Great (≥Mw8.0) megathrust earthquakes and the subduction of excess sediment and bathymetrically smooth seafloor

Ruff (1989) observed that the entrance of an “excess quantity” of sediment into a lengthy subduction (≥230–300 km) sector of subduction zone (SZ) favors the nucleation there of interplate thrust (IPT) or megathrust earthquakes of magnitude Mw8.2 or greater (Fig. 1). He conjectured that subducted sediment worked to both strengthen interplate coupling and smooth the lateral or trench-parallel distribution of coupling strength, a circumstance that promotes rupture continuation and the consequent generation of high-magnitude or great (≥Mw8.0) IPT earthquakes.

An “excess” quantity was considered a thickness adequate to nourish the building of an accretionary frontal prism. This thickness is commonly estimated at ≥1 km (von Huene and Scholl, 1991; Clift and Vannucchi, 2004; Scholl and von Huene, 2007). Subducting sediment enters the subduction channel (see Fig. 2, panel C) that physically separates the upper and lower plate (Cloos and Shreve, 1988a, 1988b; Moore et al., 2007; Collot et al., 2011). Megathrust rupturing occurs along the top or bottom or within the subduction channel.

In contrast, the subduction of bathymetrically rough seafloor would be expected to produce an uneven or heterogeneous distribution of coupling strength. This situation would condition short-duration rupturing typical of lower magnitude IPT earthquakes (see Ruff, 1989; Fig. 3). It is now generally recognized that, in fact, the subducted relief of seamounts and seamount groups, although capable of localizing rupture initiation (Bilek et al., 2003), tend to significantly modify or arrest rupture continuation (Kodaira et al., 2000; Mochizuki et al., 2008; Bilek, 2010; Singh et al., 2011; Wang and Bilek, 2011; Trehu et al., 2012; El Hariri et al., 2013; Wang and Bilek, 2014), as commonly do subducting ridges (Franke et al., 2008; Sparkes et al., 2010; von Huene et al., 2012; Kopp, 2013) and also some upper plate structures (Bejar-Pizarro et al., 2013).

With respect to the posited rupture-promoting effect of subducted sediment, Ruff (1989) lamented that, “the statistical correlation between excess sediments and great earthquake occurrence is less than compelling.” His inference that a relation did exist was drawn from a population of 19 instrumentally recorded earthquakes and a then-available but incomplete and largely inaccurate tabulation of the thickness of sediment entering SZs. After 1989 and through January 2013, an additional 16 great (≥Mw8.0), five giant (≥Mw8.5), and two super giant (≥Mw9.0) IPT earthquakes broke at 14 different trench sectors (Tables 1–6). These more recent earthquakes include the re-seized 1946 Unimak, Alaska, or Scotch Cap megathrust (Lopez and Okal, 2006). Also since 1989, accurate tables and maps of the global distribution of trench-sediment thickness have become available (von Huene and Scholl, 1991; Scholl and von Huene, 2007; Heuret et al., 2012).

To test the Ruff conjecture concerning the rupture-promoting effect of sediment thickness alone (we did not evaluate a strengthening effect, a notion that has been challenged by Wang and Bilek [2014]), we compiled improved and larger data sets of trench-axis sediment thickness and vetted instrumental (1899 through January 2013) era IPT earthquakes of magnitude ≥Mw7.5 (see Table 5) and pre-instrumental-era (1700–1898) earthquakes of estimated magnitude ≥Mw8.0 (see Table 6). Descriptively simplified catalogs of instrumental-era IPT earthquakes of magnitude ≥Mw7.5 (n = 176) and pre-instrumental megathrusts of magnitude ≥Mw8.0 (n = 12) are listed on Tables 1–4. In these tables, IPT earthquakes are linked to an occurrence area along a sector of trench distinguished by a fluctuating but average sediment fill that can be characterized as either thick or thin. Along thick sectors the average fill is >1.0 km, whereas that for thin sectors is <1.0 km. Trench sectors are shown on Figure 1. They are defined as a sector of trench axis long enough (i.e., ≥250–300 km) to support a potential high-magnitude rupture between observed rupture barriers and containing either a thin- or thick-sediment fill.

Because our purpose was to test for a definitive association of sediment thickness versus occurrence of high-magnitude megathrust earthquakes—either statistically strong, weak, or not there—we only used the two thickness bins noted above and did not divide sectors further into categories of very thin (>0.5 km) or very thick trench sequences (>2.0 km). Nonetheless, as described by Heuret et al. (2012), useful information and insights can be gained by analyzing with smaller thickness increments as well as including other physical parameters to assess the propensity of a SZ sector to break in a great megathrust earthquake.

Based on the instrumental data listed in Tables 1, 2, and 3 and the plots of Figures 4–13, a statistically supported statement (see Summary and Conclusions) can be made that large IPT earthquakes are most prone to nucleate at well-sedimented SZs. For example, despite the 7500 km shorter global length of thick-sediment trenches (thin = ∼21,500 versus thick = ∼14,000 km), at thick-sediment trenches occurred ∼53% of all earthquakes of magnitude ≥Mw8.0, ∼75% of all ruptures ≥Mw8.5, and 100% of all IPTs ≥Mw9.1 (Fig. 9; see also Heuret, 2012). When the number of thin-sediment earthquakes in the magnitude range of Mw7.5–8.4 are normalized (reduced) to compensate for the disproportionately longer occurrence length of thin- versus thick-sediment trench sectors in this magnitude range (19,000 versus 7800 km, respectively) ∼73% of all earthquakes ≥Mw8.0, ∼75% of all earthquakes ≥Mw8.5, and 100% of all earthquakes ≥Mw9.1 occurred at thick-sediment trenches.

