Articulo de Revision

Physics of the Earth and Planetary Interiors 225 (2013) 41–67
Contents lists available at ScienceDirect
Physics of the Earth and Planetary Interiors

journal homepage: www.elsevier.com/locate/pepi
Global correlations between maximum magnitudes of subduction zone

interface thrust earthquakes and physical parameters of subduction
zones
W.P. Schellart a,⇑, N. Rawlinson b
a
School of Geosciences, Monash University, Melbourne, VIC 3800, Australia
b
Research School of Earth Sciences, The Australian National University, Canberra, ACT 0200, Australia
a r t i c l e i n f o a b s t r a c t
Article history: The maximum earthquake magnitude recorded for subduction zone plate boundaries varies considerably
Received 31 December 2012 on Earth, with some subduction zone segments producing giant subduction zone thrust earthquakes (e.g.
Received in revised form 13 August 2013 Chile, Alaska, Sumatra–Andaman, Japan) and others producing relatively small earthquakes (e.g. Mariana,
Accepted 2 October 2013
Scotia). Here we show how such variability might depend on various subduction zone parameters. We
Available online 14 October 2013
Edited by G. Helffrich
present 24 physical parameters that characterize these subduction zones in terms of their geometry,
kinematics, geology and dynamics. We have investigated correlations between these parameters and
the maximum recorded moment magnitude (MW) for subduction zone segments in the period 1900–June
Keywords:
Earthquake
2012. The investigations were done for one dataset using a geological subduction zone segmentation (44
Moment magnitude segments) and for two datasets (rupture zone dataset and epicenter dataset) using a 200 km segmenta-
Subduction tion (241 segments). All linear correlations for the rupture zone dataset and the epicenter dataset
Stress (|R| = 0.00–0.30) and for the geological dataset (|R| = 0.02–0.51) are negligible-low, indicating that even
Rupture for the highest correlation the best-fit regression line can only explain 26% of the variance. A comparative
Asperity investigation of the observed ranges of the physical parameters for subduction segments with MW > 8.5
and the observed ranges for all subduction segments gives more useful insight into the spatial distribu-
tion of giant subduction thrust earthquakes. For segments with MW > 8.5 distinct (narrow) ranges are
observed for several parameters, most notably the trench-normal overriding plate deformation rate
(vOPD\, i.e. the relative velocity between forearc and stable far-field backarc), trench-normal absolute
trench rollback velocity (vT\), subduction partitioning ratio (vSP\/vS\, the fraction of the subduction
velocity that is accommodated by subducting plate motion), subduction thrust dip angle (dST), subduction
thrust curvature (CST), and trench curvature angle (aT). The results indicate that MW > 8.5 subduction
earthquakes occur for rapidly shortening to slowly extending overriding plates (3.0 6 vOPD\ 6 2.3 cm/
yr), slow trench velocities (2.9 6 vT\ 6 2.8 cm/yr), moderate to high subduction partitioning ratios
(vSP\/vS\ 6 0.3–1.4), low subduction thrust dip angles (dST 6 30°), low subduction thrust curvature
(CST 6 2.0 1013 m2) and low trench curvature angles (6.3° 6 aT 6 9.8°). Epicenters of giant earth-
quakes with MW > 8.5 only occur at trench segments bordering overriding plates that experience short-
ening or are neutral (vOPD\ 6 0), suggesting that such earthquakes initiate at mechanically highly coupled
segments of the subduction zone interface that have a relatively high normal stress (deviatoric compres-
sion) on the interface (i.e. a normal stress asperity). Notably, for the three largest recorded earthquakes
(Chile 1960, Alaska 1964, Sumatra–Andaman 2004) the earthquake rupture propagated from a zone of
compressive deviatoric normal stress on the subduction zone interface to a region of lower normal stress
(neutral or deviatoric tension). Stress asperities should be seen separately from frictional asperities that
result from a variation in friction coefficient along the subduction zone interface. We have developed a
global map in which individual subduction zone segments have been ranked in terms of their predicted
capability of generating a giant subduction zone earthquake (MW > 8.5) using the six most indicative sub-
duction zone parameters (vOPD\, vT\, vSP\/vS\, dST, CST and aT). We identify a number of subduction zones
and segments that rank highly, which implies a capability to generate MW > 8.5 earthquakes. These
⇑ Corresponding author. Tel.: +61 3 9905 1782; fax: +61 3 9905 4903.
E-mail address: wouter.schellart@monash.edu (W.P. Schellart).
0031-9201 Ó 2013 The Authors. Published by Elsevier B.V. Open access under CC BY-NC-SA license.
http://dx.doi.org/10.1016/j.pepi.2013.10.001
42 W.P. Schellart, N. Rawlinson / Physics of the Earth and Planetary Interiors 225 (2013) 41–67
include Sunda, North Sulawesi, Hikurangi, Nankai-northern Ryukyu, Kamchatka-Kuril-Japan, Aleutians-

Alaska, Cascadia, Mexico-Central America, South America, Lesser Antilles, western Hellenic and Makran.
Several subduction segments have a low score, most notably Scotia, New Hebrides and Mariana.
Ó 2013 The Authors. Published by Elsevier B.V. Open access under CC BY-NC-SA license.
1. Introduction 241 trench segments, each with a 200 km trench-parallel extent.

Such segmentation into equal-length subduction segments gives
At subduction zones oceanic lithosphere is recycled back into equal weighing to each segment in the statistical analysis. For com-
the Earth’s mantle. The process of subduction is largely driven by pleteness, we also make such investigations using a geological sub-
subducted slabs of oceanic lithosphere, which are denser than duction zone segmentation (total of 44 segments for 23 subduction
the ambient mantle and are thus pulled downward by gravity (Els- zones), which is more in accordance with the previous works cited
asser, 1971; Forsyth and Uyeda, 1975; Hager, 1984; Davies and above. The 24 parameters are related to subduction zone geometry,
Richards, 1992). The potential energy that is released during sink- kinematics, dynamics and geology. Our work shows that all the
ing is used primarily to drive flow in the mantle, to move and de- parameters have low or negligible correlations with MW that are,
form the tectonic plates, and to deform the slab. Part of this with the exception of one, all statistically insignificant at 95% con-
potential energy is also used to overcome resistance at the subduc- fidence level. Nevertheless, it will be demonstrated that very large
tion zone fault plate boundary, where part of the energy is released subduction thrust earthquakes (MW > 8.5) have only been observed
during interplate subduction zone thrust earthquakes. under specific physical conditions with relatively narrow ranges
Since the advent of plate tectonic theory it was recognized that for overriding plate deformation rate, trench migration velocity,
subduction zones differ in many aspects that relate to their geom- subduction partitioning, subduction thrust dip angle, trench curva-
etry, geology, physics and chemistry. At different subduction zones ture angle and subduction thrust curvature. The relevance of these
around the globe one might find differences in the age of the physical conditions can be explained in the framework of the phys-
downgoing plate, nature of the overriding plate (continental/oce- ical parameters that quantify MW. These findings provide new
anic), overriding plate topography, overriding plate strain (exten- understanding as to why certain subduction zone segments have
sion/shortening), trench kinematics, subduction rate, subduction produced MW > 8.5 earthquakes, which ones have the potential to
accretion/erosion rate, arc volcanism, slab dip angle, slab length, produce them in the future, and which ones are not likely to pro-
slab depth and trench curvature (e.g. Karig et al., 1976; Molnar duce them in the future. The findings also provide new under-
and Atwater, 1978; Jarrard, 1986; Gudmundsson and Sambridge, standing as to the occurrence and lateral rupture propagation of
1998; Clift and Vannucchi, 2004; Heuret and Lallemand, 2005; the three largest recorded earthquakes on Earth, namely the
Schellart, 2008). Similarly, it has been recognized that different 1960 MW 9.5 Chile earthquake, the 1964 MW 9.2 Alaska earthquake
subduction zones show differences in seismic behavior (e.g. Uyeda and the 2004 MW 9.1–9.3 Sumatra–Andaman earthquake.
and Kanamori, 1979; Ruff and Kanamori, 1980; Peterson and Seno,
1984; Ruff, 1989; Pacheco et al., 1993; Stein and Okal, 2007). For
example, several subduction zone segments have produced excep- 2. Methods
tionally large earthquakes in the last 70 yr with moment magni-
tude MW P 9.0, e.g. Alaska, Chile, Sumatra and Japan, while others 2.1. Subduction zone parameters
have not, e.g. Scotia, New Hebrides and Mariana (Fig. 1). This could
potentially be related to the relatively short period of global instru- In this paper we investigate the correlation between the maxi-
mental observations (McCaffrey, 1997, 2008; Stein and Okal, 2007), mum moment magnitude (MW) for subduction zone interplate
but it is also possible that there are essential physical ingredients thrust earthquakes and 24 physical parameters of subduction zone
that subduction zones require to be capable of producing giant characteristics. We have investigated 23 mature subduction zones
earthquakes. in terms of subduction earthquakes and values for the 24 parame-
Numerous previous works have investigated the potential ters. For several subduction zones, including Cyprus, Betic-Rif, Ven-
dependence between subduction zone thrust earthquake magni- ezuela and South Shetland, the Wadati–Benioff zone is not
tude and various subduction zone parameters, including subduct- accurately defined and/or subduction zone interface thrust earth-
ing plate age, subduction rate, sediment subduction, downdip quakes have not been recorded or could not be identified with con-
extent of seismogenic zone, forearc structure, overriding plate fidence due to uncertainty in the subduction zone thrust geometry.
velocity and overriding plate stress regime (e.g. Kelleher et al., This leaves us with 19 subduction zones for which the magnitudes,
1974; Uyeda and Kanamori, 1979; Ruff and Kanamori, 1980; Peter- velocities, rates and values for the parameters were calculated
son and Seno, 1984; Jarrard, 1986; Ruff, 1989; Pacheco et al., 1993; (Fig. 1).
McCaffrey, 1993, 1997; Scholz and Campos, 1995; Llenos and The correlations have been investigated using two different ap-
McGuire, 2007; Stein and Okal, 2007; Heuret et al., 2011). In these proaches that differ in the way that the 19 active subduction zones
previous works data are plotted for somewhat subjectively defined have been segmented. In one approach (referred to as the geolog-
subduction zone segments, where the limits of such segments have ical approach), the 19 subduction zones were divided into subduc-
some geological/structural/geometrical basis (e.g. aseismic ridge tion zone segments based in particular on the geometrical
subduction, cusp, overriding plate nature) or can be somewhat characteristics of trench curvature (i.e. arcs), the nature of the
arbitrary (such as for several South American segments). It is clear overriding plate (continental or oceanic) or the presence of aseis-
that the statistical correlation analyses performed in such studies mic ridges/plateaus at the trench, resulting in a total of 40 seg-
are influenced by the choices of subduction zone segmentation. ments. Narrow subduction zones (e.g. Scotia) are mostly
In this paper we present a global investigation to test the represented by one data-point, while wide subduction zones (e.g.
dependence of the maximum subduction zone interplate thrust South America) are divided into 2–6 segments. In the other ap-
earthquake moment magnitude (MW) on 24 subduction zone proach (in our view physically the most meaningful), each of the
parameters. We test such dependence for all active subduction 19 subduction zones was divided into individual trench segments
zones on Earth (23), which have been segmented into a total of with a length of 200 km, resulting in a total of 228 subduction
W.P. Schellart, N. Rawlinson / Physics of the Earth and Planetary Interiors 225 (2013) 41–67 43
Fig. 1. Global map showing the location of the active subduction zones. Subduction zones have been divided into 200 km segments as indicated by the colored line
segments. Color indicates the maximum subduction zone thrust earthquake recorded in that segment in the period 1900–June 2012. For a large number of segments a focal
mechanism has been plotted as obtained from the GCMT catalog or from published material (see Section 2). The map also shows the trench-parallel rupture extent of the
largest earthquakes (MW > 8.5) whose rupture extent overlaps with multiple trench segments (thick black lines for recorded earthquakes since 1900 and thick grey lines for
four ‘‘historic’’ earthquakes before 1900). Numbers indicate the MW value, while the year of the earthquake is in brackets (recorded earthquakes in bold and ‘‘historic’’
earthquakes in bold italic). Subduction zone segments: Ad – Andaman; Ak – Alaska; Am – Central America; An – Lesser Antilles; At – Aleutian; Be – Betic-Rif; Bl – Bolivia; Br –
New Britain; C-Ch – Central Chile; Cb – Calabria; Co – Colombia; Cr – San Cristobal; Cs – Cascadia; Cy – Cyprus; Ha – Halmahera; Hk – Hikurangi; Hl – Hellenic; Iz – Izu-Bonin;
Jp – Japan; Jv – Java; Ka – Kamchatka; Ke – Kermadec; Ku – Kuril; Me – Mexico; Mk – Makran; Mn – Manila; Mr – Mariana; N-Ch – North Chile; N-Hb – North New Hebrides;
N-Pe – North Peru; Na – Nankai; Pr – Puerto Rico; Pu – Puysegur; Ry – Ryukyu; S-Ch – South Chile; S-Hb – South New Hebrides; S-Pe – South Peru; Sa – Sangihe; Sc – Scotia;
Sh – South Shetland; Sl – North Sulawesi; Sm – Sumatra; To – Tonga; Ve – Venezuela.
segments. We emphasize that our rationale for the 200 km seg- relation coefficients (R) for parameter sets obtained from least
mentation is not based on the segment size of the different earth- squares linear regression analysis. Parameters with the highest
quake ruptures (which is highly variable). Our 200 km interdependence include vOPD\ and OPSC (R = 0.70), vOPD\ and
segmentation is done to properly characterize the trench-parallel vT\ (R = 0.66), vS\ and vC\ (R = 0.56), dST and dS (R = 0.62), vA\
variability in magnitude of each of the physical parameters for and TTS (R = 0.66), vA\ and TSS (R = 0.55), TTS and TSS (R = 0.89), CT
the active subduction zones on Earth. Most of the 24 physical and CST (R = 0.94), CT and |aT| (R = 0.91), CST and |aT| (R = 0.80), LUMS
parameters are not constant along individual subduction zones, and DUMS (R = 0.88), and LUMS and FBu (R = 0.76). For completeness
nor for individual arc-shaped segments of individual subduction we present the results of all 24 parameters in Figs. 3 and 4.
zones. A geological segmentation does not capture the physical For all the parameters that are velocities and rates (vOPD\, vSP\,
reality of the trench-parallel variability of individual parameters, vOP\, vT\, vS\, vC\, vA\) the trench-normal component was calcu-
while our 200 km segmentation does. Considering that we want lated. From all the 24 parameters only vOP\, vSP\, vT\ and vSP\/
to give equal weighing to all of our subduction zone segments vS\ are dependent on the choice of global reference frame. Here
(such that our statistical investigations are justified), it is most rea- we use the Indo-Atlantic moving hotspot reference frame from
sonable to make them all of equal length. We have chosen a O’Neill et al. (2005) as our preferred frame of reference, because
200 km segmentation such that the segments are not larger than in this reference frame subduction kinematics is most in agree-
the smallest subduction zones, such as Betic-Rif with 250 km ment with observed upper mantle slab structure and lower mantle
and Calabria with 300 km. Furthermore, from a practical point high-velocity anomalies (Schellart, 2011; Schellart and Spakman,
of view, they should be larger than the thickness of the oldest oce- 2012), and it fits best with plate kinematics and energy minimiza-
anic lithosphere, which is 95–106 km thick (Stein and Stein, tion arguments (Schellart et al., 2008). Furthermore, global plate
1992; McKenzie et al., 2005), because trench segments that are motion studies indicate that plate motions in Indo-Atlantic hotspot
smaller than the thickness of the subducting oceanic lithosphere reference frames are generally in best agreement with the global
will show very similar values compared to neighboring trench seg- asthenosphere anisotropy pattern below the interior of the plates
ments, and with a smaller trench segment size a larger number of (Kreemer, 2009; Conrad and Behn, 2010). We also briefly discuss
segments will have no data for MW. results using the Pacific fixed hotspot reference frame from Gripp
The 24 physical parameters that have been investigated are and Gordon (2002). We note that the Pacific hotspot frame gives
listed in Table 1 and most are illustrated in Fig. 2. They are meant significantly different plate and trench velocities at individual sub-
to cover the broad range of geometric, kinematic, dynamic and duction zones (see Schellart et al., 2008), but the parameter corre-
geological parameters of subduction zones. Because of this several lations are low as for the Indo-Atlantic hotspot frame, while the
parameters are not fully independent. This is indicated by the cor- ranges are generally larger (see Table 2). The hotspot reference
Table 1
24 Subduction zone parameters.
Parameter Units Explanation

