Earth and Planetary Science Letters 480 (2017) 147–157

Contents lists available at ScienceDirect

Earth and Planetary Science Letters

www.elsevier.com/locate/epsl

Relationships between along-fault heterogeneous normal stress and

fault slip patterns during the seismic cycle: Insights from a strike-slip
fault laboratory model
Yannick Caniven a,∗ , Stéphane Dominguez a , Roger Soliva a , Michel Peyret a ,
Rodolphe Cattin a , Frantz Maerten b
a
University of Montpellier, Géosciences Montpellier Laboratory, France
b
Schlumberger, Information Solutions Division, Montpellier, France

a r t i c l e i n f o a b s t r a c t

Article history: We use a strike-slip fault analog model to study experimentally the role played by along-fault non-
Received 29 March 2017 uniform and asymmetric applied normal stress on both coseismic slip and long-term fault behavior.
Received in revised form 26 September Our model is based on a visco-elasto-plastic multi-layered rheology that allows to produce several
2017
hundreds of scaled analog microquakes and associated seismic cycles. Uniform or heterogeneous applied
Accepted 5 October 2017
Available online 2 November 2017
normal stress along the fault plane is imposed and maintained constant during the whole experiment
Editor: R. Bendick durations. Our results suggest that coseismic slip patterns are strongly controlled by spatial normal
stress variations and subsequent accumulated shear stress along fault strike. Major microquakes occur
Keywords: preferentially in zones of major shear stress asperities. Coseismic slip distributions exhibit a pattern
seismic cycle similar to the along-fault applied normal stress distribution. The occurrence of isolated low to moderate
analog modeling microquakes where residual stresses persist around secondary stress asperities, indicates that stress
earthquakes conditions along the fault also control the whole variability of fault slip events. Moreover, when fault slip
strike-slip fault
stability conditions are modulated by normal stress distribution, our experiments suggest that the along-
tectonics
fault stress heterogeneity influences the seismic cycle regularity and, consequently, long-term fault slip
stress
behavior. Uniform applied normal stress favors irregular seismic cycles and the occurrence of earthquakes
clustering, whereas non-uniform normal stress with a single high amplitude stress asperity generates
strong characteristic microquake events with stable return periods. Together our results strengthen the
assumption that coseismic slip distribution and earthquake variability along an active fault may provide
relevant information on long term tectonic stress and could thus improve seismic hazard assessment.
© 2017 Elsevier B.V. All rights reserved.

1. Introduction ical example. The inversion of near and far field ground deforma-
tion obtained from InSAR and sub-pixel optical images correlation
One of the main measurable parameters that allow character- provided coseismic slip distribution on the fault plane and led
izing earthquake dynamics is short- and long-term fault slip kine- to a better understanding of the mechanical processes activated
matics. Studying coseismic fault slip distribution provides crucial during this earthquake (Avouac et al., 2014). From such studies,
data to better constrain earthquake mechanics (e.g. Scholz, 1982; coseismic slip distribution along the fault rupture appears usu-
King and Wesnousky, 2007), and has also strong implications on ally heterogeneous, with zones of high slip, referred as asperities,
seismic hazard assessment (e.g. Stein et al., 1997). Nowadays, re- separated by zones of low to zero slip (e.g. Peltzer et al., 2001;
cent technological advances in remote sensing measurements as Mai and Beroza, 2002; Avouac et al., 2014).
well as the development of dense and permanent geodetic net- The observed variability in earthquake fault slip distribu-
works provide a detailed analysis of surface deformation along tions can be attributed to many factors (Bizzarri, 2009) includ-
active faults. The 2013 Mw = 7.7 Balochistan earthquake is a typ- ing tectonic context, fault plane geometry (e.g. Robinson et al.,
2006), earthquake triggered processes such as fluid pressurization
(Faulkner and Rutter, 2001), inelastic deformation in the ambi-
ent rocks (e.g. Bürgmann et al., 1994; Kaneko and Fialko, 2011)
*
Corresponding author at: UMR-5243, Géosciences Montpellier Laboratory, Uni-
or interactions between nearby faults (e.g. Andrews, 1994). In-
versity of Montpellier, CC.60, 34095 Montpellier cedex 5, France.
E-mail address: yannick.caniven@gm.univ-montp2.fr (Y. Caniven). vestigations based on the analysis of earthquakes data catalog

https://doi.org/10.1016/j.epsl.2017.10.009
0012-821X/© 2017 Elsevier B.V. All rights reserved.
148 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157

Fig. 1. a) General view of the experimental set-up. The device is constituted by a computerized shear box fixed to a rigid aluminum structure. b) Mechanical and kinematical
boundary conditions. The model visco-elasto-plastic rheology is achieved by superimposing granular, elastic and viscous layers representing the upper and lower crust.
Schematic cross section of the experimental device showing its internal mechanical structure, the geometry of analog material layers and how boundary conditions (σn
and loading rate) are imposed. c) Model surface horizontal displacements (top) and cumulated shear stress (bottom). Surface displacements are quantified using sub-pixel
image correlation technique. Numerical tools are used to analyze model deformation at the surface and at depth. Examples of surface horizontal displacement measurements
acquired during a typical experiment. Each stage is separated by 6 s corresponding to 20 μm of far-field simple shear model deformation. Stick–slip behavior is observed;
instantaneous incremental fault slip events are separated by long time periods where the fault is locked and model records long wavelength elastic loading, i.e. coseismic
and interseismic stages respectively.

(e.g. Wesnousky, 2008) strongly suggest that coseismic slip dis- mic slip pattern and seismic cycle behavior. We use a scaled visco-
tribution is essentially controlled by fault strength variations. Fur- elasto-plastic multi-layered analog model for reproducing seismic
thermore, the study of specific earthquake ruptures reveal that cycles on a strike-slip fault (Caniven et al., 2015). This experimen-
accumulated stress distribution along strike is also a key pa- tal device allows to apply a uniform or a non-uniform temporally
rameter that must be taken into account (Peyrat et al., 2001; constant normal-stress (σn ) along the fault plane.
Mai and Beroza, 2002). One of the most recent examples con- The main goal of this paper is to study how spatial variations
firming these conclusions comes from the study of the 2008 Mw in normal stress on a fault can influence coseismic slip distribution
7.9 Wenchuan earthquake (Wen et al., 2012b). The simulation of and fault kinematics at seismic cycle time-scale (equivalent to a
this earthquake rupture, based on the inversion of surface geode- few months up to several thousand years in nature).
tic data, revealed that coseismic slip pattern can be explained by
spatial heterogeneities in stress loading on the fault plane inher- 2. Experimental model
ited from the interseismic period. This heterogeneous loading can
also results from spatial variability in the fault frictional behav- Experimental set-up characteristics and analog model rheolog-
ior linked to the fault rheology. Numerical modeling, based on ical properties are extensively described in a previous publication
rate-and-state friction laws (Scholz, 1998), succeed in reproduc- (Caniven et al., 2015). Hereafter, we only sum up the main points.
ing these couplings (e.g. Barbot et al., 2012). As observed in the
nature, simulated seismic ruptures appear mostly confined within 2.1. Experimental set-up and boundary conditions
strong patches (asperities) that remain locked during the interseis-
mic period. These asperities seem to play a key role not only on The experimental device consists of a rigid structure (1 m ×
the slip distribution but also in the seismic cycle behavior. The 1.5 m × 1.8 m) made of aluminum profiles supporting all mechan-
models with a single asperity embedded into a creeping zone ical and model-monitoring equipment (Fig. 1a). The main structure
produce quite regular seismic cycles, whereas multiple asperity is made of two compartments, moving in opposite direction at a
models conduct to more complex cycles. constant velocity ranging from 1 to 7 μm/s (Fig. 1b). Both com-
In the present study, we apply an original experimental ap- partments have similar sizes of 120 cm × 73 cm × 12 cm and are
proach to investigate the role of fault stress distribution on coseis- in contact along their longest dimension. They represent the two
 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157 149