It is important to emphasize that great IPT earthquakes also ruptured adjacent to thin-sediment trench sectors, a circumstance noted by Ruff (1989) and Heuret et al. (2012). This observation also includes the 2014 north Chile Mw8.2 that is not listed on Tables 1–4. Three recorded giant earthquakes—1952 Kamchatka Mw9.0, 1963 Kuril Mw8.5, and 2011 Japan (Tohoku-Oki) Mw9.0 (Table 1)—are each associated with the subduction of bathymetrically smooth or low-relief oceanic crust and where basal subduction erosion of the upper plate is prominent (von Huene et al., 1994; Scholl et al., 2011; Kopp, 2013; Wang and Bilek, 2014). The combination of underthrusting low-relief seafloor and subduction erosion evidently also works to produce a smooth trench-parallel distribution of interplate coupling, the condition posited by Ruff (1989) that promotes the lengthy (>250–300 km), along-trench rupturing characteristic of high-magnitude megathrust earthquakes. For the Tohoku-Oki Mw9.0, a stratigraphically focused surface of interplate slip contributed importantly to a near-trench displacement of 30–60 m and the launching of the horrendous Tohoku tsunami (Chester et al., 2013; Moore et al., 2013; Ujiie et al., 2013; Nakamura et al., 2013; Moore et al., 2015).

As noted above, large megathrust ruptures commonly initiate, terminate, or are significantly modulated at subducted high-relief bathymetric structures or elements (Kodaira et al., 2000; Bilek et al., 2003; Collot et al., 2004; Franke et al., 2008; Mochizuki et al., 2008; Bilek, 2010; Sparkes et al., 2010; Singh et al., 2011; Wang and Bilek, 2011; Trehu et al., 2012; von Huene et al., 2012; El Hariri et al., 2013; Wang and Bilek, 2014). The global length of ocean-margin SZs listed on Table 3 (∼35,500 km) at which megathrust earthquakes ≥Mw7.5 have occurred was divided into 48 sectors identified as a physically bordered or seismically identified trench segment ≥250–300 km in length containing a mostly axially continuous thin (<1.0 km) or thick (>1.0 km) sediment fill (Figs. 1 and 2). Heuret et al. (2012; Fig. 1) used a seismic-behavior (delimiting) sectoring approach identify 44 sectors that are similar to those shown on Figure 1.

Sectors listed on Tables 1–4 and shown on Figures 1 and 3 identify trench segments that display seismically observed rupture lengths that start and end at: (1) a subducted bathymetric element of large dimensions (e.g., an aseismic ridge, oceanic plateau, seamount group or chain, or prominent fracture zone); (2) a plate boundary, e.g., South Chile Triple Junction separating the Nazca, Antarctic, and South America plates; and (3) an abrupt change in the regional strike of the SZ, e.g., southern Peru to northern Chile, western Aleutian to Kamchatka, and Kuril to Japan (Figs. 1 and 3).

Figure 3 displays our sectoring of the ∼7000-km-long SZ bordering western South America (see also fig. 2 of Heuret et al., 2012). All sectors, thin or thick, are delimited (terminated) by bathymetric elements or a prominent change in strike of the SZ. For example, the ∼1500-km-long, sediment-flooded central section or segment of the Chile Trench is bordered to the north and south, respectively, by the underthrusting Juan Fernandez Ridge (∼33°S) and the triple junction of the South Chile Rise (∼46°S). This lengthy segment characteristically ruptures north and south of the deeply (∼2.5 km) sediment-buried Mocha-Valdivia complex of underthrusting fracture zones (FZs) (Melnick et at., 2006; Bilek, 2010; Sparkes et al., 2010). Recognizing this observation, the central segment of the Chile Trench was divided into separate north and south sectors (Figs. 1 and 3; Tables 2 and 3). Rupture termination at the subducted segment of the Mocha FZ is linked by Moreno et al. (2014) to the upward injection of fluids from the FZ into the overlying seismogenic zone.

Similarly, the laterally continuous, sediment-charged Aleutian SZ tends to break in great megathrusts along a western sector and a central sector bordered on its eastern side by the Amlia fracture zone (Fig. 1), also considered to be a subsurface source of fluid injection (Singer et al., 1996, 2007). An eastern Aleutian or Fox Island sector extending from the Amlia FZ eastward to the Alaska SZ at Unimak Pass is not known to have broken in a great IPT earthquake. But the potential to do so seems high (Butler, 2012; Ryan et al., 2012) (Fig. 1; Table 2).

Estimates of trench-floor sediment thickness were extracted from a diverse and global set of published and unpublished seismic-reflection records largely tabulated by von Huene and Scholl (1991), Scholl and von Huene (2007), and most recently by Heuret et al. (2012). Numbered, alphabetically named sectors are included on Tables 1, 2, and 3 and shown on Figures 1 and 3. The thickness recorded on all tables was estimated below the inner or landward side of the trench floor where the underlying ocean crust enters the SZ. Trench deposits are thickest here either because of the accumulation of debris shed by mass wasting from the landward trench slope (Strasser et al., 2013; see Fig. 2B) or where a landward-thickening, wedge-shaped body of trench-axis turbidite deposits is thickest (see Figs. 2A, 2C, and 12).

Our purpose in investigating the global distribution of the magnitude of megathrust earthquakes is to recognize possible empirical trends in their occurrence settings and also the source regions for large-magnitude earthquakes that launch ocean-crossing tsunamis. We restricted this compilation to megathrust or IPT earthquakes because they are most closely related to convergent plate motions and are the most potent generator of tsunamigenic events (see Polet and Kanamori, 2009). Each era of megathrust observations, i.e., the instrumental era (1899 to the present), and the historical or pre-instrumental period (1700–1898), presents challenges to identifying scalar moments, moment magnitudes, and focal mechanism of coastal and submerged forearc earthquakes. Doing so becomes more and more difficult going back in time. Thus our ability to identify older, large IPT earthquakes also becomes increasingly difficult with distant time. Accordingly, the target magnitude thresholds for earthquakes selected for consideration vary with the era. For the instrumental era, the target threshold was set at Mw ≥7.5 (Tables 5), and Mw ≥8.0 for pre-instrumental earthquakes (Table 6).

