3 Article

Geochemistry
G
Volume 4, Number 2
Geophysics 22 February 2003
1020, doi:10.1029/2001GC000207
Geosystems
AN ELECTRONIC JOURNAL OF THE EARTH SCIENCES ISSN: 1525-2027
Published by AGU and the Geochemical Society

Constraint of fault parameters inferred from nonplanar

fault modeling
Hideo Aochi and Raul Madariaga
Laboratoire de Géologie, E´cole Normale Supérieure, 24 rue Lhomond, 75231 Paris Cedex 05, France
(aochi@geologie.ens.fr; madariag@geologie.ens.fr)

Eiichi Fukuyama
National Research Institute for Earth Science and Disaster Prevention, 3-1 Tennodai, Tsukuba 305-0006, Japan
( fuku@bosai.go.jp)

[1] We study the distribution of initial stress and frictional parameters for the 28 June 1992 Landers,
California, earthquake through dynamic rupture simulation along a nonplanar fault system. We find that
observational evidence of large slip distribution near the ground surface requires large nonzero cohesive
forces in the depth-dependent friction law. This is the only way that stress can accumulate and be released
at shallow depths. We then study the variation of frictional parameters along the strike of the fault. For this
purpose we mapped into our segmented fault model the initial stress heterogeneity inverted by Peyrat et al.
[2001] using a planar fault model. Simulations with this initial stress field improved the overall fit of the
rupture process to that inferred from kinematic inversions, and also improved the fit to the ground motion
observed in Southern California. In order to obtain this fit, we had to introduce an additional variations of
frictional parameters along the fault. The most important is a weak Kickapoo fault and a strong Johnson
Valley fault.

Components: 6946 words, 11 figures, 4 videos.

Keywords: Landers earthquake; Boundary integral equation method; Fault geometry; Rupture dynamics; Ground motion.
Index Terms: 7209 Seismology: Earthquake dynamics and mechanics; 7212 Seismology: Earthquake ground motions and
engineering; 8123 Tectonophysics: Dynamics, seismotectonics; 8164 Tectonophysics: Evolution of the Earth: Stresses—crust
and lithosphere.
Received 27 July 2001; Revised 30 July 2002; Accepted 17 October 2002; Published 22 February 2003.

Aochi, H., R. Madariaga, and E. Fukuyama, Constraint of fault parameters inferred from nonplanar fault modeling, Geochem.
Geophys. Geosyst., 4(2), 1020, doi:10.1029/2001GC000207, 2003.

1. Introduction fault branching, fault curvature and other geomet-

rical complexities without major changes in its
[2] Since the studies of Harris et al. [1991] and current formulation. Thanks to rapid progress in
Harris and Day [1993], dynamic rupture propaga- the development of boundary integral equation
tion along nonplanar faults has been simulated methods (BIEM), originally proposed by Koller
using the finite difference method (FDM) [Kase et al. [1992], Cochard and Madariaga [1994] and
and Kuge, 1998, 2001; Harris and Day, 1999]. Fukuyama and Madariaga [1995], it is now pos-
Unfortunately, 3-D finite difference can not model sible to model complex fault systems consisting of

Copyright 2003 by the American Geophysical Union 1 of 16

 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

several subfaults, including branched and bent that appears from these works is that the accumu-
faults [Kame and Yamashita, 1997; Tada and lated stress is much less than the stress extrapolated
Yamashita, 1997; Aochi et al., 2000a]. These from laboratory experiments.
methods can also be applied to study spontaneous
[4] In spite of the difficulty and uncertainty of the
crack propagation in intact material [Kame and
estimation of fault parameters, we absolutely need
Yamashita, 1999]. Based on these recent studies,
them for reproducing the rupture branching phe-
Aochi and Fukuyama [2002] modeled the 1992
nomena in numerical simulations. In this paper, we
Landers earthquake, using a realistic nonplanar
focus on the question of how to constrain them
fault geometry learned from field observations, a
from observations of rupture of the Landers earth-
loading system (remote tectonic stress) derived
quake. As shown by Aochi et al. [2000a, 2000b],
from geological considerations, and a depth-
the absolute level of stress and frictional parame-
dependent slip-weakening law. In this paper, we
ters appear explicitly in nonplanar fault systems,
investigate the initial condition of the friction law
because the external tectonic forces produces het-
for the simulation of the Landers earthquake.
erogeneous stress distribution depending on fault
orientation (strike). In a recent simulation, Aochi
[3] For the purpose of understanding earthquakes,
and Fukuyama [2002] succeeded in reproducing
we have to study the role of the friction law that
the general rupture propagation of the Landers
controls rupture process. Since rupture process are
earthquake without any horizontal heterogeneity
very complex and spatially heterogeneous, fric-
of fault parameters. Rupture complexity was only
tional parameters or stress field are probably also
due to a heterogeneous stress field produced by the
very heterogeneous. We know from fracture
assumed tectonic loading forces. In the previous
physics that spontaneous rupture propagation
model [Aochi and Fukuyama, 2002], an external
requires stress release in order to propagate, so
loading force, whose direction changed regionally
that we need an opposite mechanism in order to
according to local tectonics, produced heterogene-
stop the rupture process. The effect of frictional
ity not in fault parameters but in initial stress field
parameters are well understood in numerical
along the fault system. That model successfully
experiments [Boatwright and Cocco, 1996]. On
reproduced realistic rupture transfer between faults.
the other hand, it is very difficult to determine
For instance, rupture progressed not along the
them quantitatively for a real earthquake in the
northern Johnson Valley fault, but propagated into
field. For the Landers earthquake, the stress change
the Kickapoo and the Homestead Valley faults, as
during the earthquake was very heterogeneous
shown in the map of Figure 1.
along the fault as determined from kinematic fault
models [Wald and Heaton, 1994; Bouchon et al., [5] In this paper we will test several rupture scenar-
1998a; Day et al., 1998] and dynamic modeling ios for the Landers earthquake. In a first step we
and inversion [Olsen et al., 1997; Peyrat et al., consider the variation of the friction law with depth.
2001]. The relation between shear stress and fault As proposed by Scholz [1988], cohesive force is
movement was also determined from kinematic usually assumed to be zero at the ground surface in
inversions of the 1995 Hyogoken-nanbu, Japan, some seismic rupture simulations [Yamashita and
earthquake [Ide and Takeo, 1997; Guatteri and Ohnaka, 1992]; whereas other simulations did not
Spudich, 2000; Guatteri et al., 2001]. But accord- assume any depth-dependency [Olsen et al., 1997;
ing to these authors, the resolution of these esti- Peyrat et al., 2001]. Here we will discuss the
mations is limited because of the finite frequency question of how cohesive force may affect rupture
band used in the kinematic inversion. Usual kine- propagation and strong ground motion. In the
matic inversions can only determine relative second part, we will discuss the horizontal hetero-
dynamic stress changes during rupture. In recent geneity of frictional parameters. We first assume a
works, there have been attempts to investigate the uniform external load (remote tectonic stress). We
absolute level of stress field [Bouchon et al., will investigate what kind of heterogeneity of fault
1998b; Spudich et al., 1998]. One of the results parameters is required so that rupture chooses the
2 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Camp Rock Fault