compartments of a strike-slip vertical fault. Compartment displace- describes the velocity dependence of friction, and (2) the critical
ments are controlled using a laser telemeter and image monitoring slip distance D c , referred as the slip amount necessary for friction
techniques (sub-pixel spectral correlation). For all the considered to change in response to a velocity variation. This leads to the de-
experiments in this paper, the shear sense is left-lateral. Analog termination of the critical length L c of the slipping region above
materials filling the two compartments undergo, then a localized which instability occurs, i.e. the so-called nucleation size. At first
left-lateral shear deformation at a constant loading rate. The ana- order, the model scaling is considered to be correct if (a − b) is
log model material can be loaded laterally to scale σn along the conserved between model and nature (dimensionless), and if D c
fault plane (from 50 to 1000 Pa) and also to adjust the threshold and L c scale to nature using the geometric factor L ∗ .
of static and dynamic friction (Fig. 1b). For (a − b) ≥ 0, the material is “velocity-strengthening” and the
system is intrinsically stable whereas for (a − b) ≤ 0, the material
2.2. Model rheology and scaling is “velocity-weakening” and the system becomes unstable at a crit-
ical value of σn . For our experiments, (a − b) has been estimated to
The analog model rheology intends to simulate the mechani- −0.017 ± 0.005 (Caniven et al., 2015), consistent with determined
cal behavior of an idealized continental crust. With this aim, the values for rocks in the velocity-weakening domains (e.g. Scholz,
model is formed with three superimposed layers of different ana- 1998). Estimates of D c and L c and theoretical concepts are de-
log materials (Fig. 1). The basal layer is 3 cm thick and is made tailed in the provided supplementary material. We estimate D c in
of a visco-elastic silicone compound (PDMS-SGM 36, Dow Corning the model to about 30 μm (scaling to ∼7 m using L ∗ ) consistent
Ltd., viscosity about 50 kPa s at room temperature) for simulat- with natural earthquakes values ranging from decimeters to sev-
ing the viscous deformation and strain hardening behavior of the eral meters (e.g. Fukuyama and Suzuki, 2016). The critical length
lower crust (Rudolf et al., 2016). The intermediate layer is 4 to L c in the model is estimated to about 1 mm (scaling to ∼250 m
6 cm thick and it is made of a high resilience Polyurethane foam using L ∗ ), consistent with typical values for natural earthquakes in
(Young modulus = 95 ± 10 kPa, Poisson ratio = 0.06 ± 0.02, Shear the order of hectometers to kilometers (e.g, Kaneko et al., 2016).
modulus = 45 ± 5 kPa, V p = 100 m/s, V s = 68 m/s). The foam al- Following Rosenau et al. (2009) and Corbi et al. (2011) tem-
lows for the simulation of the elastic deformation characterizing poral scaling evaluation can be achieved by defining two different
the mechanical behavior of the upper crust at the seismic cycle timescales for the interseismic and the coseismic periods (Caniven
time scale (Caniven et al., 2015). The upper-most layer is 0.2 to et al., 2015). Due to slow interseismic deformation, inertia forces
0.5 cm thick and is formed by a granular material mixture (silica are negligible and viscous forces in the silicone layer, representing
powder and graphite). It represents the uppermost few kilometers the lower crust, become dominant. Using silicone SGM36 proper-
of the shallow crust where deformation is considered as essentially ties and scaling rules, we obtain an interseismic timescale of T i∗ ∼
brittle. 2.5 × 10−9 (i.e., 1 s in the model ∼10–15 yr in nature). The coseis-
The contact zone between the two foam plates constitutes the mic timescale (T c∗ ) can be estimated considering that the Froude
√
fault plane. It measures 1 m long and extends vertically 8–10 cm number should remain constant. As a result, T c∗ = L ∗ = 2 × 10−3
from the base of the two rigid aluminum profile compartments up (1 s in the model ∼500 s in nature). As proposed by Rosenau et al.
to the model surface, through the silicone, the polyurethane foam (2009), we estimate the interseismic and coseismic stage velocities
and the granular material layers. Thus, friction properties vary ver- using the ratios between L ∗ and T i∗ and L ∗ and T c∗ . We obtain for
tically along the fault plane. In the basal silicone layer, the fault interseismic: 5 mm/h in the model versus ∼27 mm/yr in nature
zone broadens into a 3 cm thick ductile shear zone, whose resis- and for coseismic 3 × 10−3 m/s in the model versus ∼1.5 m/s in
tance is shear-rate dependent. The main frictional interface is con- nature. More details concerning the model scaling can be found in
stituted by the intermediate layer along which the two foam plates Caniven et al. (2015).
slip in opposite direction. The fault plane is coated using a colored
epoxy resin to obtain a static friction coefficient μs = 0.65 ± 0.05 2.3. Kinematics monitoring, data processing and analysis
and a dynamic friction coefficient μd = 0.42 ± 0.05. At fault tips,
two low friction patches, 3 cm long, are stuck on each side of the Model surface deformation is monitored using a digital cam-
fault plane to limit boundary effects. These patches induce a local era, located at the top of the experimental device. It records a
stable sliding of the foam plates at both tips of the locked fault, high-resolution image (21 Mpixels) every 6 s (Fig. 1). These data
restricting the stick–slip events to the main central part of the ex- are processed using a sub-pixel spectral correlation algorithm to
perimental device. derive the two components of horizontal surface displacements
Extrapolation of analog modeling results to nature is based on ( X and Y , the parallel and normal directions to the fault strike).
the scaling theory elaborated by Hubbert (1937) and later refined Typical measurements accuracy and spatial resolution reach 2 μm
by Cobbold and Jackson (1992). Using analog material and nature and 5 mm, respectively leading to a total of about 20.000 indepen-
physical properties, we estimate the geometric factor (L ∗ ) to be dent measurements every 6 s. Horizontal displacement data are,
in the range between 3 × 10−6 to 5 × 10−6 (1 cm in the model then, filtered and formatted to study model surface kinematics.
∼2–3 km in nature). Model strain and stress states evolution can be analyzed in de-
The upper crust is modeled using a polyurethane foam plate tail using the large catalog of surface displacements. Note that the
which has a low density (ρ = 40 kg m−3 ) compared to upper crust vertical component of surface displacements is not measured, and
density (2700 kg m−3 ). As a result, the lithostatic stress and the cannot be used to calculate strain and stress. However, because
normal stress along the fault plane are very low, inducing pos- the fault has a pure strike-slip kinematics, amplitude of vertical
sible model scaling distortion. We compensate this limitation by displacements are estimated to be one to two orders of magni-
imposing the normal stress σn along the fault plane (∼500 Pa) tude smaller than horizontal displacements (Okada, 1985). Thus,
so that it scales to nature (∼150 MPa) using the geometric fac- we consider that plane strain conditions (2D) can be assumed and
tor (L ∗ ∼ 4 × 10−6 ). Similarly, we select a specific polyurethane the components of model surface strain ε and stress tensors σ are,
foam whose shear modulus (G = 45 ± 5 kPa) scales to nature then, derived from the horizontal surface displacements. Assuming
(G ∼ 10–30 GPa) using the same factor (L ∗ ). linear elasticity, the stress tensor σ is obtained from the Hooke’s
Similarity between model and nature in term of slip stability law:
can be established by comparing rate-and-state frictional proper-
ties (Scholz, 1998). It concerns (1) the (a − b) parameter, which σi j = 2G εi j + λϕ i j (1)
150 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157