vOPD\ [cm/yr] Trench-normal overriding plate deformation rate (Fig. 2a) (extension/spreading is positive, shortening is negative); Most rates were calculated
using geodetically defined Euler parameters as summarized in Schellart (2008)
OPSC Overriding plate strain class (Fig. 2a) (based on Jarrard, 1986): 3 = highly compressive; 2 = intermediately compressive; -1 = mildly
compressive; 0 = neutral; 1 = mildly extensional; 2 = moderately to highly extensional; 3 = spreading; Data were extracted and updated from
Schellart (2008)
vA\ [cm/yr] Trench-normal overriding plate accretion/erosion rate (Fig. 2b), which is the rate at which material is added to the overriding plate by accretion
(positive) or removed from the overriding plate by erosion (negative); Compiled from Clift and Vannucchi (2004); For several subduction zones
no data are available
vSP\ [cm/yr] Trench-normal subducting plate velocity (Fig. 2c); Trenchward motion is positive
vOP\ [cm/yr] Trench-normal overriding plate velocity (Fig. 2c); Trenchward motion is positive
vT\ [cm/yr] Trench-normal trench migration velocity (Fig. 2c), also often referred to as subduction hinge migration velocity or rollback velocity, where
vT\ = vOP\ + vOPD\ + vA\; Migration towards the subducting plate side (rollback) is positive
vS\ [cm/yr] Trench-normal subduction velocity (Fig. 2c), which is the velocity at which the subducting plate disappears into the mantle at the trench, i.e.
vS\ = vT\ + vSP\
vC\ [cm/yr] Trench-normal convergence velocity (Fig. 2c), which is the velocity of the subducting plate relative to the overriding plate, i.e. vC\ = vSP\ + vOP\
vSP\/vS\ Subduction partitioning, which gives the fraction of the subduction velocity that is accommodated by trenchward subducting plate motion
ASP [Ma] Subducting plate age (Fig. 2d), which is the age of the subducting plate at the trench; Data were obtained from Schellart (2008)
TTS [km] Thickness of the trench sediments (Fig. 2b); Data were obtained from Clift and Vannucchi (2004) and Syracuse et al. (2010)
TSS [km] Thickness of the subducted sediments (Fig. 2b); Data were obtained from Syracuse et al. (2010)
dST [°] Subduction zone thrust dip angle (Fig. 2e), which is the dip angle of the subduction zone thrust from the trench down to 50 km depth; Values
were calculated from the location of the trench (Bird, 2003) and the location of the 50 km isodepth contour for the subducted slab
(Gudmundsson and Sambridge, 1998); For some subduction zones and segments no data are available
dS [°] Shallow slab dip angle (Fig. 2e), which is the dip angle of the slab as measured between 0 km and 125 km depth; Data were obtained from
Schellart (2008); For some subduction segments no data are available
dD [°] Deep slab dip angle (Fig. 2e), which is the dip angle of the slab as measured between 125 km and 670 km depth; Data were obtained from
Schellart (2008); For some subduction segments no data are available
DUMS [km] Upper mantle slab tip depth (Fig. 2e), which is the depth of the slab tip in the upper mantle (DUMS 6 670 km); If the slab tip is in the lower
mantle then DUMS = 670 km; Data were obtained from Schellart et al. (2008)
LUMS [km] Slab length (Fig. 2e), which is the upper mantle down-dip slab length, excluding horizontal slab segments located at the 670 km discontinuity;
Values were calculated from dS, dD and DUMS
W [km] Slab width (Fig. 2f), which is the trench-parallel extent of the slab; Data were obtained from Schellart et al. (2007, 2010)
DSE [km] Distance to the closest lateral slab edge (Fig. 2f), which is the distance from the center of the trench segment to the closest lateral edge of the
subducted slab; Data were obtained from Schellart (2008)
FBu [N/m] Upper mantle slab negative buoyancy force per meter trench length (Fig. 2c), which is the product of LUMS, average slab thickness (which is
related to slab age following the square root age law), average density contrast between slab and ambient upper mantle, and the gravitational
acceleration (g = 9.8 m/s2)
CT [m2] Subduction zone trench curvature (Fig. 2f), where CT = 1/r2 and r is the radius of trench curvature
CST [m2] Subduction zone thrust curvature, which is the curvature of the subduction zone thrust plane (from the trench down to 50 km depth), where
CST = sin(dST)/r2
aT [°] Trench segment curvature angle (Fig. 2g); Note that concave curvature towards the overriding plate is positive
|aT| [°] The absolute value of the trench segment curvature angle
frames were combined with the geodetic relative plate motion magnitude >4.5 and we consider 431,789 earthquakes that are
model from Kreemer et al. (2003), which is representative of pres- listed in the catalog between 1973 and June 2010. The ISC Bulletin
ent-day relative plate velocities. is an earthquake database from the International Seismological
Centre. We use data for the period 1904–July 2012, and consider
2.2. Earthquake data a total of 55,584 events.
The issue of magnitude heterogeneity between catalogs due to
We have extracted our earthquake data for subduction zone the use of different datasets, methods of analysis and magnitude
thrust events from four earthquake catalogs (GCMT, PAGER-CAT, scales (e.g. Pacheco and Sykes, 1992; Engdahl and Villaseñor,
NEIC PDE and ISC) as well as from a number of publications on 2002) can be significant, but in our study this effect is minimal
large subduction zone thrust events (Stauder, 1968; Kanamori, due to the nature and origin of the catalogs we have chosen, our
1970a, 1972, 1976, 1977; Abe, 1972; Wu and Kanamori, 1973; focus on large shallow subduction thrust earthquakes and the pre-
Ando, 1975; Beck and Ruff, 1987; Beck and Christensen, 1991; By- dominant use of MW (193 out of 216 for the rupture zone dataset,
rne et al., 1992; Okal, 1992; Johnson et al., 1994; Johnson and Sa- see further below, with the remaining 23 being relatively small and
take, 1999; Yagi, 2004; Matsubara et al., 2005; López and Okal, having magnitudes of 67.8). We have found that estimates of MW
2006; Satake and Atwater, 2007; Lay et al., 2009, 2010; Beavan for this class of events experience little variation across the cata-
et al., 2010). The GCMT catalog contains a global set of earthquakes logs (generally not more than 0.1), and are not sufficient to influ-
since 1976 with moment magnitude MW > 5.0. We use this catalog ence our results. This is no doubt partly due to data inheritance
for the period January 1976–June 2012, which lists 37,164 events. (e.g. ISC draws on both NEIC PDE and GCMT), but also due to
The PAGER CAT catalog is an earthquake database from the United improvement of instrumentation (in both quality and distribution)
States Geological Survey (USGS) that contains input from various and analysis techniques.
catalogs (Allen et al., 2009), and spans the period 1900–December Another characteristic of the earthquake catalogs is that the
2007 (with emphasis on earthquakes since 1973). Earthquakes quality and quantity of their entries has a strong dependence on
with a magnitude greater than 5.5 are listed and it contains a total the capabilities of the global (and in some cases regional) seismic
of 22,000 earthquakes. The NEIC PDE catalog is an earthquake data- networks that were available at the time. The value of a global net-
base from the National Earthquake Information Center (NEIC) of work of seismometers was recognized as far back as 1895, just 6 yr
the USGS starting from 1973. It includes worldwide events with after it was discovered that earthquakes could be detected at long
Fig. 2. Schematic diagrams illustrating most of the 24 subduction zone parameters that are investigated in relation to their potential influence on maximum recorded
subduction zone thrust earthquake moment magnitude (MW). (a) Trench-normal overriding plate deformation rate (vOPD\) and overriding plate strain class (OPSC). (b)
Trench-normal trench accretion rate (vA\), trench sediment thickness (TTS) and subducted sediment thickness (TSS). (c) Trench-normal subducting plate velocity (vSP\),
trench-normal overriding plate velocity (vOP\), trench-normal trench velocity (vT\), subduction velocity (vS\), convergence velocity (vC\) and slab negative buoyancy force
(FBu). Note that subduction partitioning is vSP\/vS\. (d) Subducting plate age at the trench (ASP). (e) Subduction thrust dip angle (dST), shallow slab dip angle (dS), deep slab dip
angle (dD), upper mantle slab tip depth (DUMS) and upper mantle slab length (LUMS). (f) Slab width (W), lateral slab edge distance (DSE) and trench curvature (CT). Note that
subduction thrust curvature CST is estimated as CST = CTsin(dST). (g) Trench curvature angle (aT). Note that |aT| is the absolute value of trench curvature.
distances (Agnew et al., 1976). Although the first network of stan- gions are of interest, and these will almost always be better re-
dardized instruments was in place by 1898 (Dewey and Byerly, corded and analyzed than smaller events.
1969), it wasn’t until the early 1960s that a standardized network We have searched the four earthquake catalogs for subduction
of high quality instruments came into existence in the form of the zone thrust events using the following two search criteria: (1)
World-Wide Network of Standard Seismographs or WWNSS (Oli- The event must lie within the area of a polygon formed by the
ver and Murphy, 1971) with 120 continuously recording analog 200 km trench segment, the adjoining segment of the 50 km slab
stations from 1967 onwards. In 1986, the Global Seismographic contour, and two lines that begin at each end of the trench segment
Network (GSN) was established to replace the by then obsolete and have equal angular distance between the trench segment and
WWNSS (Butler et al., 2004). GSN comprises some 150 high quality the adjoining trench segment at each end. We allow for a 15 km
digital broadband stations that transmit their recordings in real location error. The trench locations have been extracted from Bird
time. (2003) and the 50 km slab contour locations have been extracted
Given the above, it is clear that the quality and quantity of seis- from Gudmundsson and Sambridge (1998). (2) The event must
mic data recorded today far exceeds that which was available prior lie at a depth between 0.0 km and 50.0 km. For the GCMT catalog
to WWNSS, and that between the establishment of WWNSS and we have two additional search criteria: (3) The T-Axis of the focal
now, considerable improvements have been made. As such, earth- solution must have a minimum plunge of 45 degrees, and N-axis
quakes from the early 1900s will invariably be less accurately con- plunge must not exceed 15°. (4) The maximum (strike) angle be-
strained and much more sparse than entries from the last 30–40 yr. tween the trench segment and the nodal planes of the focal solu-
Although this creates challenges for studies that seek to analyze tion must not exceed 30 degrees. The last two search criteria
patterns in global seismicity over time, it is worth noting that in tend to eliminate most reverse events that are not subduction zone
our case, only the largest earthquakes from subduction zone re- interface thrust events, as well as strike-slip events and normal
Table 2
Correlations and ranges for 24 subduction zone parameters (200 km datasets).
Parameter n Total range Rupture zone dataset (RZD) Epicenter dataset (ED)
R Range for MW > 8.5 % of total P R Range for MW > 8.5 % of total P
range range
vOPD\ 216 3.2 to 15.0 0.04 3.0 to 2.3 29 0.003 0.00 3.0 to 0 16 0.011
OPSC 216 3 to 3 0.21 3 to 1 67 0.000 0.22 3 to 0 50 0.011
vA\ 216 0.5 to 0.6 0.03 0.3 to 0.6 82 0.256 0.27 0.3 to 0.6 82 0.653
vSP\ (IA) 209 3.6 to 9.3 0.26 1.0–8.0 55 0.010 0.28 1.5–8.0 50 0.143
vOP\ (IA) 209 5.9 to 6.7 0.03 2.7 to 2.7 43 0.001 0.06 1.0 to 2.7 30 0.022
vT\ (IA) 216 4.1 to 15.4 0.02 2.9 to 2.8 29 0.001 0.02 2.8 to 2.8 29 0.086
vS\ 216 0.1–22.9 0.23 2.2–9.2 31 0.001 0.19 3.5–9.2 25 0.026
vC\ 209 2.0 to 9.8 0.26 0.1–9.5 79 0.151 0.30 3.4–9.5 52 0.018
vSP\/vS\ (IA) 209 2.5 to 6.2 0.05 0.3–1.4 13 0.001 0.06 0.4–1.2 10 0.016
ASP 216 2–159 0.13 11–134 78 0.003 0.13 26–134 69 0.023
TTS 201 0.3–6.0 0.11 0.6–5.0 77 0.000 0.16 0.6–5.0 77 0.001
TSS 201 0.1–3.4 0.20 0.3–2.6 70 0.000 0.04 0.3–2.6 70 0.000
dST 205 7–53 0.01 10–30 43 0.002 0.13 10–25 32 0.019
dS 216 10–72 0.11 16–40 39 0.000 0.07 16–40 39 0.072
dD 208 25–86 0.15 29–79 82 0.091 0.10 30–76 75 0.327
DUMS 216 60–670 0.04 200–670 77 0.073 0.03 300–670 61 0.162
LUMS 216 157–1674 0.17 299–1444 75 0.017 0.16 436–1397 63 0.115
W 216 500–7850 0.29 3400–7850 61 0.000 0.17 3400–7850 61 0.007
DSE 216 100–3900 0.28 100–3300 84 0.300 0.25 100–3100 79 0.538
FBu 216 3.1 1012–1.7 1014 0.16 1.2 1013–1.4 1014 81 0.053 0.14 2.9 1013–1.4 1014 69 0.068
CT 216 1.2 1017– 0.22 6.9 1017– 18 0.014 0.18 2.6 1015– 6 0.007
5.2 1012 9.4 1013 3.2 1013
CST 205 2.7 1018– 0.15 1.5 1017– 5 0.001 0.10 7.0 1016– 2 0.004
4.0 1012 2.0 1013 7.7 1014
aT 216 21.8 to 20.0 0.18 6.3 to 9.8 39 0.001 0.21 6.3 to 5.2 28 0.015
|aT| 216 0.0–21.8 0.26 0.1–9.8 45 0.003 0.20 0.6–6.3 26 0.007
vSP\ (Pa) 209 7.0 to 11.9 0.10 0.5 to 11.3 62 0.034 0.16 2.0–11.3 49 0.030
vOP\ (Pa) 209 9.3 to 10.4 0.08 4.2 to 4.1 42 0.000 0.04 4.2 to 3.9 42 0.041
vT\ (Pa) 216 7.2 to 12.1 0.03 6.0 to 4.4 54 0.005 0.02 6.0 to 4.2 53 0.144
vSP\/vS\ 209 4.9 to 23.7 0.03 0.2 to 1.7 7 0.000 0.01 0.3–1.7 5 0.012
(Pa)
For an explanation of the 24 subduction zone parameters and their units see Table 1. Note that n is the number of data points, R is the correlation coefficient (calculated
through least squares linear regression analysis), and P is the probability for all the earthquakes with MW > 8.5 to fall by pure chance inside the range for MW > 8.5. Further
note that (IA) refers to the Indo-Atlantic moving hotspot reference frame from O’Neill et al. (2005), while (Pa) refers to the Pacific hotspot reference frame from Gripp and
Gordon (2002). The rupture zone dataset contains mostly maximum MW data points for epicenters of earthquakes located within polygons of individual trench segments
(black diamonds in Fig. 3), but also data points for trench segments that overlap with the rupture area of giant earthquakes that cover multiple trench segments (red
diamonds in Fig 3, which exclude trench segments with epicenter), which are assigned the MW value of the giant earthquake. The epicenter dataset contains only maximum
MW data points for epicenters of earthquakes located within polygons of individual trench segments (black diamonds and green circles in Fig. 3).
events. From the discussion above it will be clear that those events tively accurately determined and given in MW, and the focal mech-
extracted from the published literature and from the GCMT catalog anisms are known. From the remaining data points (76 for RZD
have the highest probability of being a subduction zone thrust with 53 MW, 8 MS, 7 MB and 8 unknown magnitude types, 86 for
earthquake. For the earthquake data originating from the GCMT ED with 59 MW, 11 MS, 8 MB and 8 unknown magnitude types)
catalog and the published literature, the focal mechanism solutions most are from the PAGER CAT catalog (69 in RZD and 78 in ED),
have been checked to make sure these are (in all likelihood) indeed with the remainder from the NEIC PDE catalog (6 in RZD and 6 in
thrust events (see Fig. 1). ED) and the ISC bulletin (1 in RZD and 2 in ED).
The earthquake data points in the dataset with the 200 km The dataset with the geologically defined trench segments con-
trench segments (Fig. 3) represent the maximum-recorded magni- tains a maximum of 37 data points in Fig. 4. We use the maximum-
tude at each subduction zone segment. In case of a giant earth- recorded magnitude (32 MW, 4 MS and 1 MB) recorded within the
quake where the rupture area overlaps with more than one geologically defined trench segment. In case the rupture area of a
trench segment, we present two data points for those overlapping giant earthquake overlaps with more than one 200 km trench seg-
segments that do not contain the epicenter (n = 25): one data point ment, then the value of the parameter is averaged for the overlap-
that is assigned the MW value of the giant earthquake (red dia- ping trench segments. Note that in Figs. 3 and 4 we plot the small
monds in Fig. 3), and one data point that is assigned the maximum number of earthquakes with magnitude type other than MW
MW of an earthquake with its epicenter located within that trench assuming that their numerical value is the same in MW. Consider-
segment polygon (green circles in Fig. 3). The data points for the ing that their magnitudes are all relatively small (67.8) this is rea-
remaining trench segments (n = 191) are plotted as black dia- sonable. Also note that for some trench segments no earthquake
monds in Fig. 3. As such, we define two datasets for the 200 km data are available (see Fig. 1).
trench segments, namely a rupture zone dataset (RZD, black dia-
monds and red diamonds in Fig. 3) and an epicenter dataset (ED, 3. Results
black diamonds and green circles in Fig. 3).
The rupture zone and epicenter datasets each contain a maxi- 3.1. Global variability in MW at subduction zones
mum of 216 earthquake data points in Fig. 3. Most originate from
the GCMT catalog and the published literature (97 and 43, respec- Fig. 1 shows the global distribution of subduction zones and
tively for RZD, 108 and 22 respectively for ED), giving a total of 140 their 200 km trench segments. Each trench segment is color coded
(RZD) and 130 (ED). For these data points, their magnitude is rela- to indicate the maximum recorded thrust earthquake that has
Fig. 3. Diagrams showing the dependence of subduction zone interplate thrust earthquake magnitude (maximum moment magnitude MW) on 24 subduction zone
parameters (see Table 1) for all active subduction zones on Earth, which have been subdivided into 200 km trench segments (n = 241). Note that earthquake data are available
for 216 of the segments. The black diamonds represent maximum MW data points for epicenters of earthquakes located within polygons of individual trench segments. In case
of a giant earthquake where the rupture area overlaps with more than one trench segment, two data points are presented for those overlapping segments that do not contain
the epicenter: one data point that is assigned the MW value of the giant earthquake (red diamonds), and one data point that is assigned the maximum MW of an earthquake
with its epicenter located within that trench segment polygon (green circles). Due to this representation we define two datasets: (1) a ‘‘rupture zone dataset’’ (RZD) with black
diamonds + red diamonds (with dashed red-black least squares linear regression best-fit line); and (2) an ‘‘epicenter’’ dataset (ED) with black diamonds + green circles (with
dashed green-black least squares linear regression best-fit line) (see also Table 2). In each diagram the light grey area and dark grey area indicate the observed range of the
particular parameter for earthquakes with MW > 8.5 using the ‘‘rupture zone dataset’’. The dark grey area is exclusively for the observed range for earthquakes with MW > 8.5
using the ‘‘epicenter dataset’’. R2 is the coefficient of determination. Note that we have chosen the following magnitudes for the largest six earthquakes: Chile 1960 – MW 9.5;
Sumatra–Andaman 2004 – MW 9.3; Alaska 1964 – MW 9.2; Japan 2011 – MW 9.0; Kamchatka 1952 – MW 8.9; Chile 2010 – MW 8.8. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 4. Diagrams showing the dependence of subduction zone interplate thrust earthquake magnitude (maximum moment magnitude MW) on 24 subduction zone
parameters (see Table 1) for all active subduction zones on Earth. Subduction zones have been divided into one to six trench-parallel segments (n = 44) based on their
geometry and geology (e.g. arc cusps, nature of overriding plate). Note that earthquake data are available for 37 of the segments. In each diagram the light grey area indicates
the observed range of the particular subduction zone parameter for earthquakes with MW > 8.5. Black line is least squares linear regression best fit line. R2 is the coefficient of
determination.
occurred there since 1900. The focal mechanism solution has been of 9.5 in central Chile (1960 earthquake) and maxima of only MW
shown for those segments where such information is available. The 5.0–5.7 in southernmost Chile. Another notable example is the
figure also presents the rupture extent of those subduction zone northwest Pacific subduction zone, where great and giant earth-
thrust earthquakes with MW > 8.5 that have occurred since 1950 quakes have occurred along the Kamchatka, Kuril and Japan seg-
in cases where it exceeds 200 km. The figure shows a large vari- ments, while for the Izu-Bonin and Mariana segments all except
ation in recorded maximum earthquake magnitude along individ- the northernmost Izu Bonin segment have MW 6 7.4. There is also
ual subduction zones, such as South America, with a maximum MW variability between subduction zones. Notably, several only show
low maxima in subduction earthquake magnitude with MW 6 7.5, R = 0.30 for vC\) are statistically significant at 89% (DSE) and 80%
including Scotia, Hellenic, Halmahera and Manila. (vSP\) and 69% (vC\) confidence level. The highest correlation for
the rupture zone dataset is R = 0.29 (again for W), which is statis-
3.2. Subduction zone parameters and correlations with MW tically significant at 76% confidence level. The second, third and
fourth highest correlation coefficients for the rupture zone dataset
Fig. 3 and Table 2 show the correlations between the 24 subduc- (R = 0.28 for DSE, R = 0.26 for vC\ and R = 0.26 for vSP\) are statisti-
tion zone parameters and MW for the 200 km trench segments re- cally significant at 73% (DSE) and 61% (vC\) and 59% (vSP\) confi-
corded for the period 1900–June 2012 (n = 201–216). The dence level. From these calculations we conclude that only the
correlation coefficients (R) and coefficients of determination (R2) parameter W for the geological dataset (Fig. 4r) has a significant
have been calculated through least squares linear regression anal- correlation with MW, because its confidence level is at least 95%.
ysis. The diagrams show that MW is in the range 4.2–9.5. What is
immediately apparent from the diagrams is the large amount of 3.3. Subduction zone parameter ranges
scatter of the data and the lack of a clear correlation for all the
parameters. The large scatter is reflected in the low correlations The ranges of values observed for the 200 km trench segments
(R = 0.26 to 0.29 for rupture zone dataset, RZD, and 0.27 to that have experienced giant subduction zone thrust earthquakes
0.30 for epicenter dataset, ED). This indicates that even for the are generally less than the total ranges for the 24 different param-
parameters with the highest correlation (slab width for RZD with eters. Note that we define giant subduction zone thrust earth-
R = 0.29 and MW increasing for increasing W, convergence velocity quakes as those with MW > 8.5. The ranges for those trench
for ED with R = 0.30 and MW increasing with increasing vC\), the segments with MW > 8.5 (n = 32 for RZD and n = 10 for ED) with re-
best-fit regression line can only explain 9% of the variance. spect to the total range (n = 201–216) differ significantly for the 24
Fig. 4 and Table 3 show the correlations between the 24 subduc- individual parameters, between 5% and 84% for RZD and between
tion zone parameters and MW for the geologically defined trench 2% and 82% for ED (Fig. 3, Table 2). Note that for simplicity, we will
segments (n = 33–37). Here, MW is in the range 5.7–9.5. For most discuss in the remainder of the text only the ranges as observed for
parameters there is still considerable scatter of the data and a lack the rupture zone dataset RZD unless specified otherwise.
of correlation. However, several parameters do show a more mod- A number of physical parameters do not show a clear distinc-
erate correlation. R is in the range 0.28 to 0.51 and R2 is in the tion between observed range for MW > 8.5 and total observed
range 0.00–0.26. The parameter with the highest correlation is slab range, including ASP, vC\, TTS, DSE, vA\, TSS, dD, LUMS, OPSC, DUMS
width (R = 0.51 with MW increasing for increasing W), for which and FBu with ranges for MW > 8.5 being 67–82%. Parameters that
the best-fit regression line can explain 26% of the variance. show a moderate distinction between observed range for
We have calculated confidence intervals for our correlations MW > 8.5 and total observed range include W (61%) and vSP\
using Fisher’s z. The highest correlation coefficient (R = 0.51) for (55%). Parameters that show a more clear distinction between ob-
W (geological dataset) is statistically significant at 97% confidence served range for MW > 8.5 and total observed range include vOPD\
level. The second, third and fourth highest correlation coefficients (29%), vOP\ (43%), vT\ (29%), vS\ (31%), vSP\/vS\ (13%), dST (43%),
for the geological dataset (R = 0.39 for DSE, R = 0.37 for vSP\ and dS (39%), CT (18%), CST (5%), aT (39%) and |aT| (45%) (Table 2).
Table 3
Correlations and ranges for 24 subduction zone parameters (geological dataset).
Parameter n Total range R Range for MW > 8.5 % of total range P