s1
s1

Emerson Fault

Homestead
Valley
Fault

Kickapoo Fault s1
s1
M w = 7.3 Johnson
Valley
Fault
40° N
Brunt Mt. Eureka
Fault Peak
Fault

10 km
30° N
120°W
Figure 1. Fault model (red lines) studied by Aochi and Fukuyama [2002]. Black lines are the trace of observed active
faults [Hart et al., 1993]. Arrows represent the assumed directions of the maximum principal stress s1 (red, spatially
rotating principal stress assumed by Aochi and Fukuyama [2002]; blue, uniform stress field used in this study).

correct fault branches, and we will compare with constitutive parameters [e.g., Olsen et al., 1997].
the previous heterogeneous external load case For nonplanar fault modeling, on the other hand,
[Aochi and Fukuyama, 2002]. We finally study we need absolute stress values [Aochi et al.,
the mapping of the heterogeneity of initial stress 2000a]. Aochi and Fukuyama [2002] constructed
obtained from a planar fault inversion [Peyrat et al., a nonplanar fault model based on the surface fault
2001] onto our segmented fault model, and com- traces [Hart et al., 1993] shown in Figure 1. It is
pare rupture propagation and seismic radiation. important to remark that the model included both
the faults that did break during the 1992 rupture
2. The Nonplanar Fault Model of Aochi and those that did not; so that the model allowed
and Fukuyama [2002] the rupture to progress along any fault segment.
The fault segments that the rupture chose
[6] For planar fault simulations, one needs to depended on the initial stress field and fault
specify the initial shear stress on the fault and its properties.
3 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

[7] For the initial stress field, Aochi and Fukuyama (a) Stress and Strength [MPa]
[2002] assumed an external load (remote tectonic 0 20 40 60 80 100 120 140
stress) based on geological studies of local tecton- 0
ics [Dokka and Travis, 1990; Unruh et al., 1994;

sp
Sowers et al., 1994]. The red arrows in Figure 1

Depth [km]
shows the maximum principal stress s1 assumed in
their simulation. The direction of s1 points in the 10 tr
NNE direction except in the southern part of the
rupture area (Johnson Valley, Kickapoo, and south-
ern Homestead Valley faults) where it is closer to
the NE direction. The depth dependence of the 20
assumed external load is shown in Figure 2c. As
confining pressure increases with depth, all com- (b) Critical Slip Displacement [m]
ponents of the principal stress are assumed to 0 1 2 3 4 5
increase proportionally to depth. Initial shear stress 0
shear stress
t0 and normal stress sn0 are then given by sp
t0 Dt b
Depth [km]
s1  s3
t0 ¼
2
sinð2qÞ ð1Þ tr
10
s1 þ s3 s1  s3 Dc slip
s0n ¼  cosð2qÞ ð2Þ
2 2 Dc
where q is the angle of the fault strike with respect
20
to the direction of s1.
[8] Total shear and normal stresses (ttotal and
sntotal) at each point on the fault are written in the
(c) Stress [MPa]
0 200 400 600
following, 0
     
ttotal ~
x; t ¼ t0 ~
x þ Dt ~
x; t ð3Þ
s1
Depth [km]

     
stotal
n
~
x; t ¼ s0n ~
x þ Dsn ~
x; t ð4Þ 10 Ds
where ~
s2
x and t are position and time. Dt and Dsn
represent increment stress during dynamic rupture s3
process and are in the form of integral equation, a 20
convolution of Green’s function and slip velocity Figure 2. Friction model used by Aochi and Fukuya-
on the fault, calculated based on BIEM [Aochi et ma [2002]. (a) Depth variation of peak strength sp and
al., 2000a]. At the beginning (t = 0), Dt and Dsn residual stress tr. (b) Depth variation of critical slip
equal to zero, whereas t0 and sn0 are given in displacement Dc. A simple slip-weakening law was
equations (1) and (2). It should be emphasized that used. (c) Depth variation of external load (three
compressional principal stress s1 > s2 > s3) and the
share and normal stresses in equations (3) and (4) deviation Dt = (s1  s3)/2.
are defined with respect to the slip vector on the
fault (strike parallel direction). Other four stress
components can be calculated in the same way as applied shear stress t overcomes peak strength
in Aochi et al. [2000a]. sp, rupture starts following a slip-weakening rela-
tionship between fault function s and fault slip Du,
[9] For the fault parameters, Aochi and Fukuyama
   