Fig. 2. a) Model rheology and boundary conditions. b) Surface horizontal displacement amplitude and vector field showing the pre-stressing of the fault plane (up) and
calculated σ y y . c) σ y y distribution (absolute value) along the fault trace from the σ y y map in (b).

with ϕ the first invariant of strain tensor,  the Kronecher’s delta, 3. Results
G and λ, the shear modulus and Lamé coefficient, respectively. By
convention, tension and right-lateral shear stress are positive. An 3.1. Studied experiment and boundary conditions
increase of σxy is, then, associated with a drop in left-lateral stress
induced by a fault slip event. To present and discuss our results, we have selected one
In all of our experiments, after a transient time period of gen- demonstrative experiment among the 51 performed to investigate
eral elastic loading (typically after 10 mm of bulk shearing), stick– model behavior. This experiment is characterized by a constant
far field left-lateral loading rate of 20 mm/h ± 2 mm/h and a
slip behavior and seismic-cycle like phases of deformation are
non-uniform and asymmetric applied normal stress σ y y along the
observed and measured. This phase is characterized by the occur-
fault trace (Fig. 2). Normal stress distribution is maintained con-
rence of instantaneous incremental slip events separated by peri-
stant during the whole experiment duration. At a local scale, σ y y
ods (lasting from a few seconds to a few minutes) during which variations along the fault depend on irregularities of the contact
the fault appears to be partially or totally locked. Fig. 1c shows geometry between the two polyurethane foam plates induced by
a short sequence, lasting 1 min, extracted from a typical experi- industrial cutting. Such fault surfaces irregularities have amplitudes
ment monitored every 6 s. The first stage corresponds to the end of of 0–5 mm in the Y -direction over wavelengths of up to 30 cm in
the previous interseismic stage during which the fault was totally the X -direction. These oscillations induce variations of σ y y when
locked. Besides, using Eq. (1), derived cumulated σxy of this prece- the foam plates are brought into contact. The amplitude of σ y y
dent stage indicates a global loading of the model surface with a along the fault plane is set using lateral screws. Strain associated
slight concentration around the fault central part. The next stage to this pre-stressing stage is quantified by subpixel correlation of
shows the occurrence of a strong coseismic event (fault slip of two photos taken before and after putting the foam plates in con-
about 200 μm) characterized by an asymmetric distribution of slip tact (Fig. 2b). σ y y value along the surface fault trace is estimated
using the measured strain field, the foam Young modulus and the
along the rupture. It is immediately followed by a model surface
Hooke’s law.
deformation pattern interpreted as a postseismic stage (Caniven et
For the present experiment, the absolute σ y y peak value
al., 2015). Surface cumulated σxy map shows a relaxation along
reaches 650 ± 30 Pa and is located between x-coordinates x =
and in the vicinity of the rupture with a maximum stress drop lo- 600 mm and x = 750 mm where σ y y is above 550 Pa (Fig. 2b
cated in the maximum slip region. Then, the fault relocks before and 2c). A second lower peak is observed between x = 280 mm
slipping again but with a lower amplitude as shown by a moder- and x = 350 mm where σ y y exceeds 400 Pa with a maximum
ate event that occurs to the left and relaxes locally the shear stress of 430 ± 30 Pa. Following the definition proposed by Schellart
in this zone. and Rawlinson (2013), this zone of highest σ y y can be considered
 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157 151

