Subduction zone megathrust earthquakes

The contact surfaces between underthrusting and overriding plates in subduction zones are called megathrust faults (Fig. 1). Over time, aside from secondary inelastic deformation or block motions of the surrounding plates, the total displacement across the entire megathrust surface must equal the relative plate convergence. This slip budget extends over the depth range from the subduction zone trench to the downdip limit of the sliding plate boundary, defined by the onset of ductile flow in the mantle wedge. The megathrust itself may involve a shear zone rather than a single surface, and repeated ruptures of the plate-boundary contact may occur on parallel faults within the shear zone. The overall plate convergence is accommodated by spatially varying stick-slip earthquake failure, episodic slow slip, and continuous aseismic sliding. Heterogeneous frictional properties of the megathrust and dynamic interactions control how any given region behaves, including the possibility that some regions of stable sliding might transition to stick-slip failure during high-strain-rate loading. Large-scale deformation of the underthrusting plate occurs in subduction zones, with the plate bending and straightening as it subducts beneath either oceanic or continental lithosphere, such that the forces acting on each megathrust and the resulting seismicity are affected by the broad tectonic configuration of each plate boundary. All of these deformation processes can produce hazards, from seismic shaking to submarine slumping and tsunami generation. Here, we focus on megathrust faulting, which produces the greatest hazards.

Based on the elastic rebound theory of earthquake occurrence and the relatively steady loading provided by long-term plate convergence in subduction zones, one might expect the stick-slip failure of megathrust faults to be at least somewhat regular. This does appear to hold for many regions, where irregular “cycles” of long intervals of interseismic strain accumulation, sudden earthquake sliding, and postseismic adjustment are apparent for well-documented sequences of great earthquakes (M > 8) in regions such as the Nankai Trough of southwest Japan, along southern Chile, and along the Kuril Islands (e.g., Ando, 1975; McCann et al., 1979; Lay et al., 1982). In many places, the historical record of great earthquakes is limited, with too few events to establish any regularity, or such limited information about early events that it is hard to establish their relationship to more recent well-documented ruptures. Nonetheless, the time interval since the last great earthquake in a given region and that event’s inferred size and spatial extent have quite successfully guided designation of “seismic gaps” as regions with potential for future great earthquakes (e.g., McCann et al., 1979; Nishenko, 1991), albeit without time predictability (e.g., Lay, 2015). Smaller events appear to be irregular in occurrence and to interact strongly with nearby ruptures, causing the seismic gap approach to be less useful than for great events. This perspective is not lacking in contention, as some researchers deny any degree of cyclicity in earthquake occurrence (e.g., Kagan and Jackson, 1995; Kagan et al., 2012), but for very large earthquakes, these objections do not appear to be valid. Regularity in great earthquake cycles has long been recognized to be complicated by intermittent triggering of adjacent megathrust segments that otherwise tend to fail separately in smaller events, notably in southwest Japan (Ando, 1975) and Ecuador-Colombia (Kanamori and McNally, 1982), with synchronization of rupture possibly occurring in supercycles of huge earthquakes on time scales that greatly exceed recorded history (e.g., Sieh et al., 2008). Such unprecedented events have low, but nonzero probability for many regions, and it is difficult to reliably evaluate the worst-case potential and its associated probability of occurrence, given the limited temporal span of the seismological and paleoseismological records.

Surprises that stem from the limited historic record for megathrust events have included great earthquakes striking in regions lacking any known prior great event and having disrupted subduction zone structure. This includes examples such as subduction of a young ridge at a triple junction where a great earthquake struck in the Solomon Islands in 2007 (e.g., Taylor et al., 2008; Furlong et al., 2009), and the great 2006 Kuril Islands rupture (e.g., Ammon et al., 2008), which struck in a region with discontinuous arc structure and an unusual forearc basin. Both areas had been argued to potentially be permanently aseismic, an assertion also still invoked for regions of subducting ridges such as the Tehuantepec and Carnegie Ridges along South America, where there is no history of great earthquakes. Confidence in such assertions must be assessed with caution, and the use of geodetic measurements to assess upper-plate strain is the most promising approach to evaluating actual seismic potential in such regions.

Great megathrust earthquakes have also now been observed to have triggering interactions with great intraplate faulting (e.g., Ammon et al., 2008; Lay et al., 2010b, 2017), and cascading failures on the relevant megathrust (e.g., Lay et al., 2010a), presenting further challenges to hazard assessment and rapid warning procedures. Very large earthquakes have also been documented to rupture the shallowest part of the megathrust (e.g., Polet and Kanamori, 2000; Lay and Bilek, 2007), where it had been thought that aseismic deformation of wet sediments on the plate boundary would preclude stick-slip sliding.

Deployment of high-quality geodetic networks in subduction zones has added important constraints on coseismic slip during nearby megathrust earthquakes, along with capturing deformation prior to and after large ruptures of the plate boundary. This has revealed the existence of slow-slip events that may be accompanied by seismic tremor or small repeating earthquakes, but the full deformation occurring over days to months can only be resolved geodetically. Deformation in the upper plate of a frictionally locked megathrust, with landward compression, can be resolved by global positioning system (GPS) and Global Navigation Satellite System (GNSS) sensors operating over years to decades, and these data sets now provide direct measurement of the interseismic process and the ability to establish whether a seismic gap is actually accumulating strain that will result in a future large earthquake or whether aseismic sliding is releasing the strain. Geodesy also enables characterization of upper-plate sliver or block motions that may result from oblique convergence. This is now recognized as quite common, and it must be accounted for when evaluating the megathrust strain energy budget (e.g., McCaffrey et al., 2000; Wallace et al., 2004; La Femina et al., 2009). Precursory sliding near a future hypocenter, postseismic slip around main-shock rupture zones, relocking of a ruptured zone, and viscoelastic deformation after an event are processes that have been sensed by onshore and, increasingly, offshore geodetic measurements, greatly expanding our view into the seismic strain accumulation and release process and the ways in which relative plate motion is accommodated across the entire megathrust (e.g., Ito et al., 2013; Yokota et al., 2016). In this overview, we highlight some of the basic seismological characteristics of major subduction zones and salient findings of recent investigations of large earthquakes and slow deformation on megathrusts. We conclude with consideration of the seismic hazards posed by megathrust earthquakes.

Earthquake occurrence reveals many attributes of subduction zone megathrusts, although it is good to keep in mind that cumulative seismic slip appears to account for less than half of total convergence across the global set of plate-boundary faults in the seismic record. Variations in earthquake size, number, and depth distribution indicate distinct properties of megathrust faults that guide comparisons of processes in different subduction zones.

Measured seismic magnitudes for shallow subduction zone earthquakes span from near zero at the low end, depending on availability of regional seismic monitoring, to the largest observed event, the 1960 Chile moment magnitude (M_w) = 9.5 earthquake. Globally, earthquake catalogs have improved with time, with fairly complete catalogs back to 1900 for events larger than M ∼ 7, and near-complete recording since the mid-1970s of events larger than M ∼ 5.0. To provide a sampling of diverse megathrust activity, we considered the seismicity characteristics for nine major subduction zones (Fig. 2; Table 1) and identified the largest observed earthquakes in each zone using two readily available earthquake catalogs, the U.S. Geological Survey National Earthquake Information Center catalog (USGS-NEIC, https://earthquake.usgs.gov), from 1900 to 2016, and the Global Centroid Moment Tensor catalog (GCMT, http://www.globalcmt.org), from 1976 to 2016. The USGS-NEIC catalog includes the largest-magnitude earthquakes over the last century, drawing upon redetermined magnitude and location estimates from the International Seismological Centre–Global Earthquake Model (ISC-GEM; http://www.isc.ac.uk/iscgem/). The GCMT catalog provides relatively uniformly determined estimates of each event’s moment tensor and seismic moment. We utilized the focal mechanisms to identify events that were most likely located on the subduction megathrust (Figs. 3–11).

We first considered instrumentally recorded, great (M ≥ 8.0), shallow plate-boundary thrust-faulting events in nine major subduction zones (Fig. 2; Table 1). For earthquake seismologists, most of these earthquakes are immediately identifiable by their years and place names, as they represent major tectonic events that have sometimes produced devastating shaking and tsunami damage. Along the Alaska-Aleutian subduction zone, six great earthquakes spanned the arc in 1964 (Alaska, M_w 9.2), 1965 (Rat Islands, M_w 8.7), 1957 (Fox Islands, M_w 8.6), 1946 (M_w 8.6), 1938 (Alaskan Peninsula, M 8.3), and 1986 (Fox Islands, M_w 8.0). The 1986 event reruptured a western portion of the much larger 1957 earthquake. The 1946 event ruptured between the 1957 and 1938 events and produced a disproportionally large tsunami, leading to it being identified as a “tsunami earthquake” by Kanamori (1972). These events have essentially filled in most of the Alaska-Aleutian plate boundary during this century. An historical great event in 1788 ruptured the 1938 zone and Kodiak Island portions of the 1964 ruptures (e.g., Briggs et al., 2014), indicating that boundaries between recent ruptures have not been persistent.