[2002] introduced a depth-dependent slip-weaken-
Du Du
sðDuÞ ¼ tr þ sp  tr 1 H 1 : ð5Þ
ing friction law, shown in Figure 2. When the Dc Dc

4 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Here tr is residual strength and Dc is critical slip- mental results [Bouchon et al., 1998; Spudich et
weakening distance. Breakdown strength drop Dtb al., 1998], as well as other geological observation
is defined by (sp  tr). H() represents the also suggested low stress along the San Andreas
Heaviside function. The slip-weakening friction fault. Regardless of the uncertainty, the depth
law was proposed theoretically and numerically variation of the frictional parameters is often used
[Ida, 1972; Palmer and Rice, 1973], it was then in the simulation, based on the rheology due to the
experimentally observed [Okubo and Dieterich, pressure and temperature with depth, as modeled in
1984; Ohnaka et al., 1987], modeled theoretically Sibson [1982], Scholz [1988] and Yamashita and
[Matsu’ura et al., 1992], and inferred from Ohnaka [1992]. Above the depth of 12 km, the
seismological modeling of actual earthquake rup- peak strength sp as a function of depth z is given
tures [Ide and Takeo, 1997; Olsen et al., 1997; by
Guatteri and Spudich, 2000]. As clearly seen in the
definition of equation (3), this criterion is basically sp ðzÞ ¼ s0 þ mf  ðPðzÞ  PH ðzÞÞ; ð6Þ
combined with total shear stress ttotal in equation
(5). In our previous paper [Aochi and Fukuyama,
where mf is frictional coefficient, P and PH are the
2002] and this study, all parameters, tr, sp and Dc,
confining pressure and hydrostatic pressure, re-
are supposed to be temporally invariable, so that it
spectively. s0 is the cohesive force, but s0 = 0 was
is enough to follow numerically shear stress
assumed in the simulation by Aochi and Fukuyama
increment Dt for practical use. On the other hand,
[2002].
it is possible to introduce a much more complex
criterion instead of equation (5). Aochi et al. [2002] [11] Aochi and Fukuyama [2002] modeled the
investigated a dynamic Coulomb law, whose rupture process of the Landers earthquake using a
parameters are temporally variable according to numerical boundary integral equation method
normal stress increment Dsn. In that case, equation (BIEM) for nonplanar faults embedded in a 3D
(5) must be combined with both equations (3) and unbounded, homogeneous elastic medium [Aochi et
(4) at the same time. However, they reported that al., 2000a]. Time step and square grid size were
increment in normal stress Dsn is generally much taken as 0.06 s and 750 m, respectively; and P- and
smaller than its absolute normal stress level (sn0) S-wave velocities were 6.20 and 3.52 km/s, respec-
when we consider a realistic situation of high tively. We used mirror sources for approximating
confining pressure in the crust, and that, as a result, the effect of the free surface. A calculation of 400
the frictional parameters do not change drastically time steps took about 4  105 s of CPU time using 8
with time. That is why, in this paper, we will to 16 CPUs with fortran90 and MPI on a COMPAQ
suppose each frictional parameters are temporally ES40 Cluster (EV6 500 MHz), although we occa-
invariable, and consider the effect of their spatial sionally ran several jobs for one simulation.
heterogeneity on rupture process.
[12] In Animation 1 (available in the HTML ver-
[10] The product of Dtb and Dc determine the sion of the article at http://www.g-cubed.org) and
fracture energy, a parameter that is the easiest to Figure A1a, we show a movie of one of the
invert from seismic observations [Aki, 1979; Guat- dynamic simulations by Aochi and Fukuyama
teri and Spudich, 2000; Peyrat et al., 2002]. The [2002]. We observe that rupture does not propagate
absolute level of stress in the friction law, sp and on the northern Johnson Valley fault, but chooses
tr, are very important because, for an assumed the Kickapoo and Homestead Valley faults instead,
external load (remote tectonic stress), we need the and then jumps to the northernmost Camp Rock
absolute stress field on any segmented, nonplanar fault. This is the most important feature of the
fault system. However, the estimation of absolute Landers earthquake that they succeeded in repro-
value of fault parameters is still unsolved. Some ducing. In the following sections, we will inves-
seismological analyses proposed that they are tigate how important were the assumptions they
much lower than that extrapolated from experi- made for rupture propagation, and then discuss

