Geomatics, Natural Hazards and Risk

ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/tgnh20

Mechanism for seismic supershear dynamic

rupture based on in-situ stress: a case study of the
Palu earthquake in 2018

Kanghua Zhang, Yishuo Zhou, Yimin Liu & Pu Wang

To cite this article: Kanghua Zhang, Yishuo Zhou, Yimin Liu & Pu Wang (2022) Mechanism
for seismic supershear dynamic rupture based on in-situ stress: a case study of the
Palu earthquake in 2018, Geomatics, Natural Hazards and Risk, 13:1, 1987-2005, DOI:
10.1080/19475705.2022.2104659

To link to this article: https://doi.org/10.1080/19475705.2022.2104659

© 2022 The Author(s). Published by Informa

UK Limited, trading as Taylor & Francis
Group.

View supplementary material

Published online: 28 Jul 2022.

Submit your article to this journal

Article views: 852

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=tgnh20
GEOMATICS, NATURAL HAZARDS AND RISK
2022, VOL. 13, NO. 1, 1987–2005
https://doi.org/10.1080/19475705.2022.2104659

Mechanism for seismic supershear dynamic rupture based

on in-situ stress: a case study of the Palu earthquake
in 2018
Kanghua Zhanga, Yishuo Zhoua, Yimin Liub
and Pu Wanga
a
State Key Laboratory of Hydraulic Engineering Simulation and Safety, School of Civil Engineering,
Tianjin University, Tianjin, China; bSchool of Mechanic Engineering, Tianjin University of Technology,
Tianjin, China

ABSTRACT ARTICLE HISTORY

The Palu earthquake, with a magnitude of 7.5, severely devas- Received 18 February 2022
tated the region in Central Sulawesi, Indonesia, but the mechan- Accepted 18 July 2022
ism of dynamic rupture considering in-situ conditions is not well
KEYWORDS
understood. The in-situ stress field and fault geometry play
Supershear rupture; in-situ
important roles in the evolution of the supershear rupture. This stress; focal mechanism
study investigated the in-situ stress field, critical factors including solution; shape ratio;
the inversion of principal stress orientation by the focal mechan- Palu earthquake
ism solution, and constraint magnitude using the improved lateral
pressure coefficient polygon. Furthermore, the dynamic rupture
process is simulated using the finite element method (FEM) con-
sidering the strike and in-situ stress magnitudes of the seismo-
genic fault. The result shows that the principal stress orientation
rotates counterclockwise about 15 from north to south, and the
northern part of the fault zone is less straight than the fault from
Palu Bay to the south, which are two decisive factors for super-
shear in the north. The damage zones considering the peak accel-
eration and co-seismic displacement are enlarged in Palu City,
which is attributed to the rupture mode. It provides plausible
explanations for this supershear event and sheds light on differ-
ent dynamic rupture processes and seismic hazards that can be
predicted by fully understanding the regional in-situ stress field.

1. Introduction
On 28 September 2018 at 10:02 (UTC), the MW7.5 earthquake struck the Indonesian
island of Sulawesi and triggered a tsunami in Palu Bay. This disaster caused the death
of two thousand individuals and destroyed nearly seventy-thousand buildings. After
the earthquake, a growing number of studies using synthetic aperture radar (SAR) or

CONTACT Yishuo Zhou 2021205174@tju.edu.cn; Yimin Liu 153973418@qq.com

Supplemental data for this article can be accessed online at https://doi.org/10.1080/19475705.2022.2104659

National Institute of Natural Hazards, MEMC, Beijing, China
ß 2022 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/
licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
1988 K. ZHANG ET AL.

Figure 1. The geological structure of the Palu region. The epicentre (yellow stars) is located south
of Palu Bay. The earthquake induced a new surface fault with co-seismic rupture (red line and dot;
Wu et al. 2021). The earthquake ruptured through Palu Bay and continued to propagate on the
identified faults (orange line and dot), and many previously mapped faults that did not rupture
(black line) belong to the Palu-Koro fault zone.

geodetic surveys have provided robust evidence of the severe damage caused by this
event (Bao et al. 2019; Socquet et al. 2019). Dynamic models of supershear earth-
quakes show that the rupture velocity of this event significantly exceeded the S wave
velocity of the region’s crust, and was the dominant cause for the event. The longer
the duration of the supershear rupture, the larger is the number of disasters that can
occur. The 1999 MW7.6 event in Izmit (Bouchon et al. 2001), the 2002 MW7.9 event
in Alaska (Walker and Shearer 2009), and the 2010 Mw6.9 Yushu earthquake (Zhu
and Yuan 2020) are examples of past supershear earthquakes. These earthquakes are
mainly strike-slip, and the Palu earthquake belongs to this model of supershear earth-
quakes. However, according to the results of space geodesy research, both the signifi-
cant supershear rupture and the largest surface deformation occurred in the Palu Bay
area, located 80 km south of the epicentre, instead of the region at the onset of seis-
mic rupture (Socquet et al. 2019). Thus, it is necessary to study the mechanism of the
seismic dynamic rupture process of the Palu earthquake.
Figure 1 shows that the Palu-Koro fault zone is a partially straight strike-slip struc-
ture located between the Makassar and Sula blocks. Wu et al. (2021) reported new
fault structures in the northern Palu Basin near Palu City, marking the previously
unmapped Palu-Koro fault, which accurately reflects the actual fault structure and
rupture range of the event. Supershear earthquakes that occur on straight faults have
ubiquitously been validated by geodetic surveys (Valkaniotis et al. 2018; Bao et al.
2019) and numerical simulation (Liu and Shi 2020). Bao et al. (2019) used teleseismic
 GEOMATICS, NATURAL HAZARDS AND RISK 1989