as a “stress asperity”. Such an asperity is described as a delim- jor microquake events that generate seismic waves during experi-
ited portion of the fault plane where the mechanical coupling ments. The recurrence-time for such strong microquakes shows a
is significantly higher than in the surroundings (Kanamori, 1981; modal distribution with a mode located at 144 s and a mean value
Lay et al., 1982). of 137 s (∼1400 yrs), which is representative for 40% of them
The duration of the presented sequence is 40 min (∼24000 yrs) (Fig. 4-A). D max of SC events ranges typically between 300 μm
from which we derived and analyzed 400 successive surface dis- and 400 μm for 50% of them and between 400 and 600 μm for
placement fields. It was extracted far enough from the beginning 20% (Fig. 4-A). As detailed in a former paper (Caniven et al., 2015),
of the experiment (90 mm of cumulative displacement) to be slip gradients (γco = D max / L) of these strong events range between
sure that the model has reached its long-term evolution, i.e. the 10−4 and 10−3 which is comparable to nature.
cycle-invariance and that these measurements were acquired sig- Most of their maximum slip (D max ) is located in a restricted
nificantly after model spin-up time (Hetland and Hager, 2006). The zone extending between x-coordinates 600 mm and 800 mm along
time sampling is 6 s (∼50–100 yrs) which corresponds to ∼33 μm the fault trace, exactly in the major stress asperity region (Fig. 4-A).
of far field horizontal displacement parallel to the fault. Because Note that some of these profiles present also a second maximum
fault plane coating covers uniformly the foam sides in contact and into the region of the second stress asperity.
does not deteriorate with time, the frictional properties along the For all these major events, slip profiles along the fault exhibit
fault plane can be considered homogeneous and constant during asymmetric shapes. In average, their right portion presents steep-
the whole experiment duration. est slopes with maximum gradient reaching γco ∼ 10−3 and very
short tapers at fault tip. On the left portion, slip gradient are
3.2. Slip event classes lower by one order of amplitude γco ∼ 10−4 and are associated
with longer fault tip tapers. Although these slip profiles present
To characterize the kinematics of each fault slip events (mi- non-linear curves with multiple local maxima, at first order, half
croquakes and creep) that occurred during the studied period, profiles can be reasonably adjusted using a linear fit (Caniven et
we focused on the fault parallel component of horizontal surface al., 2015). Fault slip profiles of SC events have, at first order, similar
displacements along the fault. For each recorded measurements, shapes, meaning that their kinematical and mechanical properties
raw data were first pre-processed to extract fault slip events with remain identical on the considered time period.
amplitudes greater than the imposed far field displacement, i.e.
∼33 μm during the 6 s sampling interval. Considering the mea- 3.2.2. Postseismic slip events (PS)
surement accuracy (∼2 μm) and rare but potential variation of On a total number of 18 detected SC events, 15 were imme-
compartments relative velocities (±10 μm), we adjusted the detec- diately followed by slip events with D max > 50 μm. In a previous
tion limit to 50 μm, i.e. slightly above the critical slip distance D c . work (Caniven et al., 2015) we showed that this stage is kinemat-
Consequently, we assumed that most of the slip events associated ically and mechanically analog to the post-seismic phase following
with slow aseismic fault creep were filtered out from the dataset. major earthquakes (Pollitz et al., 2001; Perfettini and Avouac, 2007;
Data catalog and model kinematics are summed up in Fig. 3. Fig. 3a Bruhat et al., 2011; Wen et al., 2012a). Post-seismic slip inversion
shows 64 fault slip profiles (fault parallel component of horizon- on the fault plane for one of these sequences shows a down-
tal surface displacements) that occurred successively during the ward migration of slip consistent with natural observations such
studied period. The temporal distribution of fault slip events is as, for example, the post-seismic phase of the Izmit earthquake
presented in Fig. 3b using the location of maximum slip (D max ) (Reilinger et al., 2000). During this earthquake, coseismic slip was
of each event. confined to the upper 10 km of the crust then evolved in the
To constrain time, location and size of fault slip events, we next month into aseismic slip located below the seismic rupture
combine information of slip profiles along the fault (Fig. 3a) with zone.
time-position of their maximum slip (Fig. 3b) to build a syn- These PS events are characterized by surface D max ranging be-
thetic diagram illustrating fault slip dynamics (Fig. 3c). Based tween 50 μm and 400 μm, 70% of them being between 100 μm and
on this dataset, 3 classes of slip events were defined based on 300 μm (Figs. 3a and 4-B). As for SC events, most of them broke
their magnitudes, locations and fault slip properties. Class 1 cor- the whole fault but with lower slip amplitudes. Slip gradients γ co
responds to large slip events with maximum displacements (D max ) are also lower and range between 10−5 and 10−4 . Slip profiles ap-
greater than 250 μm. They are referenced as “Strong coseismic pear less asymmetric compared to SC events. Some of them have
events” (SC) (red profiles in Fig. 3a). Class 2 is constituted by multiple local maxima and flat-topped fault slip distributions are
slip events occurring immediately after class 1 events. As shown also observed (Fig. 3a).
hereafter, they correspond to postseismic fault slip triggered by Most of D max values (55%) are located into the major stress
major microquakes. They are referred as “Postseismic events” (PS) asperity zone. Note that 40% of them are also detected 200 mm
(green profiles in Fig. 3a). Finally, Class 3 concerns single coseis- leftward between the two previous defined stress asperities. PS
mic slip events of lower magnitudes. They are referenced as “Low events appear to migrate preferentially with a leftward direction
to moderate coseismic events” (LMC) (blue profiles in Fig. 3a). (Fig. 3c) toward the longest tapers of both SC slip profiles and
For each event class, maximum slip (D max ) values and locations normal stress distribution along the fault (Fig. 3a). Due to the
as well as recurrence-time T r were determined and analyzed low sampling rate of camera measurements (6 s), we cannot con-
(Fig. 4). firm that the fault slip dynamics follow the same trend. However,
as shown in Fig. 1c, this preferential leftward migration of PS
3.2.1. Strong coseismic events (SC) maximum slip reflects the progressive relaxation of shear stress
The 18 SC events that occurred during the considered period, along the whole fault through additional slips immediately fol-
have typical D max ranging between 250 μm and 700 μm (Fig. 3a lowing SC events. Note that some PS events, whose duration is
and 4-A) and generally broke the whole fault. Based on a cam- shorter than the camera sampling rate (6 s), can be mixed with
corder (25 frame/s) and acoustic sensor records, we know that the the SC measurements and are, then, not detected as PS events.
duration of these large events generally does not exceed 100 ms This bias tends to slightly increase the calculated recurrence-
(Caniven et al., 2015). Furthermore, their slip amplitude is much times.
larger than the critical slip distance D c , as required for the nucle- These PS events extend over several successive measurements
ation of fast dynamic instabilities. They can be interpreted as ma- after the coseismic rupture, typically up to 6 to 12 s (a few hun-
152 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157

Fig. 3. a) 64 fault slip profiles (fault parallel component of horizontal surface displacements) that occurred successively during the studied period and respective maximum
slip D max (colored points) for Strong Coseismic (SC) events (red), Post-Seismic (PS) events (green), Low-to-Moderate Coseismic (LMC) events (blue). Dashed black line shows
normalized σ y y distribution along the fault trace b) D max along the fault vs Time. c) Locations of D max along the fault vs Time. Red ellipses and red arrows indicate sequences
of left-ward migration of events. Dotted black circles indicate single moderate events. The white-black shaded background indicates the σ y y distribution along the fault trace
(from dashed profile) with highest compressions in white. (For interpretation of the references to color in this figure, the reader is referred to the web version of this article.)

dred of years in nature), which highlights long wavelength defor- Hartzell, 1988; Perfettini and Avouac, 2007), could also be em-
mation signal in the surface kinematics, probably mainly controlled bedded in the recorded signal, but the lack of fine seismological
by slow postseismic viscoelastic motions at depth and afterslip records and measurements of slip dynamics doesn’t allow yet to
near the surface (Caniven et al., 2015). Because of the limita- know whether such additional seismic ruptures occur during this
tion inherent in the sampling rate, real after-shocks (Mendoza and stage or not.
 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157 153

Fig. 4. Analysis of kinematical parameters for the three event classes – A) SC events; B) PS events; C) LMC single events. Distributions of 1) D max at the model surface; 2)
D max location along the fault trace. Red line shows normalized σ y y distribution along the fault trace; 3) Recurrence-time T r of slip events. Red bar indicates mean value.
(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