5 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

how we may constrain the fault properties from the [15] Let us first compare the dynamic rupture
numerical simulations. simulations with s0 = 0, 5, 10, and 12.5 MPa.
Figure 4 shows the final slip distribution on all
3. Constraint of the Depth-Dependent segments of the Landers earthquake for each value
Friction Law of s0. We observe that slip at shallow depth
increases as the cohesive force s0 increases. Fault
[13] Although the segmented model of Aochi and slip of more than 5 m was observed on the Home-
Fukuyama [2002] reproduced the general features stead Valley and the Camp Rock faults [Hart et al.,
of rupture propagation along the fault system, there 1993], so that in order to explain the surface fault
are some clear discrepancies between their model slip, breakdown strength drop must be larger than
and observations. In their simulations, there were 5 MPa. In our simulations, large slip areas appear
several large slip areas along strike in overall near the ground surface, while artificial maximum
agreement with the asperities inferred from kine- slips at depth disappear.
matic inversion [Wald and Heaton, 1994], but the
maximum slip was located at a depth of around [16] In Figure 5, we show synthetic seismograms
12 km. Thus, their model could not produce large computed using the discrete wave number method
slip near the ground surface, although a slip of (AXITRA) for the same crustal structure as in Olsen
more than 5 m was observed in the field [Hart et et al. [1997] and Aochi and Fukuyama [2002]. For
al., 1993]. The discrepancy is due to the depth each simulation, we show synthetics at station YER
dependency of friction assumed in the simulation, which is located in the forward direction of rupture
the depth of 12 km corresponds to the depth where propagation, and where large amplitudes were
the breakdown strength drop is maximum, as observed. We observe that the amplitude of the seis-
shown in Figure 2. Clearly the assumption of zero mogram improves as the cohesive force increases
peak strength at the ground surface is incorrect. and the fault slip near surface gets larger. We also
Thus, we have to modify the depth variation of observe that the width of the pulse becomes nar-
fault properties adapting finite cohesive force s0 in rower as cohesive force increases. This is because
equation (6), so that we expect that finite stress cohesion at shallow depths causes large fault slip
could be accumulated and released near the ground and fast rupture velocities.
surface to produce much fault slip. [17] We conclude that a finite cohesive force of
[14] As discussed earlier, this is in agreement with more than 5 MPa is required in order to explain the
many observations of the Landers earthquake. For vertical heterogeneity of slip distribution, especially
example, the inversion result by Bouchon et al. the large slip near surface. Our simulation results
[1998] showed more than 30 MPa static stress drop indicate the possibility that s0 may be as large as 10
at the depth of 4 km. The numerical simulation by MPa, since the synthetic seismograms for this case
Olsen et al. [1997] and Peyrat et al. [2001] fits the observed results better than those of the other
required more than 10 MPa breakdown strength values of s0 (Figure 5). s0 = 10 MPa is consistent
drop at the ground surface as well as in the deeper with the value of 12.5 MPa determined by Olsen et
crust. Figure 3 shows the depth-variation of fault al. [1997] and Peyrat et al. [2001].
parameters we assumed in the numerical simula-
tion for different cohesive forces. In the deeper part 4. Constraints on Horizontal
of the fault, we assumed a slip-hardening friction Heterogeneity
law instead of a slip-weakening law. We assumed a
residual stress level tr equal to 0 at the surface, and
4.1. Uniform External Load
breakdown strength drop Dtb to be 20 MPa at the [18] Choosing the initial stress field has many
depth of 12 km regardless of the values of s0. degrees of freedom, although the simple assumption
Friction coefficient mf = 0.6 and hydrostatic pres- in Aochi and Fukuyama [2002] was reasonably
sure were assumed. based on previous geological studies. They found

6 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Stress and Strength [MPa]

0 20 40 60 80 100 120 140
0
(a)
sp
Depth [km]

10
s0 tr

20

0
(b)
Depth [km]

10
s0

20
Figure 3. Profile of depth-dependent frictional law with cohesion for surface cohesion (a) s0 = 5 MPa, and (b) 10
MPa, respectively.

that a heterogeneous external load (remote tectonic as shown in Figures 6a and 6b. This principal
stress) in the southern and northern parts of the direction was inferred from the focal mechanisms
model region was essential for the correct fault of the seismicity before and after the 1992 Landers
selection of the Kickapoo and Homestead Valley earthquake [Hauksson, 1994]. We assumed the
faults. The selectivity of rupture along a branching same frictional parameters as Aochi and Fukuyama
fault system is actually very sensitive to the con- [2002], that is, equation (6) with s0 = 0. In this
ditions on each branch [Aochi et al., 2000b, 2002]. uniform external load, we found that the rupture
did not jump to the Kickapoo fault, but continued
[19] Before we discuss stress or frictional hetero- along the Johnson Valley fault as shown in Ani-
geneity, let us examine the role of a hypothetical mation 2 or Figure A1b, causing a completely
uniform external load on the Landers earthquake different history of rupture propagation from the
propagation. We assume that the direction of max- one that was actually observed. We get a maximum
imum uniform principal stress s1 is NNW, as fault slip of 5.0 m and a seismic moment of 3.4 
shown by the blue arrows in Figure 1 which 1019 Nm. Thus, under a uniform external load, we
produces heterogeneous stress field on the fault have to assume that the Kickapoo and/or Home-
7 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

s 0 = 0 MPa stress field is shown in Figures 6a and 6b, and

0 possible stress drop in Figure 6c. Normal stress is
much lower on the Kickapoo fault than on the

3
northern Johnson Valley fault (Figure 6b), so that it
15 implys a ‘‘weak’’ Kickapoo fault following equa-
60 30 0 tion (7). However, the accumulated shear stress
[km]
s 0 = 5 MPa (Figure 6a) is also low on the Kickapoo fault, so
that it is not possible to produce a positive stress
drop on this fault as shown in Figure 6c. Thus
rupture can not propagate along the Kickapoo and
Homestead Valley faults in this case.
s 0 = 10 MPa [21] This feature is very interesting. The Kickapoo
fault is relatively weak because of a reduced normal
stress, but shear stress is not enough to produce slip
due to the unfavorable orientation of fault strike with
s 0 = 12.5 MPa respect to the external load s1. This difficulty does
not go away even if we introduce the normal stress
dependency dynamically into frictional parameters;
sp(~x, t) = mssn(~
x, t) and tr(~
x, t) = mdsn(~
x, t), where the
fault parameters are variable with time according to
rupture propagation.
Slip [m]
0 1 2 5 8
[22] Thus we added further heterogeneity of the
Figure 4. Comparison of final slip distribution for initial stress field as shown in Figure 7;
different values of the surface cohesive force s0 = 0, 5,      
10, and 12.5 MPa. For cohesive forces larger than sp ~
x ¼ ms s0n ~
x þa ~
x : ð9Þ
5 MPa, large slip areas appear near the surface and
artificial zones of large slip at depth disappear.
Here the space-varying parameter a(~ x) contains the
stead Valley faults are weaker than the northern following information: a strong northern Johnson
Johnson Valley fault in order to reproduce the Valley fault, and weaker and enough loaded
Kickapoo, Homestead Valley and Camp Rock
correct rupture path.
faults In this case, we fixed a breakdown strength
4.2. Heterogeneity Produced by a More drop Dtb that is a function of normal stress, so that
Complex Friction Law we impose the same heterogeneity a(~ x) to tr;
     