backprojection to reveal that the rupture rapidly reached a steady velocity of approxi-
mately 4.1 km/s, exceeding the local S-wave velocity. Fang et al. (2019) used
Advanced Land Observing Satellite-2 (ALOS-2) interferometry synthetic aperture
radar (InSAR) data along with broadband regional seismograms to demonstrate that
the earthquake was a supershear event. Socquet et al. (2019) used geodesy to conclude
that the Palu earthquake probably ruptured an extremely linear segment of the 30 km
section of the rupture south of Palu city at supershear velocities of 4.3-5.2 km/s.
Ulrich et al. (2019) revealed that a physics-based earthquake and coupled tsunami
3 D model can reasonably provide information on the mechanisms and competing
hypotheses, but the calculation of the stress magnitude depends on two stress shape
ratio assumptions of 0.5 and 0.7, without further constraints. Therefore, a systematic
study was conducted, including but not limited to the Palu event, indicating that the
shear and normal stresses on seismic faults control the rupture velocity transition on
strike-slip faults (Burridge 1973; Andrews 1976; Burridge et al. 1979; Andrews 1985;
Dunham 2007). Stress heterogeneities and fault geometry are two important factors
affecting the rupture mode.
Some researchers have summarised the regularity of rupture under non-uniform
stress distribution on faults (Hu et al. 2021), and others have also proved that supershear
rupture could propagate in a stepwise manner on curved faults (Hu et al. 2016;
Tadapansawut et al. 2021). Unfortunately, few studies have analysed natural seismic
events while simultaneously considering the fault geometry and stress of in-situ condi-
tions. One important reason is the barrier to determining in-situ stresses at the hypo-
centre depth, which is much shallower than the focal depth; therefore, the initial stress
is one of the more difficult parameters to estimate in a dynamic rupture. Generally,
empirical estimates and theoretical predictions are utilised to define the stress state in
deep rock masses. The former primarily relies on empirical fitting formulae (Hoek and
Brown 1980), physical experimental models (Zoback et al. 1987; Starr, 2011), and
numerical simulations (Parsons 2006; Matsuki et al. 2009; Zhong et al. 2018).
Regarding the theoretical calculation of in-situ stresses, Zoback et al. (1987) intro-
duced the fault frictional strength theory (i.e. stress polygon theory) and constrained
the possible maximum and minimum effective principal stresses with the frictional
strength. The results are typically inaccurate (the scale is too large). The results are usu-
ally too vast; therefore, a secondary constraint using borehole damage data is necessary.
In the Kontinentales Tiefborh program der Bundesrepublik Deutschland (KTB), Brudy
et al. (1997) established the relative relationships between two horizontal principal
stresses by examining the distribution angle of drilling-induced fractures and rock ten-
sile strengths on the borehole wall at depths of approximately 8 km, as well as estimated
the total stress tensors at depths of 7.7 and 8.6 km using the polygon theory. Using the
same method, Chang et al. (2010) constrained the deep stresses in two of the four bore-
holes in the Nankai wedge based on wellbore ruptures and drilling-induced tension
cracks in the four boreholes. The images of boreholes several hundred meters below the
seafloor captured at different times after completion were compared. Moore et al.
(2011) discovered that borehole wall damage estimations from resistivity imaging are
broader, and the horizontal principal stress estimations using stress polygon methods
1990 K. ZHANG ET AL.

are higher. Secondary constraints using borehole damage data are typically required.
However, the stress state of the hypocentre is not well understood.
Driven by these results, we first used a method of constraining the in-situ stress
using a lateral pressure coefficient polygon and a focal mechanism solution (FMS;
Wang et al. 2019). To improve the accuracy of the constraint results, a normal distri-
bution analysis was conducted on the shape ratio of the FMS. Second, the three-
sigma rule is used to further constrain the deep lateral pressure coefficient in the
Palu area, and the deep stress state of the fault plane is calculated according to the
occurrence of the fault. Finally, we used the open-source software, Pylith (Aagaard
et al. 2013) to establish a 2 D Palu fault model (mainly considering the strike feature)
and simulate the earthquake rupture process. The results show that stable supershear
rupture occurred approximately 7 s after the onset of rupture when the rupture front
reached Palu Bay, by considering both the in-situ stress field and fault structure.
Simultaneously, there is an apparent Mach cone phenomenon causing a much greater
surface displacement than in other regions, which shows a high degree of similarity
to the SAR analysis and geodetic data. The numerical simulation of earthquake rup-
ture, based on the in-situ stress and fault structure, also aids in predicting the level of
seismic damage and provides implications for the mitigation of seismic risks.

2. Analysis of the in-situ stress field in Palu Bay

2.1. Method background
In light of Anderson’s theory and the Mohr-Coulomb criterion, Zoback et al. (1987)
constrained the in-situ stresses in the crust. Based on these theories, we can derive the
relationship between the maximum and minimum effective principal stresses versus
frictional strength for a fault region without considering the cohesive force, as follows:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2
r1 S1 Pp
¼ 6 ðl2 þ 1Þ þ l (1)
r3 S3  Pp

where r1 and r3 are the maximum and minimum effective principal stresses, respect-
ively; S1 and S3 are the maximum and minimum principal stresses, respectively; Pp is
the pore water pressure; and l is the fault frictional coefficient. According to
Anderson’s fault classification theory, Zoback et al. (1987) introduced the concept of a
stress polygon and constrained the possible maximum and minimum effective principal
stresses with the frictional strength, but the use of this method is complicated.
Therefore, Wang et al. (2019) modified Equations (1) and (2) after considering the cor-
responding relationships among the principal stresses for the three basic fault types:
8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2
>
>
>
> kmax  ð Þ
l þ 1 þ l ðReverse faultÞ
2
>
>
< kmax pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2
 ðl2 þ 1Þ þ l ðStrike slip faultÞ (2)
>
> kmin  
> 1
> pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 2
>
>  ðl2 þ 1Þ þ l ðNormal faultÞ
:
kmin
 GEOMATICS, NATURAL HAZARDS AND RISK 1991

