Journal of Geophysical Research: Solid Earth

Along-strike variability of rupture duration in subduction zone earthquakes

Authors


Corresponding author: S. L. Bilek, Earth and Environmental Science Department, New Mexico Institute of Mining and Technology, 801 Leroy Pl., Socorro, NM 87801, USA. (sbilek@nmt.edu)

Abstract

[1] Subduction zone earthquakes exhibit a wide spectrum of rupture times that reflect conditions on the megathrust fault. Tsunami earthquakes are examples of slower than expected ruptures that produce anomalously large tsunamis relative to the surface-wave magnitude. One model explaining tsunami earthquakes suggests slip within patches of low rigidity material at shallow depths. Heterogeneous fault conditions, such as having patches of low rigidity material surrounded by higher strength material, should produce heterogeneous earthquake rupture parameters. Here we investigate along-strike variation in rupture duration for 427 shallow thrust earthquakes (Mw = 5.0–7.0) in the Peru, Chile, Alaska, Tonga, Kuril, Izu, and Java-Sumatra subduction zones to explore how heterogeneous seismic and tectonic characteristics, such as differences in sediment type, thickness, and roughness of subducting bathymetry, affect earthquake properties. Earthquake source parameters, including rupture durations, are estimated using multi-station deconvolution of teleseismic P and SH waves to solve for earthquake source time functions, and all events are relocated using additional depth phase information. We classify events into shallow (≤26 km) and deep (>26 km and ≤61 km) groups based on the overall mean depth and focus on the longest duration events with moment normalized rupture durations of >1 standard deviation above the mean duration for each group. We find long-duration events at all depths within the study regions except Peru and Chile. We find no correlation with incoming sediment thickness or type, and limited spatial correlation with regions of past tsunami earthquakes, regions of observed afterslip, and subducting bathymetric features.

1 Introduction

[2] Subduction zone earthquakes occurring along the seismogenic megathrust fault account for the majority (~90%) of the global seismic moment budget [e.g., Kanamori, 1986; Pacheco and Sykes, 1992]. The largest subduction zone events (Mw > 8) typically produce tsunami resulting from large seafloor displacements and hence large water column displacements. These tsunami can be devastating, as shown by the recent Mw = 9 Tohoku earthquake offshore Japan [e.g., Simons et al., 2011; Ozawa et al., 2011; Sato et al., 2011] and the Mw = 9.2 2004 Sumatra earthquake [e.g., Lay et al., 2005]. However, smaller subduction zone earthquakes can produce large tsunami as well. Tsunami earthquakes, a special class of typically shallow earthquakes of Mw 7–8, are characterized by deficient high frequency energy radiation and long rupture duration and produce tsunami that are anomalously large with respect to the surface-wave magnitude (Ms) [e.g., Kanamori, 1972; Satake and Tanioka, 1999; Bilek and Lay, 1999; Polet and Kanamori, 2000]. Tsunami earthquakes are particularly hazardous because their smaller magnitude will not trigger tsunami alerts in local communities. Understanding the conditions needed to produce these events and mapping where those conditions exist globally will improve seismic and tsunami hazard estimates.

[3] Several models have been proposed to explain the unexpectedly large tsunami for these events. Fukao [1979] suggested that the large tsunami waves following the 1963 and 1975 Kuril earthquakes resulted from large vertical displacement along high-angle splay faults in the sedimentary wedge. Others [Gutenberg, 1939; Hasegawa and Kanamori, 1987] proposed that landslides due to sediment slumping induce these large tsunami as in the 1929 Grand Banks event. Kanamori [1972] observed that earthquakes generating unusually large tsunami contained dominantly long period energy, such as the 1896 Sanriku and the 1946 Aleutian events, and first defined the class of tsunami earthquakes. Abe [1989] quantified tsunami earthquakes as events where the Mt (tsunami magnitude) is larger than Ms by more than 0.5 units. Kanamori and Kikuchi [1993] suggested that tsunami earthquakes in zones with little to no accreting sediment rupture all the way to the trench within the subducting sediments, whereas in accretionary margins, tsunami were due to large earthquakes triggering slumps within the accretionary prism. Based on waveform data for the 1992 Nicaragua, 1996 Peru, and the 1896 Sanriku (tsunami waveform data), other studies have also concluded that rupture propagated through shallow, subducted sediments [e.g., Satake, 1994; Tanioka and Satake, 1996; Satake and Tanioka, 1999; Polet and Kanamori, 2000]. Bilek and Lay [1999; 2002] linked tsunami earthquakes and other long-duration earthquakes to rupture through low rigidity material, and Geist and Bilek [2001] demonstrated that using a depth dependent rigidity model to compute slip used to model tsunami wave heights better predicts the 1992 Nicaragua tsunami amplitudes than a constant value for rigidity. If the model of shallow rupture through low strength materials is correct, it follows that all earthquakes near or within known tsunami earthquake rupture zones would have similar source properties. These other earthquakes could be too small to generate enough displacement of the overlying water column to produce a tsunami, but other seismic source parameters, such as rupture duration, should also show effects from rupture propagation in low strength materials. Okal and Newman [2001] examined this issue for a dataset of earthquakes in three subduction zones, finding an area along Peru that experienced both tsunami earthquakes and other events with low energy-moment ratios. Our current effort expands the study areas significantly and explores different source parameters.

