Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Geophys. J. Int. (2010) 181, 141–153 doi: 10.1111/j.1365-246X.2010.04511.

Source model of an earthquake doublet that occurred in a pull-apart


basin along the Sumatran fault, Indonesia

M. Nakano,1 H. Kumagai,1 S. Toda,2 ∗ R. Ando,2 T. Yamashina,1 H. Inoue1 and Sunarjo3

GJI Geodynamics and tectonics


1 NationalResearch Institute for Earth Science and Disaster Prevention, Tsukuba, Ibaraki 305-0006, Japan. E-mail: mnakano@bosai.go.jp
2 ActiveFault and Earthquake Research Center, National Institute of Advanced Industrial Science and Technology, Tsukuba, Ibaraki 305-8567, Japan
3 Meteorological and Geophysical Agency, Jakarta Pusat 10720, Indonesia

Accepted 2010 January 7. Received 2010 January 5; in original form 2009 January 6

SUMMARY
On 2007 March 6, an earthquake doublet occurred along the Sumatran fault, Indonesia. The
epicentres were located near Padang Panjang, central Sumatra, Indonesia. The first earthquake,
with a moment magnitude (M w ) of 6.4, occurred at 03:49 UTC and was followed two hours later
(05:49 UTC) by an earthquake of similar size (M w = 6.3). We studied the earthquake doublet by
a waveform inversion analysis using data from a broadband seismograph network in Indonesia
(JISNET). The focal mechanisms of the two earthquakes indicate almost identical right-
lateral strike-slip faults, consistent with the geometry of the Sumatran fault. Both earthquakes
nucleated below the northern end of Lake Singkarak, which is in a pull-apart basin between
the Sumani and Sianok segments of the Sumatran fault system, but the earthquakes ruptured
different fault segments. The first earthquake occurred along the southern Sumani segment
and its rupture propagated southeastward, whereas the second one ruptured the northern
Sianok segment northwestward. Along these fault segments, earthquake doublets, in which
the two adjacent fault segments rupture one after the other, have occurred repeatedly. We
investigated the state of stress at a segment boundary of a fault system based on the Coulomb
stress changes. The stress on faults increases during interseismic periods and is released by
faulting. At a segment boundary, on the other hand, the stress increases both interseismically
and coseismically, and may not be released unless new fractures are created. Accordingly,
ruptures may tend to initiate at a pull-apart basin. When an earthquake occurs on one of
the fault segments, the stress increases coseismically around the basin. The stress changes
caused by that earthquake may trigger a rupture on the other segment after a short time
interval. We also examined the mechanism of the delayed rupture based on a theory of a fluid-
saturated poroelastic medium and dynamic rupture simulations incorporating a rheological
velocity hardening effect. These models of the delayed rupture can qualitatively explain the
observations, but further studies, especially based on the rheological effect, are required for
quantitative studies.
Key words: Earthquake source observations; Earthquake interaction, forecasting, and pre-
diction; Continental tectonics: strike-slip and transform; Dynamics and mechanics of faulting;
Dynamics: seismotectonics; Rheology and friction of fault zones.

The convergence rate between the Indo-Australian Plate and the


1 I N T RO D U C T I O N
Eurasian Plate at Sumatra ranges from 60 mm yr−1 in the south
The Sumatran fault is a trench-parallel strike-slip fault system that to 52 mm yr−1 in the north (Prawirodirdjo et al. 2000). The slip
accommodates the oblique convergence of the Indo-Australian Plate rate on the Sumatran fault also varies from north to south. Recent
subducting beneath Sumatra, Indonesia (Fig. 1). Over its entire Global Positioning System (GPS) observations show that the right-
length of 1900 km, the fault is divided into 19 major fault segments, lateral slip rate is about 20 mm yr−1 in central Sumatra (Genrich
ranging in length from 35 to 200 km (Sieh & Natawidjaja 2000). et al. 2000). More than 20 destructive earthquakes with magnitudes
larger than six have occurred along this fault system in the past
100 yr (e.g. Pacheco & Sykes 1992; Bellier et al. 1997).
∗ Now at: Disaster Prevention Research Institute, Kyoto University, Uji, On 2007 March 6, an earthquake doublet occurred along the
Kyoto 611-0011, Japan Sumatran fault near Padang Panjang, central Sumatra (Fig. 2).


C 2010 The Authors 141
Journal compilation 
C 2010 RAS
142 M. Nakano et al.

Figure 1. Map of Sumatra showing tectonic features. Open triangles with


station codes indicate locations of the JISNET broad-band seismic stations.
The Sumatran fault trace, represented by the solid black lines on Sumatra,
is based on data collected by Sieh & Natawidjaja (2000). The trace of the
Sunda trench is based on data collected by Muller et al. (1997). The rectangle
outlined by dotted lines is the area shown in Fig. 2(a).

The first earthquake of the doublet, with a moment magni-


tude (M w ) of 6.4, occurred at 03:49 (UTC). Two hours later
(05:49 UTC), the second earthquake of similar size (M w =
6.3) occurred close to the source location of the first earth-
quake. The rapid hypocentre determinations by the National Earth-
quake Information Center (NEIC) of the U.S. Geological Survey
(USGS, Sipkin 1994) and the GEOFON global seismic mon-
itor system of the GeoForschungsZentrum Potsdam, Germany
(GFZ, http://www.gfz-potsdam.de/geofon/seismon/globmon.html)
indicated that the hypocentres of the two earthquakes were north
of Lake Singkarak at a depth of about 10 km and within a hor-
izontal distance of 10 km (Fig. 2b). The centroid moment tensor
(CMT) solutions estimated by the Global CMT (GCMT) Project
(http://www.globalcmt.org), on the other hand, placed the source
centroid locations of these earthquakes about 20 km apart. The
source centroid of the first earthquake was below the southwestern
shore of Lake Singkarak, whereas that of the second earthquake
was below a point north of the lake (Fig. 2b). The centroid depths Figure 2. (a) Tectonic features around Lake Singkarak, central Sumatra.
of both earthquakes were estimated as about 20 km. Three segments of the Sumatran fault are labeled. Years of major historical
earthquake doublets are also shown. The labels ‘A’ and ‘B’ attached to
A week after the earthquakes, Natawidjaja et al. (2007) con-
the years denote the first and second earthquakes, respectively, for each
ducted field investigations of the surface fault ruptures caused by
earthquake doublet. Red solid circles show the locations of the surface
the earthquakes. They also interviewed local people affected by rupture traces of the 2007 earthquake doublet observed by Natawidjaja
the earthquakes, and found that people who lived south of Lake et al. (2007). Open squares show populated areas. (b) Enlarged map of the
Singkarak felt a stronger shock during the first earthquake and those area around Lake Singkarak. Circles and squares indicate the epicentres
who lived north of the lake experienced the second earthquake as of the 2007 earthquake doublet estimated by NEIC and GFZ, respectively.
stronger. Natawidjaja et al. (2007) found two separated fault rup- Diamonds indicate the horizontal source centroid locations estimated by the
ture zones, one southeast and one northwest of the lake (red dots in GCMT Project. The source models of the 2007 earthquakes estimated by
Fig. 2a), which may correspond to the first and second earthquakes, this study are also shown. The thick lines denote the source faults, the stars
respectively. Surface fault ruptures of the first earthquake were ob- indicate the rupture initiation points, and the focal mechanisms are plotted at
the most probable horizontal source centroid locations. Red and blue colours
served along the Sumani segment of the Sumatran fault, southeast
denote the first and second earthquakes, respectively. The Sumatran fault
of Lake Singkarak. The surface ruptures run from the middle of the
trace is based on data collected by Sieh & Natawidjaja (2000).
southwestern shore of the lake to southeast of the lake, for a total
distance of about 15 km. The fault ruptures represent a right-lateral
movement with the strike oriented NW–SE. The maximum offset