where kmax and kmin are the maximum and minimum lateral pressure coefficients for
effective stresses, expressed as kmax ¼ rH/rv and kmin ¼ rh/rv, respectively; rH and rh
are the maximum and minimum horizontal stresses, respectively; rv is the vertical
stress in the strike-slip fault stress state; rH is r1; rh is r3; and the other symbols are
the same as defined above. The shape ratio R is an essential parameter in the inver-
sion of FMS and describes the relative level of the three principal stresses.
Additionally, it is an important tool for investigating the interactions between FMS
and the in-situ stress field at the focus. It is expressed as follows (Gephart and
Forsyth 1984):
r1 r2
R¼ (3)
r1 r3

where r2 is the intermediate principal stress and the other symbols are the same as
those defined above. Notably, R obtained by the inversion of multiple FMS satisfies
the normal distribution; thus, we can use the three-sigma rule to further constrain
the range of R (denoted as R’). According to Equation (3), by introducing the corre-
sponding relationships among the principal stresses for a particular stress state into
Equation (3), we can obtain the functional expression between the horizontal princi-
pal stresses and R’ as the coefficient:
8
>
> kmax ¼ R0 kmin þ 1R0 ðNormal faultÞ
>
>
< 1 R0
kmax ¼ kmin þ ðReverse faultÞ (4)
> 1 0 R0 R0  1
>
> R 1
>
: kmax ¼ 0 kmin þ ðStrike slip faultÞ
R 1 1  R0

The symbols in Equation (4) are the same as those defined above. The derivation
process from Equations (2) to (4) is shown in Supplement I.
From Figure 2, the constraint lines of R0 for different faulting stress states always
pass point (1, 1), and when R0 is an extreme value (0 or 1), it coincides with the
boundaries of the polygons for different faulting states. This further verifies that the
stress factor can constrain the stresses for a second time based on constraints by a
lateral pressure coefficient or a stress polygon. Figure 2 illustrates the constraint prin-
ciple of lateral pressure coefficients according to this method.

2.2. Seismogenic structure and earthquake catalogue

The Mw7.5 Palu earthquake occurred at the intersection of the Australian, Sunda,
and Philippine plates with strong tectonic activity. The geological structure of
Sulawesi Island in Indonesia is complex. The basin where Palu City is located is a
near-south-north valley formed by the long-term left-lateral strike-slip activity of the
Palu-Kolo fault, and active faults have developed in the eastern and western margins
of the basin (Bellier et al. 2006). The rupture with left-lateral strike-slip is compatible
with normal fault characteristics. The MW7.5 earthquake occurred on 28 September
2018 at 10:02 (UTC); the epicentre was located at 0.178 S, 119.84 E and the
1992 K. ZHANG ET AL.

Figure 2. Lateral pressure coefficient estimations for the three fault states using R00 . Estimations of
the lateral pressure coefficient in the Palu fault regions are determined by the stress model and R
(black shadow), and R00 calculated based on 3-sigma rule further constrains the range of the lateral
pressure coefficient (red shadow).

hypocentre depth was 13.5 km, as reported by USGS (2020), all FMSs are listed in
Supplement II.
Figure 3A shows that 82 small FMSs were collected from 1 January 1976 to 28
September 2018 from the Global Centroid Moment Tensor (GCMT) catalogue
(Dziewonski et al. 1981; Ekstr€ om et al. 2012), which are all listed in Supplement I.
These events are located in the area of 119 E–121 E, 1 N–2 S. For the sake of
simplicity, all FMSs are divided into the three regions comprising grids with a size of
2 1 , as shown in Figure 3B, which is a graphical visualisation used to display the
FMS properties of groups of earthquakes (Frohlich and Apperson 1992). The results
show that events in regions I and II mainly belong to the strike-slip type, and events
in region III are mainly normal strikes. As described in section 2.1, an FMS similar
to the standard Anderson’s faulting model is convenient for constraining the deep
stress field. Below, we constrain the lateral pressure coefficients and invert the stress
orientation for the three subregions using this method.

2.3. Inversion of the in-situ stress field in the study area

The constraint of the regional stress field embodies two fundamental parts: the inver-
sion of stress direction and the constraint of stress magnitudes. We used the joint
iterative inversion method introduced by Vavrycuk (2014). The inversion results of
the three regions, including the orientations of the three principal stresses and distri-
bution of the shape ratio, are shown in Figure 4A. In parallel, the normal distribution
 GEOMATICS, NATURAL HAZARDS AND RISK 1993

Figure 3. (A) Geological map and small earthquake distribution (red dots) of the Palu area. (B)
2 1 grid division (gray dotted line) of the study area and triangle diagram classification of the
FMS for each region; the colour bar shows the relative amount of FMS.

curve (red line in the histogram) and the three-sigma range of the stress shape ratio
(purple shadow in the histogram) were also obtained by statistical analysis. The type
of stress field in the three regions is compatible with the classified results of FMS;
notably, the stress fields of regions I and II are approximately strike-slip faulting in
Anderson’s model, and that of region III is strike-slip normal. Thus, the principal
stress orientations of the three regions could be approximated as 279.85 , 295.23 ,
and 325.75 . Additionally, the ranges of the stress shape factors and confidence inter-
vals up to 99.73% were 0.161-0.379, 0.462-0.654, and 0.384-0.625, respectively.
The magnitudes of the principal stresses can be constrained by combining them
with the lateral pressure coefficient polygon. First, we assume that the stress state is
limited by Coulomb frictional sliding, and the ratio between the two lateral pressure
coefficients are above certain values (boundary of the lateral pressure coefficient
polygon) defined by the coefficient of friction (ls) of 0.6 (Byerlee 1978).
Subsequently, the shape ratio from the FMS inversion was used to further constrain
the lateral pressure coefficient. Figure 4B and 4C show the ultimate constraint
results for each region. The constrained ranges of the lateral pressure coefficient,
which are shown below, are good guidelines for the in-situ stress magnitudes on
seismogenic faults. According to the lateral pressure coefficient calculated by the
stress polygon and rock density obtained from CRUST1.0 (Laske et al. 2013), com-
pleted stress profiles at a depth of 20 km in the Palu fault, including regions I and
II, are shown in Figure 5. Moreover, the in-situ stress magnitude of the two regions
at the hypocentre depth can be determined by these stress profiles. Table 1 presents
1994 K. ZHANG ET AL.