3.2.3. Low to moderate coseismic events (LMC) second stress asperity zone. This region corresponds also to the
This last category of events corresponds to coseismic slip events less-relaxed shear-stress zone after the SC events (Fig. 1c).
whose rupture length does not exceed the 2/3 of the model fault
length (average apparent rupture length of about ∼600 mm). On
3.3. Fault behavior
a total number of 31 events, 94% have D max < 200 μm and the
highest reaches 220 μm (Figs. 3c and 4c). 45% of the slip profiles
exhibit triangular shapes, 32% flat-topped shapes and 26% exhibit Fig. 5 shows the evolution of cumulated horizontal displace-
bell-shaped distribution. Recurrence-times range between 6 s and ment at two different sites located close to the fault trace. Each
180 s with an average value of 78 s (∼800 yrs) (Fig. 4-C). As for observation site can be considered as a continuous GPS station im-
the PS events class, a non-negligible proportion of events (14%) has plemented on one fault compartment. Site 1 and site 2 are located
short recurrence-times associated with sequences of consecutive on the major and second stress asperity, where σ y y reaches 600 Pa
events. and 400 Pa, respectively.
For LMC class, a maximum of two consecutive events was ob- Comparison between site 1 and site 2 measurements confirms
served per sequence. Usually the second event is lower than the that most of the SC events ruptured a long portion of the fault
first one and surface kinematics shows typical characteristics of and are detectable at both sites. The smaller events, such as LMC,
a post-seismic phase with a long wavelength deformation signal. generally affect only one site and are preferentially located outside
Due to the low amplitude of model surface deformation occurring of the main asperity zone.
after moderate events, it is sometimes difficult to distinguish them The interseismic periods at site 2 are characterized by a pro-
from postseismic events or other independent aseismic creeping gressive increase of cumulated displacements, meaning that the
events. That’s why we didn’t make the distinction in the analyses fault is never completely locked and that aseismic creep is occur-
and included them into the same class of LMC events. ring. At site 1, interseismic fault creep is much more limited and
Unlike the other event classes (SC and PS), only two events the fault remains locked for longer periods compared to site 2.
among 31 have D max located inside the σ y y peak zone. They ex- Site 1 clearly records stronger slip events and near complete
hibit generally flat-topped slip profiles (Fig. 3a). About ∼16% of fault locking whereas numerous smaller microquakes, as well
LMC events have their D max located to the right of the major stress as significant interseismic shallow creep events, are observed at
asperity zone and 80% are located to the left (Fig. 4-C), within the site 2.
154 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157

Fig. 5. a) Evolution of the cumulated horizontal displacement (fault-parallel component) at two different sites located close to the fault trace. The zoom on top shows both
curves over stacked, with cumulated fault displacement normalized to maximum value (Norm. Displacement). This allows to better compare the slip behavior at the two
sites. The inset right-below shows the σ y y distribution along the fault trace and locations of observation sites. b) Instantaneous horizontal displacement parallel to the fault
at the two sites.

4. Discussion heterogeneity can be explained by strength conditions of the fault

rather than by coseismic processes alone.
Compared to nature, the model is, of course, over-simplified. Rupture dynamic simulations performed by Lapusta and Liu
Consequently, the interpretation of experimental results and their (2009) are also consistent with our results. They showed that, de-
extrapolation to nature face inevitable limitations. However, as spite no long-term effect on dynamic rupture, the presence of a
demonstrated hereafter, analog modeling is a relevant approach persistent small patch of high normal compression on the fault
to better understand the physical processes and the mechanical plane influences long-term slip patterns in the model. It is shown
couplings controlling earthquake and seismic cycle dynamics (see that a high normal compressional stress patch alters the along
Rosenau et al., 2017 for a review). strike symmetry of fault properties, resulting in asymmetric fault
behavior. In absence of stress heterogeneity, aseismic events may
occur before coseismic events with variable directions of propaga-
4.1. Size of events and slip patterns tion. In the case of stress patch, these slow events become unidi-
rectional and propagate toward the stress asperity. When a slow
In the presented experiment, fault kinematics is dominated by slip reaches this patch, a seismic event nucleates because a larger
the occurrence of strong microquakes characterized by a maximum effective normal stress here implies a smaller nucleation size L c ,
coseismic slip located close to the normal-stress peak. This obser- thus favoring rupture nucleation in this asperity region. That’s why
vation is consistent with the second Amontons’s law implying that in our experiments, all the SC events localize in the major stress
frictional strength on a fault plane is proportional to the normal asperity zone. Moreover, as demonstrated in Caniven et al. (2015),
load to this plane. Therefore, slip profiles of these events share aseismic transients can occur during the interseismic phase along
similar shapes and show, at first order, similar trends compared to the fault where lower normal stress favors stable-slip and can play
the imposed normal stress distribution along the fault. a key role in the stress transfer which leads to the occurrence of
As proposed by Wesnousky (2008) for real earthquakes, coseis- SC events.
mic slip patterns in our experiments seem to be also controlled by
the variations in fault strength and accumulated stress along fault 4.2. Post-seismic slip propagation
strike. We note also that Wesnousky (2008) favors this interpreta-
tion rather than a dependence on rupture dynamic properties such The postseismic displacements observed in our experiments,
as the size of initial slip pulse (Heaton, 1990). exhibit also interesting slip patterns comparable with their equiv-
The variability of events and coseismic slip distributions along alent in nature. In our analog model, postseismic period is char-
the fault plane appear heterogeneous, similarly to what has been acterized by a non-linear time decay of fault displacement re-
widely observed for earthquakes (Wen et al., 2012b; Reilinger et lated to the viscoelastic relaxation at depth (Pollitz et al., 2001;
al., 2000; Peltzer et al., 2001). Thus, we suggest that most of this Freed and Bürgmann, 2004) and/or a deepening of slip on the
 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157 155