Data sources for the earthquakes listed on Tables 5 and 6 and also the derivative and descriptively reduced Tables 1–4, are described below:

• (1) U.S. Geological Survey (USGS) Centennial Catalogue 1900–2007 (Engdahl and Villasenor, 2002): http://www.globalquakemodel.org/what/seismic-hazard/instrumental-catalogue/.

  This catalogue is based on the methodology of Engdahl et al. (1998). We note the few exceptions in using this data source. Literature values for those earthquakes not in these data sources are noted in Tables 5 and 6.

• (2) Seismic moment tensors, scalar moments, and moment magnitudes in the digital era (1978 to present): The Harvard Centroid Moment Tensor (CMT) Project and its successor, the Global CMT Project: http://www.globalcmt.org/.

• (3) Seismic moment tensors, first-motion focal mechanisms, scalar seismic moments, moment magnitudes and their magnitude proxies in the pre-digital instrumental era (1899–1977): Cited literature values of individual earthquakes in Tables 5 and 6 as noted. Especially valuable were the earthquake magnitude summaries by Engdahl and Villasensor (2002) and reappraisals by W.H.K. Lee (http://www.globalquakemodel.org/what/seismic-hazard/instrumental-catalogue/) and E.A. Okal (personal commun., 2012). Not all presumed subduction zone IPT earthquakes have published focal mechanisms; hence, some of the smaller intraplate and intraslab earthquakes in our IPT catalogue may have been erroneously included.

• (4) September, 1899, Yakutat earthquakes: Published epicenter (Doser, 2006) and seismic scalar moment (Plafker and Thatcher, 2008).

• (5) Tsunami Data Sources in Tables 5 and 6: National Geophysical Data Center (NGDC) Global Historical Tsunami Earthquakes and Runups and references therein: http://www.ngdc.noaa.gov/hazard/tsu_db.shtml (Accessed 2009–2012).

• (6) Tsunami Laboratory, Novosibirsk, Russia: Historical Tsunami Database for the World Ocean based largely on the work of S.L. Soloviev (1930–1994): http://tsun.sscc.ru/On_line_Cat.htm.

• (7) Tsunami deposits: NGDC Tsunami Deposits and Proxies: http://www.ngdc.noaa.gov/hazard/tsudep.shtml.

• (1) Epicenters and magnitude estimates in the pre-instrumental era are based largely on seismic intensity estimates and tsunami effects from: USGS National Earthquake Information Center: Historic World Earthquakes (Instrumental and Pre-Instrumental): http://earthquake.usgs.gov/earthquakes/world/historical.php; National Geophysical Data Center (NGDC) Significant Earthquake Database: http://www.ngdc.noaa.gov/nndc/struts/form?t = 101650&s = 1&d = 1; and Utsu, Tokuji, Catalogue of Damaging Earthquakes in the World, 2002 Online: http://iisee.kenken.go.jp/utsu/index_eng.html.

For the instrumental era, earthquake epicenters were required to be located in forearc regions landward of trenches and have shallow epicentral depths (<60 km). For those earthquakes having first-motion constraints or moment tensor data available, IPT earthquakes should have focal mechanisms with nodal planes approximately parallel to the local azimuth of the trench and with one nodal plane dipping shallowly toward the volcanic arc. For earthquakes prior to the advent of global standard seismographic networks in the 1960s, these criteria are difficult to evaluate except for earthquakes that have been the subjects of individual special study.

Ambiguity exists in identifying offshore IPT earthquakes in subduction zones because, in the instrumental era, great tsunamigenic off-trench, outer-rise and/or outer-trench slope earthquakes have occurred that are only indirectly related to SZ earthquakes. Outer-rise and/or outer-trench earthquakes can produce destructive onshore strong ground motions and potentially destructive tsunami waves. Six off-trench earthquakes with Mw >8 are known to have occurred in the instrumental era. Among these, four are known to be normal-faulting earthquakes associated with plate bending seaward of trenches: 1933 Japan Mw8.6; 1977 eastern Indonesia (Sumbawa) Mw8.3; 2007 Kuril Mw8.1 and 2009 Tonga Mw8.1.

Moreover, a number of large, shallow offshore intraslab earthquakes have occurred since the 1960s—for example, the 31 May 1970 Peru Mw7.9 and the 13 January 2001 El Salvador Mw7.7. Finally, some large off-trench intraplate earthquakes have strike-slip mechanisms and are not potent tsunami sources. Examples include a pair of earthquakes in the Gulf of Alaska (Mw7.9 and 7.8) that occurred in 1987 and 1988, and two intraplate oceanic strike-slip earthquakes that occurred off the Sumatra Trench in 2012. None of these strike-slip earthquakes produced destructive tsunamis, but they remind us that this class of offshore subduction-zone earthquakes may be confused with IPT earthquakes when accurate epicenter information and focal mechanism constraints are lacking.

Large uncertainties in epicenter location are especially problematic in distinguishing off-trench earthquakes before ∼1930 when observatory timing, generally lacking radio time checks, was often poor and the number of stations and their azimuthal distribution were limited (see discussion by Okal et al., 2004). This problem of potential misidentification of large off-trench intraplate and intraslab earthquakes as IPT earthquakes is somewhat diminished by the fact that since 1930 these earthquakes are far less frequent in the instrumental era than confirmed IPT earthquakes ≥Mw7.5 (22 versus 176 in Tables 1 and 2 combined) and appear to have smaller maximum magnitudes (Mw8.6 versus 9.5). We assume that this lower rate of occurrence also applies to earthquakes before 1930. Undoubtedly, some undisclosed intraplate and intraslab earthquakes still exist in our catalogue, especially those that occurred before the 1960s and during World War II when station distribution was sparse, absolute timing of phase data was often poor, and earthquake first motions were difficult to interpret.