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
Source model of an earthquake doublet 143

of the surface ruptures is about 20 cm. The ruptures also show clear dividually and in opposite directions. We proposed a source model
vertical movement with the western side moving down. Natawidjaja to explain the observed feature of the earthquake doublet based on
et al. (2007) found fault traces of the second earthquake north interseismic and coseismic Coulomb stress changes. We further in-
of the lake along the southern half of the Sianok segment, which vestigated the mechanism of the delayed rupture in the doublet based
extend 22 km from the northern tip of Lake Singkarak towards the on a theory of a fluid-saturated poroelastic medium and dynamic
northwest. The surface fault ruptures show a right-lateral movement rupture simulations incorporating a rheological velocity hardening
with a maximum offset of about 10 cm. These observations indicate effect.
that the first and second earthquakes ruptured the southern Sumani
segment and the northern Sianok segment, respectively.
Lake Singkarak is in the middle of the Sumatran fault system in a 2 D ATA A N D WAV E F O R M I N V E R S I O N
pull-apart basin (Burchfiel & Stewart 1966) formed at the boundary METHOD
between the Sumani and Sianok fault segments (Sieh & Natawidjaja We used waveform data obtained from a broad-band seismograph
2000). The discontinuity between these faults consists of a 4.5-km- network in Indonesia (JISNET) to analyse the earthquake doublet
wide right step and is a dilatational step over. The basin may have that occurred on 2007 March 6. JISNET is operated by the Na-
formed by repeated earthquakes with right-lateral strike-slip motion tional Research Institute for Earth Science and Disaster Prevention
along the fault segments, resulting in opening of the crust between (NIED) and the Meteorological and Geophysical Agency of In-
the fault segments at the jog (Sieh & Natawidjaja 2000). At Lake donesia (BMKG) (Nakano et al. 2006). The distribution of JISNET
Singkarak, the total estimated geomorphic offset of the misaligned stations around Sumatra is shown in Fig. 1. Each JISNET station is
fault segments is ∼23 km. The dextral rate of slip estimated from equipped with a CMG-3T EBB three-component broad-band seis-
offsets of stream channels is ∼11 mm yr−1 , while the rate obtained mograph (0.02–360 s). Data from the seismographs are sampled at
from recent GPS measurements is 23 mm yr−1 (Prawirodirdjo et al. 20 Hz for each channel and transmitted to BMKG and NIED in
2000). The slip rate estimated from GPS observations represent nearly real time.
recent crustal motions, while the one estimated from geomorpho- We used the waveform inversion method of Nakano et al. (2008)
logical observations may represent the motion of long-term average. to estimate the source centroid locations and focal mechanisms of
Therefore, these estimations do not always give the same value if the two earthquakes. In this method, the inverse problem is solved
tectonic features have changed after the geomorphological features in the frequency domain for efficient computation, as follows. The
were created. Geological features indicate that the lake is no more displacement field excited by a point seismic source may be written
than a few million years old, which is consistent with the offset and in the frequency domain as (e.g. Stump & Johnson 1977)
slip rate.
Around Lake Singkarak, earthquake doublets have occurred re- 
Nm
u n (ωk ) =
  ni (ωk )
G m i (ωk ), k = 1, . . . , N f , (1)
peatedly (Fig. 2a) (Untung et al. 1985; Pacheco & Sykes 1992). In i=1
1926, the first earthquake of a doublet occurred along the Sumani
where ωk is the angular frequency;   ni (ωk ) are
i (ωk ) and G
u n (ωk ), m
segment and, a few hours later, the second earthquake of similar
size ruptured the northern Sianok segment. The magnitudes of both the Fourier transforms of the nth trace of a displacement seismo-
earthquakes have been estimated as M w ∼ 7 by data inversion of gram, the ith base of the moment function tensor, and the spatial
historical triangulation data and recent GPS survey measurements derivative of Green’s function, respectively; Nm is the number of
(Prawirodirdjo et al. 2000). The estimated surface displacement independent bases of moment tensor components; and Nf is the
associated with these earthquakes is 1.7 ± 1.0 m. In 1943, an- number of frequency components used for the waveform inversion.
other earthquake doublet occurred: The first earthquake ruptured the Eq. (1) is written as Nf sets of matrix equations
Suliti segment, and the second earthquake ruptured the Sumani seg-   k )
d(ωk ) = G(ω m(ωk ), k = 1, . . . , N f , (2)
ment several hours later. Their estimated magnitudes (Ms ) were 7.1
where d(ωk ) is the data vector consisting of  k ) is the data
u n (ωk ), G(ω
and 7.4 for the first and second earthquakes, respectively (Pacheco
kernel matrix with its elements G  ni (ωk ), and m  (ωk ) is the model
& Sykes 1992). The observed surface offsets associated with these
earthquakes were 1–2 m (Untung et al. 1985; Sieh & Natawidjaja parameter vector consisting of m i (ωk ). In this approach, the matrix
2000). equations for all frequencies are independent of each other and can
The tectonic settings of the source regions of the 1926 and 1943 be solved separately (Stump & Johnson 1977), and the computation
doublets are very similar. The discontinuity of the Suliti and Sumani is much more efficient than that for solving the inverse problem in
segments at the jog is a 4.5-km-wide right step and represents a the time domain. A double-couple focal mechanism is assumed in
dilatational step over. Lakes Diatas and Dibawah are at the segment our inversion in order to stabilize the solution by using data from
boundary, suggesting that a pull-apart basin is also evolving there. a small number of seismic stations. The source centroid location
Therefore, the occurrence of earthquake doublets may be controlled is estimated by a spatial grid search, in which we minimize the
by the tectonic settings at the segment boundary. The fault models normalized residual R defined by
 N f  2
of the doublets constructed by Prawirodirdjo et al. (2000) show  
k=1 d(ωk ) − G(ωk ) mest (ωk )
that the two fault segments associated with the individual doublets R= N f , (3)

k=1 |d(ωk )|
2
adjoin each other at the segment boundary.
In this paper, we extensively studied the earthquake doublet of where m est (ωk ) is the estimated model parameter vector m  (ωk )
2007 using various approaches to better understand the source pro- and |·| represents the length of a vector. The moment function
cesses. We used data obtained from a broadband seismograph net-  est (ωk ) corresponds
obtained by the inverse Fourier transform of m
work in Indonesia (JISNET) to estimate the source locations, focal to a bandpassed form, as we need to apply a bandpass filter to the
mechanisms, and rupture propagations of the earthquake doublet. observed waveforms before the inversion. The seismic moment and
Our analysis indicates that the two earthquakes initiated at the seg- rupture duration are estimated from the deconvolved form of the
ment boundary and ruptured the Sumani and Sianok segments in- moment function (see Nakano et al. 2008, for details).


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
144 M. Nakano et al.

Three-component seismograms obtained from stations BSI, KSI, (a)


TPI and LEM were used for the inversion of both earthquakes. 100˚E 101˚E
Data from station PPI, which is closest to the sources, were not 0˚ 0˚
used, since the waveforms were clipped during the two earthquakes.
The observed velocity seismograms were corrected for instrument
A’
response and then integrated in time to obtain the displacement
seismograms. These waveforms were bandpass filtered between 50
and 100 s and decimated to a sampling frequency of 0.5 Hz. We
used the total data length of 512 s (256 data points in each channel)
for the inversion. Green’s functions were synthesized by using the 0.
3

2
2

0.
3

0.
0.