fault plane (Caniven et al., 2015). These features have also been al., 2011). Based on the San Andreas Fault Observatory at Depth
evidenced following the 1999 Mw 7.5 Izmit earthquake (Reilinger (SAFOD) borehole data, it has been proposed that the creeping
et al., 2000). Microquake postseismic phase in our experiments behavior could be enhanced by the presence at depth of velocity-
shows similar surface slip patterns that could be attributed to strengthening material such as serpentinite (e.g. Carpenter et al.,
aseismic creep along the fault plane at depth. Indeed, although 2011; Lockner et al., 2011). After the last (and late) 2004 earth-
most of surface maximum postseismic slips locate into the major quake, Ma et al. (2008) speculated that along strike variations
stress asperity zone, a large proportion (40%) lies between the two of fault frictional properties could induce differential interseismic
dominant stress asperities that are also the location of higher slip stress accumulation patterns on the locked portion of the SAF and
amplitude of multi-maxima SC events profiles. Such an observation at the locked-creeping transition zone favoring, then, a character-
has been evidenced for the 2001 Mw 7.8 Kokoxili earthquake for istic earthquake behavior.
which maximum postseismic slip was located in the middle of two In our experiments, interseismic cumulated shear stress distri-
relatively high coseismic slip patches (Wen et al., 2012a). These au- bution preceding each strong seismic event is correlated with the
thors suggested that after-slip could be a plausible mechanism to location of the major stress asperity and surrounding normal com-
explain the observed surface kinematics coupled with viscoelastic pression pattern (Fig. 1c, first shear stress map and Fig. 4). In the
relaxation. The applied σn thus seem to also control the location range of the applied normal stress, high normal compression ap-
of observed aseismic events and the global postseismic slip pattern pears to favor stick–slip, whereas low normal compression induces
evolution observed after strong microquakes. aseismic creep, resulting into a specific pattern; a single asperity
embedded into a creeping zone.
4.3. Isolated low to moderate events Although such a pattern for the Parkfield example is controlled
by along-fault frictional properties rather than by normal stress
At first order, the kinematics of the fault is mainly dominated variations, It generates the same mechanical behavior, i.e. reg-
by the occurrence of strong characteristic slip events. However, iso- ular seismic cycles. Furthermore, fault slip behavior along real
lated LMC single events are also frequently observed. Unlike the faults can also be directly modulated by normal stress variations
two other event classes, they occur preferentially away from the linked to spatial changes of elastic properties. This is well known
major stress asperity zone, on both sides, with a preferential loca- for subduction cases where slow slip events are recorded in re-
tion into the second stress asperity zone where significant shear gions where the fault stiffness is reduced by local high fluid pore
stress persists after the postseismic phase following SC events pressure (Leeman et al., 2016). This suggests that, in addition to
(Fig. 1c). We never observed the occurrence of LMC events across the fault frictional properties, along-fault normal stress variations
the major stress asperity zone. A hypothesis has been already pro- could play a significant role in the regularity of the seismic cycle
posed to explain the occurrence of such isolated events on active as observed in our experiments.
faults, especially along the North Anatolian fault (NAF). Pondard et In the case of uniform loading, the occurrence of earthquakes
al. (2007) proposed that the occurrence of isolated events can be seems to mostly depend on the location of previous seismic events
explained by a non-uniform stress relief that persists after the oc- that redistribute the stresses along the fault (e.g. King et al., 1994;
currence of major earthquakes sequences. Stein et al., 1997; Pondard et al., 2007). Ruptures occur, then, at
Therefore, initial stress distribution along the fault appears to variable locations over the long-term system evolution and the
control not only the slip distribution of strong earthquakes but also seismic cycle appears less regular.
the location and size of all slip events.
In our experiments, such a non-uniform cumulated shear stress 4.5. Clustering
results on the one hand from the imposed normal stress variation
along the fault and, on the other hand, from the occurrence of SC Remarkably, the propagation of successive individual coseismic
events and associated postseismic slip (Fig. 1c). slip events rupturing progressively the whole fault was never ob-
Strong events nucleate where the fault is the strongest and served in the presented experiment. However, microquakes “clus-
isolated events occur where residual stresses persist, with prefer- tering” is frequent in other experiments where the normal stress
ential locations around other secondary preexisting stress hetero- distribution along the fault is more homogeneous (Caniven et al.,
geneities. 2015), i.e. without well-individualized stress asperity (Fig. S1). In
The size of isolated events into the second stress asperity de- these experiments, typical sequences of three to four successive
pends on the local slip amplitude of the precedent major coseismic microquakes rupturing the whole fault in a short time period
event. After SC events, characterized by multi-maxima slip profiles, (18 s, equal to ∼200 yr) after a long quiescence interseismic period
the next isolated events have a lower magnitude because most of (180 s, equal to ∼2000 yr) are often observed. During a clustering
the shear stress has been relaxed. sequence, microquakes migration can occur unidirectionally, bidi-
rectionally or even randomly and so, for the same experiment.
4.4. Seismic cycle regularity This mechanical behavior has been already evidenced in sev-
eral natural cases such as the NAF which experienced in the
Experimental seismic cycles appear mostly regular. Strong mi- last centuries clustering sequences with both uni-directional and
croquakes occur with a relatively constant return period, size and variable-directional earthquakes migrations (e.g. King et al., 1994;
location along the fault, consistent with the definition of a char- Stein et al., 1997; Nalbant et al., 1998; Pondard et al., 2007).
acteristic earthquake fault behavior (Schwartz and Coppersmith, Pondard et al. (2007) proposed that clustering sequences are fa-
1984). vored when stress loading is uniform along the fault as it is the
Regular seismic cycles have been also evidenced in nature for case along the most linear fault segments of the NAF. Indeed,
example along the San Andreas fault (SAF) near the locality of as noticed by Pondard et al. (2007), propagating earthquake se-
Parkfield (Bakun and Lindh, 1985). Here, between 1857 and 1966, quences do not occur every seismic cycle along a fault system,
a sequence of six moderate earthquakes occurred with a quite suggesting that the loading has to be in a particular state of stress
regular recurrence-time (∼22 yrs), size (M ∼ 6) and location. Re- (close to failure and uniform) all along a considered fault segment
markably, all these events occurred close to the transition be- to trigger propagating earthquake sequences. That’s, most probably,
tween the creeping central part of the SAF, northwest of Park- also the reason why such sequences are observed, during uniform
field, and the locked portion of the SAF in the southeast (Titus et applied normal stress analog experiments (Caniven et al., 2015).
156 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157