Because of the scarcity of appropriate seismic instruments, constraints on the magnitudes of earthquakes were not practical before ∼1899. Nonetheless great and giant earthquakes have been identified in the literature based on the criteria (A–F) listed below related to seismic moment release. The lateral extent of trench and/or near-shoreline continuity of these criteria during great and/or giant earthquakes in the instrumental era has shown that seismic moments of these pre-instrumental, very large IPT earthquakes approximately scale with earthquake rupture length. The lateral extent of these criteria as well as the “sizes” of these observables are thus useful in estimating the relative sizes of huge, pre-instrumental tsunamigenic IPT earthquakes.

Criteria A. Geologic investigations of near-shore features sensitive to earthquake-related subsidence or uplift, such as ancient “ghost” forests killed by submergence and buried marsh deposits overlain by muddy tidal deposits. Substantial subsidence inferred from these deposits and their large lateral distributions are strong indicators of prior occurrence of giant tsunamigenic IPT earthquakes. A prime example is the 26 January 1700 earthquake in Cascadia (Atwater et al., 2005) where coastal drowning evidence has been identified over a stretch of coastline more than 1000 km long. Uplifted and tilted corals have been identified offshore of Sumatra that have also been linked to seafloor motions associated with big IPT earthquakes in 1797 and 1833 (Natawidjaja et al., 2006).

Criteria B. Geologic investigations of tsunami deposits brought inland by the flow field of tsunami waves, such as “tsunami sand” deposits, inland coral and other rock boulders, and high driftwood strand lines above high tides and storm surges (Okal et al., 2003,). These deposits have been identified along tsunami-receiving shorelines in many coastal regions around the world. A tsunami-inundation interpretation of some of these deposits is controversial in some areas (e.g., Keating et al., 2011).

Criteria C. Historical eyewitness reports of tsunami inundation limits referenced to surviving cultural features and to contemporaneous positions of mean high tide. Prime examples are the recorded tsunami arrival in Japan of the 1700 tsunami source in Cascadia (Atwater et al., 2005), the giant Nankaido earthquake of 28 October 1707 (Hatori, 1981), and the eyewitness reports of the 1946 Alaska Peninsula Scotch Cap tsunami in the far field (Okal et al., 2002).

Criteria D. Historical reports of very large (>10 m), widespread near-field tsunami multi-meter run-ups and/or damaging large (>1 m) tsunami waves in the far field. The latter may be lacking if tsunami wave-field directivity is not expected to impact populated, developed shorelines, or where written records are lacking.

Criteria E. Reports of widespread near-field destruction of harbors, ports, or coastal towns along embayments and widely dispersed data in the far field.

Criteria F. Intense, widespread high seismic intensities based on the reported effects of earthquake ground shaking on humans, buildings and other man-made structures, and liquefaction.

The above criteria are incapable of enabling assessments of truly quantitative magnitudes, particularly for an IPT earthquake below an estimated magnitude of about Mw8.0. Our goal in applying them was to identify among these earthquakes that had the potential of producing ocean-crossing tsunami waves—a signature of a potential high-magnitude IPT earthquake. These earthquakes in the instrumental era are largely limited to ≥Mw∼8.5. We accordingly restricted our attention to pre-instrumental earthquakes that are comparable with the effects of ≥Mw∼8.5 IPT earthquakes in the instrumental period, for which we have more information. Based on these criteria, we identified 12 pre-instrumental earthquakes during the period 1700–1898 (Tables 4 and 6) of magnitude ≥Mw8.0. Undoubtedly, this list is incomplete owing to lack of historical data for some localities, particularly in sparsely populated and developed regions where little or no written records exist or where geological investigations of the region’s prehistory have not been attempted.

Tables 1 and 2 list the 176 instrumental era (1898 to January 2013) ≥Mw7.5 megathrust or IPT earthquakes that, respectively, ruptured at thin- (n = 115) and thick- (n = 61) sediment trenches. Table 3 tabulates just the highest magnitude megathrusts that occurred along 48 defined trench sectors, 21 of which are thick-sediment and 27 of which are thin-sediment sectors. Pre-instrumental earthquakes (n = 12) are listed on Table 4.

Figure 4 shows in separate columns the percent of megathrust earthquakes in binned magnitude ranges that ruptured separately at thin- and thick-sediment trenches (Tables 1 and 2, respectively). Earthquakes are binned into five, 0.5-Mw-wide magnitude groups: (1) Mw7.5–7.9; (2) Mw8.0–8.4; (3) Mw8.5–8.9; (4) Mw9.0–9.4; and (5) Mw9.5–9.9.

The stacked columns of Figure 5 show the relative percent of the combined binning of all megathrust earthquakes (n = 176) that occurred at thin- and thick-sediment trench sectors. For thick-sediment trenches, the relative occurrence percentages for increasing magnitude bins 1–5 are (1) 29%, (2) 41%, (3) 86%, (4) 50%, and (5) 100%. For thin-sediment trenches, the corresponding relative occurrence percentages are (1) 71%, (2) 59%, (3) 14%, (4) 50%, and (5) 0%.

The global length of thin-sediment trenches (∼21,500 km) at which IPT earthquakes ≥Mw7.5 occurred is ∼1.5 times longer (by ∼7500 km) than for thick-sediment trenches (∼14,000 km). Accordingly, thinly sedimented trenches would be expected to record a proportionally higher number of ≥Mw7.5 earthquakes. Consideration of this circumstance is in particular important for the earthquakes in bins Mw7.5–8.4 that record a much higher number of earthquakes (92 of a total of 115 for all magnitudes) at thin-sediment trenches than at thick-sediment trenches (38 of a total of 61) (Fig. 5). For these bins, the aggregate length of thin-sediment trenches is ∼2.5 times that of thick-sediment trenches (∼19,000 versus 7800 km).