Figure 4. (A) Inversion of the stress field including the orientation of three principal stresses and
the range of the shape ratio in different regions. (B) The constraint of the lateral pressure coeffi-
cient combined with the friction and shape ratio in regions II (black shadow) and III (red shadow).
(C) The constraint of the lateral pressure coefficient in region I (blue shadow).

Figure 5. Stress profiles of regions (A) I and (B) II.

the in-situ stress magnitude and orientation of the hypocentre depth based on the
lateral pressure coefficient.

2.4. Model configuration and parameter setup

Given that the seismogenic fault was nearly strike-slip in the Palu earthquake, the
real 3 D rupture problem can be idealised as a 2 D plane strain with a multi-segment
 GEOMATICS, NATURAL HAZARDS AND RISK 1995

Table 1. Constraint of lateral pressure coefficient and stress magnitudes.

Location kmin kmax rv(MPa) rh(MPa) rH(MPa) Orientation(゜)
Region I 0.360 1.347 347.446 125.07 468.10 279.85
Region II 0.467 1.80 341.526 159.51 614.65 295.23
Region III 2.309 3.126 346.844 801.1 1082.72 325.75

Figure 6. (A) The initial stress magnitudes on the multi-segment of the seismogenic fault at the
depth of the hypocentre. (B) Model configuration of the seismogenic fault; two virtual seismic sta-
tions (blue triangles) located near the nucleation zone (S1) and Palu City (S2).

curved fault. Model geometric features mainly refer to the study of surface ruptures
and preexisting geological structures (Wu et al. 2021). Figure 6B shows the integral
model and cell grid. We observed that the finite element model (FEM) has a square
geometry of 20  100 km to include the entire seismogenic fault, the details of which
are listed in Table 2. Moreover, we set two virtual seismic stations near the epicentre
(S1) and Palu City (S2) to record the seismic wave.
The mechanical parameters mainly include the initial stress, fault frictional consti-
tutive, and boundary conditions. With a good understanding of the regional stress
field, the magnitude of initial shear stress and normal stress on the fault plane can be
calculated by the orientation of principal stress and fault occurrence (Jaeger et al.
2007):

rp¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
l2 r1 þ m2 r2 þ n2 r3
(5)
s ¼ l r21 þ m2 r22 þ n2 r23  r2
2

where l, m, and n are the direction vectors of the fault plane; s and r are the shear
stress and normal stress, respectively; and the other symbols are the same as those
defined above. Given that the Palu fault is mainly located in regions I and II, where
the stress fields are approximately strike-slip faults, we considered that r1 ¼ rH, r2 ¼
1996 K. ZHANG ET AL.

Table 2. Model parameters for simulation of Palu earthquake.

Parameter Value
S wave speed, vs 3620 m/s
P wave speed, vp 6300 m/s
Density 2740 kg/m3
Dynamic friction coefficient, ld 0.1
Static friction coefficient, ls 0.6 or 0.8
Characteristic slip-weakening distance, dc 0.4 m
Total time 20 s
Dt 0.01 s
Dx 70 m

The crustal structure parameter querid from CRUST1.0 (Laske et al. 2013)

rv, and r3 ¼ rh. The density of the overlying strata can calculate the vertical stress.
Table 2 presents the strata parameters queried by the CRUST1.0 (Laske et al. 2013).
Subsequently, it is easy to determine rH and rh if the lateral pressure coefficient and
vertical stress are known. The lateral pressure coefficients were obtained by consider-
ing the boundary of the results shown in Figure 4B. Therefore, considering a hypo-
centre depth of 13.5 km, as reported by the USGS (2020), Figure 6A shows the shear
stress and normal stress on multi-segment seismogenic faults at a depth of 13.5 km.
Furthermore, the widely used friction laws mainly include the slip-weakening friction
law and the rate-state friction law. The slip-weakening friction law was proposed by
Ida (1972), which is based on the fault dislocation theory and fracture mechanics and
is mainly used to describe the dynamic rupture process of an earthquake (e.g. Day
1982; Harris and Day 1993). The rate-state friction law (Dieterich 1981; Ruina 1983),
which includes the fault slip rate or other state quantities, can completely describe
the co-seismic and inter-seismic stress accumulation processes. To simulate the
dynamic rupture process of the Palu earthquake, we assume that the fault is governed
by a form of slip-weakening friction law. This was validated to simulate dynamic rup-
ture using the open-source software Pylith in the community benchmark problem
TPV14 (Harris et al., 2018); the parameters used are listed in Table 2. This uniform
assumption is necessary for analysing the relationship between the in-situ stress field
and rupture mode in the Palu earthquake. The model borders are enclosed by an
absorbing boundary that can efficiently remove wave reflections.
8
< jdj
ls ðls  ld Þ , jdj < dc
l¼ dc (6)
:
ld , jdj  dc

where ls is the static friction coefficient, ld is the dynamic friction coefficient, jdj is
the fault slip, and dc is the characteristic slip distance.
In addition to the above setup, it is necessary to define a nucleation zone on the
fault for spontaneous rupture in the hypocentre area (red stars in Figure 6B).
Generally, there are two general nucleation approaches for dynamic rupture models,
including the time-weakening (TW; Andrews 1985) and overstressed patch
(Galiset al., 2015). We used the latter strategy, one of the most widely used methods
of nucleation (Bizzarri 2010; Liu et al. 2014), to set the nucleation zone at the hypo-
centre of the Palu earthquake. The size of the nucleation zone was 2 km, and the
 GEOMATICS, NATURAL HAZARDS AND RISK 1997

Figure 7. Snapshots of rupture propagation on the Palu fault (white line) at different times; the
colour bar of the contour map represents the particle velocities in the model.

initial shear stress of nucleation is set to be slightly above the static fault strength
(0.5% of the strength;1.5 Mpa).