[20] For the purpose of modeling a weak Kickapoo tr ~
x ¼ md s0n ~
x þa ~
x : ð10Þ
fault and a strong northern Johnson Valley fault, let
us write the frictional parameters sp and tr in the The heterogeneity introduced in equations (9) and
following way; (10) is completely different and independent from
the one originally included in equations (7) and (8).
   
sp ~
x ¼ ms s0n ~
x ð7Þ [23] Animation 3 shows rupture propagation in
    this case. Rupture successfully propagates from
tr ~
x ¼ md s0n ~
x ð8Þ the Johnson Valley fault to the Camp Rock fault
(See also Figure A1d). Besides we observe that
where ms and md are static and dynamic friction rupture transfers through a jog from the Home-
coefficients, and sn0 is the applied normal stress. stead Valley fault to the Emerson fault in contrast
Hereafter ms = 0.6 and hydrostatic pressure are to the previous examples without normal stress
assumed. Now the frictional parameters, sp and tr, dependency (Their comparison is shown in Figure
are functions of the fault position ~x. The initial A1). There we did not introduce any supplemental
8 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

YER 30 km

YER
E

W
20 cm
obs. N

S
s 0 = 0MPa s 0 = 10MPa
s 0 = 5MPa s 0 = 12.5MPa U

D
20s
Figure 5. Comparison of synthetic seismogram at the YER station for each simulation shown in Figure 4. As
cohesive force increases, the amplitude of synthetic seismograms improves. Seismograms are zero-phrase bandpass
filtered between 0.07 and 0.5 Hz.

heterogeneity. As a result, we get a maximum fault initial stress field of the Landers earthquake for a
slip of 3.67 m and a seismic moment of 5.2  1019 single planar fault through the inversion of the
N  s. observed ground motion. Since they assumed a
single planar fault, the heterogeneity of the stress
[24] In conclusion, in order to reproduce the rup-
field can be transferred into the heterogeneity of
ture transfer between different fault branches,
frictional parameters as long as the amount of
especially from the Johnson Valley to the Kickapoo
available energy to fracture energy is the same in
and Homestead Valley faults, we needed an addi-
the two models [Peyrat et al., 2002]. That is, a
tional heterogeneity a(~x) which explicitly indicates
certain relation between t0, sp and tr, that is
a weak Kickapoo fault and a strong northern
required stress excess Dteplanar ( sp  t0) and
Johnson Valley fault. This seems to be very
unlikely. Rockwell et al. [2000] investigated paleo- possible stress drop Dtplanar ( t0  tr) is
seismic data from several trenches and reported conserved. In the following, we consider how to
that the time intervals from previous event for the map their heterogeneity into our nonplanar fault
southern and northern Johnson Valley and Kick- modeling.
apoo faults were about 5000 years, while those of [26] Figure 8 briefly shows the procedure we
the Homestead Valley, Emerson and Camp Rock propose in this section. We start with the uniform
faults were more than 7000 years. external load (remote tectonic stress) as assumed
in section 4.1, and indicated by the blue allows
4.3. Stress Heterogeneity Derived From in Figure 1. For the purpose of determining the
Planar-Fault Simulations final slip distribution during the earthquake, static
[25] Olsen et al. [1997] and Peyrat et al. [2001, stress drop plays a fundamental role. We assume
2002] successfully determined the heterogeneous possible stress drop Dt based on the planar-fault
9 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Emerson Homestead Valley

Kickapoo

Camp Rock

Johnson Valley

Emerson
Camp Rock
Homestead Valley
Kickapoo

Johnson Valley

Figure 6. Initial conditions of the model used the Coulomb friction law. (a, b) Initial shear and normal stresses
obtained from a uniform external loading system, respectively. (c) Possible stress drop inferred from the above stress
field and the Coulomb friction law. On the Kickapoo fault, the applied normal stress is relatively low and peak
strength is also low, but the applied initial stress is also low and effective stress drop is very difficult.

inversion of Peyrat et al. [2001], as shown in stress change inverted from the planar model
Figure 8 and Figure 9. Thus, we define tr by the [Peyrat et al., 2001]. They are functions of fault
following expression: position ~x, strike as well as depth. We implicitly
      prohibit negative friction, so that we set tr = 0 in that
tr ~
x ¼ t0 ~
x  Dtplanar ~
x ð11Þ case.

where t0 is an assumed uniform external load [27] Although it is possible to use the same peak
(remote tectonic stress) and Dtplanar a heterogeneous strength sp as in equation (6), we already know
10 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Emerson
Kickapoo

Homestead Johnson Valley

Valley

Camp Rock

Figure 7. Initial conditions with additional heterogeneity to that in Figure 6. (a) Required stress increase, and (b)
possible stress drop. Artificial heterogeneity is given independently from the Coulomb friction law. In this case, the
Kickapoo fault is weak enough compared to the applied initial shear stress.