To approximately compensate (i.e., normalize) for the greater length of thin-sediment trenches, the 112 earthquakes occurring in the Mw7.5–8.4 bins were reduced to 46 or by a factor of 41% (7800 km/19,000 km × 100 = 41%). Figure 6 displays the binned, length-compensated relative percent of megathrust earthquakes occurring at thick- and thin-sediment trenches. For thick-sediment trenches, the binned occurrence percentages are (1) 50%, (2) 63%, (3) 86%, (4) 50%, and (5) 100%. For thin-sediment trenches, the occurrence percentages are, accordingly (1) 50%, (2) 37%, (3) 14%, (4) 50%, and (5) 0%.

Ruff’s conjecture that subduction of an excess thickness of trench-floor sediment is associated with great SZ earthquakes implies that the highest magnitude earthquakes should also cluster at well-sedimented trenches. Binned into 0.5-Mw-wide columns, Figure 7 shows the occurrence percent of the highest magnitude IPT earthquake that ruptured adjacent to each of the 48 sediment-thickness trench sectors, of which 21 are thick-sediment and 27 thin-sediment sectors (Table 3; Figs. 1 and 3). As above, the binned columns are for earthquakes of (1) Mw7.5–7.9, (2) Mw8.0–8.4, (3) Mw8.5–8.9, (4) Mw9.0–9.4, and (5) Mw9.5–9.9. For thick-sediment trenches, occurrence percentages are, respectively, (1) 29%, (2) 42%, (3) 86%, (4) 50%, and (5) 100%. For thin-sediment trenches, the corresponding occurrence percentages are (1) 71%, (2) 58%, (3) 14%, (4) 50%, and (5) 0%.

Figure 8 displays the separate plots of the occurrence percent of megathrust earthquakes that, separately, broke at or above a threshold magnitude at thick- (n = 61 earthquakes) and thin-sediment (n = 115) trench sectors, respectively.

Figure 9 displays the relative percent of all (n = 176) IPT earthquakes that ruptured at a magnitude equal to or greater than a threshold magnitude at thick- and thin-sediment trench sectors (Tables 1 and 2). The data plotted are not compensation for the disproportionately greater length of thin- versus thick-sediment trenches.

Figure 10 presents the same data plotted on Figure 9, but for Mw7.5–8.4 IPT earthquakes, the total number of earthquakes (n = 112) at thin-sediment sectors reduced to 41% (to n = 46) to compensate within this magnitude range for the significantly greater occurrence length of thin- versus thick-sediment trench sectors (∼19,000 versus 7800 km).

Figure 11 plots the occurrence percent of just the highest magnitude earthquake that ruptured along each of the 48 sediment-thickness trench sectors (Table 3; Figs. 1 and 3). No length compensation was applied to this relatively small population of megathrust earthquakes.

Figures 12 and 13 display, respectively, binned and continuous threshold plots of megathrust earthquakes ≥Mw8.0 that ruptured during the pre-instrumental recoding era (1700 through 1898). Criteria for vetting pre-instrumental era magnitude and rupture mechanism are explained in the Pre-Instrumental Era Earthquakes, 1700 to 1898 section.

In the previous sections, tabulated and graphic trends indicate that high-magnitude IPT earthquakes preferentially occur along thick-sediment trenches. These trends are further analyzed with statistical tests using the available instrumental-era data. The thick- and thin-sediment trenches defined in the Introduction section are mutually exclusive categories and amenable to the binomial test, an exact test appropriate for small samples (Conover, 1971). Under the definitions for thick- and thin-sediment trenches, the null hypothesis is that it is equally likely that high-magnitude IPT earthquakes occur in either of the trench categories. The alternative Ruff hypothesis, which is considered in this study, is that high-magnitude IPT events occur preferentially at thick sediment trenches. The p-value associated with the binomial test is the probability of obtaining any outcome as or more extreme than the one observed, if the null hypothesis were true. “Extreme” in this definition means in the direction of the alternative Ruff hypothesis. The p-value is calculated using the binomial distribution, with the total number of events (N_tot), the number of “successes” (i.e., events in thick-sediment trenches) (N_thick), and the assumed probability of success under the null hypothesis (R₀) as parameters to the distribution. From the outset, we chose a critical p-value of 0.05, below which we accept the alternative hypothesis.

For the first series of tests, the probability of success under the null hypothesis is provided by the global lengths of the two categories of trenches: 21,500 km for thin-sediment trenches versus 14,000 km for thick-sediment trenches, resulting in a proportion of thick-sediment trenches of R₀ = 0.39. Shown in Table 7 are the results of the binomial test for different cutoff magnitudes. Cumulative earthquake numbers are used, rather than the binned counts, because of the higher sample sizes and to avoid any artifacts associated with binning. Using the definition of the alternative hypothesis above (high-magnitude IPT earthquakes preferentially occurring along thick-sediment trenches), one-sided p-values are calculated (i.e., using only one side of the distribution). A more conservatively formed alternative hypothesis is that high-magnitude IPT events occur preferentially either in thin- or thick-sediment trenches, such that two-sided p-values would be calculated. Although this alternative hypothesis is not consistent with the physical model for these earthquakes described in the introduction, two-sided p-values are reported for completeness in Table 7.

For example, out of the 39 IPT earthquakes ≥Mw8.1 that occurred in the instrumental era, 21 occurred at thick-sediment trenches (fourth line, Table 7). In contrast, 15.2 events are expected under the null hypothesis using the global lengths of each sediment thickness category. The corresponding one-sided p-value is 0.043, meaning that there is a 4.3% probability of having 21 or more events if the null hypothesis were true. Because this is below our critical p-value of 0.05, we accept the alternative Ruff hypothesis that M_w ≥8.1 IPT earthquakes preferentially occur at thick-sediment subduction zones. The two-sided p-value is 0.070, meaning that there is a 7.0% probability of having 21 or more events or nine or fewer events (i.e., 30 or more IPT earthquakes along thin-sediment trenches) occurring under the null hypothesis. In this case, the null hypothesis cannot be rejected. Overall, the alternative hypothesis can be accepted for ≥Mw8.1, 8.5, and 8.7 using the one-sided p-values. For earthquakes ≥Mw8.9 and greater magnitude cutoffs, the test results are increasingly affected by sample size.