1
discrete wavenumber method (e.g. Bouchon 1979). We assumed the

0.
standard earth model ak135 (Kennet et al. 1995) for calculation of
Green’s functions. We used the hypocentre locations estimated by 1˚S 1˚S
the automatic GEOFON global seismic monitor system as the initial
locations. For the spatial grid search, we used adaptive grid spacings, km
A
starting from a horizontal grid spacing of 0.5◦ and a vertical grid 0 20 40
spacing of 10 km. In the next step, the grid spacing was reduced
to 0.2◦ horizontally and 5 km vertically. Finally, the horizontal grid 100˚E 101˚E
spacing was reduced to 0.1◦ . At each source location of the spatial (b)
grid search, the fault parameters (the dip, slip, and rake angles) A A’

Depth (km)
were searched in 5◦ steps. For each combination of source location 0 0.2
and fault and slip orientation angles, the waveform inversion was 10 0.1
carried out to estimate the best-fitting source parameters. 20
30
-60 -40 -20 0 20 40 60
Distance (km)
3 R E S U LT S (c)
We first estimated the source centroid location and focal mechanism 100.25˚E 100.5˚E 100.75˚E
of the first earthquake. Figs 3(a) and (b) show the horizontal and
vertical residual distributions, respectively, around the best-fitting 0.25˚S 0.25˚S
source location obtained from our waveform inversion. The best-
fitting source is about 10 km southwest of Lake Singkarak at a
depth of 15 km. The focal mechanism obtained at the best-fitting
source location shows strike-slip motion on a vertical fault: Two PPI
nodal planes correspond to the fault parameters (strike, dip, rake) =
0.5˚S 0.5˚S
(147, 80, 165)/(240, 75, 10). The strike of one nodal plane (147◦ ) is
similar to that of the Sumatran fault (Fig. 2). The seismic moment
of this earthquake was estimated as M 0 = 5.38 × 1018 N m, and the Lake
corresponding moment magnitude was M w = 6.4. The estimated Singkarak
km
moment function shows a step-like function with a rupture duration
of 4 s. Waveform fits between observed and synthetic seismograms 0.75˚S 0 10 20 0.75˚S
calculated for the best-fitting source parameters are shown in Fig. 4.
We obtained good fits with a normalized residual of 0.10. 100.25˚E 100.5˚E 100.75˚E
Although the fits were good at the best-fitting location, the con-
tour plot of the horizontal residual distribution shows elongation in Figure 3. (a) Contour plot of the horizontal residual distribution around
the NE–SW direction (Fig. 3a). This indicates a weak resolution for the best-fitting source of the first earthquake. The open star indicates the
the estimated source location in this direction. The weak resolution best-fitting source centroid location obtained by the waveform inversion.
may be because the stations used for our inversion and the earth- The grey star indicates the source centroid location under the assumption
quake source are aligned almost linearly in the NW–SE direction that this earthquake ruptured the Sumani segment of the Sumatran fault
as discussed in the text. Crosses denote the node points for the spatial grid
(Fig. 1). The field investigations of Natawidjaja et al. (2007) show
search. (b) Vertical cross-section of the residual distribution along the profile
that the first earthquake ruptured the Sumani segment. The focal
A–A’ shown in Fig. 3(a). (c) Source model of the first earthquake of the
mechanism estimated from the waveform inversion is also consis- doublet. The solid black line indicates the horizontal particle motion of the
tent with the rupture of this segment. The slight deviation of the event onset at station PPI. The black dashed line indicates an extrapolation
estimated source centroid location from the Sumani segment may of the particle motion to the direction of the P-wave arrival. The black and
be caused by the weak resolution in the NW–SE direction. The ac- grey stars indicate the estimated rupture initiation point and source centroid
tual source should be on this segment, and is most probably under location, respectively. The thick grey line denotes the source fault, and the
the middle of Lake Singkarak (Fig. 3c), where the residual of the open arrow indicates the direction of the rupture propagation. See text for
waveform inversion is the minimum along the segment (Fig. 3a). details.
This location is close to the one estimated by the GCMT Project
(Fig. 2b).
The rupture initiation point of this earthquake was investi-
gated from the particle motion at the event onset in the original


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
Source model of an earthquake doublet 145

BSI EW BSI NS BSI UD

KSI EW KSI NS KSI UD

TPI EW TPI NS TPI UD

LEM EW LEM NS LEM UD

Obs.
0 100 200 300 400 500 0.5 mm
Syn.
Time/s
Residual = 0.10

Figure 4. Waveform matches obtained from the waveform inversion of the first earthquake. Black and grey traces represent the observed and synthesized
seismograms, respectively. The station code and component of motion are indicated at the upper left-hand side of each seismogram.

seismograms observed at station PPI (Fig. 3c), which is about 20 km locking depth of the Sumatran fault in this region is about 20 km
northwest of Lake Singkarak. Although the seismograms at this sta- (Genrich et al. 2000). This discrepancy may also originate from the
tion were clipped during the arrival of the S wave, the onset portion weak resolution of the source depth (Fig. 5b).
of the P wave was clearly recorded in the three-component seis- We also investigated the rupture initiation point of the second
mograms. The initial horizontal motion was towards the northwest. earthquake. Fig. 5(c) shows the horizontal particle motion at the
Since the initial vertical motion was upward, the P wave arrived event onset, which was towards the northwest. Since the vertical
from the southeast. The extrapolation of the horizontal particle mo- initial motion was upward, the P wave arrived from the southeast.
tion towards the southeast intersects the surface trace of the Sumani The direction is slightly different from that of the first earthquake
segment at the northern end of Lake Singkarak (see Fig. 2b). There- (see Fig. 3c). The extrapolation of the particle motion to the south-
fore, we concluded that the first earthquake initiated at the northern east intersects with the surface trace of the Sianok segment at the
end of the Sumani segment and ruptured this segment towards the northern end of Lake Singkarak. We therefore concluded that the
southeast. second earthquake initiated at the northern end of Lake Singkarak
We also analysed the second earthquake using the same method- and ruptured the Sianok segment towards the northwest.
ology. The horizontal and vertical residual distributions obtained We estimated the relative hypocentre location between the two
from the waveform inversion around the best-fitting source location earthquakes using the differences in the arrival times at stations
are shown in Figs 5(a) and (b), respectively. The estimated focal KSI and BSI. The distance L between the hypocentres of the two
mechanism shows a strike-slip of a vertical fault: Two nodal planes earthquakes along the fault can be approximated by
correspond to (strike, dip and rake) = (145, 80, 165)/(238, 75, 10),
v
which are almost identical to those obtained for the first earthquake. L = {(t K 2 − t B2 ) − (t K 1 − t B1 )} , (4)
The strike of one nodal plane (145◦ ) is also similar to that of the 2
Sumatran fault (see Fig. 2). The seismic moment of this earthquake where t K1 and t K2 are the P-wave onset times of the first and second
was estimated as M 0 = 3.25 × 1018 Nm corresponding to a moment earthquakes, respectively, measured at station KSI; t B1 and t B2 are
magnitude of M w = 6.3, which is slightly smaller than that of the the onset times of the two earthquakes measured at station BSI; and
first earthquake. The estimated moment function shows a step-like v is the P-wave velocity. A positive L means that the hypocentre
function with a rupture duration of 4 s. The fits between observed of the first earthquake was closer to station KSI. We picked the onset
and synthesized seismograms are shown in Fig. 6. We obtained times at stations KSI and BSI from the original seismograms, and
good waveform fits for this earthquake with a normalized residual obtained t K1 − t B1 = −56.05 s for the first event and t K2 − t B2 =
of 0.08. −55.60 s for the second event. Assuming the P-wave velocity of
The best-fitting source of the second earthquake is located off the the shallow crust to be v = 5.8 km s−1 , which we adopted from the
Sumatran fault, about 20 km west of Lake Singkarak. The horizontal velocity structure shallower than 20 km in the ak135 earth model,
residual distribution shows elongation in the NE–SW direction, we obtained L = 1.3 km. Therefore, the hypocentres of these
and therefore the resolution is weak in this direction (Fig. 5a). It earthquakes were very close to each other, supporting the results of
is also evident from the field investigations of Natawidjaja et al. our particle motion analysis.
(2007) that the second earthquake ruptured the Sianok segment, We also found the effect of the rupture directivity in the orig-
and thus the source centroid is most probably northwest of Lake inal seismograms of the two earthquakes. We plotted the vertical
Singkarak (Fig. 5c), where the residual of the waveform inversion components of the original velocity seismograms of the first and
is the minimum along the Sianok segment (Fig. 5a). We note that second earthquakes (Fig. 7). Since the earthquakes share similar
the estimated centroid depth of 20 km may be too deep because the focal mechanisms, we may use the observed amplitudes to evaluate


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
146 M. Nakano et al.