vOPD\ 37 3.1 to 9.1 0.10 2.8 to 1.3 33 0.230
OPSC 37 3 to 3 0.23 3 to 0.7 62 0.084
vA\ 37 0.5 to 0.6 0.07 0.3 to 0.6 82 0.678
vSP\ (IA) 37 3.3 to 8.1 0.37 1.3–8.0 59 0.110
vOP\ (IA) 37 5.6 to 5.8 0.04 2.1 to 2.7 41 0.142
vT\ (IA) 37 2.9 to 9.0 0.22 2.8 to 2.8 47 0.230
vS\ 37 0.7–13.4 0.15 3.5–9.1 44 0.048
vC\ 35 0–9.4 0.30 1.7–9.4 82 0.340
vSP\/vS\ (IA) 37 2.0 to 1.9 0.21 0.3–1.3 25 0.064
ASP 37 2–151 0.11 21–132 75 0.142
TTS 33 0.4–6.0 0.15 0.6–5.0 79 0.030
TSS 33 0.1–3.4 0.27 0.3–2.6 70 0.042
dST 35 10–43 0.21 12–27 46 0.053
dS 37 12–58 0.21 19–39 42 0.036
dD 35 30–86 0.14 30–77 84 0.534
DUMS 37 60–670 0.02 263–670 67 0.182
LUMS 37 174–1421 0.18 367–1421 84 0.230
W 37 500–7850 0.51 3400–7850 61 0.036
DSE 37 100–3200 0.39 400–3200 90 0.085
FBu 37 3.1 1012–1.4 1014 0.28 2.4 1013–1.4 1014 85 0.085
CT 37 1.4 1016–2.5 1012 0.20 1.5 1014–2.5 1013 9 0.003
CST 35 4.7 1018–1.4 1012 0.25 4.4 1015–6.1 1014 4 0.002
aT 37 8.5 to 19.3 0.09 5.3 to 4.7 36 0.026
|aT| 37 0.1–19.3 0.28 1.2–5.3 22 0.004
vSP\ (Pa) 37 6.6 to 11.0 0.26 1.4–11.0 55 0.142
vOP\ (Pa) 37 8.4 to 8.3 0.03 4.0 to 0.8 29 0.142
vT\ (Pa) 37 6.3 to 10.3 0.20 5.8 to 2.2 48 0.230
vSP\/vS\ (Pa) 37 4.0 to 2.9 0.16 0.3–1.6 19 0.036
For an explanation of the 24 subduction zone parameters and their units see Table 1. Note that n is the number of data points, R is the correlation coefficient (calculated
through least squares linear regression analysis), and P is the probability for all the earthquakes with MW > 8.5 to fall by pure chance inside the range for MW > 8.5. Further
note that (IA) refers to the Indo-Atlantic moving hotspot reference frame from O’Neill et al. (2005), while (Pa) refers to the Pacific hotspot reference frame from Gripp and
Gordon (2002).
For each of the 24 physical parameters we have calculated the 4.2. Parameters with small (distinct) ranges for MW > 8.5
probability (P) that all the large events with MW > 8.5 fall – by pure
chance – within the selected narrow range of the total range (Ta- Parameters that show a clear distinction between observed
bles 2 and 3). For RZD, the probabilities for all 24 parameters are range for MW > 8.5 (5–45%) and total observed range (100%) in-
between 0.00 and 0.30. For those parameters with narrow ranges clude vOPD\, vT\, vOP\, vSP\/vS\, dST, dS, CT, CST, aT, |aT| and vS\
(vOPD\, vOP\, vT\, vS\, vSP\/vS\, dST, dS, CT, CST, aT, |aT|) the probabil- (Table 2).
ities are very low (P = 0.00–0.01).
4.2.1. Overriding plate deformation rate and trench velocity
4. Discussion The parameters vOPD\ and vT\ have low values and small ranges
for trench segments with MW > 8.5, with vOPD\ = 3.0 to 2.3 cm/yr
4.1. Correlations between subduction zone parameters and MW (29%) and vT\ = 2.9 to 2.8 cm/yr (29%) (Fig. 3a and f), while the
probabilities for these distributions are low (P = 0.00, Table 2).
From the data presented in Figs. 3 and 4 it is clear that none of The ranges suggest that giant subduction zone thrust earthquakes
the 24 parameters can individually explain the distribution of max- preferentially occur for those subduction zone segments for which
imum observed MW for individual subduction zone segments be- the overriding plate is either relatively neutral, experiences minor
cause of the low correlation coefficients for both 200 km datasets extension or experiences shortening, while the trench is relatively
(RZD and ED) and for the geological dataset. The correlations for stable with minor trench retreat or trench advance. Notably, giant
the geological dataset are generally somewhat higher than those earthquakes have not been recorded for trench segments that
for RZD and ED. Datasets RZD and ED give comparable (low to neg- experience rapid trench retreat (e.g. Scotia, New Hebrides, Tonga,
ligible) correlations. southwest Ryukyu), nor for those that experience considerable
The low correlations could potentially be explained due to the backarc extension/spreading (e.g. Scotia, Mariana, New Hebrides,
fact that accurate global recording of earthquake location and Tonga, southwest Ryukyu). What is also notable is that for the epi-
magnitude started only in the first half the 20th century (last center dataset ED the range for vOPD\ with MW > 8.5 is 3.0 to
60–110 yr), while the recurrence interval of very large earth- 0 cm/yr, indicating that the epicenters of giant earthquakes have
quakes can be hundreds to thousands of years. For example, the occurred only at trench segments with a neutral or shortening
recurrence interval for giant subduction zone earthquakes in deformation regime in the overriding plate. Epicenters of giant
northeast Japan is thought to be 800–1100 yr (Minoura et al., earthquakes (MW > 8.5) have not occurred at trench segments with
2001), in agreement with the recent occurrence of the 2011 MW an overriding plate that experiences extension. The probability for
9.0 Japan earthquake. the MW > 8.5 distribution in the epicenter dataset is low (P = 0.01,
Even if the lack in correlation is possibly related to the relatively Table 2), implying a physical cause for the narrow distribution of
short sampling period, it is worth discussing several parameters vOPD\ values for the epicenters of giant earthquakes.
that have received much attention in the past, including subduct- Uyeda and Kanamori (1979) proposed that giant earthquakes
ing plate age, trench sediment thickness, overriding plate velocity, occur for highly compressive subduction zones with significant
and slab dip angle (e.g. Uyeda and Kanamori, 1979; Ruff and Kana- backarc shortening, because only for such settings is there suffi-
mori, 1980; Lamb and Davis, 2003; Stein and Okal, 2007; Heuret cient elastic strain energy accumulation to permit a large rupture.
et al., 2011). Subducting plate age, trench sediment thickness and Since then others have also argued that a highly compressive stress
overriding plate velocity have very low correlations for all datasets regime is a requirement for the generation of giant earthquakes at
with |R| 6 0.16. Parameters related to subduction zone geometry subduction zones (e.g. Ruff and Kanamori, 1980; Ruff, 1989; Con-
(dST, CT, CST and |aT|) also show low correlations for all datasets rad et al., 2004; De Franco et al., 2008). In contrast, Heuret et al.
(|R| 6 0.28) but consistent trend lines (except dST), weakly suggest- (2011) argued that giant earthquakes occur for neutral overriding
ing that giant earthquakes prefer relatively straight fault planes. plates (no deformation), arguing that only a neutral stress regime
Parameters that are linked to the state of normal stress on the sub- at the subduction zone interface allows for large lateral propaga-
duction interface (vOPD\ and OPSC) also give low correlations tion of the rupture.
(|R| 6 0.23) but consistent trend lines, weakly suggesting that giant The data presented in Fig. 3a show that the occurrence of giant
earthquakes preferentially occur along subduction segments with earthquakes is more complex than previously proposed, ranging
a relatively high normal stress on the subduction zone interface. from highly compressive (deviatoric compression) with rapid over-
W and DSE have some of the highest correlations with R = 0.29 riding plate shortening rates (e.g. Japan, vOPD\ = 3.0 to 2.7 cm/
(W) and R = 0.28 (DSE) for the rupture zone dataset and R = 0.51 yr) to mostly neutral (e.g. Alaska, vOPD\ 0 for three segments,
(W) and R = 0.39 (DSE) for the geological dataset, giving a weak vOPD\ = 0.3 for easternmost segment) to mostly tensional (devia-
indication that giant earthquakes preferentially occur at wide sub- toric tension) with slow overriding plate extension (e.g. Sumatra–
duction zones and away from slab edges. Andaman, vOPD\ = 1.2–2.3 cm/yr for six segments, vOPD\ = 0.4
Despite the generally very low correlations for the three data- cm/yr for southernmost segment). What is noteworthy, is that
sets, it is concluded that several parameters do show correlations the three largest recorded earthquakes (Chile 1960, Alaska 1964,
with MW that are physically plausible and also consistent for the Sumatra–Andaman 2004), which were all characterized by very
three datasets, including CT, CST and |aT|, and vOPD\, vT\, W and large lateral rupture propagation (800–1300 km), all had their epi-
DSE. The datasets do provide additional insight into which param- center at a trench segment characterized by overriding plate short-
eters are more likely or less likely to have an influence on the max- ening and a compressive deviatoric stress regime, with the rupture
imum MW at subduction zones. This can be deduced from propagating towards the region characterized by a neutral or
comparing the range of values for each of the 24 parameters for (mildly) extensional overriding plate and a neutral or tensional
those trench segments with MW > 8.5 with the range of values for deviatoric stress regime (see Section 5.4 for more discussion).
each parameter for all trench segments (Figs. 3 and 4, Tables 2 As noted in the methods section, calculations of vT\ depend on
and 3). We will discuss these ranges in detail in Sections 4.2, 4.3 the choice of reference frame, with the Indo-Atlantic hotspot refer-
and 4.4. Note again that we will discuss in the remainder of the ence frame being our preferred frame. Calculations for vT\ in the
text only the ranges as observed for the rupture zone dataset Pacific hotspot reference frame from Gripp and Gordon (2002)
(RZD) and not for the epicenter dataset (ED) or the geological data- increase the range for MW > 8.5 to 6.0 to 4.4 cm/yr (54%,
set, unless specified otherwise. Table 2).
4.2.2. Overriding plate velocity the surface down to 125 km depth. There appears to be no physical
The range for vOP\ (2.7 to 2.7 cm/yr) for MW > 8.5 is smaller reason as to why the local dip of the slab below the thrust interface
(43%) than the total range (5.9 to 6.7 cm/yr) and it is centered seismogenic zone (i.e. below 50 km) should be of any relevance
on zero (Fig. 3e). The low values for |vOP\| and the centering on for predicting earthquake magnitude. However, dST and dS are
zero suggest that MW > 8.5 earthquakes preferentially occur for semi-dependent parameters and are moderately correlated
subduction zones with a low vOP\ and that within this relatively (R = 0.62), suggesting that the narrow range observed for dS is
small range there is no preference for very large earthquakes to oc- mostly a consequence of the narrow range observed for dST.
cur for either trenchward overriding plate motion or motion away
from the trench. This is in contrast to previous suggestions that 4.2.5. Trench curvature and subduction thrust curvature
giant earthquakes preferentially occur at subduction segments The trench curvature CT, subduction zone thrust curvature CST,
for which the overriding plate moves trenchward at a high rate trench curvature angle aT and absolute value of the trench curva-
(e.g. Uyeda and Kanamori, 1979; Peterson and Seno, 1984; Conrad ture angle |aT| for subduction segments with MW > 8.5 have narrow
et al., 2004). We note, however, that our calculations of vOP\ de- ranges (18%, 5%, 39% and 45%, respectively) (Fig. 3u–x). This sug-
pend on the choice of reference frame, as explained in the Methods gests that giant subduction zone thrust earthquakes do not occur
section, with the Indo-Atlantic hotspot reference frame having our for those subduction zone segments that have a relatively strong
preference, and thus that our conclusion from above depends on trench curvature and subduction zone thrust curvature and a high
this choice. However, although our calculations for vOP\ in the Pa- |aT|. Notably, giant earthquakes have not been reported for
cific hotspot reference frame from Gripp and Gordon (2002) do strongly curved subduction zones such as Scotia, nor have giant
make a difference in the distribution of data points, the range for earthquakes been reported to cross trench cusps with a strongly
MW > 8.5 is comparable (42%) and is centered on zero (Table 2). negative trench curvature angle (aT 0). It thus appears that rup-
tures of giant earthquakes preferentially follow a relatively planar
4.2.3. Subduction partitioning and subduction rate subduction zone thrust fault and that lateral rupture propagation
The subduction partitioning parameter is closely related to the will be halted in case of a strong increase in trench and thrust
parameters vT\ and vSP\. The range for vSP\/vS\ (MW > 8.5) is 0.3– plane curvature.
1.4, which is 13% of the total range (2.5 to 6.2) (Fig. 3i). The dis-
tribution of the data in Fig. 3i indicates that for most subduction 4.3. Parameters with intermediate ranges for MW > 8.5
segments experiencing giant earthquakes subduction is accommo-
dated predominantly by trenchward subducting plate motion and Parameters that show a moderate distinction between observed
not by trench retreat. In the Pacific hotspot reference frame the range for MW > 8.5 (55–61%) and total observed range (100%) in-
range for vSP\/vS\ (MW > 8.5) is also small (7%). clude W and vSP\ (Table 2).
The range for vS\ (MW > 8.5) is 2.2–9.2 cm/yr, which is 31% of
the total range (0.1–22.9 cm/yr) (Fig. 3g). This indicates that those 4.3.1. Slab width
subduction segments have experienced giant earthquakes for For earthquakes with MW > 8.5, slab width ranges between
which the subduction rate is moderately low to high. The fact that 3100 and 7850 km, which is 61% of the total range (500–
there are no recorded giant earthquakes at very high vS\ 7850 km), suggesting that giant earthquakes preferentially occur
(P10 cm/yr) is likely because the few subduction segments that at wide subduction zones (Fig. 3r). This is not entirely surprising,
experience very fast subduction do so due to very fast backarc considering that wide subduction zones (W > 3000) account for
opening and trench retreat (Tonga, New Hebrides, New Britain). the largest portion (36,250 km, 75%) of the global extent of sub-
Earthquakes with MW > 8.5 have not been documented for these duction zones (48,200 km). This might partly explain why W
subduction zones, which can possibly be ascribed to the low nor- shows the highest correlations of all parameters (Tables 2 and 3).
mal stresses at the subduction interface due to rapid backarc open- Furthermore, MW is mostly controlled by rupture length and
ing and trench retreat (see Section 4.2.1). The fact that there are no amount of slip. It is clear that giant earthquakes with extreme rup-
recorded giant earthquakes for subduction segments with very low ture lengths (>600 km), such as the Sumatra–Andaman 2004, Chile
vS\ (<2 cm/yr) can possibly be ascribed to the view that for slow 1960 and Alaska 1964 earthquakes, are not possible for the nar-
subduction the recurrence time increases, so that in a finite time rowest subduction zones with W < 600 km (e.g. Betic-Rif, Calabria,
slow subduction zones are less likely to experience a giant earth- North Sulawesi, Halmahera). Nevertheless, such narrow subduc-
quake (McCaffrey, 2008). tion zones might still produce giant earthquakes with MW > 8.5
in cases where the coseismic slip is very large (multiple tens of
4.2.4. Subduction thrust dip angle and shallow slab dip angle meters).
The subduction thrust dip angle dST for subduction segments One can also argue from a mechanical point of view as to why
with MW > 8.5 is 10–30°, which is 43% of the total range of values relatively narrow subduction zones might be less likely to produce
for dST (7–53°) (Fig. 3m). This suggests that giant subduction zone MW > 8.5 earthquakes. Narrow subduction zones preferentially
thrust earthquakes preferentially occur for those segments that subduct through trench retreat as the slab pulls the subduction
have a relatively gentle subduction thrust dip angle 630°. It has zone hinge and trench backward during slab rollback (Schellart
been argued before that giant earthquakes occur for gentle thrust et al., 2010). Such pull-back reduces the normal stress on the sub-
dip angles and large downdip extent of the seismogenic zone duction zone interface, which, in combination with the mantle
(Kelleher et al., 1974; Uyeda and Kanamori, 1979; Lay et al., flow patterns excited by slab rollback, promotes rapid backarc
1982), but it has also been argued that thrust dip angle and down- extension such as found for the Scotia and Hellenic subduction
dip extent of the seismogenic zone, which is closely related to dST, zones (Schellart and Moresi, 2013; Duarte et al., 2013). Thus, nar-
play no role in the generation of giant earthquakes (Pacheco et al., row subduction zones are generally less capable of sustaining the
1993; Heuret et al., 2011). The current work suggests that a gentle buildup of large elastic strain energy at the subduction zone inter-
subduction thrust dip angle (630°) is a requirement for earth- face due to the low normal stress. It therefore appears plausible
quakes with MW > 8.5. that narrow subduction zones preferentially release seismic energy
The shallow slab dip angle dS also shows a clear distinction be- through a relatively large number of relatively minor earthquakes.
tween observed range for MW > 8.5 and total observed range (39%) In contrast, wide subduction zones have a relatively stable subduc-
(Fig. 3n). For dS the dip angle is averaged over a depth range from tion zone hinge, in particular in the center, which can sustain large
normal stresses (Schellart and Moresi, 2013) and thus the buildup nounced. The work did suggest an upper limit for subducting
of large elastic strain energy, which can be released in larger plate age of 80 Ma for subduction zones that can produce giant
earthquakes. earthquakes. The 2011 MW 9.0 Japan earthquake shows that this
It should be remembered, though, that, apart from slab width, upper limit does not apply, as the Pacific plate subducting at the
there are several other factors that influence trench migration rate site of this massive earthquake is one of the oldest in the world
(e.g. slab length, plateau/aseismic ridge subduction, tearing resis- (130–134 Ma along rupture length).
tance), such that not all narrow subduction zones retreat rapidly A multivariate least-squares linear regression analysis of the
or show rapid backarc extension. It is thus likely that some rela- dependence of MW on ASP and vS\ gives very low correlations, both
tively narrow subduction zones are capable of producing very large for the rupture zone dataset (R2 = 0.05) and for the geological data-
earthquakes. Indeed, the Cascadia subduction zone is relatively set (R2 = 0.03). This provides further evidence that the combination
narrow (W = 1400 km), has a relatively neutral overriding plate of plate age and subduction rate is not a good predictor of maxi-
(with minor shortening in the north and minor extension in the mum MW.
south), and is thought to have experienced an MW 9 earthquake
in 1700 (Satake et al., 2003). 4.4.2. Accretion/erosion rate, and trench and subducted sediment
thickness
4.3.2. Subducting plate velocity Ruff (1989) proposed that subduction segments with large sed-
For vSP\ the range (1.0–8.0 cm/yr) is moderately smaller (55%) iment thicknesses at the trench and accreting subduction margins
than the total range (3.6 to 9.3 cm/yr) (Table 2). The data in promote the occurrence of great earthquakes, arguing that excess
Fig. 3d suggest that earthquakes with MW > 8.5 do not occur when trench sediments are associated with the subduction of a coherent
vSP\ < 1 cm/yr. Calculations for vSP\ in the Pacific hotspot refer- sedimentary layer, which forms a homogeneous and strong contact
ence frame give a comparable range of 62%. zone at the subduction interface. In such a conceptual model, the
homogeneity along the contact zone might allow for a large lateral
4.4. Parameters with large (indistinct) ranges for MW > 8.5 and down-dip rupture propagation, while the high strength at the
contact zone might allow for significant buildup of elastic strain
Physical parameters that do not show a clear distinction be- energy and subsequent large coseismic fault slip. It has also been
tween observed range for MW > 8.5 (68–84%) and total observed proposed that accreting subduction margins with significant
range (100%) include ASP, vC\, vA\, TTS, TSS, DSE, dD, LUMS, DUMS and trench sediment fill are generally weak with a low friction coeffi-
FBu (Table 2). cient, while subduction margins with small/negligible amounts of
trench fill are generally strong and have a high friction coefficient
4.4.1. Subducting plate age (Lamb and Davis, 2003). The occurrence of giant earthquakes at
Our compilation shows that the range in ASP is 11–134 Ma for accreting trenches might then be related to the concept that rup-
MW > 8.5, which is 78% of the total range of 2–159 Ma (Fig. 3j). This ture propagation is facilitated by the low strength of the interface,
indicates that giant earthquakes can occur at subduction zones promoting a large rupture area. The question then remains how
that consume very young to very old oceanic lithosphere. Earlier such a weak interface will allow for the buildup of significant strain
work (Ruff and Kanamori, 1980) found that subducting lithospher- energy to be released in a giant earthquake. Such buildup could
ic age, in combination with subduction rate, correlates strongly potentially occur locally at a high-friction asperity on the interface.
with earthquake magnitude. The authors argued that the largest The data in Fig. 3c, k and l, showing the relation between vA\
earthquakes occur at subduction zones that are characterized by and MW, TTS and MW and TSS and MW, provide insight into the pos-
rapid subduction of young lithosphere. The rationale was based sible variability of subduction zone interface strength for subduc-
on earlier conceptual models of the influence of oceanic litho- tion zones around the globe. The parameters have low
sphere age on subduction dynamics (e.g. Molnar and Atwater, correlations (|R| = 0.03–0.20) and subduction segments with
1978; Uyeda and Kanamori, 1979): (1) young oceanic lithosphere MW > 8.5 have indistinct ranges (70–82%) (Table 2). The data thus
is relatively buoyant and resists subduction, causing a gentle slab indicate that giant earthquakes can occur at accreting subduction
dip, and high coupling and large compressive stresses at the inter- margins with large amounts of trench fill and subducted sediments
face and in the overriding plate, resulting in overriding plate short- (e.g. southern Chile, Sumatra–Andaman) and eroding subduction
ening and large earthquakes at the interface; and (2) old oceanic segments with small amounts of trench fill and subducted sedi-
lithosphere has a high negative buoyancy force, sinks steeply into ments (e.g. Japan, Kamchatka).
the mantle during slab rollback and causes low coupling and devi- If we assume that the conceptual model in which giant earth-
atoric tension in the overriding plate and at the subduction zone quakes are facilitated by a mechanically weak subduction zone
interface, causing backarc extension and small earthquakes at the interface through large rupture propagation is correct, following
interface. Recent work has shown that such conceptual models Lamb and Davis (2003), then this suggests that both accreting sub-
do not apply to most subduction zones in nature. Subducting plate duction margins with thick piles of trench sediments and eroding
age does not correlate with backarc deformation style (i.e. shorten- subduction margins with very small amounts of trench sediments
ing, extension, neutral) and backarc deformation rate (Schellart, have a comparable (weak) subduction zone interface with a low
2008), nor does it correlate with trench velocity (e.g. Jarrard, friction coefficient. If we assume, however, that the conceptual
1986; Heuret and Lallemand, 2005; Schellart et al., 2010). These model, in which giant earthquakes with large fault slip are facili-
findings suggest that subducting plate age provides no (or only a tated by a homogeneous mechanically strong subduction zone
minor) control on normal stress, mechanical coupling and earth- interface is correct, following Ruff (1989), then this suggests that
quake magnitude at the subduction zone interface. both types of subduction margins have a comparable (strong) sub-
The 2004 MW 9.3 Sumatra–Andaman earthquake violates the duction zone interface with a high friction coefficient. Independent
age & subduction rate – earthquake size relationship both in terms of which of the two models applies best to nature, the results sug-
of subducting plate age (ASP = 57–92 Ma along rupture length) and gest that there is not a large variability in subduction interface
subduction rate (vS\ = 1.9–4.6 cm/yr along rupture length). Stein strength between different subduction zones. Our finding is in
and Okal (2007) revisited the model from Ruff and Kanamori agreement with results from von Huene and Ranero (2003), who
(1980) using an updated dataset and several historical earth- suggest that along the northern Chile margin (an eroding margin)
quakes, and indeed found that the correlation is much less pro- the subduction interface friction value is comparable to those of
accreting margins. Therefore, the global variability in maximum 5. Conceptual model to explain spatial distribution of giant
MW at subduction zones is related to other parameters. earthquakes
5.1. Equation for moment magnitude MW