2.5. Simulated fault rupture propagation

To analyse the mechanism of the supershear rupture of the Palu Earthquake in 2018,
we built a model configuration and initial parameters according to the structural fea-
tures and in-situ stress magnitudes of the seismogenic fault. The dynamic rupture
process of the Palu earthquake was simulated using the open-source software, Pylith
(Aagaard et al. 2013). The results demonstrated the process of dynamic rupture
propagation of this supershear event.
Figure 7 shows that the spontaneous rupture propagates from the nucleation of
the seismogenic fault with low velocities. After 5.26 s, the rupture front travelled to
the south of the epicentre (about 30 km), and the Mach cone, characteristic of super-
shear rupture, appeared as the particle velocities near the rupture front increased.
However, the particle velocities of rupture decrease because the stress on the next
segment is not prone to supershear. The stable supershear rupture accompanies the
Mach cone occurring at Palu Bay and continues to propagate for the next 4 s. The
complete rupture propagation from slow to supershear in our simulation is in good
agreement with the study results from the SAR and geodetic recordings (Valkaniotis
et al. 2018; Socquet et al. 2019).
Figures 8 and 9 show the velocities and accelerations recorded by the two seis-
mic stations. Notably, the seismic wave travels to the S1 station approximately 0.3 s
after nucleation. There are similar accelerations and velocities in the x and y direc-
tions, and the peak velocity and peak acceleration are 0.67 m/s and 17.39 m/s2,
respectively. In contrast, velocity and acceleration recorded by S2 in the y direction
are greater than those in the x direction because the Palu earthquake was mainly a
left-lateral strike-slip along the N11 S. As a standard seismic wave with low noise,
the velocity increased sharply to a peak value and gradually decreased to approxi-
mately zero. Owing to the impact of the Mach wave, the peak velocity of Palu City
was up to 3.52 m/s, which is approximately five times that of S1. Similarly, the
peak acceleration of S2 was twice that of S1. Additionally, owing to the seismic
1998 K. ZHANG ET AL.

Figure 8. (A) Two components of velocity recorded by the seismic station near the nucleation
zone (S1). (B) Two components of acceleration recorded by the seismic station near the nucleation
zone (S1).

Figure 9. (A) Two components of velocity recorded by the seismic station near Palu City (S2). (B)
Two components of acceleration recorded by the seismic station near Palu City (S2).

reflections caused by the continuous perturbation of the nucleation zone near the
epicentre, the S1 station recorded the wave signal again after 7.5 s, which remained
for approximately 5 s.
Compared to non-curved faults, curved faults cause different physical processes of
rupture. For example, different angles affect the propagation and arrest of ruptures
(Zhang et al. 2016). For a multi-curved fault system, the interaction between multi-
curved segments may also cause a supershear rupture on one segment of the fault
under certain conditions (Bhat et al. 2004; Kame et al. 2003; Aochi et al. 2000).
Similarly, as shown in Figures 6–9, our simulation results show that the fault struc-
ture and in-situ stress background play crucial roles in the evolution of the fault
dynamic rupture model. On the one hand, the northern part of the rupture, from the
epicentre to its intersection with the Palu Bay coast, is less straight than the rupture
from Palu City to the south, which provides a prerequisite for rupture acceleration
and stable propagation. Conversely, the stress fields in the two regions, wherein the
Palu fault was located in both, belong to strike-slip faulting states, but the principal
stress direction has a rotation about 15 between regions I and II. The strikes of
 GEOMATICS, NATURAL HAZARDS AND RISK 1999

Figure 10. (A) Variation of S values with the along-strike distance. (B) Stress ratio S in regions I
(red line) and II (blue line) varying with h between the fault strike and principal stress azimuth.

multiple segments for all faults are roughly consistent. The angle between the fault
strike and principal stress direction of region II is less than that of region I owing
to rotation.
Moreover, the larger the angle enhancement, the larger is the normal stress and
the lower is the shear stress. For planar faults with homogeneous stress, the occur-
rence of supershear ruptures depends on the stress ratio S (Dunham 2007), S ¼
(sps0)/(s0sr), where s0 is the background shear stress, and sp and sr are the peak
and residual frictional strengths, respectively. Supershear ruptures occur if S < 1.77
in two dimensions (Andrews 1976; Das and Aki 1977). As we assume that the fault
frictional strength is homogeneous based on Byleer’s law (Byerlee 1978) and that the
stress ratio S is determined by the shear stress and normal stress, S depends only on
the in-situ stress and fault occurrence according to Equation (5). For the sake of sim-
plicity, Figure 10 displays the relationship of the angle h (between the fault strike and
principal stress azimuth) with the stress ratio S under the in-stress in regions I
and II.
Figure 10A shows S values varying on the fault along-strike, which indicates that
most segments near Palu Bay prefer to undergo supershear than sub-Rayleigh rup-
ture (S > 1.77), and the opposite is true in the region near the nucleation. As
shown in Figure 10B, the angle h between the fault strike and principal stress azi-
muth is all converted within 90 . Considering the in-situ stress in the two regions,
the dependence of h and S reveals that the appropriate interval for supershear in
region I is significantly larger than that in region II. Moreover, almost all h fre-
quencies of fault segments near Palu Bay tend to undergo supershear rupture,
including all in region I and Palu Bay in region II. This could explain why seg-
ments of faults in Palu City and Palu Bay are more prone to supershear rupture
than others based on the in-situ stress field. Additionally, the high shear stress and
2000 K. ZHANG ET AL.

Figure 11. (A) Macroseismic intensity map of the Palu earthquake reported by U.S. Geological
Survey (2020), and the location of field survey by Hidayat (2019). (B) Contour map of the spatial
distribution for peak acceleration (PA) and four measuring lines (white dotted lines) along the dir-
ection of rupture propagation. The four profile lines are located near the epicentre, Pacuan Kuda,
Palu City (Bali Asih), and the southern end of the fault from north to south, respectively. (C) Four
PA profiles in the measuring line and damage zone with 20% (red line) and 40% (blue line) max-
imum peak accelerations of 25 m/s2.

extreme inhomogeneity of the stress field between regions III (strike-slip compatible
normal) and II (strike-slip) may explain the nucleation location in the northern
part of the seismogenic fault.