5. Conclusion Bizzarri, A., 2009. What does control earthquake ruptures and dynamic faulting?
A review of different competing mechanisms. In: Vinciguerra, S., Bernabé, Y.
(Eds.), Rock Physics and Natural Hazards. Birkhäuser, Basel, pp. 741–776.
To investigate the role of fault stress heterogeneity on coseis-
Bruhat, L., Barbot, S., Avouac, J.-P., 2011. Evidence for postseismic deformation of the
mic and long term fault slip kinematics, we used a scaled visco- lower crust following the 2004 Mw6.0 Parkfield earthquake. J. Geophys. Res. 116
elasto-plastic multi-layered analog model, which reproduces scaled (B8), B08401. http://dx.doi.org/10.1029/2010JB008073.
microquakes and seismic cycles on an analog strike-slip fault. Ex- Bürgmann, R., Pollard, D.D., Martel, S.J., 1994. Slip distributions on faults: effects
of stress gradients, inelastic deformation, heterogeneous host-rock stiffness, and
perimental results show that major coseismic slip events system-
fault interaction. J. Struct. Geol. 16 (12), 1675–1690. http://dx.doi.org/10.1016/
atically locate into the major stress asperity zone along the fault 0191-8141(94)90134-1.
with maximum slips located close to the normal stress peak. Inter- Caniven, Y., Dominguez, S., Soliva, R., Cattin, R., Peyret, M., Marchandon, M., Romano,
estingly, coseismic slip profiles of major slip events show similar C., Strak, V., 2015. A new multilayered visco-elasto-plastic experimental model
trends compared to the imposed normal stress distribution along to study strike-slip fault seismic cycle. Tectonics 34, 232–264. http://dx.doi.org/
10.1002/2014TC003701.
the fault. Carpenter, B.M., Marone, C., Saffer, D.M., 2011. Weakness of the San Andreas Fault
The stress distribution along the fault appears to control not revealed by samples from the active fault zone. Nat. Geosci. 4 (4), 251–254.
only strong earthquake properties but also the variability of slip http://dx.doi.org/10.1038/ngeo1089.
events that occur during the experiments. Indeed, postseismic slip, Cobbold, P.R., Jackson, M.P.A., 1992. Gum rosin (colophony): a suitable material for
thermomechanical modelling of the lithosphere. Tectonophysics 210, 255–271.
following strong microquakes, tend to propagate from the major
http://dx.doi.org/10.1016/0040-1951(92)90325-Z.
stress asperity toward both fault tips and low to moderate isolated Corbi, F., Funiciello, F., Faccenna, C., Ranalli, G., Heuret, A., 2011. Seismic variability of
events occur generally outside the main stress asperity zone, in re- subduction thrust faults: insights from laboratory models. J. Geophys. Res. 116,
gions where residual stresses persist around secondary order stress B06304. http://dx.doi.org/10.1029/2010JB007993.
Faulkner, D.R., Rutter, E.H., 2001. Can the maintenance of overpressured flu-
asperities.
ids in large strike-slip fault zones explain their apparent weakness? Ge-
Finally, in cases where fault slip stability conditions are modu- ology 29 (6), 503–506. http://dx.doi.org/10.1130/0091-7613(2001)029<0503:
lated by σn distribution, experimental results reveal that the σn CTMOOF>2.0.CO;2.
variations along strike controls the ability of the fault to gen- Freed, A.M., Bürgmann, R., 2004. Evidence of power-law flow in the Mojave desert
erate irregular or regular seismic cycles and produce clustering mantle. Nature 430 (6999), 548–551. http://dx.doi.org/10.1038/nature02784.
Fukuyama, E., Suzuki, W., 2016. Near-fault deformation and Dc during the 2016
sequences. A non-uniform σn characterized by a high amplitude Mw7.1 Kumamoto earthquake. Earth Planets Space 68, 194. http://dx.doi.org/
single stress asperity favors the occurrence of strong characteristic 10.1186/s40623-016-0570-6.
microquakes sharing similar location, magnitudes and return peri- Heaton, T.H., 1990. Evidence for and implications of self-healing pulses of slip
ods. In that case, the seismic cycle is time- and slip-predictable. in earthquake rupture. Phys. Earth Planet. Inter. 64, 1–20. http://dx.doi.org/
10.1016/0031-9201(90)90002-F.
For more homogeneous σn , the seismic cycles appear more ir-
Hetland, E.A., Hager, B.H., 2006. Interseismic strain accumulation: spin-up, cycle in-
regular and mostly unpredictable using cumulated slip and return variance, and irregular rupture sequences. Geochem. Geophys. Geosyst. 7 (5).
period alone. Microquakes magnitudes and return periods extend http://dx.doi.org/10.1029/2005GC001087.
over large ranges and episodic earthquake clustering is also ob- Hubbert, M.K., 1937. Theory of scale models as applied to the study of geologic
structures. Bull. Geol. Soc. Am. 48, 1459–1520.
served.
Kanamori, H., 1981. The nature of seismicity patterns before large earthquakes. In:
Our results reinforce the hypothesis that interseismic and co- Simpson, D.W., Richards, P.G. (Eds.), Earthquake Predictions, an International Re-
seismic slip distribution may provide relevant informations on the view. In: Maurice Ewing Series, vol. 4. American Geophysical Union, Washington,
absolute fault stress state and associated stress heterogeneities im- DC, pp. 1–19.
proving, then, seismic hazard assessment. Kaneko, Y., Fialko, Y., 2011. Shallow slip deficit due to large strike-slip earthquakes
in dynamic rupture simulations with elasto-plastic off-fault response. Geophys.
J. Int. 186, 1389–1403.
Acknowledgements Kaneko, Y., Nielsen, S.B., Carpenter, B.M., 2016. The onset of laboratory earthquakes
explained by nucleating rupture on a rate-and-state fault. J. Geophys. Res., Solid
This project has been funded by CNRS-INSU and University of Earth 121, 6071–6091. http://dx.doi.org/10.1002/2016JB013143.
King, G., Stein, R., Lin, J., 1994. Static stress changes and the triggering of earth-
Montpellier grants. We thank the editor Rebecca Bendick, the re- quakes. Bull. Seismol. Soc. Am. 84 (3), 935–953.
viewer Michele Cooke and the two other anonymous reviewers for King, G.C.P., Wesnousky, S.G., 2007. Scaling of fault parameters for continental strike-
helping to improve the manuscript, C. Romano for technical assis- slip earthquakes. Bull. Seismol. Soc. Am. 97 (6), 1833–1840. http://dx.doi.org/
tance, S. Nielsen for very useful remarks and suggestions, J. Chéry, 10.1785/0120070048.
Lapusta, N., Liu, Y., 2009. Three-dimensional boundary integral modeling of sponta-
S. Preuss, A.-M. and J.-P. Guyot for fruitful discussions. Data sup-
neous earthquake sequences and aseismic slip. J. Geophys. Res., Solid Earth 114,
porting Fig. 3 and Fig. 5 are available in supplementary material B09303.
(Table_Fig3.xlsx and Table_Fig5.xlsx, respectively). Lay, T., Kanamori, H., Ruff, L., 1982. The asperity model and the nature of large
subduction zone earthquakes. Earthq. Predict. Res. 1 (1), 3–71.
Leeman, J.R., Saffer, D.M., Scuderi, M.M., Marone, C., 2016. Laboratory observations
Appendix A. Supplementary material
of slow earthquakes and the spectrum of tectonic fault slip modes. Nat. Com-
mun. 7. http://dx.doi.org/10.1038/ncomms11104.
Supplementary material related to this article can be found on- Lockner, D.A., Morrow, C., Moore, D., Hickman, S., 2011. Low strength of deep
line at https://doi.org/10.1016/j.epsl.2017.10.009. San Andreas fault gouge from SAFOD core. Nature 472 (7341), 82–85. http://
dx.doi.org/10.1038/nature09927.
Ma, S., Custódio, S., Archuleta, R.J., Liu, P., 2008. Dynamic modeling of the 2004 Mw
References 6.0 Parkfield, California, earthquake. J. Geophys. Res. 113 (B2), B02301. http://
dx.doi.org/10.1029/2007JB005216.
Andrews, D.J., 1994. Fault geometry and earthquake mechanics. Ann. Geophys. 37 Mai, P.M., Beroza, G.C., 2002. A spatial random field model to characterize com-
(6). http://dx.doi.org/10.4401/ag-4136. plexity in earthquake slip. J. Geophys. Res. 107 (B11), 2308. http://dx.doi.org/
Avouac, J.-P., Ayoub, F., Wei, S., Ampuero, J.-P., Meng, L., Leprince, S., Jolivet, R., 10.1029/2001JB000588.
Duputel, Z., Helmberger, D., 2014. The 2013, Mw 7.7 Balochistan earthquake, Mendoza, C., Hartzell, S.H., 1988. Aftershock patterns and main shock faulting. Bull.
energetic strike-slip reactivation of a thrust fault. Earth Planet. Sci. Lett. 391, Seismol. Soc. Am. 78 (4), 1438–1449.
128–134. http://dx.doi.org/10.1016/j.epsl.2014.01.036. Nalbant, S.S., Hubert, A., King, G.C.P., 1998. Stress coupling between earthquakes
Bakun, W.H., Lindh, A.G., 1985. The Parkfield, California, earthquake prediction ex- in northwest Turkey and the north Aegean Sea. J. Geophys. Res. 103 (B10),
periment. Science 229 (4714), 619–624. http://dx.doi.org/10.1126/science.229. 24469–24486. http://dx.doi.org/10.1029/98JB01491.
4714.619. Okada, Y., 1985. Surface deformation due to shear and tensile faults in a half-space.
Barbot, S., Lapusta, N., Avouac, J.-P., 2012. Under the hood of the Earthquake ma- Bull. Seismol. Soc. Am. 75 (4), 1135–1154.
chine: toward predictive modeling of the seismic cycle. Science 336, 707–710. Peltzer, G., Crampé, F., Rosen, P., 2001. The Mw 7.1, Hector Mine, California earth-
http://dx.doi.org/10.1126/science.1218796. quake: surface rupture, surface displacement field, and fault slip solution from
 Y. Caniven et al. / Earth and Planetary Science Letters 480 (2017) 147–157 157