4.4.3. Convergence velocity
For subduction zones with MW > 8.5, vC\ = 0.1–9.5 cm/yr, which
The moment magnitude MW is generally calculated as follows
is 79% of the total range (2.0 to 9.8 cm/yr), indicating that giant
(Kanamori, 1986):
earthquakes can occur for very slow to very fast convergence
velocities (Fig. 3h). MW ¼ ðlog ES 11:8Þ=1:5 ð1Þ
with
4.4.4. Lateral slab edge proximity ES ¼ M 0 =ð2lR =DrÞ ð2Þ
For subduction zones with MW > 8.5, DSE = 100–3300 km, which
is 84% of the total range (100–3900 km), indicating that giant and
earthquakes can occur at trench segments located at distances M0 ¼ lR DS ð3Þ
ranging from very close to very far from the nearest lateral slab
edge (Fig. 3s). Heuret et al. (2011) presented a global analysis of where ES is the energy (in ergs) radiated from an earthquake source,
subduction zones and subduction zone thrust earthquakes and lR is the rigidity modulus of the material surrounding the fault, Dr
suggested that subduction zone thrust earthquakes with is the average stress drop in the earthquake, M0 is the seismic mo-
MW P 8.5 preferentially occur in the vicinity of lateral slab edges ment, D is the average slip on the fault, and S is the rupture area of
and a neutral overriding plate strain regime (no deformation). the fault plane. If we assume that lR is relatively constant for sub-
The 2011 MW 9.0 Japan giant earthquake is in disagreement with duction zones around the globe, then the equations indicate that
this conceptual model. The rupture area of the giant earthquake MW is affected only by D, S and Dr. From these three parameters,
occurred some 2400–2800 km from its closest lateral slab edge Dr is thought to be approximately constant from small earth-
in the north (Kamchatka–Aleutians cusp, Fig. 1) in a subduction quakes to giant earthquakes (e.g. Kanamori and Anderson, 1975;
setting that is classified as highly compressive (Jarrard, 1986; Shaw, 2009), although values generally fall within a range that
Heuret and Lallemand, 2005) with an overriding plate setting that stretches two orders of magnitude (Ruff, 1999). For some giant
is characterized by rapid shortening of 2–3 cm/yr (Schellart et al., earthquakes such as the 2011 MW 9.0 Japan earthquake reported
2007). A number of giant historical subduction thrust earthquakes stress drops are relatively large, including 2–10 MPa (Simons
have also been reported in the central parts of wide subduction et al., 2011) and 20 MPa (Hasegawa et al., 2011). For the 2010
zones (Fig. 1), such as in southern Sumatra (Zachariasen et al., MW 8.8 Chile earthquake a depth-averaged stress drop of 4 MPa
1999) and in the central Andes (Beck and Ruff, 1989; Dorbath and a peak static shear stress drop of 17 MPa at the main slip asper-
et al., 1990; Chlieh et al., 2011). ity have been reported (Luttrell et al., 2011). Also, for giant earth-
In earlier work, Schellart (2008), Schellart et al. (2011) and quakes D is very large, with meters to tens of meters of fault slip.
Schellart and Moresi (2013) proposed a physical mechanism where The most important component that makes an earthquake a giant
overriding plate deformation and trench retreat are related to W earthquake, however, is fault rupture area (Ruff, 1989). For giant
and DSE, with slow trench migration and overriding plate shorten- earthquakes the rupture area can be up to a few hundred thousand
ing occurring in the center of wide subduction zones (due to the square km. Below we will discuss how the subduction zone param-
relative immobility of the subduction zone hinge in the center), eters vOPD\, vT\, vSP\/vS\, dST, CT, CST, aT and vS\ might affect the
and trench retreat and overriding plate extension occurring near magnitude of the parameters D, Dr, and S.
lateral slab edges (slab segments near lateral slab edges can roll-
back due to efficient toroidal mantle return flow). Such a model 5.2. Fault slip and stress drop
can explain the large trench-normal compressive stresses, shorten-
ing and low trench velocities in the central Andes and Japan, and Coseismic slip on a subduction zone thrust fault plane requires
might also explain the giant (historic) subduction zone earth- the buildup of (recoverable) elastic strain energy. Such strain en-
quakes observed in these regions. The model can also explain the ergy is (at least partially) released during an earthquake resulting
rapid trench retreat and rapid backarc opening in northern Tonga, in a stress drop. If the stress drop is not unusually high for a giant
southern New Hebrides, southwest Ryukyu and Scotia, and might earthquake, then its strain is roughly the same as that for small
also explain the absence of any giant earthquakes in these regions earthquakes but distributed over a larger area. The larger slip in
(MW > 8.5). However, as stressed in Schellart et al. (2011), although the giant earthquake could then be a consequence of a larger vol-
close proximity of a trench segment to a lateral slab edge is a ume storing the strain. It appears, however, that at least for the two
requirement, it does not guarantee rapid trench retreat and back- most recent giant earthquakes the stress drop is relatively high
arc extension, because there are many other circumstances (e.g. (see Section 5.1), in particular at the main slip asperity, which is
subduction of buoyant ridges, plateaus and spreading ridges) that most likely the result of stored high elastic strain. Recent investiga-
affect vT\ and vOPD\. As such, one can also expect giant earth- tions also indicate a near-complete stress drop for the three most
quakes to occur in the vicinity of lateral slab edges, as has indeed recent giant earthquakes (Hardebeck, 2012). High elastic strain en-
been observed, e.g. the 1964 MW = 9.2 Alaska earthquake (Fig. 1). ergy stored around the subduction zone plate interface can only be
sustained by high friction coefficients at the subduction zone inter-
face, or significant compressive normal stress on the fault plane (or
4.4.5. Deep slab dip, and upper mantle slab length, depth and negative a combination thereof) to keep it temporarily locked.
buoyancy It has been suggested that the amount of sediments in the
The remaining parameters that show a lack of distinction be- trench, and the style of accretion at the subduction zone, deter-
tween observed range for MW > 8.5 and total range include dD, LUMS, mines to a large extent the subduction zone friction coefficient,
DUMS and FBu (75–82%). This indicates that giant earthquakes can with small amounts of trench sediments and trench erosion caus-
occur for subduction segments with gentle to steeply dipping deep ing a high coefficient and large amounts of sediments and trench
slab segments (in the 125–670 km depth range) with a highly var- accretion causing a low coefficient (e.g. Lamb and Davis, 2003). A
iable slab length, depth extent and negative buoyancy force. high friction coefficient is capable of sustaining larger shear stres-
ses at the subduction zone interface than a low friction coefficient, face thrust earthquakes occur has a limited depth range. The
keeping all else equal. Thus it appears plausible that high-friction downdip extent of the seismogenic zone and the brittle-ductile
subduction zone segments can sustain large elastic strains, and transition are generally thought to depend on the thermal state
can cause a large earthquake with large coseismic slip and a signif- of the subduction zone, and the maximum depth for subduction
icant stress drop, while those segments with a low friction coeffi- zones around the world is estimated at 30–70 km (Pacheco et al.,
cient cannot. However, as discussed in Section 4.4.2, the data 1993; Tichelaar and Ruff, 1993; Heuret et al., 2011) with an aver-
show that for MW > 8.5 subduction segments can be rapidly accret- age of 50 km. The absolute minimum depth of the seismogenic
ing (vA\ = 0.3–0.6 cm/yr, e.g. northern Sumatra–Andaman, south- zone is the depth of the trench, as, for example, was the case for
ern Chile) with significant trench sediments, or rapidly eroding the 2011 MW 9.0 Japan earthquake (Ide et al., 2011; Kido et al.,
(vA\ = 0.3 cm/yr, e.g. North Japan, Kamchatka) with minor trench 2011). So if we assume a seismogenic zone from 8 km depth
sediments (Figs. 3c, 4c). This suggests that friction coefficients at (trench depth) down to 50 km depth, then the down-dip extent
subduction zone thrust faults worldwide are generally comparable of the seismogenic zone is 162–242 km for dST = 10–15° (e.g. east-
for both accreting trench segments and eroding trench segments. ern Alaska) but only 53–84 km for dST = 30–53° (e.g. New Hebri-
This still leaves the question as to whether friction coefficients des). The data for dST are all low (10–30°) for MW > 8.5 (Figs. 3m,
are generally high or generally low at the subduction zone inter- 4m), implying that a gentle dST and a large down-dip extent of
face. A low subduction interface strength is implied by investiga- the seismogenic zone are indeed required for the generation of
tions of fault rock properties from the subduction zone interface giant earthquakes. Note that we did not specifically investigate
(Moore and Lockner, 2007), regional seismic studies of subduction the seismogenic zone downdip length, as it follows directly from
zones (Magee and Zoback, 1993; Wang et al., 1995; Luttrell et al., dST, the trench depth (which is relatively constant) and our
2011), studies of thrust wedge tapers (Suppe, 2007), regional heat assumption that for all subduction zones the seismogenic zone
flow studies of the subduction zone forearc (Springer, 1999; Grev- continues down to 50 km depth.
emeyer et al., 2003) and geodynamic modeling studies of subduc- The trench-parallel extent of the rupture plane is expected to be
tion (King and Hager, 1990; Moresi and Solomatov, 1998; Duarte influenced by many subduction zone parameters. One can intui-
et al., 2013). tively expect a large trench-parallel rupture extent for subduction
From the paragraph above it appears plausible that global var- zones that have a generally straight trench and planar subduction
iability of the friction coefficient at the subduction zone interface zone thrust interface (i.e. low curvature), because a laterally prop-
is relatively small and that friction coefficients are generally very agating rupture plane will most likely have more difficulty follow-
low. Thus, the friction coefficient might only play a minor or even ing a highly concave subduction zone interface such as for Scotia or
a negligible role in determining the global variability in average a convex arc cusp. As such, one can expect low values of CT, CST and
coseismic slip distance and coseismic stress drop. It might be that aT for MW > 8.5. This is indeed observed (Fig. 3u–w). One can also
the variability in deviatoric normal stress on the subduction zone expect a smooth top surface of the subducting plate entering the
thrust fault largely determines variability in fault slip and stress trench, because subduction of bathymetrically elevated features
drop: Subduction thrust faults with a relatively high normal stress such as aseismic ridges, plateaus, seamounts or fracture zones
(deviatoric compression with overriding plate shortening, Fig. 5) might form barriers to the lateral propagation of a rupture plane
would promote large elastic strain energy to buildup that can be (e.g. Ruff, 1989; Lay et al., 1982; Kelleher and McCann, 1976; Con-
released in a giant earthquake with a large displacement and a treras-Reyes and Carrizo, 2011). Indeed, the three largest earth-
considerable stress drop. Conversely, subduction thrust faults with quakes in recorded history (Chile 1960, Alaska 1964 and
a relatively low normal stress (deviatoric tension with overriding Sumatra–Andaman 2004) with very large lateral rupture propaga-
plate extension, Fig. 5) would promote smaller elastic strain energy tion (800–1300 km) occurred at subduction segments with very
to buildup, which can be released in a small earthquake with minor significant trench sediment thicknesses, implying a smooth sub-
fault slip and a normal stress drop. duction zone interface. Elevated features, in particular aseismic
A relatively high normal stress state on the subduction interface ridges and fracture zones, might also form nucleation points for
for subduction segments with MW > 8.5 agrees with the observa- giant earthquakes, as recent studies suggest (Contreras-Reyes
tions for vOPD\ as no significant trench-normal overriding plate and Carrizo, 2011; Carena, 2011; Müller and Landgrebe, 2012).
extension/spreading occurs (vOPD\ = 3.0 to 2.2 cm/yr) for subduc- Contreras-Reyes and Carrizo (2011) argue that subduction of such
tion segments with MW > 8.5. And notably, vOPD\ 6 0 for subduc- features, if elevated, locally increases the normal stress and cou-
tion segments with epicenters of MW > 8.5 earthquakes (Fig. 3a). pling on the subduction zone interface.
For those subduction zones that do have rapid backarc extension/
spreading (e.g. Scotia, New Hebrides, Tonga, southern Ryukyu) 5.4. The largest recorded subduction zone earthquakes
large earthquakes with MW > 8.5 have not been reported.
The discussion on the requirement of relatively large normal We will now turn our discussion to the largest recorded sub-
stress on the subduction interface for MW > 8.5 earthquakes does duction zone thrust earthquakes. The three largest recorded earth-
have an apparent caveat. Indeed, it has been proposed that large quakes are the 1960 MW 9.5 Central Chile earthquake, the 1964 MW
lateral rupture propagation (which is required for MW > 8.5 earth- 9.2 Alaska earthquake and the 2004 Sumatra–Andaman MW 9.1–
quakes, see next section) can only occur for a relatively smooth 9.3 earthquake, which are characterized by extreme lateral rupture
subduction zone interface and a low mechanical coupling (Lamb propagation (1000, 800 and 1300 km, respectively). What
and Davis, 2003; Heuret et al., 2011), which implies a low friction these earthquakes have in common is that they show unilateral
coefficient and a low normal stress. A potential solution will be rupture propagation (Figs. 6–8). What these earthquakes further
provided in Section 5.4 for the three largest recorded earthquakes. have in common is that the rupture started in a region bordered
by an overriding plate segment characterized by active trench-nor-
5.3. Rupture area mal shortening and that the rupture propagated unilaterally to-
wards a region bordered by an overriding plate segment with
A large fault rupture area requires a large trench-normal active trench-normal extension or a neutral strain regime. As
(downdip) extent and trench-parallel extent of the rupture plane. shown schematically in Fig. 5 using the Mohr–Coulomb failure cri-
A large trench-normal extent is promoted by a gentle thrust dip terion, a subduction segment bordered by an overriding plate char-
angle, because the seismogenic zone where subduction zone inter- acterized by trench-normal shortening (in the forearc, intra-arc
Fig. 5. Schematic cross-sections and diagram illustrating the normal stress on the subduction zone interface (rn(sf)) for two different tectonic settings in the overriding plate.
(a) Cross-section illustrating overriding plate shortening with the development of thrust faults, implying a relatively high rn(sf) (deviatoric compression). r1⁄ and r3⁄ are the
maximum and minimum principal stress; rn⁄ and s⁄ are the normal stress and shear stress on the thrust fault; rn⁄⁄ is the normal stress on the plane (green dashed line)
dipping parallel to the subduction zone fault. (b) Cross-section illustrating overriding plate extension with the development of normal faults, implying a relatively low rn(sf)
(deviatoric tension). r01 and r03 are the maximum and minimum principal stress; r0n and s0 are the normal stress and shear stress on the normal fault; r00n is the normal stress
on the plane (green dashed line) dipping parallel to the subduction zone fault. (c) Normal stress-shear stress diagram showing the Coulomb failure criterion (red line, where s
is the shear stress, lF is the coefficient of internal friction, rn is the normal stress and C is the cohesion) and two Mohr circles. Large Mohr circle on the right is for the
shortening regime in (a) and small Mohr circle on the left is for the extensional regime in (b). Note that rh is the maximum horizontal stress, rv is the vertical stress, hf is the
fracture angle, and h is the angle between the minimum principal stress and the green dashed line in (a, b). (For interpretation of the references to color in this figure legend,
the reader is referred to the web version of this article.)
and backarc) experiences relatively high normal stress on the sub- imately horizontal (Fig. 5b and c). The normal stress (r00n ) on a plane
duction zone interface (deviatoric compression). In contrast, a sub- dipping parallel to the subduction fault will be relatively low (devi-
duction segment bordered by an overriding plate characterized by atoric tension), indicating that the normal stress (rn(sf)) on the
trench-normal extension experiences lower normal stress on the subduction zone fault itself will also be relatively low (low devia-
subduction zone interface. For an overriding plate that experiences toric compression or deviatoric tension) (Fig. 5b and c). In conclu-
shortening, the maximum deviatoric principal stress r1⁄ is approx- sion, for the three largest earthquakes the rupture started in a
imately horizontal, and the minimum deviatoric principal stress region of high normal stress on the subduction zone interface
r3⁄ is the vertical stress (rv) (Fig. 5a and c). The normal stress on and propagated to a zone of lower normal stress on the subduction
a plane dipping parallel to the subduction fault (rn⁄⁄ in Fig. 5a zone interface.
and c) will be relatively high (deviatoric compression), indicating The high mechanical coupling in the regions of rupture initia-
that the normal stress on the subduction zone fault itself (rn(sf) tion is thus not necessarily because of a local high friction coeffi-
in Fig. 5a) will also be relatively high. For an overriding plate that cient at the interface, but is (at least partly) because of a high
experiences extension, the maximum deviatoric principal stress compressive normal stress at the subduction zone interface. With
r01 = rv, and the minimum deviatoric principal stress r03 is approx- increasing normal stress on the subduction fault an increasingly
Fig. 6. Tectonic map showing the subduction setting of the 1960 MW 9.5 Chile earthquake. White arrows with black outline indicate plate velocities, while black arrows at the
subduction zone plate boundary indicate trench-normal trench velocities. Numbers indicate velocity in cm/yr. Velocities calculated in the Indo-Atlantic hotspot references
frame from O’Neill et al. (2005). White dotted line indicates 50 km depth contour of the top of the slab (from Gudmundsson and Sambridge, 1998). Red dashed line outlines
the approximate rupture area of the 1960 MW 9.5 subduction zone thrust earthquake based on Plafker and Savage (1970) and Moreno et al. (2009). Location of the MW 8.1
earthquake is from Cifuentes (1989), while focal mechanism and location of the MW 9.5 earthquake is from Moreno et al. (2009). Structures in the Guanacos fold and thrust
belt region are simplified from Folguera et al. (2007). Background bathymetry and topography are from Sandwell and Smith (2009). Colored lines along subduction zone
trench indicate active deformation regime in the overriding plate (red – shortening; green – neutral). Blue triangles indicate active volcanoes. (For interpretation of the
references to color in this figure legend, the reader is referred to the web version of this article.)
larger shear stress is required to allow slip along a fault plane. (low lF), or because of a local high normal stress on the fault plane
Thus, the deviatoric compressive normal stress keeps the subduc- compared to the surrounding. For the three greatest earthquakes,
tion zone interface locked for a relatively long time and allows and possibly other earthquakes, it appears plausible that the high
the buildup of high elastic strain energy, which can be released mechanical coupling at the asperity (i.e. the epicenter zone) is a
during a giant earthquake with a large fault slip and considerable case of relatively high normal stress on the subduction fault. Thus,
stress drop, while a large lateral rupture propagation is allowed to- variation in mechanical coupling could be explained by variation in
wards regions with a lower normal stress on the subduction inter- normal stress on a subduction fault along which the friction coef-
face. A dominant role of the normal stress, rather than the friction ficient is relatively constant but with high normal stress at an
coefficient, has been proposed for determining the seismic cou- asperity (which we can call a normal stress asperity), rather than
pling coefficient at subduction zones worldwide (Scholz and Cam- an asperity with a relatively high friction coefficient compared to
pos, 2012). the surrounding (which we can call a frictional asperity). For the
The data for all the giant earthquakes with MW > 8.5 show that three giant earthquakes a frictional asperity scenario appears less
their epicenters all occur at trench segments with vOPD\ 6 0 likely than a scenario with a normal stress asperity because these
(Fig. 3a, black diamonds with MW > 8.5), supporting our conceptual three examples are characterized by sediment-filled trenches and
model. In the asperity model (e.g. Lay et al., 1982), it is generally high trench accretion rates along the entire rupture zone length
thought that an asperity is a local area on the fault plane with a (TTS = 2–3, 2 and 5 km and vA\ = 0.3, 0.2 and 0.6 cm/yr for
high strength (i.e. high mechanical coupling). At such an asperity, Chile, Alaska and Sumatra–Andaman, respectively), suggesting a
the strength is relatively high compared to the average strength laterally homogeneous subduction zone interface.
on the fault plane. Such a high strength can be because of a high One can also use geodynamic models of subduction and
friction coefficient (high lF) on a local part of the fault plane com- structural geological observations of the overriding plate regions
pared to the surrounding region with lower friction coefficients at the hypocenters of the three largest recorded earthquakes
Fig. 7. Tectonic map showing the subduction setting of the 1964 MW 9.2 Alaska earthquake. Red dashed line outlines the approximate rupture area of the earthquake based
on Plafker (1965), Kanamori (1970b) and Christensen and Beck (1994). Epicenter location is from Plafker (1965), while focal mechanism solution is from Christensen and Beck
(1994). Thrust structures of the KIfz, Pfz and CSEf are simplified from Bruhn et al. (2004) and Gulick et al. (2007). Fold patterns in the Cook Inlet region are simplified from
Bruhn and Haeussler (2006). Fold and thrust structures north of the Denali fault are simplified from Bemis et al. (2012). Grey shaded area indicates extent of Yakutat terrane
that has been thrust under Alaska (from Koons et al. (2010) based on data from Eberhart-Phillips et al. (2006)). Background bathymetry and topography are from Sandwell
and Smith (2009). Colored lines along subduction zone trench indicate active deformation regime in the overriding plate (red – shortening; green – neutral). For an
explanation of other symbols and notation see figure caption of Fig. 6. CSEf – Chugach-St. Elias fault; KIfz – Kayak Island fault zone; PB-HBfz – Patton Bay-Hanning Bay fault
zone; Pfz – Pamplona fault zone. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
(Figs. 6–8) to distinguish between the possible existence of a fric- compressive overriding plate stress regime with very slow
tional asperity or a normal stress asperity at the subduction zone shortening in the northernmost part to a neutral stress regime
interface. Using geodynamic subduction models, Schellart and with regions of no active deformation or possible very slow exten-
Moresi (2013) found that low-medium friction at the subduction sion further south. Subduction is dominated by trenchward sub-
zone interface causes horizontal deviatoric compression at a short ducting plate motion (vSP\/vS\ = 0.79–0.83) and moderately high
length scale normal to the trench (forearc region, within 125 km subduction rates (vS\ = 6.5–6.8 cm/yr), along a very gentle
from the trench), while high compressive deviatoric normal stres- subduction zone thrust (dST = 12–16°) with minor trench and
ses at the subduction zone interface affect the forearc and the thrust curvature (CT = 6.9 1017–3.7 1014 m2, CST = 1.9
backarc region at a length scale up to 500–600 km from the trench. 1017–2.0 1014 m2) and small trench curvature angles
From investigating the spatial extent of deformation in the overrid- (aT = 1.9° to 4.6°). The values of these physical subduction zone
ing plate for the three biggest recorded earthquakes (Figs. 6–8) one parameters show favorable conditions for generating giant
can observe that at the epicenter location, trench-normal shorten- earthquakes.
ing is observed up to 450 km (Chile), 400 km (Sumatra) and The epicenter of the main earthquake with MW 9.5 occurred on
600 km (Alaska) from the trench. The large distances for these 22 May 1960 at 38.0°S and was preceded by an initial MW 8.1
three cases therefore suggest that the high deviatoric stress in earthquake that occurred to the north on 21 May 1960 with an epi-
the overriding plate is the result of a normal stress asperity at center at 37.2°S (Cifuentes, 1989). The main shock showed south-
the subduction zone interface, not a frictional asperity. ward rupture propagation as far south as the intersection of the
Chile spreading ridge with the subduction zone. The tectonic set-
5.4.1. The 1960 MW 9.5 Chile earthquake ting in the Andean mountain range in the overriding plate shows
The 1960 MW 9.5 Chile earthquake occurred at a 1000 km long there is a major north to south transition at 37–38°S from active
subduction segment along a rupture plane that is thought to ex- Andean shortening and mountain building to absence of shorten-
tend from 37°S to 46°S (Fig. 6) (Plafker and Savage, 1970; Cifu- ing (Dewey and Lamb, 1992; Folguera et al., 2004; Folguera et al.,
entes, 1989; Moreno et al., 2009). The subduction segment is 2006a). Structural fieldwork points to shortening and transpressive
retreating slowly westward (vT\ = 1.2–1.4 cm/yr) with a mildly deformation in the Antinir-Copahue fault system zone down to
southward (vT\ = 0.2–1.6 cm/yr) with slow overriding plate short-