Along Mexico and Middle America, the largest seismically recorded earthquakes have been much smaller, and they occurred along the coast of Mexico—an M 8.1 earthquake in 1932 (Jalisco) and two M_w 8 earthquakes in 1985 (Michoacan) and 1995 (Colima). However, the largest documented historical event occurred along the coast from Guerrero to Tehuantepec in 1787 with M ∼ 8.6 (Suárez and Albini, 2009), apparently rupturing across multiple zones of more recent smaller events, so the limited time span of the seismological catalog must be kept in mind. The largest recorded earthquake in Peru was the 2001 (Arequipa, M_w 8.4) event, which ruptured within approximately two-thirds of the zone of the 1868 (Arica, M ∼8.5–9.0) earthquake. Other great Peruvian earthquakes occurred in 1940 (M_w 8.2), 1966 (M_w 8.1), 1942 (M_w 8.1), and 2007 (Pisco, M_w 8.0). The Chile subduction zone has experienced 10 great events: 1960 (Valdivia, M_w 9.5), 2010 (Maule, M_w 8.8), 1922 (Atacama, M 8.5), 2015 (Illapel, M_w 8.3), 1906 (Valparaíso, M 8.2), 2014 (Iquique, M_w 8.2), 1960 (Bio-Bio, M_w 8.1), 1943 (Coquimbo, M 8.1), 1985 (Valparaíso, M_w 8.0), and 1995 (Antofagasta, M_w 8.0). Of these, only the 1943 and 2015 events rupture zones overlapped significantly. The region south of the 2014 Iquique event last ruptured in 1877 (Iquique, M ∼8.8), and this remains a major seismic gap. The 1922 rupture zone along Atacama likely has substantial strain accumulation relative to surrounding regions, as geodetic measurements indicate that the megathrust is pervasively locked in the region (Métois et al., 2014).

Along the Tonga-Kermadec arc, the largest known interplate thrust event occurred in 1976 (Kermadec, M_w 8.0). The 2009 Samoa earthquake (M_w 8.1) began with interplate normal faulting but triggered M_w 8.0 thrusting on the northern Tonga megathrust (Beavan et al., 2010; Lay et al., 2010b; Fan et al., 2016). Other great shallow events in the region in 1917 (Kermadec, M 8.2), 1919 (Tonga, M 8.1), and 1917 (Samoa, M 8.0) all appear to have been intraplate events (Meng et al., 2015b). The subduction zones along Vanuatu, Solomon Islands, Sulawesi, Taiwan, and the Philippines all have maximum recorded shallow thrusting earthquake magnitudes of about M ∼8.1. Maximum magnitudes for recorded events along the Marianas subduction zone have been less than M 8. While it is conceivable that very-low-probability events might rupture the entire megathrust contact in each of these island-arc regions, it may be that the apparently low coupling and high fraction of aseismic convergence preclude this from happening, and seismic hazard assessments can focus on ruptures of the size historically observed. In contrast, the Sumatra subduction zone has experienced several great thrust earthquakes since 2004, in 2004 (Sumatra-Andaman, M_w 9.15), 2005 (Nias, M_w 8.6), and 2007 (Bengkulu, M_w 8.4). Multiple historical great thrust earthquakes of up to M ∼9 struck the region prior to 1900 as well. Great intraplate strike-slip ruptures with M_w 8.6 and 8.2 struck in 2012 in the Indo-Australian deformation zone in the oceanic plate south of Sumatra.

The Honshu, Japan, subduction zone has hosted three great thrust earthquakes since 1900, with the largest being the 2011 (Tohoku, M_w 9.1) event in 2011, followed by the 1968 (Tokachi-oki, M_w 8.2) and 1960 (Sanriku, M_w 8.0; possibly an overestimated magnitude) earthquakes. The 2011 event appears to have reruptured a significant portion of the 1896 Meiji Sanriku tsunami earthquake zone on the shallow portion of the megathrust near the trench from 38°N to 40.3°N (e.g., Yamazaki et al., 2018; Lay, 2017). Other significant northern Japan earthquakes occurred in 1933 (Sanriku outer-rise normal faulting, M 8.4) and 1923 (Kanto oblique thrust along the Sagami Trough, M 8.1). Nine great thrust earthquakes have occurred along the Hokkaido-Kuril-Kamchatka subduction zone, including an M_w 9.0 event in 1952 (Kamchatka), and the 1963 (Kuril Islands, M_w 8.5), 1923 (Kamchatka, M 8.4), 2006 (Central Kuril Islands, M_w 8.3), 2003 (Hokkaido, M_w 8.3), 1994 (Kuril Islands M_w 8.3), 1958 (Kuril Islands, M_w 8.3), 1952 (Hokkaido, M_w 8.1), and 1918 (Central Kuril Islands, M 8.1) events. The 2006 Kuril Islands event appears to have triggered the 2007 (M_w 8.1) outer-rise normal fault rupture (Ammon et al., 2008; Lay et al., 2009), given that it immediately activated outer-rise faulting aftershocks. The 2003 Hokkaido rupture zone closely overlaps the 1952 zone.

The variation in largest recorded earthquake size between regions is very substantial, and many efforts have been made to establish the factors influencing the distribution of great earthquakes, as discussed later. Within the seismological record, only a few recent great events have reruptured a region with seismological recordings of prior great earthquake activity (1986 Aleutians, 2015 Illapel, 2003 Hokkaido). Because the prior events (1957, 1943, 1952, respectively) all have limited seismological data and relatively poor constraints on their rupture characteristics, it is very difficult to directly evaluate the stability of repeated great ruptures of a given portion of the relevant megathrust.

It is important to recognize that only ∼118 yr of reliable seismological records exist, and many earlier great events have been documented around the Pacific subduction zones by historical accounts and paleoseismic analyses (e.g., Beck et al., 1998; Kelleher, 1972; Lay et al., 1982; McCann et al., 1979; Scholz and Campos, 1995, 2012). In some cases, such as for the 1837 and 1737 events in the 1960 Chile rupture zone, the size of the historic events is being reevaluated based on detailed examination of the record, modifying inferences of segmentation and regularity (e.g., Cisternas et al., 2017). The Cascadia subduction zone is known to have had many great earthquakes, but the most recent was in 1700, and so it is not represented by the seismological catalogs. Similarly, the Himalayan front hosts great thrust events, but none has occurred during the seismological record. Southwest Japan did have well-characterized great earthquakes in 1944 and 1946, and documentation and field observations exist for numerous events in the same region dating back over 1000 yr (e.g., Ando, 1975).

During the relatively short 40 yr time interval spanned by the GCMT catalog, the largest-magnitude megathrust earthquakes occurred in Sumatra (2004 M_w 9.2), northern Japan (2011 M_w 9.1), and central Chile (2010 M_w 8.8). In Figures 3–11, we show GCMT faulting mechanisms for likely megathrust events (those having shallow-dipping thrust fault solutions) for our subset of nine subduction zones, along with regional M_w-frequency distributions for shallow GCMT events. The maps display the variation in megathrust geometry and width among the major subduction zones. Table 1 lists details of the regions and magnitude distributions from this catalog.

In all regions except for the Marianas, we found b values < 1 for both the total and megathrust populations (Table 2). The numbers of events in the populations and the magnitude ranges spanned by the data sets (see Figs. 3–11) are limited for the GCMT catalog, and most magnitude distributions only have well-sampled distributions up to M_w ∼ 7.5 in the figures (lower in the Marianas). These data limitations give rise to significant uncertainty in the b values of ∼0.05, so regional differences are suggested, but not well resolved. Apart from the Marianas and Tonga, the b values for the megathrust events (0.7–0.8) were found to be systematically lower than for the full populations (0.76–0.88) by an average of 0.07. The Tonga subduction zone has b values of 0.92 and 0.93 for the full data set and the megathrust, respectively, while the sparse Marianas data set has corresponding b values of 1.03 and 1.07, respectively. Reduction of the b value for the quasi-two-dimensional, single-fault-plane thrust faulting data set relative to the three-dimensional total distribution is plausibly the result of having more faulting geometries in the greater volumes sampled by the full data set, although the differences are marginally resolved. The high thrust-fault b values for the Tonga (Fig. 11) and Marianas (Fig. 8) zones, which have steeply dipping and narrow megathrust interfaces, are likely due to inclusion of many thrust faulting intraplate events by our criteria, since intraslab compression events are known to occur in these regions (e.g., Meng et al., 2015a).

Previous studies of b values in subduction zones have also found b < 1 in shallow megathrust zones. Along the northern Japan subduction zone, Tormann et al. (2015) estimate b values of 0.5–0.9 at depths between the surface and ∼100 km, although they did not specifically separate out events occurring on the megathrust zone versus upper-plate or intraslab events. Nishikawa and Ide (2014) computed b values for all of the subduction zones considered here, although with more spatial subdivisions within each subduction zone and no separation by mechanism. Given the data and methodological differences in b value calculation, comparison of absolute values is difficult. However, their lowest b values (∼0.8 for parts of Central America and Peru) correspond to relatively low b values in our analysis, and their highest b values, exceeding 1, were also found for the Marianas and Tonga-Kermadec regions.