(a) tion of rupture propagation, whereas the rupture of the second one
100˚E 101˚E propagated northwestward and short-period waves were amplified
0˚ 0˚ in that direction. Similar features were also observed in the hori-
zontal components, although they are not shown here. The effect of
B’
the rupture directivity becomes smaller as the periods of waves get
longer than the rupture duration. In our inversion we used the seis-
mograms filtered between 50 and 100 s, which are much longer than
0.
1 the rupture duration (4 s); thus, the effect of the rupture directivity
0.
3 is negligible in the inversion results.

2
2
3

0.
0.
0.

We summarize the results of our waveform analyses of the doublet


0.1
earthquakes as follows: The first earthquake initiated below the
northern end of Lake Singkarak and ruptured the Sumani segment
1˚S 1˚S
southeastward (Fig. 3c). Two hours later, the second earthquake
B initiated at a location close to the hypocentre of the first one, and
km the rupture propagated along the Sianok segment northwestward, in
0 20 40 the opposite direction to that of the first earthquake (Fig. 5c).

100˚E 101˚E
(b) 4 S O U RC E M O D E L O F T H E
E A RT H Q UA K E D O U B L E T
B B’
Depth (km)

0 0.2
10 0.1 4.1 Coseismic stress changes
20 0.1 The proximity of the two earthquakes in time and space suggests that
30
-60 -40 -20 0 20 40 60 the second earthquake was triggered by the stress changes caused
Distance (km) by the first one. Triggering of seismic activity by large earthquakes
is generally evaluated by the Coulomb failure criterion (e.g. King
(c) et al. 1994). The change in the static Coulomb failure function
100.25˚E 100.5˚E 100.75˚E
CFF caused by an earthquake is given by

0.25˚S 0.25˚S CFF = σs + μ (p − σn ), (5)


where σ s is the shear stress change on a given fault plane (positive
in the direction of fault slip), σ n is the fault-normal stress change
(positive when the fault is clamped),  p is the pore pressure change,
and μ is the effective coefficient of friction. When CFF is positive,
PPI a failure on the given fault plane tends to occur, whereas a negative
0.5˚S 0.5˚S
value of CFF indicates that failure is suppressed. It has now been
widely recognized that static stress transfer plays a governing role
Lake in interactions of earthquakes, including aftershock activity (Harris
Singkarak 1998; Stein 1999).
km We calculated CFF around Lake Singkarak associated with the
0.75˚S 0 10 20 0.75˚S first earthquake using the Coulomb 3.1 program (Lin & Stein 2004;
Toda et al. 2005). For simplicity, we assumed a rectangular vertical
fault as the geometry of the input fault. The length of the fault
100.25˚E 100.5˚E 100.75˚E
along strike was assumed to be 26 km, which we estimated from the
Figure 5. (a) Contour plot of the horizontal residual distribution around distance between the epicentre location of our waveform analysis
the best-fitting source of the second earthquake of the doublet. (b) Vertical and the southern end of the surface fault ruptures (Natawidjaja
cross-section of the residual distribution along the profile B–B’ shown in et al. 2007). The fault extended vertically from the surface to 20
Fig. 5(a). (c) Source model of the second earthquake. Symbols here are the km depth. The seismic moment was set at 5.38 × 1018 N m, from
same as those in Fig. 3. the result of our waveform inversion. A linear taper with width of
10 km in the horizontal direction was applied to the slip distribution
the effect of the rupture directivity. Because the difference in the on the fault, in which the seismic moment was kept as the input
hypocentres of the two earthquakes is very small compared with the value. Poisson’s ratio, Young’s modulus, and the effective coefficient
distances to the sources from the stations, the effect of the difference of friction were assumed to be 0.25, 80 GPa, and 0.6, respectively.
in the hypocentre locations is negligible except at station PPI. At The fault and slip orientation angles of the receiver fault were set at
the stations located southeast of the source (KSI, TPI and LEM), those of the second earthquake. We did not consider pore pressure
the maximum amplitudes of the first earthquake at the individual change here. The calculated spatial Coulomb stress changes at a
stations were larger than those of the second one. At the station to depth of 15 km caused by the first earthquake (Fig. 8a) and the
the northwest (BSI), on the other hand, the maximum amplitude stress profile along the fault plane of the second earthquake (Fig. 8b;
of the second earthquake was larger, even though its magnitude thick solid line) indicate that Coulomb stress increased along the
was smaller than that of the first one. This may be attributed to the Sianok segment at the time of rupture on the Sumani segment. This
rupture directivity. The rupture of the first earthquake propagated modelling of the Coulomb stress field was relatively consistent for
southeastward and short-period waves were amplified in the direc- different combinations of effective coefficient of friction, dip angle,


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
Source model of an earthquake doublet 147

BSI EW BSI NS BSI UD

KSI EW KSI NS KSI UD

TPI EW TPI NS TPI UD

LEM EW LEM NS LEM UD

Obs.
0 100 200 300 400 500 0.25 mm
Syn.
Time/s
Residual = 0.08

Figure 6. Waveform matches obtained from the waveform inversion of the second earthquake. Black and grey traces represent the observed and synthesized
seismograms, respectively. The station code and component of motion are indicated at the upper left-hand side of each seismogram.

1st event (03:49) 2nd event (05:49)

BSI V 0.2 mm/s

PPI V 14 mm/s

KSI V 4 mm/s

TPI V 0.5 mm/s

LEM V 0.5 mm/s

0 250 500 0 250 500


Time/s Time/s
Figure 7. Comparison of the original vertical velocity seismograms of the first earthquake (left-hand column) with those of the second one (right-hand
column). The rows are ordered from top to bottom according to the alignment of the stations from northwest to southeast. The station code is indicated at the
left-hand side of each row. The vertical scale of the seismograms is the same for both events. Note that the waveforms at station PPI were clipped during the
two earthquakes.


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
148 M. Nakano et al.

(a) y

x
-l/2 l/2

(b)

0.2

Stress change / Δσ
0.0
ΔCFF
-0.2

-0.4
Pore pressure
-0.6

10-4 10-3 10-2 10-1 100 101


Nondimensional time (t/tr)

Figure 9. (a) Schematic representation of the fault system used in our


calculation of stress changes in a poroelastic medium. The thick solid line
indicates the source fault. The filled star indicates the location where the pore
pressure and the Coulomb stress changes are calculated. (b) Black and grey
lines indicate the pore pressure and CFF. The values of pore pressure and
CFF are normalized by the stress drop σ of the main rupture. The time
is normalized by the characteristic relaxation time tr . See text for details.