3. Discussion
In this study, the lateral pressure coefficient polygon was improved by the stress poly-
gon, and the stress shape ratio was used to constrain the in-situ stress at the hypocentre
depth of the 2018 Mw7.5 Palu earthquake. Simultaneously, an equal proportion fault
model is constructed to simulate the dynamic rupture process of this event by consider-
ing the geometric structure. The results show that the rupture mode of this event is con-
trolled by the stress state of the fault plane, and only a stable super-shear rupture is
formed after the rupture front travels to the Palu Bay area, which is consistent with the
results of space geodesy (Socquet et al. 2019). According to simulation results, the
dynamic mechanism of this event is revealed by a sufficient understanding of the in-situ
stress field and fault geometry. The key parameters of earthquake disasters, including
the peak acceleration and surface displacement, are discussed below.

3.1. Peak acceleration of the co-seismic area

Seismic damage can be enlarged when supershear ruptures engender Mach waves that
are directly related to the ground motion. Through the simulation results, we
extracted the maximum amplitude of the resultant acceleration (peak acceleration)
 GEOMATICS, NATURAL HAZARDS AND RISK 2001

Figure 12. (A) Along-track displacement from ALOS-2 SAR reported by Bao et al. (2019). The arrow
labeled as ‘track’ indicates the direction of the measurement. The red star denotes the National
Earthquake Information Center (NEIC) epicentre of Palu. (B) Simulated result of the displacement in
the y direction (approximately along-strike) and absolute displacement on the fault; blue and
orange colours represent south and north displacements, respectively.

from all the steps of the dynamic rupture. To study seismic damage, we analysed the
peak acceleration in the study area. Figure 11 shows that standardised acceleration
was used to compare the relative damage in different regions. The 2 D model can
only simulate the acceleration at the depth of the hypocentre rather than the peak
ground acceleration (PGA) of shallow ground. We selected four different measuring
lines perpendicular to the fault strike to obtain the acceleration profile in the epi-
centre north of Palu Bay, Palu City, and south of the Palu fault.
Notably, the peak acceleration contour range increases significantly from north to
south, which represents the zone of co-seismic damage. In particular, significant
expansion of the seismic damage zone with a peak acceleration above 5 m/s2, located
between lines II and III (where the stable supershear rupture occurs as the simulation
result), and consistent with the point of supershear rupture, can enlarge the damage
in most cases (Das 2010; Vallee and Dunham 2012; Liu et al. 2014; Xu et al. 2021).
In reality, the supershear rupture caused by the heterogeneous stress on the fault seg-
ments is the decisive factor affecting the distribution of the damage zone in this
event. Additionally, our dynamic rupture simulation could not describe the damage
near the shoreline area in detail (for example, Qa Pagerawu and Bali Asih), which is
due to the homogeneous material used in our model, and did not account for field
effects by heterogeneous material properties, such as sand. This problem should be
addressed and optimised in future work.
2002 K. ZHANG ET AL.

3.2. Co-seismic displacement

Co-seismic displacements are vital for characterising earthquake-induced faults and
understanding earthquake dynamics. Bao et al. (2019) reported along-track displace-
ments from the ALOS-2 SAR offsets and bathymetry after the Palu earthquake. As
shown in Figure 12A, the along-track displacements are almost parallel to the fault
strike and show that the maximum slip is located near the city of Palu and a clear
difference in slip magnitude exists between the northern and southern segments.
Many subsequent studies have used geodetic data, teleseismic data, and other satellite
optical images to show similar results (Valkaniotis et al. 2018; Bao et al. 2019).
Similarly, simulated final displacements in the y direction (approximate along-
strike) after the termination of the rupture were extracted to analyse the co-seismic
displacement. As shown in Figure 12B, the maximum displacement in the y-direc-
tion was located near Palu City. As a result of the supershear rupture, the average
displacement along the fault in the northern part of the Palu fault is significantly
higher than that in the southern part. The relative co-seismic displacement based
on numerical simulations shows a high degree of similarity with the SAR and geo-
detic data. The co-seismic displacement on the fault near Palu City is significantly
larger than that at the epicentre, and the maximum displacement near Palu City is
up to 7 m, which is also similar to the field measurement reported by Valkaniotis
et al. (2018). Therefore, the simulation of the Palu earthquake validated that the
level of seismic hazards, including the peak acceleration and displacement, could
be predicted by dynamic simulation with a full understanding of the regional in-
situ stress field. In future studies, a 3 D simulation can be considered to fully and
quantitatively evaluate the actual surface displacement and peak ground acceler-
ation instead of a relative qualitative evaluation. Simultaneously, it is essential to
establish a complete stress profile and fine 3 D fault structure for accurate simula-
tion studies.

4. Conclusion
1. The lateral pressure coefficient polygon combined with the shape ratio R0 based
on the three-sigma rule, referred to as the stress polygon, is a novel and conveni-
ent way to repeatedly constrain the lateral pressure coefficient.
2. The orientation of the regional principal stress in the Pallu fault zone rotates
counterclockwise about 15 from north (region II) to south (region I). Moreover,
the northern part of fault zone is less straight than the rupture from Palu Bay to
the south, which is a joint influence on the stable supershear in the north.
3. The supershear rupture of the Palu earthquake enlarged the zone of peak acceler-
ation and surface displacement to 7 m near Palu City. The severity of damage in
Palu City is associated with the significance of the peak acceleration and surface
displacement attributed to the supershear.

The simulation results of the Palu earthquake, considering the in-situ stress
condition and the fault structure, provide an explanation for initiating the super-
shear propagation of this event. This further provides the possibility for locating
 GEOMATICS, NATURAL HAZARDS AND RISK 2003

the seismic failure zone before the large earthquake and the prediction of the fault
dynamic rupture process, and also provides a reference for disaster prevention
and mitigation.