We also used the GCMT M_w-frequency relations for M_c = 5.2 to evaluate overall seismic productivity variations during the 1976–2016 time period using both the full set of events, as well as only thrust faulting events. The Japan and Tonga-Kermadec subduction zones are most productive overall, with ∼1.5 earthquakes of any mechanism per kilometer length along strike of the subduction zone. Sumatra is next, with ∼0.7 events/km, followed by Kuril-Kamchatka (0.6 events/km), Chile (0.5 events/km), Mexico–Middle America (0.4 events/km), Marianas and Alaska-Aleutians (0.3 events/km), and Peru (0.2 events/km).

Japan and Tonga-Kermadec also produce the most megathrust events, at ∼0.9 and ∼0.7 events/km, respectively. Kuril-Kamchatka is next at 0.4 events/km, and several others group together at ∼0.2–0.3 events/km (Chile, Sumatra, Mexico–Middle America, and Alaska-Aleutians). Peru and Marianas produced the fewest megathrust events in this time period, at ∼0.09 and 0.08 events/km, respectively.

The regional differences in seismic productivity are also manifested in large event aftershock productivity, with lower values found in Mexico–Middle America, Peru, and Chile and higher values found in Tonga, Sumatra, Japan, and Southwest Pacific island arcs (e.g., Singh and Suárez, 1988; Wetzler et al., 2016). In general, island-arc megathrusts that host great earthquakes tend to be more productive than continental arcs.

We compiled cumulative seismic moment from the GCMT database for all thrust-mechanism earthquakes and for all likely megathrust events with strike aligned with the trench. Seismic moment is predominantly associated with the likely megathrust events. For Chile, Japan, and Sumatra, the three subduction zones with M_w ≥ 8.8 events during the 1976 to 2016 period, cumulative moment is largely concentrated in the 15–25 km depth range (Fig. 12). The histograms are dominated by the largest earthquakes, namely, the 2004 M_w 9.2 Sumatra-Andaman, 2010 M_w 8.8 Maule, Chile, and 2011 M_w 9.1 Tohoku, Japan, earthquakes, and the moment for these events is actually distributed over a depth range rather than at the GCMT centroid depth. The other subduction zones have less cumulative seismic moment in the GCMT time period, and we found more diversity in the depth range of peak moment in each region (Fig. 13). For example, both the Alaska and Peru subduction zones had a peak moment in the 20–30 km range, but the Mexico–Middle America zone seismic moment peaked at shallower depth, between 10 and 20 km. Both the Kuril and Tonga subduction zones have recorded multiple peaks in moment depth distribution. Kuril moment peaks have been recorded at both 5–10 km and 20–30 km depths. There is another peak in the Kuril zone that is much deeper, between 60 and 65 km, although the corresponding thrust-mechanism events have strikes inconsistent with the local trench orientation, so they are likely intraslab events. The Tonga subduction zone includes a shallow peak at 10–15 km, and a deeper one at 40–45 km; again, the deeper one is likely to be from intraslab compressional events (e.g., Meng et al., 2015b). We believe that all of the significant interplate seismic moment release is at depths less than 50 km in our selected regions, and less than 60 km globally.

In recent decades, many observations have indicated depth dependence along megathrusts for a variety of seismic characteristics. The recognition of tsunami earthquakes (Kanamori, 1972) that produced much larger tsunamis than would be expected for a typical earthquake with the same surface wave magnitude (M_S) prompted focused research on depth-dependent characteristics. These rare shallow ruptures tend to have anomalous properties, with relatively low rupture velocity, low radiated energy, and large slip that generates a strong tsunami. Such events have struck along the Aleutians, Japan, the Kuril Islands, Peru, Nicaragua, El Salvador, New Zealand, Java, and Sumatra, so they are widespread, albeit infrequent events (e.g., Kanamori, 1972; Polet and Kanamori, 2000; Bilek and Lay, 2002).

In addition, the occurrence of several great (M > 8) earthquakes since 2004 has provided excellent recordings by large numbers of broadband seismographs, leading to observations of large slip but very little in the way of coherent bursts of high-frequency radiation in the shallowest part of the relevant subduction zone, but many bursts of high-frequency radiation originating in the deeper regions of the seismogenic zone, where there is lower coseismic slip (e.g., Lay et al., 2012). In the following sections, we outline several of the depth-dependent seismic characteristics that have been noted for megathrust earthquakes.

Tsunami earthquake ruptures have significantly longer rupture durations than is typical for similar-sized events. For example, the 1992 M_w 7.7 Nicaragua tsunami earthquake had source time function duration estimates of 100–150 s (e.g., Kanamori and Kikuchi, 1993; Ihmlé, 1996b), and the 2010 M_w 7.8 Mentawai event had duration estimates of 80–125 s (Yue et al., 2014a; Newman et al., 2011).

Motivated by these long source duration observations for the tsunami earthquakes, Bilek and Lay (2002) examined over 500 earthquakes (M_w 5.0–7.5) within 14 circum-Pacific subduction zones and found that some events, but not all, in the shallowest 15 km of the megathrust had long moment-scaled source durations (to an equivalent M_w = 6.0 reference), similar to scaled durations of seven tsunami earthquakes. Results for individual subduction zones indicate several events with long durations (10–20 s) located in the uppermost 15 km, with events deeper than 15 km having shorter (mean ∼5–6 s) durations. El Hariri et al. (2013) examined relocated subduction zone earthquakes with improved depth constraints and classified the events into shallow (<26 km) and deep (>26 km) groups based on overall longer moment-scaled durations of the shallow group. El Hariri et al. (2013) also focused on regional comparisons, finding weak to no correlation between longer duration and reported amounts of subducted sediment, presence of bathymetric features, or regions of observed afterslip, all factors that have been proposed to affect rupture complexity and duration.

More recent work suggests some depth variation in duration estimates using global data sets. Denolle and Shearer (2016) examined spectral ratios for over 900 thrust earthquakes (M > 5.5) and found weak dependence of scaled source duration with increasing depth on the fault. Similarly, Ye et al. (2016c) found longer-duration earthquakes in the shallowest (<18 km) portion of the subduction megathrust using finite fault determinations for over 100 M_w > 7 earthquakes. Tsunami earthquakes, as well as a few other shallow events, were found to have scaled durations (to an equivalent M_w = 6.0 reference) of between ∼8 and 19 s, i.e., significantly larger than the average of the rest of the events in their study (∼5.6 ± 1.5 s).

Recent observations of depth dependence in high-frequency radiation (e.g., Lay et al., 2012) suggest that other seismic characteristics, such as the short-period body wave magnitude (m_b), may also vary with depth relative to the long-period M_w. Tsunami earthquakes, given their shallow slip with limited high-frequency radiation, can be identified by their relatively low m_b relative to M_w measurement (e.g., Kanamori, 1972; Kanamori and Kikuchi, 1993; Lay et al., 2011a). Rushing and Lay (2012) examined the differences between m_b and M_w for a large catalog of likely megathrust events with M_w ≥ 5 in several subduction zones (Chile, Japan, Sumatra, Kuril, Aleutians, Sumba, Peru). They found positive m_b-M_w perturbations for deeper megathrust events, reflecting enhanced deep short-period radiation in Japan, northern Sumatra, and central Chile. Little to no apparent depth trend was observed in the scattered data for the other subduction zones in their study.

Recent well-recorded large and great earthquakes have displayed extensive diversity in rupture characteristics, with differences in radiated energy depending on depth of the rupture patch, both bilateral and unilateral rupture expansion, and highly variable slip distribution complexity. Hundreds of events in subduction zones now have finite-fault solutions, derived from a mix of teleseismic, strong motion, geodetic, and tsunami observations. The USGS-NEIC routinely computes high-quality solutions for large shallow earthquakes (Hayes, 2017), and many contributions are collected in the SRCMOD database (http://equake-rc.info/SRCMOD/). Some solutions are well resolved, others have severe trade-offs between kinematic constraints and model parameterization. Point source time functions are routinely determined by the SCARDEC method (e.g., Vallée and Douet, 2016). Efforts are under way to synthesize the accumulated data sets (e.g., Kanamori, 2014; Lay, 2015; Ye et al., 2016b, 2016c; Denolle and Shearer, 2016; Meier et al., 2017; Melgar and Hayes, 2017; Hayes, 2017), and we will not attempt to summarize the multitude of studies. Large shallow coseismic slip occurred in the 2015 Illapel (e.g., Li et al., 2016; Melgar et al., 2016), 2011 Tohoku (e.g., Lay et al., 2011b; Iinuma et al., 2012; Ozawa et al., 2012; Satake et al., 2013; Romano et al., 2014; Bletery et al., 2014; Melgar and Bock, 2015; Lay, 2017), 2010 Maule (e.g., Vigny et al., 2011; Yue et al., 2014b; Yoshimoto et al., 2016; Maksymowicz, et al., 2017), and 2004 Sumatra (e.g., Ammon et al., 2005; Rhie et al., 2007; Fujii and Satake, 2007) events, accompanying slip on the downdip portions of the megathrusts. In other cases, such as the 2014 Iquique, Chile (e.g., Lay et al., 2014; Hayes et al., 2014b), 2012 Nicoya, Costa Rica (e.g., Yue et al., 2013; Liu et al., 2015), 2003 Tokachi-oki, Japan (e.g., Miyazaki and Larson, 2008; Romano et al., 2010), and 2007 Pisco, Peru, ruptures (e.g., Lay et al., 2010a; Sladen et al., 2010), slip was concentrated on the central or deeper portion of the rupture zone, with no shallow coseismic slip.