4.2 Delayed rupture


The deformation of a fluid-infiltrated poroelastic medium is a mech-
anism that may explain the delayed triggering of an earthquake
(e.g. Nur & Booker 1972). Because the pore fluid in a poroelastic
Figure 8. (a) The Coulomb stress changes (CFF) at 15 km depth on the medium diffuses down the pressure gradient, the pore pressure, and
fault of the second earthquake caused by the first earthquake. The yellow accordingly the Coulomb stress, changes with time after an earth-
and white stars indicate the epicentres of the first and second earthquakes,
quake (e.g. Piombo et al. 2005). Hudnut et al. (1989) and Horikawa
respectively. The white solid and black dotted lines indicate the fault models
of the first and second earthquakes, respectively. The green lines indicate
(2001) explained delayed triggering of earthquakes based on a lin-
the surface fault traces of the Sumatran fault, after Sieh & Natawidjaja ear, quasi-static, elasticity theory of a 2-D fluid-saturated porous
(2000). (b) CFF along the source fault of the second earthquake. Thick medium developed by Rice & Cleary (1976). We calculated changes
grey, thick black, and thin grey solid lines indicate CFF assuming effective in the pore pressure and CFF due to the first earthquake of the
coefficients of friction of 0.4, 0.6 and 0.8, respectively, with the fault dipping 2007 doublet by following the method of Rice & Cleary (1976) and
at 90◦ . Black and grey dotted lines indicate CFF assuming the fault dips Li et al. (1987). Fig. 9(a) shows schematically the fault system we
of 80◦ and 70◦ , respectively, with an effective coefficient of friction of 0.6. used in our calculation. We assumed ν = 0.25, ν u = 0.34 and B =
The dash–dotted line indicates CFF assuming an effective coefficient of 0.85 for the poroelastic medium, where ν and ν u are Poisson’s ratios
friction of 0.6 and a fault dip of 90◦ obtained at a depth of 10 km. when the medium is deformed under drained and undrained condi-
tions, respectively, and B is Skempton’s coefficient, which is defined
as the ratio of the induced pore pressure to the change in applied
stress for the undrained condition (Wang 2000). These values were
and the depth at which Coulomb stress was estimated (Fig. 8b). obtained from Westerly granite samples (table 1 of Rice & Cleary
These results suggest that the second earthquake was triggered by (1976). We also assumed μ = 0.6 for the effective coefficient of
the stress changes caused by the first one. friction.
We also estimated the effect of the Coulomb stress changes on Fig. 9(b) shows the temporal changes of pore pressure and CFF
the faults of the earthquake doublet due to the 2005 Nias-Simeulue at the source location of the second earthquake, which is located
earthquake (M w = 8.6). For this estimation we used the source cen- 4 km away from the fault of the first one. In this figure the stress
troid location and fault parameters obtained for the Nias-Simeulue changes are shown in non-dimensional units normalized by the
earthquake by the GCMT project and its fault dimension of 300 × stress drop σ of the main rupture of the first earthquake. The time
200 km2 (Konca et al. 2007). We obtained that the Coulomb stress is also normalized by the characteristic relaxation time tr = l 2 /16c,
changes on the faults of the doublet were decreases of less than where l is the fault length and c is the fluid diffusivity of the medium.
5 KPa. These changes are one order of magnitude smaller than that Since the hypocentre of the second earthquake is within the region of
of the first event of the doublet, and thus may be negligible. dilatation, the pore pressure initially decreases from its static value.


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
Source model of an earthquake doublet 149

It then gradually recovers to its static value because of diffusion of ture propagates along Fault 1. When the dynamic stress change on
pore fluids from the surrounding region. Because of the decrease Fault 2 caused by the rupture on Fault 1 is large enough, rupture on
in the pore pressure, CFF is initially less than zero. As the pore Fault 2 may be triggered. The fault length, strike, and slip direction
pressure recovers, CFF increases, which may cause the delayed were fixed during our calculations.
triggering of a second earthquake. The relaxation becomes evident In previous studies of dynamic fault rupture propagations
after t/tr ∼ 0.01 and CFF becomes positive after t/tr ∼ 0.06 (e.g. Harris et al. 1991; Harris & Day 1993, 1999; Duan & Oglesby
(Fig. 9b). Assuming c = 0.22 × 10−4 m2 s−1 estimated for Westerly 2006) the rupture of the second event is triggered immediately af-
granite samples (Rice & Cleary 1976) and a fault length l = 26 km, ter the passage of seismic waves of the first event. These studies
we obtain tr ∼ 1012 s and the time when CFF becomes positive assume the linear slip weakening friction law (e.g. Ida 1972; Ando
td = tr × 0.06 ∼ 3 × 1011 s, which are much longer than the observed & Yamashita 2007). In addition to the slip weakening friction, we
delay of 2 hr. As the hypocentre region may be heavily fractured introduced a rheological velocity hardening effect commonly ob-
because of repeated earthquakes, the fluid diffusivity may be much served in rock frictions (e.g. Dieterich 1979), using the power-law
higher there. If we assume c = 5 m2 s−1 , which is comparable to form of creep behaviour in frictional properties. The shear traction
that of sandstone (Hudnut et al. 1989), we have tr ∼ 107 s and td ∼ τ on the fault surface is given by
141 hr, which are still longer than the observed delay.
τ = f (d)σn + ηv p , (6)
We investigated the effect of dynamic stress changes on the
delayed triggering by solving the temporal evolution of dynamic where d and v are the slip and slip velocity, respectively; f (d) is
fault ruptures. We employed an elastodynamic boundary integral the coefficient for the slip weakening friction; σ n is the effective
equation method (BIEM) for the simulation of dynamic mode II normal stress; η is the pre-exponent factor and p is the exponent.
fault rupture in a 2-D infinite, homogeneous, and isotropic elas- We first assumed only the slip weakening friction (i.e. ignoring
tic medium. We used the method developed by Tada & Madariaga rheological effect) in our simulations. The friction coefficients at
(2001) and Ando et al. (2004). Two strike-slip faults were aligned the static and dynamic states were assumed to be 0.6 and 0.2, re-
with an offset representing a dilatational step over (Fig. 10a). The spectively. The length of critical slip displacement Dc was assumed
initial rupture is nucleated on Fault 1, at the step over, and the rup- to be 0.5 m. The shear strength of the faults relative to the level of
initial shear stress applied to the faults is specified by a parameter
(a) Tp , which is defined as
y
τ p − τr
Tp = , (7)
τ0 − τr
Fault 2 Fault 1
w where τ p , τ r and τ 0 are, respectively, the shear strength of the static
x state, the shear strength when fault slip is larger than Dc , and the
initial shear stress. See Ando & Yamashita (2007) for the definitions
-l 0 l of τ p , τ r and τ 0 . Note that the stress drop due to fault slip is given by
(b) τ 0 − τ r . Ando & Yamashita (2007) used Tp = 2 for the simulations
of rupture propagation along a branched fault. Using this value in
8

our study, the rupture on Fault 2 was not triggered by the dynamic
Slip velocity (m/s)

Fa
6

stress changes caused by the rupture on Fault 1. Smaller values


lt
2

of Tp , namely a higher level of the initial stress, may be required.