Acknowledgements
The authors would like to thank the reviewers for their valuable suggestions and comments on
the Ph.D. Qiang Su(Nanjing University) about data visualization.

Funding
This research was funded by Tianjin Science Foundation for Distinguished Young Scientists of
China (2021YJSB131) and National Natural Science Foundation of China (41804089).

Data availability statement

The data that support the findings of this study are available from the corresponding author,
Yishuo Zhou or Yimin Liu, upon reasonable request.

References
Aagaard BT, Knepley MG, Williams CA. 2013. A domain decomposition approach to imple-
menting fault slip in finite-element models of quasi-static and dynamic crustal deformation.
J Geophys Res Solid Earth. 118(6):3059–3079.
Andrews DJ. 1976. Rupture velocity of plane strain shear cracks. J Geophys Res. 81(32):
5679–5687.
Andrews DJ. 1985. Dynamic plane-strain shear rupture with a slip-weakening friction law cal-
culated by a boundary integral method. Bull Seismol Soc. Am. 75(1):1–21.
Aochi H, Fukuyama E, Matsu’ura M. 2000. Spontaneous rupture propagation on a nonplanar
fault in 3-D elastic medium. Pure Appl Geophys. 157:2003–2027.
Bao H, Ampuero J-P, Meng L, Fielding EJ, Liang C, Milliner CWD, Feng T, Huang H. 2019.
Early and persistent supershear rupture of the 2018 magnitude 7.5 Palu earthquake. Nat
Geosci. 12(3):200–205.
Bhat H, Dmowska R, Rice J, Kame N. 2004. Dynamic slip transfer from the denali to tot-
schunda faults, Alaska: testing theory for fault branching. Bull Seismol Soc Am. 94(6B):
S202–S213.
Bellier O, Sebrier M, Seward D, Beaudouin T, Villeneuve M, Putranto E. 2006. Fission track
and fault kinematics analyses for new insight into the Late Cenozoic tectonic regime
changes in West-Central Sulawesi (Indonesia). Tectonophysics. 413(3–4):201–220.
Bizzarri A. 2010. How to promote earthquake ruptures: different nucleation strategies in a
dynamic model with slip-weakening friction. Bull Seismol Soc Am. 100(3):923–940.
Bouchon M, Bouin MP, Karabulut H, Toksoz MN, Dietrich M, Rosakis AJ. 2001. How fast is
rupture during an earthquake? New insights from the 1999 Turkey earthquakes. Geophys
Res Lett. 28(14):2723–2726.
Brudy M, Zoback MD, Fuchs K, Rummel F, Baumg€artner J. 1997. Estimation of the complete
stress tensor to 8 km depth in the KTB scientific drill holes: implications for crustal
strength. J Geophys Res. 102(B8):18453–18475.
Burridge R. 1973. Admissible speeds for plane-strain self-similar shear cracks with friction but
lacking cohesion. Geophys J Int. 35(4):439–455.
2004 K. ZHANG ET AL.

Burridge R, Conn G, Freund LB. 1979. The stability of a rapid mode II shear crack with finite
cohesive traction. J Geophys Res. 84(B5):2210–2222.
Byerlee JD. 1978. Friction of rocks. PAGEOPH. 116(4–5):615–626.
Chang CD, McNeill LC, Moore JC, Lin WR, Conin M, Yamada Y. 2010. In situ stress state in
the Nankai accretionary wedge estimated from borehole wall failures. Geochem Geophys
Geosyst. 11:Q0AD04.
Das S. 2010. Earthquake supershear rupture speeds preface. Tectonophysics. 493(3–4):213–215.
Das S, Aki K. 1977. A numerical study of two-dimensional spontaneous rupture propagation.
Geophys J Int. 50(3):643–668.
Day SM. 1982. Three-dimensional finite difference simulations of fault dynamics: rectangular
faults with fixed rupture velocity. Bull Seismol Soc Am. 72:705–727.
Dieterich JH. 1981. Mechanical behavior of crustal rocks. Washington, USA: American
Geophysical Union.
Dunham EM. 2007. Conditions governing the occurrence of supershear ruptures under slip-
weakening friction. J Geophys Res. 112(B7):B07302.
Dziewonski AM, Chou TA, Woodhouse JH. 1981. Determination of earthquake source param-
eters from waveform data for studies of global and regional seismicity. J Geophys Res.
86(B4):2825–2852.
Ekstr€om G, Nettles M, Dziewo nski AM. 2012. The global CMT project 2004–2010: centroid-
moment tensors for 13,017 earthquakes. Phys Earth Planet Inter. 200-201:1–9.
Fang J, Xu C, Wen Y, Wang S, Xu G, Zhao Y, Yi L. 2019. The 2018 Mw 7.5 Palu earthquake:
a supershear rupture event constrained by InSAR and broadband regional seismograms.
Remote Sens. 11(11):1330.
Frohlich C, Apperson KD. 1992. Earthquake focal mechanisms, moment tensors, and the con-
sistency of seismic activity near plate boundaries. Tectonics. 11(2):279–296.
Gephart JW, Forsyth DW. 1984. An improved method for determining the regional stress ten-
sor using earthquake focal mechanism data: application to the San Fernando Earthquake
sequence. J Geophys Res. 89(B11):9305–9320.
Harris RA, Day SM. 1993. Dynamics of fault interaction: parallel strike-slip faults. J Geophys
Res. 98(B3):4461–4472.
Harris RA, Barall M, Aagaard B, Ma S, Roten D, Olsen K, Duan BC, Liu DY, et al. 2018. A
Suite of Exercises for Verifying Dynamic Earthquake Rupture Codes. Seismol Res Lett.
89(3):1146–1162.
Hidayat R. 2019. Correlation of geological conditions and levels of damage in Palu earthquake.
IOP Conf Ser: Mater Sci Eng. 650:012022.
Hoek E, Brown ET. 1980. Underground excavations in rock. London (Britain): The Institute of
Mining and Metallurgy.
Hu F, Oglesby D, Chen X. 2021. The effect of depth-dependent stress in controlling free-sur-
face-induced supershear rupture on strike-slip faults. J Geophys Res: Solid Earth. 126:
e2020JB021459.
Hu F, Xu J, Zhang Z, Chen X. 2016. Supershear transition mechanism induced by step over
geometry: supershear by step over. J Geophys Res: Solid Earth. 121(12):8738–8749.
Ida Y. 1972. Cohesive force across the tip of a longitudinal-shear crack and Griffith’s specific
surface energy. J Geophys Res. 77(20):3796–3805.
Jaeger J, Cook N, Zimmerman R. 2007. Fundamental of rock mechanics. London (Britain):
Chapman and Hall.
Kame N, Rice J, Dmowska R. 2003. Effects of prestress state and rupture velocity on dynamic
fault branching. J Geophys Res. 108(B5):2265.
Laske G, Masters G, Ma Z, Pasyanos M. 2013. Update on CRUST1.0 – a 1-degree global model
of earth’s crust. Geophys Res Abstracts. 15:2658.
Liu C, Bizzarri A, Das S. 2014. Progression of spontaneous in-plane shear faults from sub-
Rayleigh to compressional wave rupture speeds. J Geophys Res: Solid Earth. 119(11):
8331–8345.
 GEOMATICS, NATURAL HAZARDS AND RISK 2005