Several subduction zones have experienced aseismic slip transients both before and after the main shocks. The 2011 Tohoku and 2014 Iquique earthquakes had large numbers of migrating repeating earthquakes during active foreshock sequences, with the repeating earthquakes providing a proxy for shallow aseismic slip that migrated toward the eventual main-shock hypocenters (e.g., Kato et al., 2012; Ruiz et al., 2014; Hayes et al., 2014b; Kato and Kakagawa, 2014; Meng et al., 2015a). Socquet et al. (2017) describes GPS data analysis that supports the presence of a slow slip event in the 2014 Iquique region during the 8 mo prior to the main shock. Aseismic slip triggered by great earthquakes is also common (e.g., Chlieh et al., 2007; Hsu et al., 2006; Ozawa et al., 2011; Hayes et al., 2014a; Uchida et al., 2015). Aseismic, or slow-slip, transients are not limited to regions of recent great earthquakes, as these have been observed in both shallow and deep portions of the seismogenic zone in many subduction zones around the globe (e.g., Dragert et al., 2001; Obara et al., 2004; Brudzinski et al., 2007; Schwartz and Rokosky, 2007; Correa-Mora et al., 2009; Jiang et al., 2012; Saffer and Wallace, 2015; Wallace et al., 2016).

Many of the great earthquakes in the last 39 yr have occurred in previously recognized seismic gaps (McCann et al., 1979; Nishenko, 1991), although some only partially filled the gap, leaving residual strain accumulation to be released in future events (Fig. 14). Targeted instrumentation of specific identified seismic gaps prior to great earthquakes in 2010, 2014, and 2015 along Chile, in 2005 and 2007 along Sumatra, in 2016 along Ecuador, and in other locations, has resulted in unprecedented seismic and geodetic data sets for recent events. However, even given the general constraint of the plate-boundary slip budget, the diversity of recent large earthquakes on megathrusts has presented surprises in terms of event location, size, and hazard for multiple plate boundaries. The 2004 Sumatra earthquake ruptured over a much greater length (>1300 km) of the plate boundary than had been considered likely (e.g., Shearer and Bürgmann, 2010), and the 2011 Tohoku, Japan, earthquake ruptured with unexpectedly huge slip, extending to the trench, along with rerupturing regions of recent, smaller earthquakes (e.g., Lay and Kanamori, 2011; Fujiwara et al., 2011; Iinuma et al., 2012; Lay, 2017). Both events generated devastating tsunamis and reaffirmed the potential for events larger than have been documented in recorded history to strike in various regions.

The 2015 Illapel earthquake ruptured the Chilean megathrust close to previous earthquakes in 1880 and 1943, and so the 2015 event has been suggested to be a “characteristic earthquake” for this segment of the margin (e.g., Tilman et al., 2016; Klein et al., 2017). In contrast, the 2014 Iquique earthquake spanned only about 20% of the northern Chile seismic gap, with most of the 1877 M ∼ 8.8 earthquake rupture zone remaining unbroken (e.g., Hayes et al., 2014b; Meng et al., 2015a; Cesca et al., 2016). Earthquakes along northern Japan were common in the last century, mainly in the M 7–8 magnitude range, but the 2011 M_w 9 Tohoku earthquake ruptured a region where the last giant event occurred in A.D. 869, along with multiple rupture zones of previous individual ruptures (e.g., Koper et al., 2011; Lay, 2015). The 2010 M_w 8.8 Maule earthquake ruptured a seismic gap in a region that last ruptured in 1835 in an M ∼8.5 earthquake (e.g., Moreno et al., 2010), but the main slip extended beyond that rupture, overlapping the rupture zone of a large event in 1928, but not the adjacent 1985 rupture zone. Along the Sumatra megathrust, the great earthquakes in 2005 and 2007 reruptured regions that had previously broken in 1797, 1833, 1861, and 1907 (e.g., Lay et al., 2005; Konca et al., 2008; Chlieh et al., 2008). The 2007 Peru earthquake occurred in a region last possibly ruptured in 1687 or 1746 and only partially filled a seismic gap between the 1996 and 1974 events (Pritchard and Fielding, 2008). The 2007 Solomon Islands earthquake ruptured in a very complex triple-junction plate-boundary zone, with no prior record of large earthquakes (Taylor et al., 2008; Furlong et al., 2009; Chen et al., 2017). Along the central Kuril Islands, the 2006 M_w 8.3 earthquake ruptured a portion of the megathrust between the 1963 M_w 8.5 and 1952 M_w 9 earthquakes identified as a seismic gap that may have partially failed in 1918 (e.g., Lay et al., 2009).

Observations of slip processes for subduction zone megathrusts have expanded greatly beyond early measurements of earthquake rupture front velocities, which were on the order of 75%–95% of the shear wave velocity in the slip zone expansion of the earthquake (e.g., Kanamori and Brodsky, 2004). Tsunami earthquakes appear to have very low rupture velocities, such as the 1–1.5 km/s rupture velocity of the 1992 Nicaragua tsunami earthquake (e.g., Kikuchi and Kanamori, 1995; Ihmlé, 1996a, 1996b) and the 1.25–1.5 km/s rupture velocity for the 2010 Mentawai event (e.g., Lay et al., 2011a; Newman et al., 2011). Kanamori et al. (2010) and Bilek et al. (2011, 2016) suggested that other events in the shallow region of tsunami earthquakes may also exhibit low rupture velocities. Variable rupture speeds may exist within an individual earthquake, such as the 2010 Maule (Kiser and Ishii, 2011) event, and rupture velocities may lie along a wide continuum between “typical” and “slow” speeds, such as the ∼1.5–2 km/s rupture velocity estimated for the 2013 Santa Cruz Islands earthquake (Lay et al., 2013; Hayes et al., 2014a). Supershear earthquakes have not been observed in the shallow portion of subduction zones, possibly because the events tend to have mixed mode ruptures.

At rupture velocities much slower than for tsunami earthquakes, a family of slip processes, including low-frequency earthquakes (LFEs) and very low-frequency earthquakes (VLFEs), involves slip events with velocities orders of magnitude lower than typical earthquake rupture velocities (Ide et al., 2007a). LFEs are deficient in high-frequency energy, with dominant signals around 1–8 Hz, durations of ∼0.3 s, and magnitudes of ∼10¹¹ Nm, with locations and mechanisms reflecting slip on the megathrust (e.g., Shelly et al., 2007; Ide et al., 2007b). In many cases, these LFEs occur as swarms that comprise much of the seismic nonvolcanic tremor observed in subduction zone settings (e.g., Ide et al., 2007b). LFEs were initially detected and located on the deeper portion of seismogenic zones (e.g., Katsumata and Kamaya, 2003; Shelly et al., 2006; Brown et al., 2009; Bostock et al., 2012), but they have now also been observed in shallow portions of accretionary prisms (e.g., Obana and Kodaira, 2009; Yamashita et al., 2015; Nakamura, 2017).

VLFEs represent even slower slip, with durations of ∼20 s, magnitudes up to ∼10¹⁴ Nm, and frequencies of ∼0.02–0.05 Hz, i.e., even more deficient in high-frequency energy than LFEs (e.g., Ide et al., 2007a). Similar to LFEs, these events appear to be located throughout the seismogenic zone (Fig. 1), from the deep transition between the seismogenic and aseismic zones (e.g., Ito et al., 2007, 2009) to the updip end of the seismogenic zone, in some cases, very close to the trench (e.g., Ito and Obara, 2006; Asano et al., 2008; Matsuzawa et al., 2015; Araki et al., 2017), and within the accretionary prism (e.g., Obara and Ito, 2005; Ito and Obara, 2006).

At the slowest end of the slip spectrum, there are slow slip events (SSEs), which are observed geodetically along many subduction zones. These slip events can be quite large, reaching an equivalent seismic moment of 10¹⁸–10²⁰ Nm, with durations over 10⁵–10⁷ s (days to months; e.g., Ide et al., 2007a). SSEs are often, but not always, accompanied by nonvolcanic tremors and are grouped into the episodic tremor and slip (ETS) category (e.g., Dragert et al., 2001; Kostoglodov et al., 2003; Obara et al., 2004; Hirose and Obara, 2006; Schwartz and Rokosky, 2007; Beroza and Ide, 2011). These were also initially observed in the vicinity of the deepest portions of the seismogenic zone (Fig. 15), but more recent observations indicate that shallower megathrust SSEs occur as well (Saffer and Wallace, 2015, and references therein), including very close to the trench (Araki et al., 2017).