4

Fa
ult

Therefore, we searched the range of Tp within which the rupture on


1
2

25 Fault 2 was triggered by the rupture on Fault 1. We assumed P-wave


20 velocity, rigidity, and Poisson’s ratio of 6 km s−1 , 30 GPa and 0.25,
0

15
-2

respectively. The stress drop was assumed to be constant at 10 MPa


10

(s)
10
-
)

e
m

Tim along the two faults. The fault length l and step-over width w were
(k

0
e
nc

10

5
ta
is
D

assumed to be 26 and 4 km, respectively. The rupture initiation point


20

0
(c) on Fault 1 was set at the jog (x = 0 in Fig. 10a). We simulated the
ruptures on the faults using a time window of 40 s. Fig. 10(b) shows
Slip velocity (m/s)

4 the slip velocity on the fault planes as a function of the location and
x=-11.4 km
p=0.5 η=0 time and assumes Tp = 1.1. When we assumed Tp = 1.1 and 1.2,
η=1.13 the rupture on Fault 2 was triggered immediately after the arrival of
2 η=0.5
η=1.0 η=1.12 rupture-stopping phase from Fault 1. If we assumed Tp larger than
1.2, namely a higher peak strength compared to the initial stress,
0 the rupture on Fault 2 was not triggered.
0 10 20 30 40 Next, we added the rheological effect to our simulations. In the
Time (s) following calculations, we assumed Tp = 1.1. Accordingly, the be-
haviour of fault ruptures depends on the parameters η and p. We
Figure 10. (a) Schematic diagram showing the fault system used in our investigated the fault-slip behaviour assuming p = 0.3, 0.5, 0.8 and
calculation of dynamic fault ruptures. Thick lines indicate the alignment
1. The slip-velocity functions for various values of η at the centre
of the faults of the first (‘Fault 1’) and second (‘Fault 2’) earthquakes,
respectively. (b) Spatio-temporal distribution of slip velocity on the fault
of Fault 2 assuming p = 0.5 are shown in Fig. 10(c). As the value of
planes, in which Tp = 1.1 and η = 0 were assumed. (c) The slip velocity η increases, the rupture initiation delays on Fault 2. A delay longer
functions on Fault 2 for the effective viscosities η = 0, 0.5, 1.0, 1.12 and than 20 s was observed for η = 1.12. Assuming η = 1.13, the rupture
1.13. The functions were obtained at x = −11.4 assuming p = 0.5 and on Fault 2 was not observed within the calculated time window. The
Tp = 1.1. delayed rupture was clearly observed in the calculations assuming


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
150 M. Nakano et al.

p = 0.3 and 0.5. The delay became shorter for p = 0.8, and was (1973). This indicates that the Sumatran fault is locked above a
hardly observed for p = 1. depth of about 20 km and aseismic slip occurs below that level dur-
Our calculations indicated that the delay in the rupture on the ing interseismic periods. We therefore estimated the interseismic
second event occurred when we used both the slip-weakening fric- Coulomb stress changes by assigning secular slip during an inter-
tion and the power-law creep. Therefore, the rheological velocity seismic period on the deep extensions of the faults below the locking
hardening effect may also be a mechanism of the delayed rupture depth. The coseismic stress changes due to faulting were computed
of the second earthquake. However, we obtained a delay of up to by assigning slip on the faults above the locking depth, while zero
20 s, which is much shorter than the observed delay of 2 hr. A slip was assumed on the deeper extensions. The Coulomb stress
longer delay may occur if we set the parameters properly. How- changes were evaluated for right-lateral strike-slip faulting with the
ever, we were not able to carry out such calculations, which require same strike as the fault segments depicted in Fig. 11.
a longer time window and thus extensive computer memory and Figs 11(a) and (b) show schematic views of the interseismic and
time. coseismic Coulomb stress changes. During an interseismic period,
the Coulomb stress increases both along the faults and at the segment
boundary as a result of crustal loading (Fig. 11a). After earthquakes
4.3 Model of an earthquake doublet at a segment occur along the faults, the Coulomb stress decreases on and around
boundary the faults because the causative stress is released by the faulting
As described above, earthquake doublets have occurred repeatedly (Fig. 11b). At the segment boundary, on the other hand, the Coulomb
around Lake Singkarak, which suggests that there is a common stress increases again. This occurs because the lobes of the increase
mechanism that generates the doublets during sequential earthquake of Coulomb stress due to faulting are distributed asymmetrically
cycles. We modelled the interseismic and coseismic static Coulomb with respect to the fault plane (see Fig. 8a). Because the faults are
stress changes at a segment boundary of a fault system as follows. aligned with only a small offset, the areas of stress increase due to the
Two vertical fault segments representing a right-lateral strike-slip slip of the two faults interfere positively with each other, resulting
motion are aligned in a half-space, as shown in Fig. 11. The discon- in a coseismic stress increase at the segment boundary. Thus, the
tinuity of the faults at the segment boundary is a right step and thus stress there increases both interseismically and coseismically, and
represents a dilatational step over. During an interseismic period, the remains at a high level unless new fractures are created. Accordingly,
crust is loaded by the motion of the oceanic plate. In this period, the ruptures may tend to initiate at a segment boundary representing a
faults are locked, that is, the slip shallower than the locking depth is dilatational step over.
zero. At extensions of the faults below the locking depth, slip may When an earthquake occurs along one fault segment, a rupture
occur aseismically at a steady rate (Savage & Burford 1973; Scholz of the other segment may be triggered after a short time interval
2002). Genrich et al. (2000) showed that the interseismic crustal by stress changes caused by the first event. The earthquakes may
movement around the Sumatran fault, as estimated from GPS mea- constitute a doublet, in which the ruptures propagate in opposite
surements, was well explained by the model of Savage & Burford directions.

Figure 11. Schematic diagrams of the Coulomb stress changes around a segment boundary of a fault system representing a dilatational step over. (a) Continuous
Coulomb stress accumulations during an interseismic period. (b) Coseismic Coulomb stress changes associated with faulting on the fault segments. The left-
hand panels show spatial distributions of the Coulomb stress changes. The black lines indicate strike-slip fault segments. The right-hand figures display the
Coulomb stress changes as a function of time. The purple and green lines indicate the Coulomb stress on the fault (annotated as ‘1’ in the left-hand panel) and
at the segment boundary between the faults (annotated as ‘2’ in the left-hand panel), respectively.


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
Source model of an earthquake doublet 151