Liu C, Shi Y. 2020. Stress partition on the Palu-Koro fault controlling the 2018 Mw 7.5 Palu
earthquake and its seismic hazards. Geochem Geophys Geosyst. 22:e2020GC009552.
Matsuki K, Nakama S, Sato T. 2009. Estimation of regional stress by FEM for a heterogeneous
rock mass with a large fault. Int J Rock Mech Min Sci. 46(1):31–50.
Moore JC, Chang C, McNeill L, Thu MK, Yamada Y, Huftile G. 2011. Growth of borehole
breakouts with time after drilling: implications for state of stress, NanTroSEIZE transect,
SW Japan. Geochem Geophys Geosyst. 12(4):Q04D09.
Parsons T. 2006. Tectonic stressing in California modeled from GPS observations. J Geophys
Res. 111(B3):B03407.
Ruina A. 1983. Slip instability and state variable friction laws. J Geophys Res. 88(B12):
10359–10370.
Socquet A, Hollingsworth J, Pathier E, Bouchon M. 2019. Evidence of supershear during the
2018 magnitude 7.5 Palu earthquake from space geodesy. Nat Geosci. 12(3):192–199.
Starr J. 2011. Closure stress gradient estimation of the Marcellus shale from seismic data. In
SEG technical program expanded abstracts 2011. San Antonio, USA, p. 1789–1793.
Tadapansawut T, Okuwaki R, Yagi Y, Yamashita S. 2021. Rupture process of the 2020
Caribbean earthquake along the oriente transform fault, involving supershear rupture and
geometric complexity of fault. Geophys Res Lett. 48(1):e2020GL090899.
U.S. Geological Survey. 2020. Earthquake lists, maps, and statistics. [accessed 2020 Mar 18].
https://www.usgs.gov/natural-hazards/earthquake-hazards/lists-maps-and-statistics.
Ulrich T, Vater S, Madden EH, Behrens J, van Dinther Y, van Zelst I, Fielding EJ, Liang C,
Gabriel A-A. 2019. Coupled, physics-based modeling reveals earthquake displacements are
critical to the 2018 Palu, Sulawesi Tsunami. Pure Appl Geophys. 176(10):4069–4109.
Valkaniotis S, Ganas A, Tsironi V, Barberopoulou A. 2018. A preliminary report on the M7.5
Palu earthquake co-seismic ruptures and landslides using image correlation techniques on
optical satellite data.
Vallee M, Dunham EM. 2012. Observation of far-field Mach waves generated by the 2001
Kokoxili supershear earthquake. Geophys Res Lett. 39(5):L05311.
Vavrycuk V. 2014. Iterative joint inversion for stress and fault orientations from focal mecha-
nisms. Geophys J Int. 199:69–77.
Walker KT, Shearer PM. 2009. Illuminating the near-sonic rupture velocities of the intraconti-
nental KokoxiliMw7.8 and Denali faultMw7.9 strike-slip earthquakes with global P wave
back projection imaging. J Geophys Res. 114(B2):B02304.
Wang P, Wang CH, Yang RH, Hou ZY, Wang H. 2019. Preliminary investigation on the deep
rock stresses prediction method based on stress polygon and focal mechanism solution.
Rock Soil Mech. 40:4486–4496.
Wu D, Ren Z, Liu J, Chen J, Guo P, Yin G, Ran H, Li C, Yang X. 2021. Coseismic surface
rupture during the 2018 Mw 7.5 Palu earthquake, Sulawesi Island, Indonesia. Geol Soc Am
Bull. 133(5-6):1157–1166.
Xu JK, Zhang ZG, Chen XF. 2021. The effects of sediments on supershear rupture.
Tectonophysics. 805:228777.
Zhong S, Jiang Q, Feng XT, Liu JG, Li SJ, Qiu SL, et al. 2018. A case of in-situ stress measure-
ment in Chinese Jinping underground laboratory. Rock Soil Mech. 39:356–366.
Zhu S, Yuan J. 2020. Physical mechanism for severe seismic hazard in the 2010 Yushu, China,
earthquake (Mw¼ 6.9): insights from FEM simulations. Geomat Nat Hazards Risk. 11(1):
2123–2146.
Zhang L, Shibazaki B, Liao WL, Li JG, Wang QL. 2016. Controlling factors analysis of
dynamic rupture propagation simulation of curved fault based on Boundary integral equa-
tion method. Chin J Geophys- Chin Ed. 59:981–991.
Zoback MD, Mastin L, Barton C. 1987. In-situ stress measurements in deep boreholes using
hydraulic fracturing, wellbore breakouts, and stonely wave polarization. In Stockholm,
Swede: ISRM International Symposium,