Choy and Boatwright (1995) determined radiated energy from global shallow earthquakes (M > 5.8) and used it to compute apparent stress (given by the ratio of radiated energy to seismic moment, E_R/M₀, multiplied by the average rigidity). They found that subduction zone thrust events have the lowest values (∼0.29 MPa) of apparent stress for any tectonic category, although with regional variations between 0.15 MPa (Kermadec) and 0.42 MPa (Peru-Ecuador).

Newman and Okal (1998) determined radiated energy and energy-to-moment ratios for over 50 teleseismic earthquakes, highlighting the 1992 Nicaragua, 1994 Java, and 1996 Peru tsunami earthquakes as having distinctly low values for moment-scaled radiated energy relative to other earthquakes in their data set. Okal and Newman (2001) examined other earthquakes within these three tsunami earthquake-producing regions, finding additional “slow” earthquakes present only near the 1960 Peru tsunami earthquake zone. Convers and Newman (2011) expanded the catalog of Newman and Okal (1998) by determining estimates of radiated seismic energy for 342 earthquakes from 1997 to 2010, finding shallow subduction zone thrust earthquakes generally deficient in radiated energy. They found additional energy-deficient events in the rupture zones of the 1992 Nicaragua and 2006 Java tsunami earthquakes. Venkataraman and Kanamori (2004) and Ye et al. (2016c) also demonstrated that tsunami earthquakes have low moment-scaled radiated energy compared to other subduction zone earthquakes.

For events within a given region, recent great earthquakes that have been well recorded by large seismic networks have a depth dependence in the frequency of radiated seismic energy (e.g., Lay et al., 2012). The deeper parts of the megathrust produce significantly more high-frequency energy than the shallower megathrust for most of the largest earthquakes in recent decades, such as the 2010 Maule Chile M_w 8.8 (e.g., Kiser and Ishii, 2011; Koper et al., 2012), 2011 Tohoku M_w 9.1 (e.g., Koper et al., 2011; Ide et al., 2011; Ishii, 2011), and the 2015 M_w 8.3 Illapel, Chile, event (e.g., Melgar et al., 2016; Yin et al., 2016). Ye et al. (2016c) documented modest depth dependence in high-frequency radiation for the 100+ large events, with lower spectral decay rates for deeper events on subduction zone megathrusts.

Allmann and Shearer (2009) determined stress drop for over 1700 global earthquakes, finding little depth variation in the full global catalog, but regionally significant depth-dependent stress drops. They also found along-strike variations, in particular, very low values along portions of the Central America subduction zone and near the hypocenter of the 2004 Sumatra earthquake. Similar to the depth dependence of source duration discussed in the “Source Duration” subsection herein, Denolle and Shearer (2016) found very weak depth dependence for stress drop and scaled energy using their global catalog. They observed regional variations, with lower stress drop in general in western Pacific subduction zones as compared to eastern Pacific zones, and very low stress drops for earthquakes along the Alaska-Aleutian subduction zone. Ye et al. (2016c) found little depth dependence of stress drop for major and great megathrust events deeper than 15 km, but there is an order of magnitude scatter in stress drop values with little regional coherence. Our compilations of these various global studies also suggest weak depth dependence in both E_R/M₀ and stress drop, with large scatter (Figs. 16 and 17).

On regional scales, spatial variations in stress drop for small earthquakes are observed and may be useful for probing the stress state of a megathrust. Uchide et al. (2014) described stress drops computed for over 1500 small earthquakes within the Tohoku region of Japan prior to the 2011 M_w 9.1 earthquake. They found increasing stress drop at 30–60 km depths, and strong along-strike variations that appeared to link areas of high-stress-drop events with zones of little coseismic slip in 2011, south of the region of maximum main-shock slip. Bilek et al. (2016) used spectral ratio techniques to compute corner frequencies, seismic moment, and stress drop for a series of earthquakes that occurred within and adjacent to the 1992 Nicaragua tsunami earthquake rupture zone. Earthquakes from 2005 to 2006 that occurred within the 1992 rupture zone exhibited significantly lower corner frequencies and stress drop than those events that occurred adjacent to the 1992 rupture zone, suggesting that regional variations in the megathrust zone that affect rupture processes may persist for many years afterward, if not permanently.

Compilations of aftershock locations relative to areas of high slip during large earthquakes of all types suggest that aftershocks tend to locate at boundaries between areas of high slip and low slip and/or at edges of large-slip regions (e.g., Das and Henry, 2003; Wetzler et al., 2018). This observation has been supported by aftershock catalogs for many of the M > 8 earthquakes, such as the recent Sumatra (e.g., Hsu et al., 2006), Chile (Hayes et al., 2014b; Yue et al., 2014b; Li et al., 2016), and Tohoku (e.g., Asano et al., 2011) earthquakes. The 2013 Santa Cruz Islands earthquake appears to have had few aftershocks on the megathrust fault, and this event may have triggered an adjacent aseismic slip episode following the main shock (Hayes et al., 2014a). The 2014 Iquique earthquake produced many aftershocks, but the aftershocks had low b values, despite the occurrence of a large aftershock (Hayes et al., 2014b). There is a weak tendency for high-stress-drop megathrust ruptures with a given moment to have lower aftershock production, which is attributed to the smaller main-shock rupture dimensions (Wetzler et al., 2016). The source dimensions of the main slip zone for the 2014 Iquique event were indeed unusually small for a great earthquake.

Aftershocks occur off of the main megathrust rupture plane as well. Normal faulting aftershocks have been observed after many large megathrust ruptures (Fig. 18), particularly those that rupture very shallowly near a trench (e.g., Kanamori, 1971; Christensen and Ruff, 1988; Lin and Stein, 2004; Ammon et al., 2008; Lay et al., 2009; Asano et al., 2011; El Hariri and Bilek, 2011; Bilek et al., 2011; Ide et al., 2011; Kato et al., 2011; Lange et al., 2012; Rietbrock et al., 2012; Yue et al., 2014b; Wetzler et al., 2017; Sladen and Trevisan, 2018). Some of these intraplate extensional events are within the outer rise, seaward of the trench, or on plate-bending related faults below the megathrusts, while others are found within the upper plate adjacent to the zone of high slip (e.g., Farías et al., 2011; Hicks and Rietbrock, 2015). Events with rupture deeper on the megathrust activate relatively few intraplate aftershocks (Wetzler et al., 2017; Sladen and Trevisan, 2018), but in some cases, they have large afterslip on the updip region of the megathrust (Fig. 18).

As demonstrated in the previous section, earthquake observations now span a wide range of slip speed and complexity. Here, we outline a range of possible factors leading to these diverse slip observations.

Key questions about where slip can occur in subduction zones often start with an examination of seismic coupling. This can be defined and estimated in a variety of ways (e.g., Wang and Dixon, 2004), but it fundamentally involves a short time window estimate of how much of the plate motion is released in fast processes that generate seismic energy (e.g., Scholz and Campos, 2012). Various studies have estimated seismic coupling for large segments of subduction zones in order to compare it with other parameters, such as slab age and dip, to understand the large-scale controls on coupling. With the availability of longer earthquake catalogs, more detailed seismic observations of individual earthquakes, and increasingly widespread geodetic observations, other studies have focused on smaller spatial scales to examine coupling and the relationship to seismic activity. We summarize both scales of observations here.

Uyeda and Kanamori (1979) described end-member subduction models with distinct back-arc processes. They noted that zones with active back-arc spreading, such as the Marianas, have fewer large and great earthquakes than zones without back-arc spreading, such as the great earthquake-producing Chilean margin. As an explanation for these end-member behaviors, they suggested each zone has a different level of plate coupling caused by either time evolution of the subduction geometry and progressive decoupling or differences in the motion of the overriding plate.

In their broader review of plate coupling in many subduction zones around the globe, Scholz and Campos (1995) suggested that the strongest coupling occurs where large megathrust normal forces exist, controlled by plate age, slab length and dip angle, and velocity of plate motion. Subduction zones with active back-arcs, and old, steeply dipping subducting lithosphere such as the Marianas, southern Tonga, and Izu-Bonin have the lowest normal force acting on the interface in this model, and so they have the weakest coupling and smaller maximum-magnitude earthquakes. In contrast, the Chile, Peru, Sumatra, Alaska, and Cascadia subduction zones have higher normal forces and plate coupling. Scholz and Campos (2012) revisited this model with updated seismic catalogs and coupling estimates finding a similar correlation, with the larger earthquakes occurring in regions of higher normal forces and higher coupling. They also took advantage of the large number of geodetically based coupling estimates to resolve significant along-strike coupling variations within individual subduction zones.