For the second series of test, the probability of success under the null hypothesis is given by the cumulative lengths of trench sectors where Mw7.4–8.5 events have occurred in the instrumental era: 19,000 km for thin-sediment trenches versus 7800 km for thick-sediment trenches, resulting in a thick-sediment proportion of R₀ = 0.29 (Table 8). As before, both one- and two-sided p-values are reported, although we only use the one-sided values for testing. Under this null hypothesis, the alternative, Ruff hypothesis that IPT earthquakes occur along thick-sediment trenches is accepted for events ≥Mw7.7–8.7, inclusively. The null hypothesis cannot be falsified for earthquakes ≥Mw7.5 or for ≥Mw8.9, the latter most likely because the sample size is too small. In summary, statistical analysis using the binomial test supports the conclusion that, for the most part, large-magnitude IPT earthquakes preferentially occur at thick-sediment trenches.

The tabulated, plotted, and statistically analyzed data presented in Tables 1, 2, and 3 were compiled to test with modern data sets the validity of the posit by Ruff (1989) that excess sediment entering a subduction zone (SZ) favors nucleation of higher-magnitude megathrust or IPT (interplate thrust) earthquakes. This conjecture reasons that the entrance of an “excess” thickness (i.e., thick enough to form a frontal prism by sediment accretion) of sediment into the SZ works to strengthen and smooth the lateral distribution of interplate coupling—thus favoring the continuation of along-trench rupturing and the generation of high-Mw earthquakes.

At the time of his surmise, the notion of sediment subduction was little recognized as a significant tectonic process or that a subduction channel filled with sediment and tectonically eroded debris separated the underthrusting oceanic plate from the rock framework of the overlying convergent margin (Scholl et al., 1980; Cloos and Shreve 1988a, 1988b; von Huene and Scholl, 1991). Since the early 1980s, geophysical, drilling, and geologic investigations of the submerged forearc documented that at most (75–80%) convergent margins over a long period of time (i.e., 5–10 Myr) the greater part (75% or higher) of the incoming sediment enters the subduction channel and continues landward beneath the rock framework of the convergent margin (von Huene and Scholl, 1991; Clift and Vannucchi, 2004; Scholl and von Huene, 2007).

This process of sediment subduction is shown on Figure 2C for the Ecuadorian SZ. Here the thickness of sediment entering the subduction channel below the base of the overlying plate is ∼1 km. Farther to the south (33–45°S), where the super giant 1960 Mw9.5 megathrust ruptured the south-central Chile SZ (Table 2; Fig. 3), trench sediment is thicker (∼2.5 km) and more laterally continuous (∼750 km). Beneath the margin, the estimated long-term (∼5 Myr) volume rate of sediment subduction is ∼150 km³/Myr/km of convergent margin (Scholl and von Huene, 2007; Table 2). At an orthogonal convergent rate of ∼75 km/Myr, transport of this volume requires an average thickness of sediment in the subduction channel of ∼2.0 km (see fig. 1 of Melnick et al., 2006).

Farther south still (50–57°S), along the southern sector of the Chile Trench (Fig. 3), the thickness of sediment entering the subduction channel is ∼3 km (Polonia et al., 2007). The trench fill is thick enough to completely bury 2–3-km-high seamounts entering the SZ (Fig. 14). Burial of this relief effects “smoothing” of interplate roughness that Ruff (1989) argues promotes rupture continuation and the generation of megathrusts of high magnitude. Arguing in principal similarly, Wang and Bilek (2014) observe that underthrust relief favors fault creep and either rupture termination or significant changes in the rupture process.

The plotted data of Figures 5–7 and 9–11 also document that great and giant IPT earthquakes nucleate in absence of a thick layer of subducting sediment. Accepting the statistically supported correctness of the Ruff conjecture, this observation implies that lateral coupling smoothness favoring rupture continuation can also be effected where seafloor of low average bathymetric relief underthrust the margin and where the added tectonic debris of basal and frontal subduction erosion significantly thickens the subduction channel (Figs. 15 and 16; von Huene et al., 1994; Kopp, 2013).

The plots of Figures 4 and 8 show that for both thin- and thick-sediment trenches, most IPT earthquakes ≥Mw7.5 ruptured at magnitudes less than Mw7.7–7.9. Both plots document an expected progressive decrease in number of earthquakes occurring at increasingly higher magnitudes. Of the 61 earthquakes that nucleated at thick-sediment and 115 at thin-sediment SZs, the occurrence number and percent at and above any threshold magnitude are higher for sediment-rich trenches than for sediment-poor trenches. Interplate thrust magnitudes exceeding Mw9.0 are unknown for sediment-thin sectors, whereas three super giant earthquakes (1960 Chile Mw9.5, 1964 Alaska Mw9.2, and 2004 Sumatra Mw9.2) nucleated at thickly sedimented SZs (Tables 2 and 3).

It seems significant, as shown by the separate thin- and thick-sediment plots of Figures 4 and 8, and also the combined or relative occurrence plots for all recorded earthquakes (n = 176) shown on Figures 9–11, that ruptures >Mw9.0 have only been recorded at well-sedimented SZs (see also Heuret et al., 2012). For higher-magnitude earthquakes, in particular for ≥Mw8.3, the number of earthquakes is small (n = 20), and the significance of the statistics of small numbers becomes a statistical concern (see Summary and Conclusions). This circumstance, about which nothing can be done, is true for IPT earthquakes ≥Mw8.3, occurring at both thick (n = 11) and thin (n = 9) trenches (Fig. 9; Tables 1 and 2).