ening in the easternmost part and a neutral overriding plate in the
west (Fig. 7). Subduction is dominated by trenchward subducting
plate motion (vSP\/vS\ = 0.76–0.92) and moderately high subduc-
tion rates (vS\ = 5.0–6.7 cm/yr), along a very gentle subduction
zone thrust (dST = 10–15°) with minor-moderate trench and thrust
curvature (CT = 2.2 1015–1.8 1013 m2, CST = 5.5 1016–
3.8 1014 m2) and small trench curvature angles (aT = 1.0° to
2.3°).
The epicenter of the 1964 earthquake was at 61°N 147.5°W
(Plafker, 1965), which is in the Prince William Sound region at the
northeastern end of the Aleutian-Alaska subduction zone. The
earthquake showed unilateral rupture propagation of 800 km to-
wards the Kodiak Island region in the southwest. The tectonic set-
ting in both the overriding plate and subducting plate shows
considerable variability from northeast to southwest across the re-
gion of the rupture zone. In the northeast underthrusting and flat-
slab subduction of the Yakutat terrane takes place, while in the
southwest normal subduction of Pacific plate oceanic lithosphere
occurs (Ferris et al., 2003; Eberhart-Phillips et al., 2006). The
boundary between the plateau and the oceanic lithosphere strikes
NW–SE and the subducted part of this boundary is thought to
pass below the easternmost part of the Cook Inlet (Eberhart-Phil-
lips et al., 2006), which is located 100–130 km to the west of
the 1964 earthquake epicenter. The earthquake epicenter is lo-
cated within a zone of active compressive to transpressive defor-
mation with the Chugach-St Elias thrust zone to the east (Bruhn
et al., 2004), forearc shortening and thrusting in the offshore region
to the south and southeast including the Pamplona fault zone and
Kayak Island fault zone (Carlson and Molnia, 1977; Bruhn et al.,
2004; Gulick et al., 2007), reverse faulting in the Patton Bay-Han-
ning Bay fault zone (Plafker, 1965; Ferguson et al., 2011), strike-
slip faulting and folding in the Cook inlet region (Bruhn and
Haeussler, 2006) and thrusting north of the Denali fault (Bemis
et al., 2012). Active transpressive deformation in the region of
the epicenter is evident from upper plate earthquake activity with
a combination of thrust-type and strike-slip type earthquake focal
Fig. 8. Tectonic map showing the subduction setting of the 2004 MW 9.3 Sumatra– mechanisms pointing towards dextral shear of 0.5 cm/yr in com-
Andaman subduction zone thrust earthquake. Red dashed line outlines the bination with NW-SE oriented (trench-normal) shortening of
approximate rupture area of the MW 9.3 earthquake based on Ammon et al. 0.3 cm/yr (Leonard et al., 2008). Towards the southwest subduc-
(2005) and Shearer and Bürgmann (2010). Epicenter location is from Lay et al.
tion of normal oceanic lithosphere of the Pacific plate occurs, and
(2005), while moment tensor solution is from the GCMT catalog (note that centroid
location is to the west of epicenter). Structures in the Andaman backarc region are
the overriding plate here is characterized by a relatively neutral
simplified from Curray (2005), while ABf is from Singh et al. (2010). Background strain regime without significant permanent (non-elastic) defor-
bathymetry and topography are from Sandwell and Smith (2009). Semi-transparent mation as suggested by the low level of overriding plate seismicity
grey zone in Indian Ocean region indicates diffuse plate boundary between Indian on Kodiak Island, the southwest Kenai Peninsula, surrounding off-
plate and Australian plate. Colored lines along subduction zone trench indicate
shore regions, and the Alaska Peninsula (Doser et al., 2002). The
active deformation regime in the overriding plate (red – shortening; green –
neutral; blue – extension). For an explanation of other symbols and notation see rupture of the giant 1964 earthquake thus started in a region of
figure caption of Fig. 6. ABf – Aceh Basin fault; Bf – Battee fault; =WAf – West overriding plate shortening due to compression and transpression,
Andaman fault. (For interpretation of the references to color in this figure legend, implying relatively high deviatoric compressive normal stresses on
the reader is referred to the web version of this article.)
the subduction zone interface, while the rupture propagated
southwestward into a region characterized by a neutral overriding
38°S in the eastern orogenic front of the main cordillera, but ab- plate, implying low deviatoric normal stresses on the subduction
sence of shortening and a neutral overriding plate south of 38° zone interface.
with locally potential minor extension (Folguera et al., 2004,
2006a,b). The initial and main earthquakes thus occurred in the re- 5.4.3. The 2004 MW 9.3 Sumatra–Andaman earthquake
gion of the transition zone in the overriding plate with compres- The 2004 MW 9.3 Sumatra–Andaman earthquake occurred at a
sion and transpression, implying a deviatoric compressive normal 1300 km subduction segment (Lay et al., 2005; Stein and Okal,
stress on the subduction zone interface, while the rupture propa- 2005; Ni et al., 2005) with an approximately stationary trench
gated southward into a region characterized by a neutral strain re- (vT\ = 0.7 to 0.7 cm/yr) with a mildly compressive overriding
gime or possibly minor overriding plate extension, implying lower plate stress regime in the south (with slow overriding plate short-
normal stresses (neutral or minor deviatoric tension) on the ening) changing to a mildly-moderately tensional overriding plate
interface. stress regime towards the Andaman islands in the north with an
extending overriding plate due to the opening of the Andaman
5.4.2. The 1964 MW 9.2 Alaska earthquake backarc basin east of the Andaman-Nicobar Islands (Fig. 8). Sub-
The 1964 MW 9.2 Alaska earthquake occurred at an 800 km duction is dominated by trenchward subducting plate motion
long subduction segment (Plafker, 1965) that is retreating slowly (vSP\/vS\ = 0.85–1.25) and low subduction rates (vS\ = 1.9–
4.6 cm/yr), along a gentle subduction zone thrust (dST = 12–20°) 9.5 cm/yr), along a moderately dipping subduction zone thrust
with minor-moderate trench and thrust curvature (CT = 1.1 (dST = 25–30°) with minor trench and thrust curvature (CT = 1.8
1014–9.4 1013 m2, CST = 2.9 1015–2.0 1013 m2) and 1016–4.4 1014 m2, CST = 9.1 1017–2.0 1014 m2) and
small-intermediate trench curvature angles (aT = 1.1–9.8°). a small trench curvature angle (aT = 0.2–2.7°).
The epicenter of the 2004 earthquake is at 3.3°N 96.0°E, which The 2010 MW 8.8 Chile (Maule) earthquake occurred at a
is west of northern Sumatra, and the earthquake showed unilateral 550 km subduction segment (Vigny et al., 2011) that is retreating
northward rupture propagation towards the Nicobar and Andaman slowly westward (vT\ = 1.2–1.4 cm/yr) with a mildly to moderately
Islands as far as 14°N (Lay et al., 2005). The tectonic setting in the compressive overriding plate stress regime with no deformation to
overriding plate shows considerable variability from south to moderate shortening rates (vOPD\ 0.3 to 0 cm/yr). Subduction is
north, with a continental overriding plate in the south (north dominated by trenchward subducting plate motion (vSP\/vS\ =
Sumatra segment) and an oceanic overriding plate (actively open- 0.76–0.80) and moderately high subduction rates (vS\ = 5.8–
ing Andaman backarc basin) north of 6°N (Andaman-Nicobar seg- 6.7 cm/yr), along a gentle subduction zone thrust (dST = 16°) with
ment). The north Sumatra segment is characterized by active minor-moderate trench and thrust curvature (CT = 3.8 1014–
shortening in the forearc, intra-arc and backarc, as indicated by 1.2 1013 m2, CST = 2.1 1014–7.0 1014 m2) and a small
geological and geophysical studies: Seismic reflection studies of trench curvature angle (aT = 3.8° to 2.2°).
the offshore forearc region show active shortening along south- The 2011 MW 9.0 Japan earthquake occurred at a 400–450 km
west-dipping reverse faults (Mosher et al., 2008); Geological stud- subduction segment (Ide et al., 2011; Ozawa et al., 2011) that is
ies of the onshore intra-arc zone show active thrust faulting and advancing westward (vT\ = 2.9 to 2.8 cm/yr) with a highly com-
reverse faulting in the Aceh and Tripa regions of northernmost pressive overriding plate stress regime with fast overriding plate
Sumatra (Sieh and Natawidjaja, 2000); Seismological data show shortening (vOPD\ = 3.0 to 2.7 cm/yr). Subduction is dominated
dextral centroid moment tensor solutions for earthquakes along by trenchward subducting plate motion (vSP\/vS\ = 1.20) and mod-
the Sumatra fault in the intra-arc region and reverse/thrust fault- erately high subduction rates (vS\ = 6.7 cm/yr), along a gentle sub-
ing centroid moment tensor solutions for earthquakes in the back- duction zone thrust (dST = 16°) with minor-moderate trench and
arc region (McCaffrey, 2009). In addition, the Euler parameters thrust curvature (CT = 3.6 1016–2.2 1013 m2, CST = 1.0
from Bird (2003) that quantify the relative motion between the 1016–6.2 1014 m2) and a small trench curvature angle
Burma plate (BU, located west of the Andaman spreading ridge (aT = 0.2–5.2°).
and the Sumatra fault) and the Sunda plate (SU, to the east) point A difference between the Chile 2010 and Japan 2011 earth-
towards active overriding plate shortening (vOPD\ 0.4 cm/yr) quakes and the largest three reported in Sections 5.4.1, 5.4.2 and
along the trench segment that borders the 2004 earthquake epi- 5.4.3 is that the former two show bilateral rupture propagation.
center. The Andaman-Nicobar segment is bordered by the actively The trench-parallel rupture extent was also more limited (450–
opening Andaman backarc basin. This basin is not a standard back- 550 km). These differences might be (partly) due to the compres-
arc basin (with most opening occurring in a direction normal to the sive setting at the subduction zone interface along the entire rup-
trench) as it formed mostly by transtension along a complex dex- ture length for both the Chile 2010 and Japan 2011 earthquakes,
tral fault system between the arc ‘‘sliver plate’’ (Burma plate) which would likely have suppressed lateral rupture propagation.
and the main overriding plate (Sunda plate) (Curray, 2005). The This is in contrast to the Chile 1960, Alaska 1964 and Sumatra–
driving mechanism is generally thought to be the highly oblique Andaman 2004 quakes, for which the epicenter regions character-
convergence between the Indian and Sunda plates. Although most ized by deviatoric compression were flanked on one side by a neu-
of the opening indeed occurs in a direction parallel to the trench, tral or deviatoric tensional normal stress regime at the subduction
the BU-SU relative plate motion parameters from Bird (2003) indi- zone interface, promoting lateral rupture propagation towards
cate that there is also a trench-normal component pointing to these regions, but on the other side by a compressive stress regime
trench-normal backarc opening. Calculations indicate that (i.e. north of 37°S for Chile, Fig. 6, and east of 145°W for Alaska,
vOPD\ 1.2–2.3 cm/yr along the Andaman-Nicobar segment. The Fig. 7) or a trench cusp (at 2°N 96°E for Sumatra–Andaman),
rupture of the 2004 earthquake thus started in the region of over- likely suppressing lateral rupture propagation towards these
riding plate shortening due to compression and transpression, regions.
implying a deviatoric compressive normal stress on the subduction
zone interface (Fig. 5a), while the rupture propagated northward
into a region characterized by minor-moderate trench-normal 6. Predicted spatial distribution for giant earthquakes with
overriding plate extension, implying lower normal stresses (devia- MW > 8.5
toric tension) on the subduction zone interface (Fig. 5b).
It has been proposed by McCaffrey (2008) that any subduction
5.4.4. Other giant earthquakes zone segment can produce a giant subduction zone thrust earth-
We will now discuss three more giant subduction zone thrust quake with MW P 9, given enough time (i.e. hundreds to thou-
earthquakes in the light of their tectonic setting and the eight sands of years). Although the datasets presented here indicate
physical parameters (vOPD\, vT\, vSP\/vS\, dST, CT, CST, aT and vS\) that many physical parameters do not discriminate against earth-
that provide the strongest constraints on the likelihood of giant quake size (e.g. vA\, vC\, ASP, DSE, TTS, TSS, dD, LUMS, DUMS, FBu), several
earthquakes occurring. These include the 1952 MW 8.8–9.0 Kam- parameters appear to do so, most notably vOPD\, vT\, dST, vSP\/vS\,
chatka earthquake (Kanamori, 1976; Okal, 1992; Johnson and Sa- CT, CST and aT. As such, we argue that there are certain subduction
take, 1999), the 2010 MW 8.8 Chile earthquake (Vigny et al., zones and segments that are not capable of producing giant earth-
2011), and the 2011 MW 9.0 Japan earthquake (Ide et al., 2011; quakes. Below we will apply our findings to some of the better-
Ozawa et al., 2011; Simons et al., 2011). quantified historic giant subduction zone thrust earthquakes, for
The 1952 MW 8.8–9.0 Kamchatka earthquake occurred at a which the rupture extent and MW are approximately known. We
600 km subduction segment (Johnson and Satake, 1999) that is will investigate if these events fall within the ranges for the dis-
retreating moderately eastward (vT\ = 2.3–2.6 cm/yr) with a neu- criminating parameters as suggested by our 1900–2012 earth-
tral overriding plate stress regime with no deformation. Subduc- quake dataset. We will also discuss which subduction zone
tion is dominated by trenchward subducting plate motion segments are more likely or less likely to produce a giant thrust
(vSP\/vS\ = 0.72–0.75) and high subduction rates (vS\ = 9.2– earthquake in the future in the light of our findings on controlling
Table 4
Ranges for seven subduction zone parameters (200 km datasets) for four historic giant earthquakes (for entire rupture zone extent).
Parameter Cascadia 1700 MW 9 South Sumatra 1833 MW 8.8–9.2 South Peru 1868 MW 8.8–9.2 North Chile 1877 MW 8.8
vOPD\ 0.4 to 0.6 0 1.1 to 0.7 1.5 to 1.3
vT\ (IA) 1.9–2.7 0.9 to 0.7 1.4 to 0.8 1.0 to 0.4
vSP\/vS\ (IA) 0.3–0.5 1.1–1.2 1.2–1.4 1.1–1.2
dST 11–16 14–15 15–18 15–16
CT 3.1 1014–2.2 1013 1.5 1015–1.4 1013 2.6 1016–4.1 1013 1.5 1014–1.1 1012
CST 7.0 1015–8.7 1014 4.0 1016–3.3 1014 6.6 1017–1.1 1013 3.8 1015–3.0 1013
aT 7.7 to 5.2 0.4–4.0 7.2 to 2.3 11.6 to 1.5
For an explanation of the subduction zone parameters and their units see Table 1. Note that (IA) refers to the Indo-Atlantic moving hotspot reference frame from O’Neill et al.
(2005).
subduction zone parameters. Note that the current work does not terms of the six subduction zone parameters that appear to provide
provide any constraints on when a giant earthquake might take a control on the spatial occurrence of giant subduction zone thrust
place. earthquakes (vOPD\, vT\, dST, vSP\/vS\, CST and aT). Note that we
have excluded CT because of the high interdependence of CT and
6.1. Historic giant earthquakes CST (R = 0.94, see Section 2.1). For the lowest possible score
(S = 0), the values of the six parameters of a particular trench seg-
Historic giant subduction earthquakes with MW > 8.5 include ment are all outside the range observed for MW > 8.5 earthquakes
the 1700 MW 9 Cascadia earthquake (Satake et al., 2003), the (using the rupture zone dataset for ranges). For the highest possi-
1833 MW 8.8–9.2 southern Sumatra earthquake (Zachariasen ble score (S = 6), the values are all inside the range.
et al., 1999), and the 1868 MW 8.8–9.2 southern Peru and 1877
MW 8.8 northern Chile earthquakes just north and south of the 6.3. Subduction segments with high probability of MW > 8.5
Arica bend (Beck and Ruff, 1989; Dorbath et al., 1990; Chlieh earthquakes
et al., 2011; Okal et al., 2006). The lateral rupture extent of these
four historic earthquakes is approximately known (see Fig. 1). In general, the map in Fig. 9 shows a large number of segments
The values of the subduction zone parameters vOPD\, vT\, dST, with high scores. From a total of 241 subduction zone segments,
vSP\/vS\, CST and aT for the rupture zone segments of these four his- 222 have been ranked (19 could not be ranked due to a lack of
toric earthquakes are shown in Table 4. data), and of these 105 segments have the highest score (S = 6).
The vOPD\, vT\ and dST values for the four historic earthquakes This indicates that, on a global scale, at least 44% of the active sub-
all fall within the range for the 1900–2012 subduction thrust duction zone segments posses six out of six important physical
earthquakes with MW > 8.5 as presented in Fig. 3 and Table 2. What characteristics, predicting they are capable of generating giant sub-