5 DISCUSSION 1999 İzmit earthquake on the NAFZ, the rupture propagated be-
yond the segment boundaries because step-over widths were small
Our waveform analyses of the earthquake doublet along the Suma- (1–2 km), as shown by Lettis et al. (2002). The step-over widths
tran fault indicate that the earthquakes nucleated below the segment between the Sumani and Sianok segments and between the Suliti
boundary and ruptured the two adjacent fault segments. We pre- and Sumani segments of the Sumatran fault are about 4.5 km
sented a model to explain this feature based on the interseismic and (Sieh & Natawidjaja 2000), which may be large enough to arrest
coseismic Coulomb stress changes. We can find a similar example ruptures.
of an earthquake rupture that initiated at a fault segment bound- We examined the mechanism of delayed ruptures based on pore
ary of a dilatational step over. The İzmit earthquake (M w = 7.4) pressure changes in a poroelastic medium and the dynamic stress
on 1999 August 17 in Turkey. This earthquake occurred along the changes. Assuming a porous medium corresponding to sandstone,
North Anatolian Fault Zone (NAFZ). The tectonic setting of the the time scale of the pore pressure change is two orders of magnitude
fault system is similar to that of the Sumatran fault. The NAFZ is larger than the observed delay. Our dynamic rupture simulations
a 1500-km-long fault system with right-lateral, strike-slip motion showed that a delayed rupture occurred if we employed the rheo-
(Barka & Kadinsky-Cade 1988). GPS observations show that the logical velocity hardening effect using the power-law form of creep
slip rate on the fault is 20–30 mm yr−1 (e.g. McClusky et al. 2000). behaviour in frictional properties. A similar rheological effect, in
This fault system can be separated into a number of fault segments, the form of the logarithm of the slip velocity, is used in the rate and
and pull-apart basins are found at the segment boundaries. The rup- state friction law (e.g. Dieterich 1994). Belardinelli et al. (1999)
ture of the İzmit earthquake initiated at the boundary between the pointed out that the rate and state friction law can explain a delayed
Sapanca and Gölcük segments of the NAFZ. It then propagated rupture of a subevent (nearly 20 s after the first rupture) during the
bilaterally along both fault segments and along other segments be- 1980 Irpinia earthquake in Italy. We believe that our study is the
yond the eastern Sapanca segment, resulting in a single large event first that has simulated a delayed rupture by using a fully dynamic
(e.g. Yagi & Kikuchi 2000; Delouis et al. 2002; Sekiguchi & Iwata rupture model. However, we obtained a delay of up to 20 s, which
2002). is two orders of magnitude smaller than the observed delay. Limi-
Barka & Kadinsky-Cade (1988) thoroughly investigated the his- tations of available computer power prevented us from performing
torical activity of each fault segment along the NAFZ. Sixteen extensive dynamic simulations using time windows long enough to
earthquakes of M > 6 occurred along the NAFZ between 1939 and reproduce the observed delay. Clearly, further studies are required
1967. After 1967, there was little seismic activity along the NAFZ to investigate the mechanism that governs delayed triggering of
until the İzmit earthquake of 1999. Dewey (1976) recalculated the earthquakes.
epicentres of earthquakes that occurred in northern Anatolia be-
tween 1930 and 1972. Poor seismic data quality during this period
limited the accuracy of these epicentre locations to a few tens of
kilometres at best. Accordingly, if a recalculated epicentre was at
6 C O N C LU S I O N S
a segment boundary, within that range of the accuracy, we consid-
ered it to be an earthquake that initiated at a segment boundary. On 2007 March 6, an earthquake doublet occurred near Lake
We found five such earthquakes at segment boundaries of dilata- Singkarak, which is in a pull-apart basin between the Sumani and
tional step overs along the NAFZ. In particular, earthquakes on 1943 Sianok segments of the Sumatran fault system, in Indonesia. Our
November 26 and 1944 February 1 initiated at a segment boundary study showed that the first earthquake initiated at the northern end
representing a dilatational step over near Bayramuren, and these of the Sumani segment, and the rupture propagated along this fault
earthquakes ruptured the segments east and west of the boundary, southeastward. The second earthquake initiated at a location close
respectively. These observations support our proposal that ruptures to that of the first one, and its rupture propagated along the Sianok
tend to initiate at segment boundaries. segment northwestward. Earthquake doublets for which two adja-
Initiation of earthquake ruptures at segment boundaries is also cent fault segments have ruptured sequentially have occurred re-
supported by numerical simulations. Duan & Oglesby (2006) sim- peatedly near Lake Singkarak. Our study of the Coulomb stress
ulated dynamic ruptures on two parallel, strike-slip faults repre- changes in the region of the fault segment boundary showed that
senting a dilatational step over. In their computations, the rupture stress increases both interseismically and coseismically at the seg-
initiation point is not fixed: its location depends on the state of ment boundary. Accordingly, the stress below a pull-apart basin that
stress immediately before an earthquake, which is determined by the has formed at a segment boundary remains high until new fractures
stress changes due to previous earthquakes and interseismic load- form, and earthquakes tend to initiate there. When an earthquake
ing. Their simulations showed that the coseismic stress changes, occurs along one such fault segment, after a short time interval the
which are non-uniform along the two faults, accumulate after re- stress changes caused by that earthquake trigger rupture on the other
peated ruptures and the ruptures of the two fault segments initiate at segment. This pair of earthquakes may constitute a doublet in which
the jog after the system approaches a steady state. Segall & Pollard the first and second ruptures propagate in opposite directions. We
(1980) also showed that ruptures tend to initiate at jogs representing also investigated the mechanism of the delayed rupture based on
dilatational step overs because the normal stress decreases around pore pressure changes in a poroelastic medium and the dynamic
such jogs. stress changes. Although these models can qualitatively explain
The conditions that govern rupture propagation along adja- the delayed rupture, we were not able to reproduce the observed
cent fault segments were investigated by Lettis et al. (2002) and delay, and further studies especially based on the dynamic stress
Wesnousky (2006). They showed from field observations that rup- changes are required. Our detailed waveform analysis of the earth-
tures can propagate beyond segment boundaries where step overs quake doublet, the model of interseismic and coseismic Coulomb
are less than 1–2 km, whereas step overs wider than 4–5 km ar- stress changes, and the examination of the delayed rupture provide
rest ruptures. Numerical studies (e.g. Harris et al. 1991; Harris new clues to understand source processes of earthquake doublets at
& Day 1993, 1999) agree with these observations. During the segment boundaries.


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
152 M. Nakano et al.