that this did not work in the case of the uniform al., 2001], but also a small additional heterogeneity
external load. Thus, in order for rupture to progress b(~
x).
correctly, we also have to change peak strength        
sp – or strictly speaking, the stress increase (Dte) sp ~
x ¼ t0 ~
x þ Dtplanar
e
~
x þb ~
x ð12Þ
from the assumed initial stress field (sp  t0).
Actually, we not only give the stress increase That is necessary because, without a small b(~
x), the
Dteplanar of the planar fault simulation [Peyrat et rupture does not propagate. In the simulations the
11 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

shear stress by uniform external load

initial shear stress by Peyrat et al.(2001)
Strike [km]
80 70 60 50 40 30 20 10

Depth [km]
10

-12-8 -4 0 4 8 12
stress [MPa]

planar
(11) τ r (ξ) = τ0(ξ) − ∆τ (ξ)

planar
(12) σ p (ξ) = τ0(ξ) + ∆τ e (ξ) + β(ξ)
Figure 8. Procedure for mapping stress heterogeneity determined from a planar fault model into a segemented,
nonplanar fault. A uniform external load (remote stress) imposes a heterogeneous initial shear stress t0 on the fault as
shown in the left panel on the top (see also Figure 9a). First we constrain the residual stress tr by the stress drop
Dt planar (see Figure 9b) based on the initial condition of Peyrat et al. [2001] (right panel on the top), as shown in
equation (5). In a second step, we adjust the peak strength sp by the required stress excess Dteplanar of the planar fault
simulation, as shown in equation (6). Here Dteplanar = 12.5  Dt planar. We further need further heterogeneity b which
indicates a week Kickapoo fault and a strong northern Johnson Valley fault (See Figure 9c).

rupture never reached the branching point of the arbitrary initial stress t0, otherwise it is also
Kickapoo fault, because the nonplanarity of the possible to begin with sp given by equation (7)
fault perturbs the stress field and actually reduces depending on applied normal stress, and constrain
shear stress with respect to the planar case. Thus the other parameters t0 and tr with the deviations
we had to increase the available potential strain Dt planar and Dteplanar.
energy release by increasing b(~ x). An even more
[28] Animation 4 (see also Figure A1d) shows
critical problem in nonplanar fault modeling is
rupture propagation for this model. Rupture suc-
that we need to introduce additional information
so that rupture chooses the Kickapoo fault when it cessfully propagates along the Kickapoo and
reaches the branching point. As shown in the Homestead Valley faults, then it transfers to the
previous section 4.1, when the frictional parame- last segment, the Emerson and Camp Rock faults,
ters (sp and tr) are the same on both subfaults, and finally it is gradually arrested. The process of
rupture clearly prefers the northern Johnson Valley rupture propagation is very similar to that of Wald
fault. Thus, in order to obtain the correct fault and Heaton [1994] and Peyrat et al. [2001].
selection, a weak Kickapoo fault is necessary. As However, even though the stress drop Dt planar is
a result, we assume, in the expression of b(~ x), that the same as theirs, the amount of final slip in our
the northern Johnson Valley fault is strong, and simulation is smaller due to the perturbation of the
that the Kickapoo, southern Homestead Valley stress field by the geometry of the fault system.
and Camp Rock faults are weak. Now we find The maximum fault slip is 3.6 m, while theirs
that frictional parameters sp and tr are different reached 6 m. We got a seismic moment of 5.3 
from equations (7) and (8). Furthermore, since we 1019 Nm, which is somewhat less than the 7.0 
give derivative additional stress Dt planar and 1019 Nm obtained in the planar simulation [Peyrat
Dteplanar in equations (11) and (12), regional stress et al., 2001].
given by external load is hidden behind the
heterogeneity inverted from the planar fault sim- [29] In Figure 10, we compare synthetic seismo-
ulation. Thus we can actually use an almost grams obtained from our simulation with the

12 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Camp Rock Emerson

Homestead Valley
Kickapoo
Johnson Valley

Figure 9. Initial stress distribution constrained by the mapping of the planar fault simulation [Peyrat et al., 2001].
(a) Initial shear stress produced by the uniform external loading system. (b) Possible stress drop, the difference
between the initial shear stress and the residual stress drop (t0  tr), mapped from the planar-fault simulation. (c)
Required stress increase, the difference between the peak strength and the initial shear stress (sp  t0), obtained after
adding artificial heterogeneity.

observed ground displacement records and those from the fault. At YER, located in the forward
produced by the previous model of Animation 1 direction of rupture propagation, the amplitude of
or Figure A1d [Aochi and Fukuyama, 2002]. We the synthetics does not fit the observation well,
show recordings at the YER, SVD, and PFO because, as we explained above, fault slip is still
stations which are located in different directions underestimated in our model. Ground motion at

13 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

YER E

W
20 cm YER
obs. N 50 km

synthetic 1 S SVD
synthetic 2
U PFO

20 sec D

SVD PFO
E E

W W
obs. 10 cm N obs. 2 cm N

synthetic 1 S synthetic 1 S
synthetic 2 synthetic 2
U U

D D
20 sec 20 sec

Figure 10. Synthetic seismograms at YER, SVD, and PFO stations (synthetic 2). At SVD and PFO stations,
synthetic waveforms becomes much closer to the observed ones because the slip distribution is much closer to that
constrained by planar-fault simulation. As a reference, we also show synthetic seismograms obtained from the
uniform stress model of Animation 1 (synthetic 1). Seismograms are zero-phase bandpass filtered between 0.07 and
0.5 Hz.