[4] Earlier studies found that sediments can act to increase earthquake magnitude [Ruff, 1989a], duration, and tsunami occurrence [Kanamori, 1972; Polet and Kanamori, 2000]. Ruff [1989a] related the occurrence of great earthquakes (Mw > 8) to the presence of thick trench sediments that increase the contact surface on the interface resulting in large rupture areas. Also, subduction of thick sediments has been attributed to the occurrence of aseismic slow-slip events in Sumatra [Hsu et al., 2006; Kreemer et al., 2006; Chlieh et al., 2007] and Chile [Sallares and Ranero, 2005].

[5] Subducting plate bathymetry has been previously linked to levels of rupture heterogeneity. Earthquake magnitude and moment release patterns can be influenced by local roughness on the incoming plate and the interaction with the overriding plate at the plate boundary [e.g., Kelleher and McCann, 1976; Cloos, 1992; Scholz and Small, 1997; Abercrombie et al., 2001; Bilek et al., 2003; Robinson et al., 2006; Mochizuki et al., 2008; Das and Watts, 2009; Wang and Bilek, 2011]. However, no clear link has been made between plate bathymetry and earthquake durations, so we specifically examine the rupture durations in areas of both rough and smooth plate bathymetry to determine if a relationship exists.

[6] Thus, one goal of this work is to explore regional variations in earthquake rupture properties to test the hypothesis that earthquakes in areas of past tsunami earthquakes have similar long-duration character indicative of rupture through specific fault conditions. Because the catalog of tsunami earthquakes is limited by time and space, we conduct a global subduction zone survey to identify and map long-duration earthquakes across a wide range of magnitudes. We then compare source parameters (rupture duration and focal depths) to geophysical data available for each subduction zone to determine if there are spatial correlations with physical characteristics of the margin or to past reported slow rupture processes. We compare the earthquake parameters to sediment thickness and type as well as to subducting plate bathymetry, as these parameters are defined along many margins and have been previously suggested to affect earthquake rupture.

2 Data

[7] We focus on source parameters of subduction zone thrust earthquakes, expanding on the works of Bilek et al. [2004, 2011] and Bilek [2007; 2009] with an updated catalog that spans 1989–2009 (one 2010 event in Alaska) within the regions outlined in Figure 1 (see also Table 1). We use earthquakes less than 60 km depth that are located on or close to the interface and have thrust-faulting mechanisms as determined by the Global Centroid Moment Tensor (GCMT) catalog [e.g., Ekström et al., 2005] with Mw 5.0–7.0. Event magnitude is limited because of the later use of waveform modeling techniques that assume a single point source rupture model.

Figure 1.

Map of study regions (boxes). Red stars indicate location of known tsunami earthquakes, and gray circles show location of the earthquakes included in this study.

Table 1. Tectonic Parameters, Margin Type and Date Range, and Number of Events in Each of Our Study Regionsa
RegionMargin typeDip (deg)Age (Ma)Convergence rate (cm/yr)Sediment thickness (km)Sediment volume subduction rate (km3/Myr)19DateMagnitude range
  1. a

    Tectonic parameters are from the following: 1, von Heune and Scholl [1991]; 2, Clift and Vannucchi [2004]; 3, Schlüter et al. [2002]; 4, Kopp et al. [2006]; 5, Jarrard [1986]; 6, Tichelaar and Ruff [1991]; 7, Patzwahl et al. [1999]; 8, Ryan and Scholl [1993]; 9, Muller et al. [2008]; 10, DeMets et al. [1994]; 11, Bevis et al., 1995; 12, Seno et al. [1993]; 13, Tregoning et al. [1994]; 14, Simons et al. [2007]; 15, Angermann et al. [1999]; 16, Kelemen et al. [2003]; 17, Singer et al. [2007]; 18, Turner et al. [1997]; 19, Scholl and von Huene [2007]; 20, Syracuse et al., 2010; 21, Bailey, 1996; 22, Masson et al. 1990; 23, Dean et al. [2010]; 24, Schweller et al. [1981]; 25, Bangs and Cande [1997]. NF, normal-faulting.