Konca, A.O. et al., 2007. Rupture kinematics of the 2005 M w 8.6 Nias-
AC K N OW L E D G M E N T S
Simeulue earthquake from the joint inversion of seismic and geodetic
All the figures presented here were drawn using Generic Mapping data, Bull. seism. Soc. Am., 97, S307–S322.
Tools (GMT, Wessel & Smith 1995). We greatly appreciate valuable Lettis, W., Bachhuber, J., Witter, R., Brankman, C., Randolph, C.E., Barka,
comments from M. Cocco and four anonymous reviewers. A., Page, W.D. & Kaya, A., 2002. Influence of releasing step-overs on
surface fault rupture and fault segmentation: examples from the 17 August
1999 İzmit earthquake on the north Anatolian fault, Turkey, Bull. seism.
Soc. Am., 92, 19–42.
REFERENCES
Li, V.C., Seale, S.H. & Cao, T., 1987. Postseismic stress and pore pressure
Ando, R. & Yamashita, T., 2007. Effects of mesoscopic-scale fault struc- readjustment and aftershock distributions, Tectonophysics, 144, 37–54.
ture on dynamic earthquake ruptures: dynamic formation of geomet- Lin, J. & Stein, R.S., 2004. Stress triggering in thrust and subduction
rical complexity of earthquake faults, J. geophys. Res., 112, B09303, earthquakes and stress interaction between the southern San Andreas
doi:10.1029/2006JB004612. and nearby thrust and strike-slip faults, J. geophys. Res., 109, B02303,
Ando, R., Tada, T. & Yamashita, T., 2004. Dynamic evolution of a fault doi:10.1029/2003JB002607.
system through interactions between fault segments, J. geophys. Res., McClusky, S. et al., 2000. Global Positioning System constraints on plate
109, B05303, doi:10.1029/2003JB002665. kinematics and dynamics in the eastern Mediterranean and Caucasus,
Barka, A.A. & Kadinsky-Cade, K., 1988. Strike-slip fault geometry in J. geophys. Res., 105, 5695–5719.
Turkey and its influence on earthquake activity, Tectonics, 7, 663– Muller, R.D., Roest, W.R., Royer, J., Gahagan, L.M. & Sclater, J.G., 1997.
684. Digital isochrons of the world’s ocean floor, J. geophys. Res., 102,
Belardinelli, M.E., Cocco, M., Cooutant, O. & Cotton, F., 1999. Redistri- 3211–3214.
bution of dynamic stress during coseismic ruptures: evidence for fault Nakano, M. et al., 2006. Source estimates of the May 2006 Java earthquake,
interaction and earthquake triggering, J. geophys. Res., 104, 14 925– EOS, Trans. Am. geophys. Un., 87, 493–494.
14 945. Nakano, M., Kumagai, H. & Inoue, H., 2008. Waveform inversion in the fre-
Bellier, O., Sebrier, M., Pramumijoyo, S., Beaudouin, T., Harjono, H., Bahar, quency domain for the simultaneous determination of earthquake source
I. & Forni, O., 1997. Paleoseismicity and seismic hazard along the great mechanism and moment function, Geophys. J. Int., 173, 1000–1011,
Sumatran fault (Indonesia), J. Geodyn., 24, 169–183. doi:10.1111/j.1365-246X.2008.03783.x.
Bouchon, M., 1979. Discrete wave number representation of elastic wave Natawidjaja, D.H., Tohari, A., Subowo, E. & Daryono, M.R., 2007. Western
fields in three-space dimensions, J. geophys. Res., 84, 3609–3614. Sumatra earthquakes of March 6, 2007, The West Sumatra Earthquakes
Burchfiel, B.C. & Stewart, J.H., 1966. “Pull-apart” origin of the central of March 6, 2007, EERI Special Earthquake Report, May 2007, 1–5.
segment of Death Valley, California, Geol. soc. Am. Bull., 77, 439–442. Nur, A. & Booker, J.R., 1972. Aftershocks caused by pore fluid flow?,
Delouis, B., Giardini, D., Lundgren, P. & Salichon, J., 2002. Joint inversion Science, 175, 885–887.
of InSAR, GPS, teleseismic, and strong-motion data for the spatial and Pacheco, J.F. & Sykes, L.R., 1992. Seismic moment catalog of large shallow
temporal distribution of earthquake slip: application to the 1999 İzmit earthquakes, 1900 to 1989, Bull. seism. Soc. Am., 82, 1306–1349.
mainshock, Bull. seism. Soc. Am., 92, 278–299. Piombo, A., Martinelli, G. & Dragoni, M., 2005. Post-seismic fluid flow and
Dewey, J.W., 1976. Seismicity of northern Anatolia, Bull. seism. Soc. Am., Coulomb stress changes in a poroelastic medium, Geophys. J. Int., 162,
66, 843–868. 507–515.
Dieterich, J., 1979. Modeling of rock friction 1. Experimental results and Prawirodirdjo, Y., Bock, Y., Genrich, J.F., Puntodewo, S.S.O., Rais, J.,
constitutive equations, J. geophys. Res., 84, 2161–2168. Subarya, C. & Sutisna, S., 2000. One century of tectonic deformation
Dieterich, J., 1994. A constitutive law for rate of earthquake production and along the Sumatran fault from triangulation and Global Positioning Sys-
its application to earthquake clustering, J. geophys. Res., 99, 2601–2618. tem surveys, J. geophys. Res., 105, 28 343–28 361.
Duan, B. & Oglesby, D.D., 2006. Heterogeneous fault stresses from previ- Rice, J.R. & Cleary, M.P., 1976. Some basic stress diffusion solutions for
ous earthquakes and the effect on dynamics of parallel strike-slip faults, fluid-saturated elastic porous media with compressible constituents, Rev.
J. geophys. Res., 111, B05309, doi:10.1029/2005JB004138. Geophys., 14, 227–241.
Genrich, J.F., Bock, Y., McCaffrey, R., Prawirodirdjo, L., Stevens, C.W., Savage, J.C. & Burford, R.O., 1973, Geodetic determination of relative plate
Puntedewo, S.S.O., Subarya, C. & Wdowinski, S., 2000. Distribution motion in central California, J. geophys. Res., 78, 832–845.
of slip at the northern Sumatran fault system, J. geophys. Res., 105, Scholz, C.H., 2002. The Mechanics of Earthquakes and Faulting, 2nd edn,
28 327–28 341. Cambridge University Press, Cambridge.
Harris, R.A., 1998. Introduction to special section: stress triggers, stress Segall, P. & Pollard, D.D., 1980. Mechanics of discontinuous faults, J. geo-
shadows, and implications for seismic hazard, J. geophys. Res., 103, phys. Res., 85, 4337–4350.
24 347–24 358. Sekiguchi, H. & Iwata, T., 2002. Rupture process of the 1999 Kocaeli,
Harris, R.A. & Day, S.M., 1993. Dynamics of fault interaction: parallel Turkey, earthquake estimated from strong-motion waveforms, Bull. seism.
strike-slip faults, J. geophys. Res., 98, 4461–4472. Soc. Am., 92, 300–311.
Harris, R.A. & Day, S.M., 1999. Dynamic 3D simulations of earthquakes Sieh, K. & Natawidjaja, D., 2000. Neotectonics of the Sumatran fault, In-
on en echelon faults, Geophys. Res. Lett., 26, 2089–2092. donesia, J. geophys. Res., 105, 28 295–28 326.
Harris, R.A., Archuleta, R.J. & Day, S.M., 1991. Fault steps and the dynamic Sipkin, S.A., 1994. Rapid determination of global moment-tensor solutions,
rupture process: 2D numerical simulations of a spontaneously propagating Geophys. Res. Lett., 21, 1667–1670.
shear fracture, Geophys. Res. Lett., 18, 893–896. Stein, R.S., 1999. The role of stress transfer in earthquake occurrence,
Horikawa, H., 2001. Earthquake doublet in Kagoshima, Japan: rupture of Nature, 402, 605–609.
asperities in a stress shadow, Bull. seism. Soc. Am., 91, 112–127. Stump, B.W. & Johnson, L.R., 1977. The determination of source proper-
Hudnut, K.W., Seeber, L. & Pacheco, J., 1989. Cross-fault triggering in ties by the linear inversion of seismograms, Bull. seism. Soc. Am., 67,
the November 1987 Superstition Hills earthquake sequence, southern 1489–1502.
California, Geophys. Res. Lett., 16, 199–202. Tada, T. & Madariaga, R., 2001. Dynamic modelling of the flat 2-D crack
Ida, Y., 1972. Cohesive force across the tip of a longitudinal-shear crack and by a semi-analytic BIEM scheme, Int. J. Numer. Methods Eng., 50, 227–
Griffith’s specific surface energy, J. geophys. Res., 77, 3796–3805. 251.
Kennet, B.L.N, Engdahl, E.R. & Buland, R., 1995. Constraints on seismic Toda, S., Stein, R.S., Richards-Dinger, K. & Bozkurt, S.B., 2005. Fore-
velocities in the Earth from traveltimes, Geophys. J. Int., 122, 108–124. casting the evolution of seismicity in southern California: animations
King, G.C.P., Stein, R.S. & Lin, J., 1994. Static stress changes and the built on earthquake stress transfer, J. geophys. Res, 110, B05S16,
triggering of earthquakes, Bull. seism. Soc. Am., 84, 935–953. doi:10.1029/2004JB003415.


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS
Source model of an earthquake doublet 153

Untung, M., Buyung, N., Kertapati, E., Undang, & Allen, C.R., Wesnousky, S.G., 2006. Predicting the endpoints of earthquake ruptures,
1985. Rupture along the great Sumatran fault, Indonesia, during the Nature, 444, 358–360, doi:10.1038/nature05275.
earthquakes of 1926 and 1943, Bull. seism. Soc. Am., 75, 313– Wessel, P. & Smith, W.H.F., 1995. New version of the generic mapping tools
317. released, EOS, Trans. Am. geophys. Un., 76, 329.
Wang, H.F., 2000. Theory of Linear Poroelasticity with Applications to Yagi, Y. & Kikuchi, M., 2000. Source rupture process of the Kocaeli, Turkey,
Geomechanics and Hydrogeology, Princeton University Press, Princeton, earthquake of August 17, 1999, obtained by joint inversion of near-field
New Jersey. data and teleseismic data, Geophys. Res. Lett., 27, 1969–1972.


C 2010 The Authors, GJI, 181, 141–153
Journal compilation 
C 2010 RAS

You might also like