T. Iwata, and K. Irikura (2000).
system data with a multifault model, Bull. Seism. Soc. Am. 91, 1028–
Source characterization of inland earthquakes in Japan using source
1043.
inversion results, in Proc. 12th World Conf. Earthquake Engineering,
Yoshida, S., K. Koketsu, B. Shibazaki, T. Sagiya, T. Kato, and Y. Yoshida
Aukland, New Zealand, 30 January–4 February (CD-ROM).
(1996). Joint inversion of near- and far-field waveforms and geodetic
Miyamachi, H., K. Iwakiri, H. Yakiwara, K. Goto, and T. Kakuta (1999).
data for the rupture process of the 1995 Kobe earthquake, J. Phys.
Fine structure of aftershock distribution of the 1997 northwestern Ka-
Earth 44, 437–454.
goshima earthquakes with a three-dimensional velocity model, Earth
Zeng, Y., and C. H. Chen (2001) Fault rupture process of the 20 September
Planet Space 51, 233–246.
1999 Chi-Chi, Taiwan, earthquake, Bull. Seism. Soc. Am. 91, 1088–
Nakamura, H., and T. Miyatake (2000). An approximate expression of slip
1098.
velocity time function for simulation of near-field strong ground mo-
Zeng, Y., K. Aki, and T. Teng (1993). Mapping of the high-frequency
tion, Zisin 53, 1–9 (in Japanese with English abstract).
source radiation for the Loma Prieta earthquake, California, J. Geo-
Nakahara, H., T. Nishimura, H. Sato, and M. Ohtake (1998). Seismogram
phys. Res. 98, 11,981–11,993.
envelope inversion for the spatial distribution of high-frequency en-
ergy radiation from the earthquake fault: application to the 1994 far
east off Sanriku earthquake, Japan, J. Geophys. Res. 103, 855–867. Disaster Prevention Research Institute
Nakahara, H., T. Nishimura, H. Sato, M. Ohtake, S. Kinoshita, and H. Kyoto University
Hamaguchi (2002). Broadband source process of the 1998 Iwate pre- Gokasho, Uji, Kyoto 611-0011, Japan
fecture, Japan, earthquake as revealed from inversions analyses of miyake@egmdpri01.dpri.kyoto-u.ac.jp
seismic waveforms and envelopes, Bull. Seism. Soc. Am. 92, 1708–
1720. Manuscript received 22 August 2002.