Because of the importance of slab dip angle on normal forces and coupling, various studies have examined dip angles in the context of other subduction zone parameters. For example, a comparison of slab dip in over 150 geometrically simple subduction zone transects suggests that the absolute motion of the overriding plate has the strongest correlation with slab dip, much larger than factors such as the age and thermal structure of the slab, the slab pull force, or convergence rate (Lallemand et al., 2005). Back-arc spreading is only found in areas where the deeper portion of the slab dips at greater than 50°, and back-arc shortening is prevalent in areas with slab dips of less than 31° (Lallemand et al., 2005). Curvature of the megathrust is strongly correlated with average dip angle of the megathrust, but it is also influenced by dip of the deeper slab, which may cause shear strength to vary across the plate boundary. More shallowly dipping, flatter boundaries are able to support larger earthquakes due to more homogeneous strength (e.g., Bletery et al., 2016). The nature of the upper plate, being either continental or oceanic, also appears to be important, as are factors such as the presence of back-arc spreading and trench roll-back (Scholz and Campos, 2012).

Fluids play a critical role in many subduction zone processes, from fault slip to volcanic activity (e.g., Peacock, 1990). We focus here on the aspects of fluids that influence slip processes on the megathrust, from great earthquakes to the slow-slip and tremor episodes observed in many regions. Fluids enter the shallow seismogenic system either within the pore space of the subducting igneous crust and sediment, or bound in hydrous minerals. Normal faulting associated with bending of the subducting plate is likely to produce enhanced hydration of the crust and upper mantle seaward of the subduction zone (e.g., Ranero et al., 2003; Naliboff et al., 2013; Korenaga, 2017; Grevemeyer et al., 2018). These fluids are released either through compaction in the shallowest 5–7 km or in dehydration reactions at higher pressures and temperatures (e.g., Saffer and Tobin, 2011). The fluids move through the system by a variety of pathways, including faults, fractures, and permeable strata within the megathrust and overriding plate (e.g., Carson and Screaton, 1998), or they remain trapped by low-permeability sediments, affecting the seismogenic zone by increasing fluid pressures (e.g., Saffer and Bekins, 2002). Excess fluid pressures within the megathrust zone modify the effective normal stress, which is linked to overall fault strength and generation of both typical earthquakes and slower-slip processes (e.g., Saffer and Tobin, 2011).

Because high fluid pressure acts to reduce effective normal stress, fluids play an important role in the earthquake process. Various studies have provided evidence for significant fluid content and high fluid pressures in the megathrust zone, often based on bright spots in seismic reflection profiles and high Vp/Vs seismic velocity ratios, which serve as a proxy for high fluid pressures. We note only a few of these studies here, along with the seismic implications of these observations. Tobin and Saffer (2009) estimated high pore-fluid pressures and low effective stress along the shallow Nankai subduction zone using detailed seismic reflection data. Similarly, Park et al. (2014) attributed along-strike variability in seismic reflection characteristics along the Nankai subduction zone to differences in fluid amounts along the megathrust. Bangs et al. (2015) used three-dimensional (3-D) seismic imaging to identify fluid-rich segments along the plate interface offshore central Costa Rica. Audet et al. (2009) used seismic receiver functions and Vp/Vs to suggest that the megathrust zone along the northern Cascadia margin has low permeability, acting to seal high fluid pressures of the oceanic crust, and leading to significant overpressures in this zone. Audet and Schwartz (2013) found along-strike variations in upper-plate Vp/Vs and inferred fluid pressure variations along a small segment of Costa Rica, with lower Vp/Vs in the northwestern portion of the Nicoya Peninsula relative to the higher Vp/Vs and higher fluid pressures along the southeastern segment of the peninsula. They suggested that these upper-plate variations influence the seismogenic zone behavior.

These variations in fluid content and pressure are also linked to variations in slip processes. Along the Nankai subduction zone, the high fluid pressures in the shallow zone may inhibit near-trench slip (Tobin and Saffer, 2009), and the along-strike variability at depth may allow for nucleation of tsunami earthquakes in fluid-poor regions, with rupture into fluid-rich zones later in the event (Park et al., 2014). Along central Costa Rica, the seismicity appears to preferentially occur in the fluid-poor zones (Bangs et al., 2015), and along the Nicoya Peninsula, Costa Rica, the areas of low fluid pressures host small-magnitude earthquakes with higher apparent stresses than are found in the high-fluid-pressure areas (Audet and Schwartz, 2013; Stankova-Pursley et al., 2011).

Other studies have compared seismic observations of Vp/Vs and fluid pressure estimates to geodetic coupling and rupture details of specific earthquakes. For example, Moreno et al. (2014) examined a portion of the 2010 Maule Chile rupture zone and found a positive correlation between areas of high Vp/Vs and low geodetically determined coupling. Huang and Zhao (2013) and Yamamoto et al. (2014) found significant Vp/Vs heterogeneity in the shallow megathrust zone of the 2011 Tohoku rupture, suggesting a high Vp/Vs ratio in the very near-trench area of high coseismic slip, as well as downdip of the hypocenter of the event, with low Vp/Vs just updip of the hypocenter.

Many observations of slow slip and tremors in subduction zones correspond to areas that appear to have high fluid pressures. Early efforts to identify and locate nonvolcanic tremors in the Nankai subduction zone resulted in locations at depths where pressures and temperatures are such that fluids can be generated through dehydration (Obara, 2002). Rogers and Dragert (2003) emphasized the potential importance of fluids in their initial reporting of Cascadia episodic tremor and slip. Tremor signals along the Nicoya Peninsula, Costa Rica margin, correlate with observations of seafloor fluid-flow measurements at the surface of the forearc, which is explained by slow-slip motions in the shallow portion of the megathrust causing the observed transient flow in the upper-plate sediments (Brown et al., 2005).

LFEs have also been linked to high fluid pressures. Shelly et al. (2006) located LFEs within tremors along the Nankai subduction zone in an area of high Vp/Vs, indicative of high pore-fluid pressures. Recent locations of LFEs in Nankai correlate variations in LFE location with Vp and Vs anomalies. Nakajima and Hasegawa (2016) found gaps in LFE occurrence in areas where they observed small Vp and Vs anomalies, indicative of a well-drained and low-pore-pressure megathrust; the LFEs are instead concentrated in areas of higher Vp and Vs anomalies, serving as proxies for higher-pore-fluid pressure.

Similar correlations exist for slow-slip events and high fluid pressures. Seismic images suggestive of a high-fluid-pressure zone in the area of slow earthquakes along the Nankai subduction zone prompted Kodaira et al. (2004) to link the slow-slip event generation with fluids. Song et al. (2009) documented a very low-seismic-velocity layer in the same region as several slow-slip events along the Mexico subduction zone; they postulated that the low velocities are the result of a region with high pore-fluid pressure. Seismic reflection data suggest high fluid pressures in slow-slip zones of the Hikurangi margin (Bell et al., 2010). Audet and Schwartz (2013) identified similar correlations among high Vp/Vs, higher fluid pressures, and slow slip in the southeastern Nicoya segment of the Costa Rica subduction zone. The high fluid pressures, and the related decrease in the effective normal stress are commonly invoked as necessary for these slow-slip processes in the both the shallow (e.g., Saffer and Wallace, 2015) and deep portions of the seismogenic zone (e.g., Audet and Kim, 2016).

Observations of finer-scale spatial and temporal distribution of fluid pressure are now being made and related to slow slip and tremor processes. Kato et al. (2010) found an asymmetric pattern of highest Vp/Vs in the area of slow slip along the Tokai segment of Japan, with lower Vp/Vs and fluid pressures in the deeper segment of the fault where the LFEs and tremor occur. Frank et al. (2015) suggested variable fluid pressures over fairly short time and spatial scales based on acceleration and deceleration of LFE activity in Mexico adjacent to motion during the 2006 slow-slip event.

Another important input into the subduction zone is sediment, as the quantity and mineralogical content can affect shallow slip processes. Comparisons between locations of great earthquakes and sediment thickness highlight the first-order observation that great earthquakes occur in regions with thick sediment inputs (Ruff, 1989). This observation led to the idea that thick subducted sediment packages act to smooth the plate-boundary zone, resulting in uniform seismic coupling that allows earthquakes to rupture unimpeded for large distances on the megathrust (e.g., Ruff, 1989). This model has persisted throughout the last three decades of seismicity and improved sediment thickness estimates (e.g., Heuret et al., 2012), with Scholl et al. (2015) finding the majority of great earthquakes occur in subduction zone segments with sediment thickness greater than 1 km (Fig. 19). Seno (2017) observed that a thickness of subducted sediment greater than 1.3 km is found for regions with M_w ≥ 9.0 earthquakes, and such events do not occur if the thickness is less than 1.2 km.