is more, for the 1833 Sumatra earthquake the other values also fall duction zone thrust earthquakes with MW > 8.5. Another 42 seg-
within the range, as well as those for the 1868 southern Peru quake ments have the second highest score (S = 5). The intermediate
and the 1700 Cascadia quake, except that they have a slightly low- scores and low scores all have lower numbers, with n = 35 for
er minimum aT (7.2° and 7.7°, respectively) than that for the S = 4, n = 17 for S = 3, n = 14 for S = 2, n = 8 for S = 1 and n = 1 for
1900–2012 MW > 8.5 earthquakes (6.3°). For the northern Chile S = 0.
1877 earthquake the maxima for CT and CST are slightly higher than If we look at Fig. 9 in more detail, then the map shows that all
those for the 1900–2012 MW > 8.5 earthquakes, while the mini- subduction segments, for which an MW > 8.5 earthquake has been
mum aT (11.6°) is significantly lower than that for the 1900– recorded (a total of 32 segments), have the highest score (S = 6), as
2012 MW > 8.5 quakes (6.3°). The trench segment with expected. The map further shows several other subduction zone
aT = 11.6° sits at the northern extent of the 1877 rupture zone regions with high scores (S = 5, 6). These regions include most of
and its high curvature might have prevented propagation of the the South American subduction zone, from the Chile Ridge triple
rupture past this zone of high curvature. junction to northern Bolivia, (average S = 5.8, range = 4–6), the en-
tire Aleutians-Alaska subduction zone (average S = 5.9, range = 5–
6.2. Predicting future locations of giant subduction earthquakes 6), the entire Sunda subduction zone except for its northernmost
segment in northern Andaman (average S = 5.9, range = 5–6), the
The compilations presented in Fig. 3 and Table 2 imply that sev- Japan-Kamchatka segment (average S = 5.2, range = 4–6), the
eral of the subduction zone parameters provide a control on the northern Ryukyu-Nankai segment (average S = 5.4, range = 4–6),
spatial occurrence of giant subduction zone thrust earthquakes, the entire Cascadia subduction zone (average S = 5.6, range = 5–
including vOPD\, vT\, dST, vSP\/vS\, CT, CST and aT. The observed 6), most of the Central America-Mexico subduction zone (average
ranges of these parameters for segments with MW > 8.5 are rela- S = 5.8, range = 4–6), the Makran subduction zone (average
tively narrow (Fig. 3), and the probability that they would result S = 5.8, range = 5–6), the Lesser Antilles-Puerto Rico subduction
from mere chance is very low (P = 0.001–0.003 for RZD, Table 2). zone (average S = 5.1, range = 4–6) and the southern Kermadec-
The parameters suggest that those subduction segments with rapid Hikurangi subduction segment (average S = 5.3, range = 4–6). Sev-
backarc opening (vOPD\ > 3 cm/yr), rapid trench retreat (vT\ > eral smaller subduction zones also have high scores, including
3 cm/yr), a steep subduction thrust (dST > 30°), low partitioning the North Sulawesi subduction zone (average S = 6), Manila (three
(vSP\/vS\ < 0.3) or high curvature (CT > 1 1012 m2, CST > 2 middle segments with S = 5) the Calabria subduction zone (average
1013 m2, |aT| > 10°) are incapable of producing MW > 8.5 earth- S = 5.5, range = 5–6) and central Hellenic (two segments with S = 5
quakes. In Fig. 9 we present a global map of the active subduction and 6).
zones, where the 200 km trench segments have been ranked in If we take the conceptual model developed in Section 5.4 for the
terms of their predicted capability of generating a giant subduction three largest subduction zone earthquakes and apply it to other
zone earthquake with MW > 8.5. A high score means a high pre- subduction zone regions shown in Fig. 9 with high scores, then sev-
dicted capability and vice versa. We have ranked the segments in eral regions jump out due to their comparable tectonic setting.
Fig. 9. Global map of the active subduction zones, where the 200 km trench segments have been ranked in terms of their predicted capability of generating a giant subduction
zone earthquake with MW > 8.5. The segments have been ranked in terms of six subduction zone parameters: trench-normal overriding plate deformation rate (vOPD\),
trench-normal trench velocity (vT\), subduction thrust dip angle (dST), subduction partitioning (vSP\/vS\), subduction thrust curvature (CST) and trench curvature angle (aT).
For the lowest possible score (S = 0), the values of the six parameters of a particular trench segment are all outside the ranges observed for MW > 8.5 earthquakes (using the
rupture zone dataset for ranges, see Fig. 3 and Table 2), implying a low risk of producing an MW > 8.5 earthquake. For the highest possible score (S = 6), all six values are inside
the ranges, implying a high risk of producing an MW > 8.5 earthquake. Abbreviations for the subduction zone segments are explained in the figure caption of Fig. 1.
These regions include the Hikurangi-southern Kermadec subduc- Alaska 1964 and Sumatra–Andaman 2004 giant earthquakes, one
tion segment and the Central America subduction segment. Other could expect a giant subduction earthquake with an epicenter at
regions could include the Nankai-northeastern Ryukyu subduction the subduction zone plate interface in the southwest Hikurangi re-
segment, the western Hellenic subduction segment, the Lesser gion, and unilateral rupture propagation towards the northeast.
Antilles-Puerto Rico subduction zone and the Manila subduction Fig. 9 shows relatively high scores (S = 4–5) for the two southern-
zone. Below we will describe the Hikurangi-southern Kermadec most Hikurangi segments, and the highest scores for the next three
subduction segment and the Central America subduction segment segments to the north (S = 6). Wallace et al. (2009) documented
in more detail. high interseismic coupling coefficients (0.8–1.0) in the south but
lower ones (0.1–0.2) in the central and northern Hikurangi region,
6.3.1. The Hikurangi-southern Kermadec subduction zone segment suggesting higher elastic strain buildup in the south.
The Hikurangi-southern Kermadec subduction segment forms
the southern part of the Tonga-Kermadec-Hikurangi subduction 6.3.2. The Central America subduction zone segment
zone with west-northwestward subduction of the Pacific plate be- The Central America subduction segment forms part of the
low the Australian plate (Fig. 10). The Kermadec trench and north- Mexico-Central America subduction zone with northeastward sub-
ernmost Hikurangi trench are flanked by the Kermadec arc and the duction of the Cocos plate below the North American plate in the
Havre Trough, an active backarc basin with extension at 1–2 cm/yr northwest, the Caribbean plate in the middle and the Panama plate
in the south (Wright, 1993; Bird, 2003; Power et al., 2012). The in the southeast (Fig. 11). In the southeast the buoyant Cocos Ridge
central Hikurangi margin is flanked by the Axial Ranges and the indents and subducts below the Panama plate, causing active over-
Taupo volcanic zone to the west, the latter of which is an active riding plate shortening in the inner forearc (Fila Costena thrust
continental backarc basin with active extension at 1.5 cm/yr in belt) and outer forearc (e.g. Burica Peninsula) as indicated by struc-
the north and decreasing southward to <0.5 cm/yr at 39°S (Wallace tural investigations (Fisher et al., 2004; Sitchler et al., 2007; Morell
et al., 2004). Further to the south, the southernmost Hikurangi et al., 2011) and geodetic investigations (LaFemina et al., 2009). It
margin is characterized by overriding plate shortening west of also causes backarc shortening where thrusting is observed at the
the Axial Ranges with active northwest-southeast shortening in Panama-Caribbean plate boundary (Silver et al., 1990), and at the
the Kapiti–Manawatu fault system of the southeastern Wanganui northern boundary of the inactive arc and in the Limon backarc re-
Basin (Lamarche et al., 2005; Barnes et al., 2010). Geodetic investi- gion north of the inactive arc (Morell et al., 2012). Towards the
gations predict an oblique shortening rate of 0.1–0.4 cm/yr here northwest there is structural, geological, geomorphological and
(Wallace et al., 2004). geochemical evidence for active slow extension (0.3–0.6 cm/yr)
The trench-parallel gradient in overriding plate deformation in the intra-arc region in Nicaragua (Phipps Morgan et al., 2008).
implies relatively high normal stress (deviatoric compression) on The overriding plate deformation implies relatively high normal
the subduction zone interface in the southwest changing to rela- stress (deviatoric compression) on the subduction zone interface in
tively low normal stress (deviatoric tension) towards the northeast the southeast changing to relatively low normal stress (deviatoric
(Fig. 10). In analogy with the tectonic settings of the Chile 1960, tension) towards the northwest. In analogy with the tectonic
Fig. 10. Tectonic map showing the setting of the Hikurangi-southern Kermadec subduction segment. White arrows with black outline indicate plate velocities, while black
arrows at the subduction zone plate boundary indicate trench-normal trench migration velocities. Numbers indicate velocity in cm/yr. Velocities calculated in the Indo-
Atlantic hotspot references frame from O’Neill et al. (2005) using the geodetic relative plate motion model from Kreemer et al. (2003). White dotted line indicates 50 km
depth contour of the top of the slab (from Gudmundsson and Sambridge, 1998). Background bathymetry and topography are from Sandwell and Smith (2009). Semi-
transparent red area indicates predicted region within which the epicenter of a giant earthquake will occur, while red dotted arrow indicates predicted rupture propagation
direction. Fault patterns on and immediately off-shore North Island are simplified from Barnes et al. (2010). Colored lines along subduction zone trench indicate active
deformation regime in the overriding plate (red – shortening; blue – extension). Blue triangles indicate active volcanoes. KMfs – Kapiti–Manawatu fault system, Tvz – Taupo
volcanic zone. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)
settings of the Chile 1960, Alaska 1964 and Sumatra–Andaman partitioning. The Tonga subduction segment is unlikely to produce
2004 giant earthquakes, one could expect a giant earthquake with MW > 8.5 thrust earthquakes mainly due to its very rapid backarc
an epicenter in southeast Costa Rica where the Cocos Ridge sub- opening and rapid trench retreat (average S = 3.2, range = 1–4).
ducts (deviatoric compression), and unilateral rupture propagation The Mariana subduction segment, with moderate scores (average
towards the Nicaragua subduction segment in the northwest with S = 4.0, range = 2–6), is also unlikely to produce MW > 8.5 thrust
low or tensional deviatoric normal stresses on the interface. earthquakes, even though it does have one trench segment with
a ranking of S = 6 and three with a ranking of S = 5. These high val-
6.4. Subduction segments with low probability of MW > 8.5 ues are for two isolated trench segments and two adjoining seg-
earthquakes ments that are surrounded by segments with lower values, and
thus large lateral rupture propagation is unlikely. Finally, the small
The map predicts that the Scotia subduction zone is the least Halmahera subduction zone has low scores (average S = 1.3,
likely of all subduction zones to produce MW > 8.5 subduction zone range = 1–2) and is also unlikely to produce MW > 8.5 earthquakes
thrust earthquakes (average S = 1, observed range = 0–2) due to its due to its rapid trench retreat, steep slab dip and high curvature.
rapid backarc opening, rapid trench retreat, steep thrust dip, low
partitioning and high curvature (Fig. 9). The map further shows 7. Conclusions
that the New Britain-San Cristobal-New Hebrides subduction zone
has low to moderate scores (1–4). In particular, it suggests that the In this paper we have presented global subduction zone data-
New Hebrides subduction segment is unlikely to produce MW > 8.5 sets for the maximum recorded subduction zone thrust earth-
earthquakes (average S = 2.1, observed range = 1–3) due to its rapid quakes that have occurred at active subduction zones around the
backarc opening, rapid trench retreat, steep thrust dip, and low globe in the period January 1900–June 2012. The datasets illustrate
Fig. 11. Tectonic map showing the setting of the Central America subduction segment. Fault pattern in the Fila Costena is simplified from Fisher et al. (2004), while the
extensional fault pattern is simplified from Phipps Morgan et al. (2008). Background bathymetry and topography are from Sandwell and Smith (2009). Semi-transparent red
area indicates predicted region within which the epicenter of a giant earthquake will occur, while red dashed arrow indicates predicted rupture propagation direction.
Colored lines along subduction zone trench indicate active deformation regime in the overriding plate (red – shortening; green – neutral; blue – extension). For an
explanation of other symbols and notation see figure caption of Fig. 10. (For interpretation of the references to color in this figure legend, the reader is referred to the web
version of this article.)
how the observed spatial variability in MW might depend on 24 dif- ments show distributions of data points that are not random
ferent subduction zone parameters. We draw the following main but are distinct for the above mentioned parameters (Fig. 3)
conclusions from our work: with distinct (narrow) ranges for segments with MW > 8.5.
4. The distinct (narrow) ranges for segments with MW > 8.5 give
1. For our (preferred) two datasets, the rupture zone dataset and low values for vOPD\ (i.e. shortening or relatively neutral
epicenter dataset (segmentation into 200 km trench segments, overriding plate) and vT\ (trench advance or slow trench
maximum of 216 segments with sufficient data), all least- retreat), and high values for vSP\/vS\ (subduction dominated
squares linear regression correlations are negligible-low by trenchward subducting plate motion) (Fig. 3). Such values
(|R| = 0.00–0.30) and statistically insignificant (Fig. 3). The promote relatively high deviatoric normal stresses on the
dataset with geologically defined subduction zone segments subduction zone interface (deviatoric compression) (Fig. 5a
(maximum of 37 segments with sufficient data) shows negligi- and c), promoting the buildup of high elastic strain energy
ble-moderate correlations (|R| = 0.02–0.51) with the highest that can be released in a giant earthquake with a high fault
correlation for slab width (Fig. 4); only slab width shows a sta- slip (meters to tens of meters) with stress drops of MPa to
tistically significant correlation at 95% confidence level. tens of MPa. High values for vOPD\ and vT\ (i.e. rapid exten-
2. For a number of parameters, including slab age, trench accre- sion and trench retreat), and low values for vSP\/vS\ (i.e. sub-
tion/erosion rate, trench sediment thickness and upper man- duction dominated by trench retreat), promote relatively
tle slab negative buoyancy force, the low correlations are low deviatoric normal stresses on the subduction zone inter-
likely due to the absence of any physical relation between face (i.e. deviatoric tension or neutral stress) (Fig. 5b and c),
the parameters and earthquake potential of the subduction promoting the buildup of only minor elastic strain energy.
segments. 5. The distinct (narrow) ranges for segments with MW > 8.5
3. For several parameters, including overriding plate deforma- give low values for dST, CST and aT (Fig. 3). Such values pro-
tion rate, trench velocity, subduction partitioning, subduc- mote large rupture propagation parallel to the dip of the
tion thrust dip angle, subduction thrust curvature and subduction zone thrust interface (dST) and laterally along
trench curvature angle, the low correlations possibly result the trench (CST and aT), thus favoring a large earthquake
from the short period of global instrumental observations, rupture area (of the order 105 km2), which is characteris-
which can be an order of magnitude shorter than the longest tic of giant earthquakes. Strongly curved subduction zones
recurrence interval of giant subduction zone earthquakes. (either concave or convex) and steep subduction thrust dips
Indeed, a comparative investigation of the observed ranges likely limit the lateral propagation of the earthquake rup-
of the physical parameters for subduction segments with ture, while steep dips also limit the downdip extent of
MW > 8.5 and the observed ranges for all subduction seg- the rupture plane.
6. Our research has shown that epicenters of giant earthquakes References