YER is affected by directivity much more than [31] Modeling heterogeneity along fault strikes is
by the details of rupture propagation. On the much more difficult. We found that the hetero-
other hand, at SVD and PFO where rupture geneity produced by simple variations of normal
directivity is small, we observe that the syn- stress derived from simple tectonic stress models
thetics have improved from the previous simu- could not explain the observations without an
lation [Aochi and Fukuyama, 2002] and become additional heterogeneity of internal origin. Then
similar to the observed records. That is because we observed that the stress heterogeneity inferred
we constrained stress drop (t0  tr) so that the by the planar fault inversions [Peyrat et al., 2001]
final slip distribution has significantly improved improved the details of the rupture process and
(see also Figures A1a and A1d). the fit to the seismograms observed around the
fault system. In every case, we needed to adjust
frictional parameters such as a weak Kickapoo
5. Summary fault and a strong northern Johnson Valley fault in
order to obtain the correct selection of rupture
[30] In this paper, we tried to determine frictional propagation. We remark that the improvement
parameters for the dynamic rupture simulation of obtained by the planar fault modeling is inde-
the 1992 Landers earthquake along a nonplanar pendent of the mathematical expressions, equa-
fault system. We found that a finite cohesive force tions (6), (7), and (8), assumed in the other part of
s0 (order of 10 MPa) is required in order to this study. Actually, we do not need much infor-
reproduce large slip areas near the surface. Other- mation about absolute stress, equations (1) and
wise, stress is never accumulated or released (2), since we use only the deviation of stress. It is
around the surface, so that it inhibits large slip clear that the state of regional stress is hidden by
near the surface which is sharp contrast with the heterogeneity inverted from the planar fault
geological and geodetic observations. models. It will be a key point whether or not we
14 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

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

Camp Rock Emerson Kickapoo
t=2 s
0
Johnson
Homestead Valley
Valley
15 km
9s

16 s

23 s
0 1 2 3 4 5

Fault Slip [m]

Figure A1. Snapshots of each simulation shown in this study. (a) The previous simulations of Aochi and Fukuyama
[2002]. Its movie is in Animation 1. (b) The case of uniform loading stress (Animation 2). Rupture propagates on the
Johnson Valley fault. (c) The model including applied normal stress in frictional parameters (Animation 3). (d) The
model with initial condition similar to that of the planar fault simulation (Animation 4).

can find this kind of information for realistic Aochi, H., E. Fukuyama, and M. Matsu’ura, Selectivity of
spontaneous rupture propagation on a branched fault, Geo-
rupture models.
phys. Res. Lett., 27, 3635 – 3638, 2000b.
Aochi, H., R. Madariaga, and E. Fukuyama, Effect of nor-
mal stress during rupture propagation along nonplanar
Appendix: Snapshots of Dynamic faults, J. Geophys. Res., 107(B2), 2038, doi:10.1029/
Simulation 2001JB000500, 2002.
Boatwright, J., and M. Cocco, Frictional constraints on crustal
[32] Simulation results are given as movie files in faulting, J. Geophys. Res., 101, 13,895 – 13,909, 1996.
Bouchon, M., M. Campillo, and F. Cotton, Stress field asso-
this manuscript. In Figure A1, we further show ciated with the rupture of the 1992 Landers, California,
their snapshots in static images for convenience. earthquake and its implications concerning the fault strength
at the onset of the earthquake, J. Geophys. Res., 103,
21,091 – 21,097, 1998a.
Acknowledgments Bouchon, M., H. Sekiguchi, K. Irikura, and T. Iwata, Some
characteristics of the stress field of the 1995 Hyogo-ken
[33] We would like to thank Ruth Harris, an anonymous Nanbu (Kobe) earthquake, J. Geophys. Res., 103, 24,271 –
reviewer and Rick O’Connell whose comments help us to 24,282, 1998b.
improve this manuscript. For numerical simulations, we used Cochard, A., and R. Madariaga, Dynamic faulting under rate-
the parallel computer at the Département de Simulation dependent friction, PAGEOPH, 142, 419 – 445, 1994.
Physique et Numérique de l’Institut de Physique du Globe Day, S., G. Yu, and D. J. Wald, Dynamic stress changes during
de Paris (IPGP). This research was supported by the profect earthquake rupture, Bull. Seismol. Soc. Am., 88, 512 – 522,
‘‘Rupture et changement d’échelle’’ of ACI Catastrophes 1998.
Naturelles of the Ministère de la Recherche, France. Dokka, R. K., and C. J. Travis, Role of the eastern California
shear zone in accommodating Pacific-North American plate
motion, Geophys. Res. Lett., 17, 1323 – 1326, 1990.
References Fukuyama, E., and R. Madariaga, Integral equation method for
plane crack with arbitrary shape in 3d elastic medium, Bull.
Aki, K., Characterization of barriers of an earthquake fault, Seismol. Soc. Am., 85, 614 – 628, 1995.
J. Geophys. Res., 84, 6140 – 6148, 1979. Guatteri, M., and P. Spudich, What can strong-motion data tell
Aochi, H., and E. Fukuyama, Three-dimensional nonplanar us about slip-weakening fault-friction laws?, Bull. Seismol.
simulation of the 1992 Landers earthquake, J. Geophys. Soc. Am., 90, 98 – 116, 2000.
Res., 107(B2), 2035, doi:10.1029/2000JB000061, 2002. Guatteri, M., G. Beroza, and P. Spudich, Inferring rate and
Aochi, H., E. Fukuyama, and M. Matsu’ura, Spontaneous state friction parameters from a rupture model of the 1995
Rupture Propagation on a Non-planar Fault in 3-D Elastic Hyogo-ken Nanbu (Kobe) Japan earthquake, J. Geophys.
Medium, PAGEOPH, 157, 2003 – 2027, 2000a. Res., 106, 26,511 – 26,522, 2001.