ERS SAR data. C. R. Acad. Sci., Ser. IIA, Earth Planet. Sci. 333, 545–555. http:// lithospheric deformation. Tectonophysics 684, 12–22. http://dx.doi.org/10.1016/
dx.doi.org/10.1016/S1251-8050(01)01658-5. j.tecto.2015.11.028.
Perfettini, H., Avouac, J.-P., 2007. Modeling afterslip and aftershocks following Schellart, W.P., Rawlinson, N., 2013. Global correlations between maximum mag-
the 1992 Landers earthquake. J. Geophys. Res. 112 (B7). http://dx.doi.org/ nitudes of subduction zone interface thrust earthquakes and physical parame-
10.1029/2006JB004399. ters of subduction zones. Phys. Earth Planet. Inter. 225, 41–67. http://dx.doi.org/
Peyrat, S., Olsen, K., Madariaga, R., 2001. Dynamic modeling of the 1992 Lan- 10.1016/j.pepi.2013.10.001.
ders earthquake. J. Geophys. Res., Solid Earth 106 (B11), 26467–26482. http:// Scholz, C.H., 1982. Scaling laws for large earthquakes: consequences for physical
dx.doi.org/10.1029/2001JB000205. models. Bull. Seismol. Soc. Am. 72 (1), 1–14.
Pollitz, F.F., Wicks, C., Thatcher, W., 2001. Mantle flow beneath a continental strike- Scholz, C.H., 1998. Earthquakes and friction laws. Nature 391, 37–42. http://
slip fault: postseismic deformation after the 1999 hector mine earthquake. Sci- dx.doi.org/10.1038/34097.
ence 293 (5536), 1814–1818. http://dx.doi.org/10.1126/science.1061361. Schwartz, D.P., Coppersmith, K.J., 1984. Fault behavior and characteristic earth-
Pondard, N., Armijo, R., King, G.C.P., Meyer, B., Flerit, F., 2007. Fault interactions quakes: examples from the Wasatch and San Andreas fault zones. J. Geophys.
in the Sea of Marmara pull-apart (North Anatolian Fault): earthquake cluster- Res. 89 (B7), 5681–5698. http://dx.doi.org/10.1029/JB089iB07p05681.
ing and propagating earthquake sequences. Geophys. J. Int. 171 (3), 1185–1197. Stein, R.S., Barka, A.A., Dieterich, J.H., 1997. Progressive failure on the North Ana-
http://dx.doi.org/10.1111/j.1365-246X.2007.03580.x. tolian fault since 1939 by earthquake stress triggering. Geophys. J. Int. 128 (3),
Reilinger, R.E., et al., 2000. Coseismic and postseismic fault slip for the 17 August 594–604. http://dx.doi.org/10.1111/j.1365-246X.1997.tb05321.x.
1999, M = 7.5, Izmit, Turkey earthquake. Science 289 (5484), 1519. http:// Titus, S.J., Dyson, M., DeMets, C., Tikoff, B., Rolandone, F., Bürgmann, R., 2011. Ge-
dx.doi.org/10.1126/science.289.5484.1519. ologic versus geodetic deformation adjacent to the San Andreas fault, central
Robinson, D.P., Brough, C., Das, S., 2006. The Mw 7.8, 2001 Kunlunshan earthquake: California. Geol. Soc. Am. Bull. 123, 794–820. http://dx.doi.org/10.1130/B30150.1.
extreme rupture speed variability and effect of fault geometry. J. Geophys. Wen, Y., Li, Z., Xu, C., Ryder, I., Bürgmann, R., 2012a. Postseismic motion after the
Res. 111 (B8). http://dx.doi.org/10.1029/2005JB004137. 2001 Mw 7.8 Kokoxili earthquake in Tibet observed by InSAR time series. J. Geo-
Rosenau, M., Corbi, F., Dominguez, S., 2017. Analogue earthquakes and seismic cy- phys. Res., Solid Earth 117, B08405. http://dx.doi.org/10.1029/2011JB009043.
cles: experimental modeling across timescales. Solid Earth 8, 597–635. http:// Wen, Y.-Y., Oglesby, D.D., Duan, B., Ma, K.-F., 2012b. Dynamic rupture simulation of
dx.doi.org/10.5194/se-8-597-2017. the 2008 Mw 7.9 Wenchuan Earthquake with heterogeneous initial stress. Bull.
Rosenau, M., Lohrmann, J., Oncken, O., 2009. Shocks in a box: an analogue model Seismol. Soc. Am. 102 (4), 1892–1898. http://dx.doi.org/10.1785/0120110153.
of subduction earthquake cycles with application to seismotectonic forearc Wesnousky, S.G., 2008. Displacement and geometrical characteristics of earthquake
evolution. J. Geophys. Res., Solid Earth 114, B01409. http://dx.doi.org/10.1029/ surface ruptures: issues and implications for seismic-hazard analysis and the
2008JB005665. process of earthquake rupture. Bull. Seismol. Soc. Am. 98 (4), 1609–1632. http://
Rudolf, M., Boutelier, D., Rosenau, M., Schreurs, G., Oncken, O., 2016. Rheological dx.doi.org/10.1785/0120070111.
benchmark of silicone oils used for analog modeling of short- and long-term