with MW > 8.5 only occur at trench segments with vOPD\ 6 0
(epicenter dataset in Fig. 3a with black diamonds and green Abe, K., 1972. Mechanism and tectonic implications of the 1966 and 1970 Peru
earthquakes. Physics of the Earth and Planetary Interiors 5, 367–379.
circles), suggesting that such earthquakes initiate at locked Agnew, D., Berger, J., Buland, R., Farrell, W., Gilbert, F., 1976. International
segments of the subduction zone interface that have a rela- deployment of accelerometers: a network for very long period seismology.
tively high normal stress (deviatoric compression) on the EOS, Transactions American Geophysical Union 57, 180–188. http://dx.doi.org/
10.1029/EO057i004p00180.
subduction zone interface (i.e. a normal stress asperity). Note Allen, T.I., Marano, K., Earle, P.S., Wald, D.J., 2009. PAGER-CAT: a composite
that the normal stress asperities should be seen separately earthquake catalog for calibrating global fatality models. Seismological
from asperities that result from variation in friction coeffi- Research Letters 80, 50–56.
Ammon, C.J., Ji, C., Thio, H.-K., Robinson, D., Ni, S., Hjorleifsdottir, V., Kanamori, H.,
cient along the subduction zone interface (i.e. a frictional
Lay, T., Das, S., Helmberger, D., Ichinose, G., Polet, J., Wald, D., 2005. Rupture
asperity). It is evident from the Mohr–Coulomb failure crite- process of the 2004 Sumatra–Andaman Earthquake. Science 308, 1133–1139.
rion that with increasing normal stress an increase in shear http://dx.doi.org/10.1126/science.1112260.
stress is required to allow slip along the subduction zone Ando, M., 1975. Source mechanisms and tectonic significance of historical
earthquakes along the Nankai Trough, Japan. Tectonophysics 27, 119–140.
thrust interface (Fig. 5). The deviatoric normal stress on the Barnes, P.M., Lamarche, G., Bialas, J., Henrys, S., Pecher, I., Netzeband, G.L., Greinert,
subduction interface results from the relative motion J., Mountjoy, J.J., Pedley, K., Crutchley, G., 2010. Tectonic and geological
between the subduction zone hinge and the overriding plate, framework for gas hydrates and cold seeps on the Hikurangi subduction
margin, New Zealand. Marine Geology 272, 26–48. http://dx.doi.org/10.1016/
with convergence between the two inducing deviatoric com- j.margeo.2009.03.012.
pression and divergence causing deviatoric tension. It fol- Beavan, J., Wang, X., Holden, C., Wilson, K., Power, W., Prasetya, G., Bevis, M.,
lows that the spatial probability of the occurrence of giant Kautoke, R., 2010. Near-simultaneous great earthquakes at Tongan megathrust
and outer rise in September 2009. Nature 466, 959–963. http://dx.doi.org/
subduction thrust earthquakes is therefore linked to the 10.1038/nature09292.
large-scale dynamics of the subducting plate, overriding Beck, S.L., Christensen, D.H., 1991. Rupture process of the February 4, 1965, Rat
plate, slab and ambient mantle. This is because slab roll- Island earthquake. Journal of Geophysical Research 96, 2205–2221.
Beck, S.L., Ruff, L.J., 1987. Rupture process of the great 1963 kurile islands
back/roll-forward processes and subduction-induced mantle earthquake sequence: asperity interaction and multiple event rupture. Journal
flow patterns provide significant control on the relative of Geophysical Research 92, 14123–14138.
motions of the subduction zone hinge and overriding plate Beck, S.L., Ruff, L.J., 1989. Great earthquakes and subduction along the Peru trench.
Physics of the Earth and Planetary Interiors 57, 199–224.
(Schellart and Moresi, 2013), and also play an important role
Bemis, S.P., Carver, G.A., Koehler, R.D., 2012. The Quaternary thrust system of the
in determining trench curvature and slab curvature (Schell- northern Alaska range. Geosphere 8, 196–205. http://dx.doi.org/10.1130/
art et al., 2007). Another feature of major importance is the GES00695.1.
presence of buoyant features on the subducting plate, such Bird, P., 2003. An updated digital model of plate boundaries. Geochemistry
Geophysics Geosystems 4, 1027. http://dx.doi.org/10.1029/2001GC000252.
as plateaus, which affect the local dip angle of the slab, and Bruhn, R.L., Haeussler, P.J., 2006. Deformation driven by subduction and microplate
which can cause local overriding plate shortening and com- collision: geodynamics of Cook Inlet basin, Alaska. Geological Society of
pression at the subduction zone interface, thereby forming America Bulletin 118, 289–303. http://dx.doi.org/10.1130/B25672.1.
Bruhn, R.L., Pavlis, T.L., Plafker, G., Serpa, L., 2004. Deformation during terrane
a potential nucleation zone for epicenters of giant accretion in the Saint Elias orogen, Alaska. Geological Society of America
earthquakes. Bulletin 116, 771–787. http://dx.doi.org/10.1130/B25182.1.
7. The current work does not provide any insight into the tem- Butler, R., Lay, T., Creager, K., Earl, P., Fischer, K., Gaherty, J., Laske, G., Leith, B., Park,
J., Ritzwoller, M., Tromp, J., Wen, L., 2004. The global seismographic network
poral probability of giant subduction zone thrust earth- surpasses its design goal. EOS, Transactions American Geophysical Union 85,
quakes, but it does give insight into their spatial 225–232. http://dx.doi.org/10.1029/2004EO230001.
probability. A number of trench segments of active subduc- Byrne, D.E., Sykes, L.R., Davis, D.M., 1992. Great thrust earthquakes and aseismic slip
along the plate boundary of the Makran subduction zone. Journal of
tion zones have parametric values for vOPD\, vT\, vSP\/vS\, Geophysical Research 97, 449–478.
dST, CST and aT that are in the same range as those trench seg- Carena, S., 2011. Subducting-plate topography and nucleation of great and giant
ments that have experienced MW > 8.5 earthquakes (Fig. 9). earthquakes along the South American trench. Seismological Research Letters
82, 629–637.
These include several subduction segments for which histor-
Carlson, P.R., Molnia, B.F., 1977. Submarine faults and slides on the continental
ical giant earthquakes (MW > 8.5) have been reported, in par- shelf, northern gulf of Alaska. Marine Geotechnology 2, 275–290.
ticular Cascadia 1700, southern Sumatra 1833, northern Chlieh, M., Perfettini, H., Tavera, H., Avouac, J.-P., Remy, D., Nocquet, J.-M.,
Chile 1877 and southern Peru 1868. There are also segments, Rolandone, F., Bondoux, F., Gabalda, G., Bonvalot, S., 2011. Interseismic
coupling and seismic potential along the Central Andes subduction zone.
including Hikurangi-southern Kermadec and Central Amer- Journal of Geophysical Research 116, B12405. http://dx.doi.org/10.1029/
ica (Figs. 10 and 11), that have comparable tectonic settings 2010JB008166.
to those segments that have experienced the three largest Christensen, D.H., Beck, S.L., 1994. The rupture process and tectonic implications of
the great 1964 Prince William Sound earthquake. Pure and Applied Geophysics
recorded earthquakes (Chile 1960, Alaska 1964 and Suma- 142, 29–53.
tra–Andaman 2004, Figs. 6–8), which are characterized by Cifuentes, I.L., 1989. The 1960 Chilean earthquakes. Journal of Geophysical Research
unilateral rupture propagation along the subduction zone 94, 665–680.
Clift, P., Vannucchi, P., 2004. Controls on tectonic accretion versus erosion in
interface from a region of compressive normal stress towards subduction zones: implications for the origin and recycling of the continental
a region of neutral stress or deviatoric tension. If the concep- crust. Reviews of Geophysics 42. http://dx.doi.org/10.1029/2003RG000127,
tual model as derived from the largest three earthquakes RG2001.
Conrad, C.P., Behn, M.D., 2010. Constraints on lithosphere net rotation and
also applies to these regions, then we predict that they will asthenospheric viscosity from global mantle flow models and seismic
experience a giant subduction zone thrust earthquake in anisotropy. Geochemistry Geophysics Geosystems 11. http://dx.doi.org/
the future. 10.1029/2009GC002970, Q05W05.
Conrad, C.P., Bilek, S., Lithgow-Bertelloni, C., 2004. Great earthquakes and slab pull:
interaction between seismic coupling and plate-slab coupling. Earth and
Acknowledgments Planetary Science Letters 218, 109–122.
Contreras-Reyes, E., Carrizo, D., 2011. Control of high oceanic features and
subduction channel on earthquake ruptures along the Chile–Peru subduction
We thank two anonymous reviewers and the editor George zone. Physics of the Earth and Planetary Interiors 186, 49–58. http://dx.doi.org/
Helffrich for their constructive and thoughtful comments. This re- 10.1016/j.pepi.2011.03.002.
Curray, J.R., 2005. Tectonics and history of the Andaman Sea region. Journal of Asian
search was supported by Future Fellowship FT110100560 and Dis-
Earth Sciences 25, 187–232.
covery grants DP110103387 and DP120102983 from the Davies, G.F., Richards, M.A., 1992. Mantle convection. Journal of Geology 100, 151–
Australian Research Council awarded to WPS. 206.
De Franco, R., Govers, R., Wortel, R., 2008. Nature of the plate contact and Johnson, J.M., Satake, K., 1999. Asperity distribution of the 1952 great Kamchatka
subduction zones diversity. Earth and Planetary Science Letters 271, 245–253. earthquake and its relation to future earthquake potential in Kamchatka. Pure
http://dx.doi.org/10.1016/j.epsl.2008.04.019. and Applied Geophysics 154, 541–553.
Dewey, J., Byerly, P., 1969. The early history of seismometry (to 1900). Bulletin of Johnson, J.M., Tanioka, Y., Ruff, L.J., Satake, K., Kanamori, H., Sykes, L.R., 1994. The
the Seismological Society of America 59, 183–228. 1957 great Aleutian earthquake. Pure and Applied Geophysics 142, 3–28.
Dewey, J.F., Lamb, S.H., 1992. Active tectonics of the Andes. Tectonophysics 205, 79– Kanamori, H., 1970a. Synthesis of long-period surface waves and its application to
95. earthquake source studies-Kuril Islands earthquake of October 13, 1963. Journal
Dorbath, L., Cisternas, A., Dorbath, C., 1990. Assessment of the size of large and great of Geophysical Research 75, 5011–5027.
historical earthquakes in Peru. Bulletin of the Seismological Society of America Kanamori, H., 1970b. The Alaska earthquake of 1964: radiation of long-period
80, 551–576. surface waves and source mechanism. Journal of Geophysical Research 75,
Doser, D.I., Brown, W.A., Velasquez, M., 2002. Seismicity of the Kodiak Island region 5029–5040.
(1964–2001) and its relation to the 1964 great Alaska earthquake. Bulletin of Kanamori, H., 1972. Tectonic implications of the 1944 Tonankai and the 1946
the Seismological Society of America 92, 3269–3292. Nankaido earthquakes. Physics of the Earth and Planetary Interiors 5, 129–139.
Duarte, J.C., Schellart, W.P., Cruden, A.R., 2013. Three-dimensional dynamic Kanamori, H., 1976. Re-examination of the Earth’s free oxcillations excited by the
laboratory models of subduction with an overriding plate and variable Kamchatka earthquake of November 4, 1952. Physics of the Earth and Planetary
interplate rheology. Geophysical Journal International 195, 47–66. http:// Interiors 11, 216–226.
dx.doi.org/10.1093/gji/ggt257. Kanamori, H., 1977. The energy release in great earthquakes. Journal of Geophysical
Eberhart-Phillips, D., Christensen, D.H., Brocher, T.M., Hansen, R., Ruppert, N.A., Research 82, 2981–2987.
Haeussler, P.J., Abers, G.A., 2006. Imaging the transition from Aleutian Kanamori, H., 1986. Rupture process of subduction-zone earthquakes. Annual
subduction to Yakutat collision in central Alaska, with local earthquakes and Review Earth and Planetary Sciences 14, 293–322.
active source data. Journal of Geophysical Research 111, B11303. http:// Kanamori, H., Anderson, D.L., 1975. Theoretical basis of some empirical relations in
dx.doi.org/10.1029/2005JB004240. seismology. Bulletin of the Seismological Society of America 65, 1073–1095.
Elsasser, W.M., 1971. Sea-floor spreading as thermal convection. Journal of Karig, D.E., Caldwell, J.G., Parmentier, E.M., 1976. Effects of accretion on the
Geophysical Research 76, 1101–1112. geometry of the descending lithosphere. Journal of Geophysical Research 81,
Engdahl, E.R., Villaseñor, A., 2002. Global Seismicity: 1900-1999. International 6281–6291.
handbook of earthquake and engineering seismology, 81A, 665–690. Kelleher, J., McCann, W., 1976. Buoyant zones, great earthquakes, and unstable
Ferguson, K., Armstrong, P.A., Haeussler, P.J., Arkle, J.C., 2011. Rock Uplift above the boundaries of subduction. Journal of Geophysical Research 81, 4885–4896.
Yakutat Megathrust on Montague Island, Prince William Sound, Alaska. Kelleher, J., Savino, J., Rowlett, H., McCann, W., 1974. Why and where great thrust
American Geophysical Union Fall Meeting, San Francisco, pp. T33A–2384. earthquakes occur along island arcs. Journal of Geophysical Research 79, 4889–
Ferris, A., Abers, G.A., Christensen, D.H., Veenstra, E., 2003. High resolution image of 4899.
the subducted Pacific (?) plate beneath central Alaska, 50–150 km depth. Earth Kido, M., Osada, Y., Fujimoto, H., Hino, R., Ito, Y., 2011. Trench-normal variation in
and Planetary Science Letters 214, 575–588. http://dx.doi.org/10.1016/S0012- observed seafloor displacements associated with the 2011 Tohoku-Oki
821X(03)00403-5. earthquake. Geophysical Research Letters 38, L24303. http://dx.doi.org/
Fisher, D.M., Gardner, T.W., Sak, P.B., Sanchez, J.D., Murphy, K., Vannucchi, P., 2004. 10.1029/2011GL050057.
Active thrusting in the inner forearc of an erosive convergent margin, Pacific coast, King, S.D., Hager, B.H., 1990. The relationship between plate velocity and trench
Costa Rica. Tectonics 23, TC2007. http://dx.doi.org/10.1029/2002TC001464. viscosity in Newtonian and power-law subduction calculations. Geophysical
Folguera, A., Ramos, V.A., Hermanns, R.L., 2004. Neotectonics in the foothills of the Research Letters 17, 2409–2412.
southernmost central Andes (37°_–38_°S): evidence of strike-slip displacement Koons, P.O., Hooks, B.P., Pavlis, T., Upton, P., Barker, A.D., 2010. Three-dimensional
along the Antiñir-Copahue fault zone. Tectonics 23, TC5008. http://dx.doi.org/ mechanics of Yakutat convergence in the southern Alaskan plate corner.
10.1029/2003TC001533. Tectonics 29, TC4008. http://dx.doi.org/10.1029/2009TC002463.
Folguera, A., Ramos, V.A., González Díaz, E.F., Hermanns, R., 2006a. Miocene to Kreemer, C., 2009. Absolute plate motions constrained by shear wave splitting
Quaternary deformation of the Guañacos fold-and-thrust belt in the Neuquén orientations with implications for hot spot motions and mantle flow. Journal of
Andes between 37°S and 37°300 S. Geological Society of America Special Paper Geophysical Research 114, B10405. http://dx.doi.org/10.1029/2009JB006416.
407, 247–266. http://dx.doi.org/10.1130/2006.2407(11). Kreemer, C., Holt, W.E., Haines, A.J., 2003. An integrated global model of present-day
Folguera, A., Zapata, T., Ramos, V.A., 2006b. Late Cenozoic extension and the plate motions and plate boundary deformation. Geophysical Journal
evolution of the Neuquén Andes. Geological Society of America Special Paper International 154, 8–34.
407, 267–285. http://dx.doi.org/10.1130/2006.2407(12). LaFemina, P., Dixon, T.H., Govers, R., Norabuena, E., Turner, H., Saballos, A., Mattioli,
Folguera, A., Ramos, V.A., Zapata, T., Spagnuolo, M.G., 2007. Andean evolution at the G., Protti, M., Strauch, W., 2009. Fore-arc motion and Cocos Ridge collision in
Guanacos and Chos Malal fold and thrust belts (36°300 _–37°s). Journal of Central America. Geochemistry Geophysics Geosystems 10, Q05S14. http://
Geodynamics 44, 129–148. http://dx.doi.org/10.1016/j.jog.2007.02.003. dx.doi.org/10.1029/2008GC002181.
Forsyth, D.W., Uyeda, S., 1975. On the relative importance of the driving forces of Lamarche, G., Proust, J.-N., Nodder, S.D., 2005. Long-term slip rates and fault
plate motion. The Geophysical Journal of the Royal Astronomical Society 43, interactions under low contractional strain, Wanganui Basin, New Zealand.
163–200. Tectonics 24, TC4004. http://dx.doi.org/10.1029/2004TC001699.
Grevemeyer, I., Diaz-Naveas, J.L., Ranero, C.R., Villinger, H.W.Ocean Drilling Program Lamb, S., Davis, P., 2003. Cenozoic climate change as a possible cause for the rise of
Leg 202 Scientific Party, 2003. Heat flow over the descending Nazca plate in the Andes. Nature 425, 792–797.
central Chile, 32S to 41S: observations from ODP Leg 202 and the occurrence of Lay, T., Kanamori, H., Ruff, L., 1982. The asperity model and the nature of large
natural gas hydrates. Earth and Planetary Science Letters 213, 285–298. http:// subduction zone earthquakes. Earthquake Prediction Research 1, 3–71.
dx.doi.org/10.1016/S0012-821X(03)00303-0. Lay, T., Kanamori, H., Ammon, C.J., Nettles, M., Ward, S.N., Aster, R.C., Beck, S.L., Bilek,
Gripp, A.E., Gordon, R.G., 2002. Young tracks of hotspots and current plate velocities. S.L., Brudzinski, M.R., Butler, R., DeShon, H.R., Ekstrom, G., Satake, K., Sipkin, S.,
Geophysical Journal International 150, 321–361. 2005. The Great Sumatra–Andaman Earthquake of 26 December 2004. Science
Gudmundsson, O., Sambridge, M.J., 1998. A regionalized upper mantle (RUM) 308, 1127–1133. http://dx.doi.org/10.1126/science.1112250.
seismic model. Journal of Geophysical Research 103, 7121–7136. Lay, T., Kanamori, H., Ammon, C.J., Hutko, A.R., Furlong, K., Rivera, L., 2009. The
Gulick, S.P.S., Lowe, L.A., Pavlis, T.L., Gardner, J.V., Mayer, L.A., 2007. Geophysical 2006–2007 Kuril Islands great earthquake sequence. Journal of Geophysical
insights into the Transition fault debate: propagating strike slip in response to Research 114, B11308. http://dx.doi.org/10.1029/2008JB006280.
stalling Yakutat block subduction in the Gulf of Alaska. Geology 35, 763–766. Lay, T., Ammon, C.J., Kanamori, H., Rivera, L., Koper, K.D., Hutko, A.R., 2010. The 2009
http://dx.doi.org/10.1130/G23585A.1. Samoa–Tonga great earthquake triggered doublet. Nature 466, 964–968. http://