As with all of these factors, spatial variations in sediment thickness may be relevant for understanding the details of slip patterns. Various diagenetic and metamorphic processes that affect the friction and effective stress in the shallow megathrust materials play important roles in defining the updip edge of seismicity (Moore and Saffer, 2001). Thick and strong sediments in the 2004 Sumatra earthquake rupture zone may have allowed very shallow rupture for that event, in contrast to the deeper slip of the 2005 Nias earthquake (Geersen et al., 2013). Large variations of subducted sediment along-strike in northern Hikurangi may affect the locations of shallow slow-slip events there (Bell et al., 2010; Eberhart-Phillips and Bannister, 2015). Based on determination of seismic velocities in the location of 2011 Tohoku rupture, Zhao et al. (2011) found that high coseismic slip and foreshock locations are concentrated in the high-seismic-velocity areas; the nearby low-seismic-velocity patches, inferred to be sediments and high fluid content, host aseismic creep and slow thrust events. Beyond the sediment thickness and composition, fluids expelled from subducted sediments also appear to be an important component in the slow-slip process (e.g., Peacock, 2009; Peng and Gomberg, 2010; Audet and Kim, 2016).

Sediment type is also important for controlling slip behavior. Many studies have focused on the frictional stability of various clay-rich and carbonate sediments observed entering subduction zones to identify the conditions under which velocity weakening behavior is possible (e.g., den Hartog et al., 2012; den Hartog and Spiers, 2013; Ikari et al., 2013; Kurzawski et al., 2016). Geersen et al. (2013) inferred that the clay-rich sediment composition in the 2004 Sumatra earthquake rupture zone was as important as the thickness in controlling the shallow rupture. Ikari et al. (2015a, 2015b) demonstrated that samples of the smectite-rich fault material obtained from the shallow drilling of the 2011 Tohoku earthquake zone can deform in the laboratory at a variety of rates, from fast seismic slip to slow SSE speeds, to produce the range of slip behaviors observed in that zone.

Yet another important input into the subduction zone is the overall roughness of the subducting plate. This can be smoothed with a thick sediment blanket, or it can be rough, marked with seamounts, ridges, and other topographic features. As mentioned in the previous section, thick sediment subduction and smooth plates have been linked with the occurrence of great earthquakes. The way in which the rough plate affects earthquake rupture appears to be more variable. From large seamounts entering the Japan Trench (e.g., Nishizawa et al., 2009), to smaller seamounts leaving scars in the forearc in Costa Rica (e.g., Ranero and von Huene, 2000), and large seamounts and ridges entering the subduction zone along Peru, Chile, and elsewhere (e.g., Spence et al., 1999; Robinson et al., 2006; Sparkes et al., 2010; Marcaillou et al., 2016), these features are significant enough to likely affect megathrust earthquakes, although exactly how is an active debate. Kelleher and McCann (1976) correlated locations of great earthquakes with the relatively smooth bathymetry portions of subduction zones, finding only moderate-magnitude earthquakes occurring in areas with significant subducting bathymetry such as aseismic ridges. They suggested that subduction of these large topographic features leads to increased buoyancy at those locations, which modifies coupling in those zones, reducing the ability to produce great earthquakes. On the other end of the spectrum, Cloos (1992) suggested that many subduction zone thrust earthquakes are the result of subducted seamounts acting to strongly couple with the overriding plate, producing asperities and earthquake slip, although the ability to produce large earthquakes would reflect the amount of sediment acting as seamount cover to seismogenic depths (Cloos and Shreve, 1996).

In order to understand how subducted topography might affect earthquakes, we need to understand the fate of the topography within the subduction zone itself. One possibility is that both normal stress and seismic coupling increase because of the extra mass and buoyancy associated with a subducted seamount (Scholz and Small, 1997). Using analogue sandbox models of seamount subduction, Dominguez et al. (2000) demonstrated significant deformation in the upper layers of sand as the topographic feature passed at depth. Following in the spirit of the sandbox models, Wang and Bilek (2011) presented a conceptual model of strong geometric features such as seamounts significantly deforming the surrounding upper plate with a complex network of fractures, supported by geologic evidence of highly fractured exhumed upper-plate zones, seismic imaging, and recent numerical models (Ding and Lin, 2016). These models suggest different seismic behavior in areas of subducting topography, where these features could act as either regions of high slip, or regions that arrest slip during a megathrust earthquake.

The models are variously supported by a range of documented cases of these features acting either as barriers to rupture or as regions of high slip. The ideas developed by Kelleher and McCann (1976) of great earthquakes avoiding areas of seamount subduction persist with consideration of updated earthquake catalogs (Wang and Bilek, 2014). Instead of producing large earthquakes, several of these regions exhibit high levels of creep (low seismic coupling), which are a proposed consequence of seamount deformation of the upper-plate model (Wang and Bilek, 2014). Yang et al. (2013) used numerical simulations to show that subducted seamounts can act as rupture barriers over a wide range of seamount sizes and normal stress conditions. In regions adjacent to subducted topography, several great earthquakes have been observed, leading to many studies suggesting that subducted topography acts as a barrier to further rupture. For example, Kodaira et al. (2000) imaged a subducted seamount at depth in the Nankai subduction zone that they suggest acted as a barrier in the 1946 M_w 8.3 Nankai earthquake rupture. Other examples of significant subducted topography acting as rupture barriers include along the South American margin (e.g., Perfettini et al., 2010; Sparkes et al., 2010; Geersen et al., 2015), Sumatra margin (e.g., Chlieh et al., 2008; Henstock et al., 2016), and Japan Trench (e.g., Simons et al., 2011; Duan, 2012). In some cases, the subducting topography may initially act as a barrier and then fail later in the rupture, producing significant slip, as proposed for the 2001 M_w 8.4 Peru earthquake (Robinson et al., 2006). Given the position of the subducted topography with depth, these features might also affect the downdip earthquake rupture behavior as well, as suggested by Marcaillou et al. (2016) as a means for impeding updip rupture of the 1942 Ecuador M 7.8 earthquake.

Other small- to moderate-magnitude earthquakes have produced slip in subducting areas with rough topography. Using high-quality seismic surveys and monitoring efforts, Mochizuki et al. (2008) located several repeating M ∼ 7 earthquakes along a portion of the Japan Trench megathrust landward of a subducted seamount, not at the seamount position at depth. Several studies have suggested that subducted seamounts acted as asperities in past Costa Rica M ∼ 7 earthquake ruptures (e.g., Protti et al., 1995; Husen et al., 2002; Bilek et al., 2003), M ∼ 7 earthquakes along the Hikurangi margin (e.g., Bell et al., 2014), as well as the 2010 Maule, Chile, earthquake (Hicks et al., 2012).

Subducting topography has also been linked with other seismic processes, such as small-magnitude earthquake swarms, nonvolcanic tremor, and slow-slip events. Earthquake swarms along South American and Japanese megathrusts appear to correlate with locations of subducted topography (Holtkamp and Brudzinski, 2011). Tréhu et al. (2012) found evidence for small-magnitude earthquake clusters in areas of subducted seamounts along the Cascadia subduction zone. Linear streaks of nonvolcanic tremors in the Nankai subduction zone correlate with the path of previously subducted seamounts (Ide, 2010). Bell et al. (2010) used seismic reflection images of the Hikurangi margin to suggest that subducted seamounts allow for lower effective stress and promote the occurrence of slow slip, while Kodaira et al. (2004) suggested a similar connection between subducted ridges and slow slip in the Nankai subduction zone.

The upper plate may also impact megathrust slip behavior. Correlations between forearc basins and the high-slip zones of many megathrust earthquakes led Wells et al. (2003) to suggest that subsidence associated with these basins might be linked with areas of subduction erosion (requiring high coupling), affecting the megathrust stress state. Similarly, correlations between large negative trench-parallel gravity anomalies and the locations of great earthquakes may also connect forearc structure with megathrust processes (Song and Simons, 2003). Gravity anomalies are also associated with displacement in great events like the 2011 Tohoku earthquake, where positive gravity anomalies, representing differences in upper-plate geology, may have influenced the frictional conditions on the plate interface (Bassett et al., 2016). Geology and strength of the margin prism may also impact the amount of slip and seafloor deformation, based on numerical models showing high slip and increased tsunami generation beneath compliant prisms (Lotto et al., 2017). Upper-plate faulting may also provide limits on megathrust earthquake ruptures, as suggested for Ecuador (Collot et al., 2004) and Chile (Moreno et al., 2012).

Upper-plate variations may also affect nonvolcanic tremor occurrence. Along Cascadia, along-strike variations in tremor location and recurrence interval are spatially linked with variations in upper-plate geologic terranes (Brudzinski and Allen, 2007). Wells et al. (2017) showed a correlation between upper-plate faults and reduced levels of tremor activity along Cascadia, suggesting that the upper-plate faults allow fluid pressures to drain from the fault at depth, limiting tremor occurrence.

Due to their size and frequent occurrence, earthquakes on megathrust faults constitute significant seismic and tsunami hazards in subduction zones around the world. Slab deformation and partitioned block faulting of the upper plate also contribute to hazards in many regions.