15 of 16
 Geochemistry 3
Geophysics
Geosystems G aochi et al.: constraint of fault parameters 10.1029/2001GC000207

Harris, R. A., and S. M. Day, Dynamics of fault interaction: Ohnaka, M., Y. Kuwahara, and K. Yamamoto, Constitutive
Parallel strike-slip faults, Geophys. Res. Lett., 98, 4461 – relations between dynamic physical parameters near a tip
4472, 1993. of the propagating slip zone during stick-slip shear failure,
Harris, R. A., and S. M. Day, Dynamic 3D simulations of Tectonophys., 144, 109 – 125, 1987.
earthquakes on en echelon faults, Geophys. Res. Lett., 26, Olsen, K. B., R. Madariaga, and R. J. Archuleta, Three-Dimen-
2089 – 2092, 1999. sional dynamic simulation of the 1992 Landers Earthquake,
Harris, R. A., R. J. Archuleta, and S. M. Day, Fault step and Science, 278, 834 – 838, 1997.
the dynamic rupture process: 2-D numerical simulations of a Palmer, A. C., and J. R. Rice, The growth of slip surfaces in
spontaneously propagating shear fracture, Geophys. Res. the progressive failure of over-consolidated clay, Proc. R.
Lett., 18, 893 – 896, 1991. Soc. London Ser. A, 332, 527 – 548, 1973.
Hart, E. W., W. A. Bryant, and J. A. Treiman, Surface faulting Peyrat, S., K. B. Olsen, and R. Madariaga, Dynamic modeling
associated with the June 1992 Landers earthquake, Califor- of the 1992 Landers Earthquake, J. Geophys. Res., 106,
nia, Calif. Geol., 46, 10 – 16, 1993. 26,467 – 26,482, 2001.
Hauksson, E., State of stress from focal mechanisms before Peyrat, S., K. B. Olsen, and R. Madariaga, La dynamique des
and after the 1992 Landers earthquake sequence, Bull. Seis- tremblements de terre vue à travers le séisme de Landers du
mol. Soc. Am., 84, 917 – 934, 1994. 28 Juin 1992, C. R. Acad. Sci. Paris, 330, 235 – 248, 2002.
Ida, Y., Cohesive force across the tip of a longitudinal-shear Rockwell, T. K., S. Lindvall, M. Herzberg, D. Murbach,
crack and Griffith’s specific surface energy, J. Geophys. Res., T. Dawson, and G. Berger, Paleoseismology of the Johnson
77, 3796 – 3805, 1972. Valley, Kickapoo, and Homestead Valley Faults: Clustering
Ide, S., and M. Takeo, Determination of constitutive relations of earthquakes in the eastern California shear zone, Bull.
of fault slip based on seismic wave analysis, J. Geophys. Seismol. Soc. Am., 90, 1200 – 1236, 2000.
Res., 102, 27,379 – 27,391, 1997. Scholz, C. H., The brittle-plastic transition and the depth of
Kame, N., and T. Yamashita, Dynamic nucleation process of seismic faulting, Geol. Rund., 77, 319 – 328, 1988.
shallow earthquake faulting in a fault zone, Geophys. J. Int., Sibson, R. H., Fault zone models, heat flow, and the depth
128, 204 – 216, 1997. distribution of earthquakes in the continental crust of the
Kame, N., and T. Yamashita, Simulation of the spontaneous United States, Bull. Seismol. Soc. Am., 72, 151 – 163, 1982.
growth of a dynamic crack without constraints on the crack Sowers, J. M., J. R. Unruh, W. R. Lettis, and T. D. Rubin,
tip path, Geophys. J. Int., 139, 345 – 358, 1999. Relationship of the Kickapoo Fault to the Johnson Valley
Kase, Y., and K. Kuge, Numerical simulation of spontaneous and Homestead Valley Faults, San Bernardino Country, Ca-
rupture processes on two non-coplanar faults: The effect of lifornia, Bull. Seismol. Soc. Am., 84, 528 – 536, 1994.
geometry on fault interaction, Geophys. J. Int., 135, 911 – Spudich, P., M. Guatteri, K. Otsuki, and J. Minagawa, Use of
922, 1998. Fault Striations and Dislocation Models to Infer Tectonic
Kase, Y., and K. Kuge, Rupture propagation beyond fault dis- Shear Stress during the 1995 Hyogo-ken Nanbu (Kobe)
continuities: Significance of fault strike and location, Geo- Earthquake, J. Geophys. Res., 88, 413 – 427, 1998.
phys. J. Int., 147, 330 – 342, 2001. Tada, T., and T. Yamashita, Non-hypersingular boundary inte-
Koller, M. G., M. Bonnet, and R. Madariaga, Modeling of gral equations for two-dimensional non-planar crack analy-
dynamical crack propagation using time-domain boundary sis, Geophys. J. Int., 130, 269 – 282, 1997.
integral equations, Wave Motion, 16, 339 – 366, 1992. Unruh, J. R., W. R. Lettis, and J. M. Sowers, Kinematic Inter-
Madariaga, R., On the relation between seismic moment and pretation of the 1992 Landers Earthquake, Bull. Seismol.
stress drop in the presence of stress and strength heteroge- Soc. Am., 84, 537 – 546, 1994.
neity, J. Geophys. Res., 84, 2243 – 2250, 1979. Wald, D. J., and T. H. Heaton, Spatial and temporal distribu-
Matsu’ura, M., H. Kataoka, and B. Shibazaki, Slip-dependent tion of slip for the 1992 Landers, California, earthquake,
friction law and nucleation processes in earthquake rupture, Bull. Seismol. Soc. Am., 84, 668 – 691, 1994.
Tectonophysics, 211, 135 – 148, 1992. Yamashita, T., and M. Ohnaka, Precursory surface deformation
Okubo, P. G., and J. H. Dieterich, Effects of physical fault expected from a strike-slip fault model into which rheologi-
properties on frictional instabilities produced on simulated cal properties of the lithosphere are incorporated, Tectono-
faults, J. Geophys. Res., 89, 5817 – 5827, 1984. physics, 211, 179 – 199, 1992.

16 of 16