Hager, B.H., 1984. Subducted slabs and the geoid: constraints on mantle rheology dx.doi.org/10.1038/nature09214.
and flow. Journal of Geophysical Research 89, 6003–6015. Leonard, L.J., Mazzotti, S., Hyndman, R.D., 2008. Deformation rates estimated from
Hardebeck, J.L., 2012. Coseismic and postseismic stress rotations due to great earthquakes in the northern Cordillera of Canada and eastern Alaska. Journal of
subduction zone earthquakes. Geophysical Research Letters 39, L21313. http:// Geophysical Research 113, B08406. http://dx.doi.org/10.1029/2007JB005456.
dx.doi.org/10.1029/2012GL053438. Llenos, A.L., McGuire, J.J., 2007. Influence of fore-arc structure on the extent of great
Hasegawa, A., Yoshida, K., Okada, T., 2011. Nearly complete stress drop in the 2011 subduction zone earthquakes. Journal of Geophysical Research 112, B09301.
MW 9.0 off the Pacific coast of Tohoku Earthquake. Earth Planets Space 63, 703– http://dx.doi.org/10.1029/2007JB004944.
707. http://dx.doi.org/10.5047/eps.2011.06.007. López, A.M., Okal, E.A., 2006. A seismological reassessment of the source of the 1946
Heuret, A., Lallemand, S., 2005. Plate motions, slab dynamics and back-arc Aleutian ‘tsunami’ earthquake. Geophysical Journal International 165, 835–849.
deformation. Physics of the Earth and Planetary Interiors 149, 31–51. http://dx.doi.org/10.1111/j.1365-246X.2006.02899.x.
Heuret, A., Lallemand, S., Funiciello, F., Piromallo, C., Faccenna, C., 2011. Physical Luttrell, K.M., Tong, X., Sandwell, D.T., Brooks, B.A., Bevis, M.G., 2011. Estimates of
characteristics of subduction interface type seismogenic zones revisited. stress drop and crustal tectonic stress from the 27 February 2010 Maule, Chile,
Geochemistry Geophysics Geosystems 12, Q01004. http://dx.doi.org/10.1029/ earthquake: implications for fault strength. Journal of Geophysical Research
2010GC003230. 116, B11401. http://dx.doi.org/10.1029/2011JB008509.
Ide, S., Baltay, A., Beroza, G.C., 2011. Shallow dynamic overshoot and energetic deep Magee, M.E., Zoback, M.D., 1993. Evidence for a weak interplate thrust fault along
rupture in the 2011 MW 9.0 Tohoku-Oki Earthquake. Science 332, 1426–1429. the northern Japan subduction zone and implications for the mechanics of
http://dx.doi.org/10.1126/science.1207020. thrust faulting and fluid expulsion. Geology 21, 809–812.
Jarrard, R.D., 1986. Relations among subduction parameters. Reviews of Geophysics Matsubara, M., Yagi, Y., Obara, K., 2005. Plate boundary slip associated with the
24, 217–284. 2003 Off-Tokachi earthquake based on small repeating earthquake data.
Geophysical Research Letters 32, L08316. http://dx.doi.org/10.1029/ Ruff, L.J., 1999. Dynamic stress drop of recent earthquakes: variations within
2004GL022310. subduction zones. Pure and Applied Geophysics 154, 409–431.
McCaffrey, R., 1993. On the role of the upper plate in great subduction zone Ruff, L., Kanamori, H., 1980. Seismicity and the subduction process. Physics of the
earthquakes. Journal of Geophysical Research 98, 11953–11966. Earth and Planetary Interiors 23, 240–252.
McCaffrey, R., 1997. Influence of recurrence times and fault zone temperatures on Sandwell, D.T., Smith, W.H.F., 2009. Global marine gravity from retracked Geosat
the age-rate dependence of subuction zone seismicity. Journal of Geophysical and ERS-1 altimetry: ridge segmentation versus spreading rate. Journal of
Research 102, 22839–22854. Geophysical Research 114, B01411. http://dx.doi.org/10.1029/2008JB006008.
McCaffrey, R., 2008. Global frequency of magnitude 9 earthquakes. Geology 36, Satake, K., Atwater, B.F., 2007. Long-term perspectives on giant earthquakes and
263–266. http://dx.doi.org/10.1130/G24402A. tsunamis at subduction zones. Annual Review Earth and Planetary Sciences 35,
McCaffrey, R., 2009. The tectonic framework of the Sumatran subduction zone. 349–374. http://dx.doi.org/10.1146/annurev.earth.35.031306.140302.
Annual Review Earth and Planetary Sciences 37, 345–366. http://dx.doi.org/ Satake, K., Wang, K., Atwater, B.F., 2003. Fault slip and seismic moment of the 1700
10.1146/annurev.earth.031208.100212. Cascadia earthquake inferred from Japanese tsunami descriptions. Journal of
McKenzie, D., Jackson, J., Priestley, K., 2005. Thermal structure of oceanic and Geophysical Research 108, 2535. http://dx.doi.org/10.1029/2003JB002521.
continental lithosphere. Earth and Planetary Science Letters 233, 337–349. Schellart, W.P., 2008. Overriding plate shortening and extension above subduction
Minoura, K., Imamura, F., Sugawa, D., Kono, Y., Iwashita, T., 2001. The 869 Jogan zones: a parametric study to explain formation of the Andes mountains.
tsunami deposit and recurrence interval of large-scale tsunami on the Pacific Geological Society of America Bulletin 120, 1441–1454. http://dx.doi.org/
coast of Northeast Japan. Journal of Natural Disaster Science 23, 83–88. 10.1130/B26360.1.
Molnar, P., Atwater, T., 1978. Interarc spreading and Cordilleran tectonics as Schellart, W.P., 2011. A subduction zone reference frame based on slab geometry
alternates related to the age of subducted oceanic lithosphere. Earth and and subduction partitioning of plate motion and trench migration. Geophysical
Planetary Science Letters 41, 330–340. Research Letters 38, L16317. http://dx.doi.org/10.1029/2011GL048197.
Moore, D.E., Lockner, D.A., 2007. Comparative deformation behavior of minerals in Schellart, W.P., Moresi, L., 2013. A new driving mechanism for backarc extension
serpentinized ultramafic rock: application to the slab-mantle interface in and backarc shortening through slab sinking induced toroidal and poloidal
subduction zones. International Geology Reviews 49, 401–415. http:// mantle flow: results from dynamic subduction models with an overriding plate.
dx.doi.org/10.2747/0020-6814.49.5.401. Journal of Geophysical Research: Solid Earth 118, 3221–3248. http://dx.doi.org/
Morell, K.D., Fisher, D.M., Gardner, T.W., La Femina, P., Davidson, D., Teletzke, A., 10.1002/jgrb.50173.
2011. Quaternary outer fore-arc deformation and uplift inboard of the Panama Schellart, W.P., Spakman, W., 2012. Mantle constraints on the plate tectonic
Triple Junction, Burica Peninsula. Geochemistry Geophysics Geosystems 116, evolution of the Tonga-Kermadec-Hikurangi subduction zone and the South Fiji
B05402. http://dx.doi.org/10.1029/2010JB007979. Basin. Australian Journal of Earth Sciences 59, 933–952. http://dx.doi.org/
Morell, K.D., Kirby, E., Fisher, D.M., van Soest, M., 2012. Geomorphic and 10.1080/08120099.2012.679692.
exhumational response of the Central American Volcanic Arc to Cocos Ridge Schellart, W.P., Freeman, J., Stegman, D.R., Moresi, L., May, D., 2007. Evolution and
subduction. Journal of Geophysical Research 117, B04409. http://dx.doi.org/ diversity of subduction zones controlled by slab width. Nature 446, 308–311.
10.1029/2011JB008969. http://dx.doi.org/10.1038/nature05615.
Moreno, M.S., Bolte, J., Klotz, J., Melnick, D., 2009. Impact of megathrust geometry Schellart, W.P., Stegman, D.R., Freeman, J., 2008. Global trench migration velocities
on inversion of coseismic slip from geodetic data: application to the 1960 Chile and slab migration induced upper mantle volume fluxes: constraints to find an
earthquake. Geophysical Research Letters 36, L16310. http://dx.doi.org/ Earth reference frame based on minimizing viscous dissipation. Earth-Science
10.1029/2009GL039276. Reviews 88, 118–144. http://dx.doi.org/10.1016/j.earscirev.2008.01.005.
Moresi, L., Solomatov, V., 1998. Mantle convection with a brittle lithosphere; Schellart, W.P., Stegman, D.R., Farrington, R.J., Freeman, J., Moresi, L., 2010. Cenozoic
thoughts on the global tectonic styles of the Earth and Venus. Geophysical tectonics of western North America controlled by evolving width of Farallon
Journal International 133, 669–682. slab. Science 329, 316–319. http://dx.doi.org/10.1126/science.1190366.
Mosher, D.C., Austin Jr., J.A., Fisher, D., Gulick, S.P.S., 2008. Deformation of the Schellart, W.P., Stegman, D.R., Farrington, R.J., Moresi, L., 2011. Influence of lateral
northern Sumatra accretionary prism from high-resolution seismic reflection slab edge distance on plate velocity, trench velocity and subduction
profiles and ROV observations. Marine Geology 252, 89–99. http://dx.doi.org/ partitioning. Journal of Geophysical Research 116, B10408. http://dx.doi.org/
10.1016/j.margeo.2008.03.014. 10.1029/2011JB008535.
Müller, R.D., Landgrebe, T.C.W., 2012. The link between great earthquakes and the Scholz, C.H., Campos, J., 1995. On the mechanism of seismic decoupling and back arc
subduction of oceanic fracture zones. Solid Earth 3, 447–465. http://dx.doi.org/ spreading at subduction zones. Journal of Geophysical Research 100, 22103–
10.5194/se-3-447-2012. 22115.
Ni, S., Kanamori, H., Helmberger, D., 2005. Energy radiation from the Sumatra Scholz, C.H., Campos, J., 2012. The seismic coupling of subduction zones revisited.
earthquake. Nature 434, 582. Journal of Geophysical Research 117, B05310. http://dx.doi.org/10.1029/
Okal, E.A., 1992. Use of the mantle magnitude Mm for the reassessment of the 2011JB009003.
moment of historical earthquakes – I: shallow events. Pure and Applied Shaw, B.E., 2009. Constant stress drop from small to great earthquakes in
Geophysics 139, 17–57. magnitude-area scaling. Bulletin of the Seismological Society of America 99,
Okal, E.A., Borrero, J.C., Synolakis, C.E., 2006. Evaluation of tsunami risk from 871–875. http://dx.doi.org/10.1785/0120080006.
regional earthquakes at Pisco, Peru. Bulletin of the Seismological Society of Shearer, P., Bürgmann, R., 2010. Lessons learned from the 2004 Sumatra–Andaman
America 96, 1634–1648. http://dx.doi.org/10.1785/0120050158. megathrust rupture. Annual Review Earth and Planetary Sciences 38, 103–131.
Oliver, J., Murphy, L., 1971. WWNSS: seismology’s global network of observing http://dx.doi.org/10.1146/annurev-earth-040809-152537.
stations. Science 174, 254–261. Sieh, K., Natawidjaja, D., 2000. Neotectonics of the Sumatran fault, Indonesia.
O’Neill, C., Müller, D., Steinberger, B., 2005. On the uncertainties in hot spot Journal of Geophysical Research 105, 28295–28326.
reconstructions and the significance of moving hot spot reference frames. Silver, E.A., Reed, D.L., Tagudin, J.E., Heil, D.J., 1990. Implications of the north and
Geochemistry Geophysics Geosystems 6, Q04003. http://dx.doi.org/10.1029/ south Panama thrust belts for the origin of the Panama orocline. Tectonics 9,
2004GC000784. 261–281.
Ozawa, S., Nishimura, T., Suito, H., Kobayashi, T., Tobita, M., Imakiire, T., 2011. Simons, M., Minson, S.E., Sladen, A., Ortega, F., Jiang, J., Owen, S.E., Meng, L.,
Coseismic and postseismic slip of the 2011 magnitude-9 Tohoku-Oki Ampuero, J.-P., Wei, S., Chu, R., Helmberger, D.V., Kanamori, H., Hetland, E.,
earthquake. Nature 475, 373–376. http://dx.doi.org/10.1038/nature10227. Moore, A.W., Webb, F.H., 2011. The 2011 Magnitude 9.0 Tohoku-Oki
Pacheco, J.F., Sykes, L.R., 1992. Seismic moment catalogue of large shallow earthquakes, Earthquake: mosaicking the Megathrust from Seconds to Centuries. Science
1900 to 1989. Bulletin of the Seismological Society of America 82, 1306–1349. 332, 1421–1425.
Pacheco, J.F., Sykes, L.R., Scholz, C.H., 1993. Nature of seismic coupling along simple Singh, S.C., Hananto, N.D., Chauhan, A.P.S., Permana, H., Denolle, M., Hendriyana, A.,
plate boundaries of the subduction type. Journal of Geophysical Research 98, Natawidjaja, D., 2010. Evidence of active backthrusting at the NE Margin of
14133–14159. Mentawai Islands, SW Sumatra. Geophysical Journal International 180, 703–
Peterson, E.T., Seno, T., 1984. Factors affecting seismic moment release rates in 714. http://dx.doi.org/10.1111/j.1365-246X.2009.04458.x.
subduction zones. Journal of Geophysical Research 89, 10233–10248. Sitchler, J.C., Fisher, D.M., Gardner, T.W., Protti, M., 2007. Constraints on inner
Phipps Morgan, J., Ranero, C.R., Vannucchi, P., 2008. Intra-arc extension in Central forearc deformation from balanced cross sections, Fila Costeña thrust belt, Costa
America: links between plate motions, tectonics, volcanism, and geochemistry. Rica. Tectonics 26, TC6012. http://dx.doi.org/10.1029/2006TC001949.
Earth and Planetary Science Letters 272, 365–371. http://dx.doi.org/10.1016/ Springer, M., 1999. Interpretation of heat-flow density in the Central Andes.
j.epsl.2008.05.004. Tectonophysics 306, 377–395.
Plafker, G., 1965. Tectonic deformation associated with the 1964 Alaska earthquake. Stauder, W., 1968. Mechanism of the Rat Island earthquake sequence of February 4,
Science 148, 1675–1687. 1965, with relation to island arcs and sea-floor spreading. Journal of
Plafker, G., Savage, J.C., 1970. Mechanism of the Chilean earthquakes of May 21 and Geophysical Research 73, 3847–3858.
22, 1960. Geological Society of America Bulletin 81, 1001–1030. Stein, S., Okal, E.A., 2005. Speed and size of the Sumatra earthquake. Nature 434,
Power, W., Wallace, L., Wang, X., Reyners, M., 2012. Tsunami hazard posed to New 581–582.
Zealand by the Kermadec and southern New Hebrides subduction margins: an Stein, S., Okal, E.A., 2007. Ultralong period seismic study of the December 2004
assessment based on plate boundary kinematics, interseismic coupling, and Indian Ocean earthquake and implications for regional tectonics and the
historical seismicity. Pure and Applied Geophysics 169, 1–36. http://dx.doi.org/ subduction process. Bulletin of the Seismological Society of America 97, S279–
10.1007/s00024-011-0299-x. S295. http://dx.doi.org/10.1785/0120050617.
Ruff, L.J., 1989. Do trench sediments affect great earthquake occurrence in Stein, C.A., Stein, S., 1992. A model for the global variation in oceanic depth and heat
subduction zones? Pure and Applied Geophysics 129, 263–282. flow with lithospheric age. Nature 359, 123–129.
Suppe, J., 2007. Absolute fault and crustal strength from wedge tapers. Geology 35, Wallace, L.M., Reyners, M., Cochran, U., Bannister, S., Barnes, P.M., Berryman, K.,
1127–1130. http://dx.doi.org/10.1130/G24053A.1. Downes, G., Eberhart-Phillips, D., Fagereng, A., Ellis, S., Nicol, A., McCaffrey, R.,
Syracuse, E.M., van Keken, P.E., Abers, G.A., 2010. The global range of subduction Beavan, R.J., Henrys, S., Sutherland, R., Barker, D.H.N., Litchfield, N., Townend, J.,
zone thermal models. Physics of the Earth and Planetary Interiors 183, 73–90. Robinson, R., Bell, R., Wilson, K., Power, W., 2009. Characterizing the
http://dx.doi.org/10.1016/j.pepi.2010.02.004. seismogenic zone of a major plate boundary subduction thrust: Hikurangi
Tichelaar, B.W., Ruff, L.J., 1993. Depth of seismic coupling along subduction zones. Margin, New Zealand. Geochemistry Geophysics Geosystems 10, Q10006.
Journal of Geophysical Research 98, 2017–2037. http://dx.doi.org/10.1029/2009GC002610.
Uyeda, S., Kanamori, H., 1979. Back-arc opening and the mode of subduction. Wang, K., Mulder, T., Rogers, G.C., Hyndman, R.D., 1995. Case for very low coupling
Journal of Geophysical Research 84, 1049–1061. stress on the Cascadia subduction fault. Journal of Geophysical Research 100,
Vigny, C., Socquet, A., Peyrat, S., Ruegg, J.-C., Métois, M., Madariaga, R., Morvan, S., 12907–12918.
Lancieri, M., Lacassin, R., Campos, J., Carrizo, D., Bejar-Pizarro, M., Barrientos, S., Wright, I.C., 1993. Pre-spread rifting and heterogeneous volcanism in the southern
Armijo, R., Aranda, C., Valderas-Bermejo, M.-C., Ortega, I., Bondoux, F., Baize, S., Havre Trough back-arc basin. Marine Geology 113, 179–200.
Lyon-Caen, H., Pavez, A., Vilotte, J.P., Bevis, M., Brooks, B., Smalley, R., Parra, H., Wu, F.T., Kanamori, H., 1973. Source mechanism of February 4, 1965, Rat Island
Baez, J.-C., Blanco, M., Cimbaro, S., Kendrick8§, E., 2011. The 2010 MW 8.8 Maule earthquake. Journal of Geophysical Research 78, 6082–6092.
Megathrust Earthquake of Central Chile, Monitored by GPS. Science 332, 1417– Yagi, Y., 2004. Source rupture process of the 2003 Tokachi-oki earthquake
1421. http://dx.doi.org/10.1126/science.1204132. determined by joint inversion of teleseismic body wave and strong ground
von Huene, R., Ranero, C.R., 2003. Subduction erosion and basal friction along the motion data. Earth Planets Space 56, 311–316.
sediment-starved convergent margin off Antofagasta, Chile. Journal of Zachariasen, J., Sieh, K., Taylor, F.W., Edwards, R.L., Hantoro, W.S., 1999.
Geophysical Research 108, 2079. http://dx.doi.org/10.1029/2001JB001569. Submergence and uplift associated with the giant 1833 Sumatran subduction
Wallace, L.M., Beavan, J., McCaffrey, R., Darby, D., 2004. Subduction zone coupling earthquake: evidence from coral microatolls. Journal of Geophysical Research
and tectonic block rotations in the North Island, New Zealand. Journal of 104, 895–919.
Geophysical Research 109, B12406. http://dx.doi.org/10.1029/2004JB003241.

Articulo de Revision

Uploaded by

Copyright:

Available Formats

Articulo de Revision

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Articulo de Revision

Uploaded by

Copyright:

Available Formats

Physics of the Earth and Planetary Interiors 225 (2013) 41–67

Contents lists available at ScienceDirect

Physics of the Earth and Planetary Interiors

Global correlations between maximum magnitudes of subduction zone

include Sunda, North Sulawesi, Hikurangi, Nankai-northern Ryukyu, Kamchatka-Kuril-Japan, Aleutians-

1. Introduction 241 trench segments, each with a 200 km trench-parallel extent.

Parameter Units Explanation

Parameter n Total range R Range for MW > 8.5 % of total range P

5.1. Equation for moment magnitude MW

southward (vT\ = 0.2–1.6 cm/yr) with slow overriding plate short-

6. Our research has shown that epicenters of giant earthquakes References

You might also like