With recent improvements in geodetic coverage by GPS and GNSS stations on the upper plate of subduction zones, detailed images of heterogeneous patterns of interseismic slip deficit (or “locking”) are now available. These illustrate the complexity of fault locking and creep in many subduction zones, and they provide a basis for refining the regional hazard. Comparisons between prior fault locking maps and recent coseismic slip in large events have led to a major advance, as relatively good correlations are observed between regions of high coseismic slip and strong plate locking when both are well resolved. For example, geodetic and paleogeodetic observations dating back several decades suggest heterogeneous patterns of strong coupling along the Sumatra margin, and the strongly coupled zones served as the high-slip regions in the 2005 M_w 8.7 Nias, and 2007 M_w 8.4 and M_w 7.9 Sumatra earthquakes (Chlieh et al., 2008). Similar observations of strong locking found in regions of high coseismic slip exist for the 2007 M_w 8.0 Pisco, Peru (Perfettini et al., 2010), 2010 M_w 8.8 Maule, Chile (Moreno et al., 2010), 2011 M_w 9 Tohoku (e.g., Ozawa et al., 2012), 2012 M_w 7.8 Nicoya, Costa Rica (e.g., Protti et al., 2014), and the 2016 M_w 7.8 Ecuador (Ye et al., 2016a) earthquakes.

These correlations between the geodetically determined locked patches and areas of high slip during great earthquakes have implications for seismic and tsunami hazard assessments along subduction zones. However, land-based geodetic measurements have very poor resolution of near-trench slip deficit, as was found for the 2011 Tohoku earthquake (e.g., Lay, 2017). A major limitation is the paucity of offshore geodetic observations in most subduction zones; without offshore observations, resolution of the spatial extent of locking and strain accumulation is very limited. An example where offshore observations now exist in reasonable density for this to be overcome is along the Nankai region of southwest Japan. Yokota et al. (2016) presented coupling estimates based on offshore measurements, suggesting wide areas of the shallow Nankai subduction zone have high slip deficit and are strongly locked, which expands the regions where shallow, tsunamigenic slip is possible. Seafloor measurements by GPS and acoustic methods (Sato et al., 2011; Kido et al., 2011), along with ocean bottom pressure sensors, provided invaluable data for resolving the shallow slip in the 2011 Tohoku earthquake. The greatly expanded cabled S-net offshore Honshu now provides continuous monitoring of moderate-slip earthquakes, and data assimilation methods will enable early tsunami warning of unprecedented quality.

The previous section outlined advances made in correlating areas of strong geodetic locking with the areas of high coseismic slip in great earthquakes; however, it is also valuable to evaluate low coupling zones. Areas thought to involve largely aseismic creep, in some cases linked to areas of rough subducting topography (e.g., Wang and Bilek, 2014), can only be confirmed as deforming this way by geodetic measurements. We cannot ignore the possibility than any subduction megathrust, including those with highly variable structure, can host large earthquakes, even if historical seismicity does not reveal the potential. One such example is along the Solomon Islands subduction zone, where a triple junction marks the subduction of the Australia and Solomon Islands–Woodlark Basin plates beneath the Pacific plate. This region was largely aseismic prior to the 2007 M_w 8.1 earthquake, which produced slip along the plate-boundary interface for both the Australian and Solomon Islands–Woodlark Basin plates, crossing the fracture zone boundary between these two plates and marking the first observed rupture across a triple junction (Taylor et al., 2008; Furlong et al., 2009).

The near-trench region of the megathrust was once considered to be an unlikely area to host large coseismic slip because of unfavorable frictional conditions, yet examples such as the massive coseismic slip in the near-trench region of the 2011 Tohoku earthquake (e.g., Lay et al., 2011b; Ide et al., 2011; Ozawa et al., 2012; Iinuma et al., 2012) provide dramatic evidence to the contrary. Very shallow megathrust slip also occurred in the 2006 M_w 8.3 Kuril earthquake, an area also lacking in previous interplate seismicity (Ammon et al., 2008; Lay et al., 2009).

Quantification of the stress transfer and earthquake triggering potential is another important consideration for assessing seismic hazard in subduction zones. Earthquake triggering by transfer of stress, either dynamic or static, is recognized in fault zones around the world (e.g., Hill et al., 1993; King et al., 1994; Lin and Stein, 2004; Freed, 2005). Many subduction zone examples exist, such as along the Sumatra margin, with its history of great earthquakes in the eighteenth and nineteenth centuries. Modern great earthquake activity in this region began with the 2004 M_w 9.2 Sumatra underthrusting event. Following the 2004 event, other great megathrust earthquake ruptures progressed southward along the margin in the 2005 M_w 8.6 Nias (e.g., Konca et al., 2007) and the 2007 M_w 8.5 and M_w 7.9 Sumatra earthquakes (e.g., Konca et al., 2008). This sequence brackets a locked region along the Mentawai Islands offshore of Padang, which last ruptured in 1797, and where a future great megathrust earthquake could occur.

The 2006–2007 Kuril Islands sequence is another example of coupled great earthquakes, although with diverse focal mechanisms. The first event in the sequence, the 2006 M_w 8.4 event, occurred on the megathrust, whereas the second M_w 8.1 event in 2007 was a normal-faulting, outer-rise event that produced stronger shaking in Japan due to higher moment-scaled radiated energy (e.g., Ammon et al., 2008). Slip from the 2006 earthquake, as well as from a large 1963 compressional earthquake in the trench slope region, led to increased static stress on optimally oriented normal faults in the region of the 2007 event (Raeesi and Atakan, 2009).

The 2009 Samoa-Tonga earthquake doublet also involved interaction between two separate faults, but with a pattern reversed in time relative to the 2006–2007 Kuril Islands sequence. In the Samoa-Tonga case, a M_w 8.1 normal-faulting, outer-rise event triggered rupture of the Tonga megathrust with cumulative M_w of 8.0 within minutes, compounding tsunami runup and fatalities (Lay et al., 2010b). This style of triggering is particularly troublesome, because the secondary underthrusting events in the sequence occurred within tens of seconds, hampering the ability to quickly assess the tsunami potential.

The hazard posed by tsunami earthquakes is now well recognized, but it remains difficult to characterize in most regions. As mentioned herein, it is very challenging to establish whether the shallow portion of a megathrust is frictionally locked and accumulating strain using on-land geodetic data. This is particularly true for wide megathrusts. It is also difficult to establish the updip limit of rupture in large megathrust ruptures to judge whether there is potential for tsunami earthquakes to occur subsequently, as occurred in the 2010 Mentawai earthquake updip from the 2007 Sumatra earthquakes. The tendency for very shallow ruptures to be depleted in short-period seismic wave radiation can cause delayed tsunami hazard reaction, even when shaking alerts coastal communities to an earthquake occurrence. The best strategy for mitigating this hazard appears to be offshore continuous seafloor deformation monitoring (particularly cabled seafloor pressure sensors) and regional real-time source inversion procedures connected to an effective tsunami warning and evacuation system.

Strong shaking from offshore events constitutes another major hazard, and even less response time is available than for tsunami response. Continuous early warning systems that evaluate direct P-wave shaking with continuous updates may provide tens of seconds of warning time prior to strong S and surface-wave shaking, but the weak onset of the 2011 Tohoku earthquake indicates the challenges of evaluating how large an earthquake will grow after saturation of early high-sensitivity sensors. Continuous seismo-geodesy, combining strong motion sensors and real-time GPS, appears to be the most promising on-land method (e.g., Geng et al., 2013; Crowell et al., 2013; Melgar et al., 2013), with offshore cabled seafloor monitoring being even more valuable, although expensive.

In many regions, current instrumentation is deficient to achieve robust tsunami and seismic warning systems, and efforts to enhance seismic and geodetic networks are needed. Societal preparation, enforced building codes, and seismic retrofitting of older structures are critical and essential elements for mitigating the hazard posed by megathrust earthquakes.

Megathrust faults on converging plate boundaries host Earth’s largest earthquakes, and the seismic and tsunami events in these regions constitute some of Earth’s primary natural hazards. Over the past 50 yr, great progress has been made in understanding the fundamental plate tectonics driving motions on megathrusts, but recent events such as the 2004 Sumatra and 2011 Japan events have still shocked even earthquake scientists with the devastating potential of megathrust faulting. Megathrust faults have diverse behavior that is partly understood in terms of overall tectonic setting, but questions remain concerning causes of frictional variations, possible loading rate dependence of frictional failure, the degree of shear stress reduction during faulting, and even behavior of some regions with long repose between events. Advances in seismology and geodesy now provide information about the seismic cycle for some regions, and over time improved assessments of seismic potential appear to be viable. Offshore geodetic monitoring will provide critical information about the along-dip and near-trench strain accumulation process and the occurrence of shallow slow-slip events and stable sliding that may prevent strain buildup. Coupled offshore and onshore seismic and geodetic monitoring combined with real-time processing have the potential to establish early warning systems for megathrust tsunami and strong shaking, but this will require extensive global investment in instrumentation and operations facilities around the world.

T. Lay’s research on earthquakes is supported by National Science Foundation (NSF) grant EAR-1245717. S. Bilek’s research on earthquakes is supported by NSF grant OCE-1434550. Reviews by Guest Editor L. Wallace and two anonymous reviewers helped us to improve the manuscript.