The combined or relative occurrence plots of earthquakes binned on Figures 5–7, and as continuous threshold diagrams on Figures 9–11, display a pronounced overall trend of increasing relative occurrence of higher-magnitude earthquakes at thick-sediment trenches. Correspondingly, at thin-sediment trenches, the overall relative number of earthquakes decreases with increasing magnitude. These prominent trends are equally apparent on threshold occurrence plots of just the highest magnitude earthquakes at thin- (n = 27) and thick- (n = 21) sediment trenches (Figs. 7 and 11).

Although few in number (n = 12), occurrence-magnitude plots of pre-instrumental IPT earthquakes also reveal a trend of increasing number of higher-magnitude earthquakes rupturing at thick-sediment trench sectors (Figs. 12 and 13).

At thin-sediment SZs, the overall trend of decreasing relative earthquake occurrence reverses at ∼Mw9.0. Trend reversal to a peak of 50% is displayed by the binned data of Figures 5–7. At the continuous threshold plots of Figures 9–11, the occurrence peak (40%) is at ≥Mw8.9, reflecting the circumstance that data points are plotted at odd Mw magnitudes. Because few earthquakes are involved, the statistically less robust plot of pre-instrumental IPT earthquakes on Figure 13 nonetheless displays the same reversal trend of decreasing relative occurrence for higher Mw earthquakes (Table 4). The reversal in trend is predominantly owing to the nucleation of giant IPT earthquakes off northern Chile and northern Peru—for example, in the instrumental era, the recent 2014, Iquique Chile Mw8.2 (not included in listed earthquakes on Table 1) (Schurr et al., 2014).

For instrumental-era IPT earthquakes, the prominent trend reversal at thin-sediment trenches reflects the nucleation of the 1952 Kamchatka Mw9.0 and, in 2011, the devastating Mw9.0 Tohoku-Oki megathrust off northern Japan (Table 1). Other, but lower magnitude great megathrusts (i.e., Mw8.2–8.5) ruptured the thin-sediment south Peru, northern Chile, and northern and southern Kuril SZs. Along these trench sectors, but in particular for the 1952 Kamchatka and 2011 Japan earthquakes, lengthy (>300–500 km) widths of the underthrusting seafloor are bathymetrically smooth, exhibiting only widely spaced subducting seamount groups, fracture zones, and ridges. Figures 15 and 16 illustrate the general smoothness of Pacific plate entering the southern Kamchatka and northern Japan SZs (see also Wang and Bilek, 2014). We infer that asperity smoothing is further enhanced at these SZs because basal subduction erosion adds a landward thickening volume of eroded debris to the subduction channel (Klaeschen et al., 1994; von Huene et al., 1994; von Huene and Ranero, 2003; Heuret et al., 2012). For northern Japan, the surface of interplate slip beneath the frontal prism was stratigraphically localized, a circumstance that conditioned near-trench rupture continuation and tsunamigenesis (Chester et al., 2013; Moore et al., 2013; Nakamura et al., 2013; Moore et al., 2015) (see Summary and Conclusions below).

For instrumental IPT earthquakes ≥w7.5, the plots of Figures 4–11 of megathrust size versus sediment thickness document a nearly continuous greater relative occurrence of higher-magnitude earthquakes at thick- (≥1 km) versus thin- (≤1 km) sediment sectors. Although much less numerically based, the same overall trend is observed in the plots of pre-instrumental era earthquakes on Figures 12 and 13. For all instrumental era earthquakes (n = 176) uncompensated for the disproportionally longer global length of thin-sediment trenches, the greater relative occurrence of higher-magnitude megathrusts at thick-sediment trenches crosses over from less to greater than 50% at ∼Mw8.0 (Fig. 9). For just the highest magnitude earthquakes recorded along thin- and thick-sediment trenches (n = 48), the crossover to greater than 50% earthquake occurrence at sediment-thick sectors is similar at ∼Mw7.8 (Fig. 11). This general observation was noted by Ruff (1989).

For the length-compensated number of earthquakes occurring in the magnitude range of Mw7.5–8.4, the greater relative occurrence of higher-magnitude earthquakes at thick-sediment trenches begins at less than Mw7.5. Except for a reversing dip at ≥Mw8.9, the relative higher occurrence at thick-sediment trenches is effectively continuous to IPT earthquakes greater than Mw9.0 (Fig. 10). Equally prominent is a reversal in this trend at approximately Mw8.9–9.0. The reversal is shown by the binned data of Figures 5–7 and the continuous threshold plots of Figures 9–11). The turnaround is virtually entirely due to the 1952 southern Kamchatka and 2011 northern Japan or Tohoku-Oki Mw9.0 IPT earthquakes. Both rupture areas are underthrust by seafloor of low relief and a landward thickening, roughness-smoothing subduction channel expanded by inclusion of tectonically eroded debris (Figs. 15 and 16; Klaeschen et al., 1994; von Huene et al., 1994; Hueret et al., 2012).

Equally significant for the tsunami-launching consequence of the Tohoku-Oki earthquake are the stratigraphic findings at International Ocean Discovery Program (IODP) Site C0019, which penetrated the frontal prism (Chester et al., 2013; Moore et al., 2013) and DSDP Site 436 located nearby but seaward of the Japan Trench (Nakamura et al., 2013). These drilling sites determined that a ∼5-m-thick layer of pelagic, abyssal clay overlying chert stratigraphically localized the surface of displacement beneath the frontal prism and conditioned the 30–60 m of slip that launched the disastrous Tohoku-Oki tsunami. It is unknown if the same stratigraphically controlled surface extends deeper and landward beneath the submerged forearc, but it certainly could. If nothing else, with respect to generating great megathrust earthquakes and high-magnitude near- and far-field tsunamis, the Tohoku-Oki drilling results demonstrate the contributing importance of the stratigraphy and lithologic makeup of subducted sediment where laterally extensive seafloor of low average roughness enters the subduction zone.

As noted, the predominant occurrence of great megathrust earthquakes at sediment-charged SZs statistically supports the correctness of Ruff’s (1989) surmise. But, accordingly, it can also be inferred that a condition of homogeneous coupling can be achieved at poorly sedimented SZs that are erosive and underthrust by low-relief ocean crust. If charged stratigraphically with a weak stratigraphic horizon, the potential up-dip slip beneath the lower trench slope can be dangerously high. These observations also underwrite the validity of the argument of Wang and Bilek (2014) that lateral coupling evenness is disrupted by underthrusting bathymetric relief, a circumstance favoring interplate creep and rupture termination rather than continuation.

The megathrust occurrence data of Tables 1 and 2 only list instrumental earthquakes of magnitude Mw7.5 and greater that ruptured at the numbered trench sectors shown on Figure 1. Many other thin- and thick-trench sectors exist at which no earthquake ≥Mw7.5 has been instrumentally recorded (see Fig. 3 and its caption). For thick-sediment trench sectors 250–300 km in length that are not underthrust by rupture-inhibiting ridges, plateaus, or seamount chains, a concern must be stated that a future IPT of magnitude Mw8.0 or much higher is plausible (Table 2). For example, the Makran SZ of southern Pakistan and Iran seems capable of unleashing a giant (>Mw8.5) megathrust earthquake (Byrne et al., 1992; Kopp et al., 2000; Smith et al., 2013). These worrisome sectors, identified on Figure 1 by horizontally striped, sediment-thick trench sectors, include the eastern Aleutian or Fox Island sector (Butler, 2012; Ryan et al., 2012), the Andaman-Myanmar or far-northern Sumatra sector (Moore et al., 1980; Steckler et al., 2008), the Lesser Antilles or Barbados sector (Biju-Duval et al., 1982; Westbrook et al., 1988; Mascle and Moore, 1990), the Hellenic Trench (Taymaz et al., 1990), and the Tierra del Fuego or southern sector of the Chile Trench (Polonia et al., 2007). Although well sedimented, the Hikurangi (Scherwath et al., 2010) and Manila Trenches (Hayes and Lewis, 1984; Lewis and Hayes, 1984) are underthrust by uneven relief and characterized by strain releasing creep; hence they seem unlikely to rupture in a great megathrust (Wang and Bilek, 2014).

It has long been noticed that megathrusts ≥Mw8.0 uncommonly, if at all, occur along intra-oceanic arcs, for example the Izu-Bonin-Mariana (IBM), Tonga-Kermadec, Solomon, and Vanuatu SZs (Fig. 1; Tables 1 and 3). These SZs are fronted by thin-sediment trenches.

However, great and giant megathrust ruptures are characteristic of the intra-oceanic Aleutian SZ. Three great megathrusts have nucleated there since 1957—from west to east, the 1965 western Aleutian or Rat Island Mw8.7 and the 1957 Mw8.6 and 1986 Mw8.0 of the Andreanof or central Aleutian Trench (Tables 2 and 3). It seems plausible that the frequent rupturing of great and giant Aleutian earthquakes reflects the circumstance that the Aleutian Trench is flooded with a ∼2.0-km-thick wedge of turbidite beds (Fig. 2A) and widely spaced high bathymetric relief. The sediment body is sourced from glaciated eastern Alaskan drainages (Scholl, 1974; Scholl et al., 1983, 1987; McCarthy and Scholl, 1985).

As emphasized by Butler (2012) and Ryan et al. (2012), since Russian exploration began in the mid-1700s, the ∼700-km-long Fox Island or eastern sector of the Aleutian Trench has not ruptured in a great earthquake. The Fox Island sector lies between ∼172.5°W and 162.5°W and separates the thick-sediment eastern Aleutian and western Alaska sectors (Fig. 1). Bathymetrically smooth ocean crust underthrusts the Fox Island sector. Progressive landward migration of the arc magmatic front since ca. 45 Ma implies that subduction erosion also provides thickening material to the subduction channel (Vallier et al., 1994).

These circumstances, and the implications of the discovery of a massive paleotsunami deposit on the island of Kauai reported by Butler et al. (2014), warn that in the coming decades the eastern or Fox Island sector may well (again?) rupture in a great or giant megathrusts, as have all other well-sedimented Aleutian Trench sectors. The corresponding generation of a trans-oceanic tsunami similar to the ones launched by the 1957 central Aleutian or Andreanof sector Mw8.6 and the 1946 western Alaska or Alaska Peninsula Mw8.6 megathrusts (Tables 2 and 3) will target the Hawaiian Islands (Fryer et al., 2004; Lopez and Okal, 2006; Butler et al., 2014; von Huene et al., 2014). Considerable damage can also be expected along the west coast of North America and coastal and island communities to the south.

With gratitude and thanks we acknowledge the many conversational inputs and guidance provided by the members of the USGS Tsunami Source Working Group (TSWG), Menlo Park, CA. The TSWG, convened in the wake of the 2004 Banda Aceh disaster, by the 2nd author and its first chair, is presently guided by our colleague and astute interlocutor Walter Mooney. Our understanding of the many factors involved in determining the occurrence areas of high magnitude megathrusts benefited greatly from discussions with Kelin Wang, Roy Hyndman, Nathan Bangs, Jean-Yves Collot, and other informed colleagues. We are especially grateful to Emile Okal and Willie Lee for many insightful deliberations that helped us certify that the large instrumental and pre-instrumental era earthquakes compiled in our tables have the most up-to-date magnitudes and are indeed megathrust ruptures. Discerning and knowledgeable reviews by Casey Moore and an anonymous reader were critical to improving the science content, focus, and scholarship of the final manuscript.