AlaskaAccretionary19 (eastern segment); 25 (central segment)555 (eastern segment);8,9 60 (central segment); 49 (western segment);5.8–6.6 (central segment);10 7.2 (western segment);Sediment filled (eastern segment); ~1–1.4 (central segment); ~1 to few hundred meters (western segment)16,17,1981,0001989–20105.5–6.9
Tonga-KermadecErosive1,223586–113 (Kermadec); 108–85 (Tonga)95 (Kermadec); 24 (Tonga)10.11~0.1–0.4 1,2,18,19Tonga, 53,040; Kermadec, 21,8401990–20085.7–6.9
IzuErosive1,2225140–13396.1–4.712~0.2–0.41,2,1928,8001990–20095.7–6.9
KurilErosive1,219 (Kamchatka); 22 (Kuril)5100–108 (Kamchatka); 108–130 (Kuril)97.8 (Kamchatka); 7.1(Kuril)10~ 1 km (near 51°N); ~0.5–0.6 (everywhere else)19,20,2154,0001990–20095.7–6.9
JavaW. Java, accretionary; E. Java, erosive 1,4165100–13096.7–7.113,14W. Java, 1.6; E. Java, 0–0.519,22W. Java, 11,250; E. Java, 27,0001990–20095.4–6.8
Java (NF)Erosive4     1994 and 20065.2–5.6
SumatraAccretionary1,3,416570–9895.5–6.3141.5–2.51,2,235850–25,5001992–20095.8–7.0
PeruErosive2,1914520 (northern segment); 45 (southern segment)97.2–7.6100.1–0.72,19,20,2468,0001991–20095.7–6.7
ChileErosive1,29–25 (Northern segment); 14 ± 4 (Central segment); 12 ± 3 (Southern segment)6,750 (Northern segment), 37 (Southern segment)96.0–6.6150.1–0.5 (northern segment); 1.5–2.2 (southern segment)24,2519,1251990–20095.7–6.8

[8] For each event, we estimate source time functions (STF) and focal depths using teleseismic (30°–90° epicentral distance) broadband three-component seismograms available through the IRIS Data Management Center. Teleseismic data is chosen because of the fairly simple Green's function. We use a time window of 30 s before to 60 s after the P and SH arrivals so we include the entire direct phase as well as depth phases (pP and sP), which are important for depth constraints. For approximately 35% of events, only P waveforms were used because no suitable SH waveforms were available due to their poor quality or low signal to noise ratio. The 427 events used in the analysis have a minimum of five waveforms and a maximum of 20 waveforms covering at least three quadrants to provide adequate azimuthal coverage around the focal sphere, thus minimizing the trade-offs between the depth and the STF shape.

3 Methods

3.1 Source Time Function and Depth Estimates

[9] We use a multi-station deconvolution of teleseismic P and SH waveforms to determine STFs and focal depths for events in our dataset [e.g., Ruff and Kanamori, 1983; Ruff, 1989b; Ruff and Tichelaar, 1990; Ruff and Miller, 1994]. Synthetic Green's functions are computed for each waveform at each station (Figure 2a) at a depth range 3–50 km using the GCMT focal mechanism information and a simple velocity model with a water layer over a uniform half-space of P-wave velocity Vp = 6.7 km/s, a reasonable half-space velocity approximation for the model. The synthetic Green's functions are then deconvolved from the observed seismograms, resulting in an STF to represent the earthquake source. The deconvolution is performed over a range of depths, with the preferred STF and depth selected based on minimizing the misfit with the observed data (Figure 2) [e.g., Ruff and Kanamori, 1983; Ruff, 1989b; Ruff and Tichelaar, 1990; Ruff and Miller, 1994; Bilek et al., 2004]. We use the preferred STF to estimate rupture duration (RD), defining it as the start to end time of the pulse that contains the major moment release, excluding the later smaller oscillations that may be a result of inaccuracies in the Green's function (Figure 2b) [e.g., Bilek et al. 2004]. Depth uncertainties are estimated using the range of depths with low misfit around the absolute minimum (Figure 2c). We use the source time functions for the depths within these bounds to determine a range of source durations, then normalize this range to estimate the uncertainties for the rupture duration.

Figure 2.

Example of waveform modeling for source time functions (STF) and depth. Seismograms for earthquake in the Sumatra subduction zone (30 March 2005, 16:19:41:77 UTC, Mw 6.3) (a) Observed seismograms (black) and synthetic seismograms (red) for optimal depth of 26 km. Station name, wave type, and station azimuth provided for each waveform. (b) STF for optimal solution. Rupture duration (RD) measured from zero to end of major moment release. Normalized rupture duration (NRD) scales RD by the seismic moment of an Mw 6.0 event. (c) Misfit between observed and synthetic seismograms, with optimal solution at minimum misfit.

[10] In order to compare events of different magnitudes, we scale the measured duration by the cubic root of the event seismic moment (Mo) and normalize it by the Mo of an event with Mw 6.0 [e.g., Kanamori and Anderson, 1975; Houston et al., 1998; Bilek and Lay, 1999]:

display math(1)

where NRD is the normalized rupture duration, RD is the rupture duration measured from the STF, inline image is the seismic moment of the event, and inline image = 1.1618 N-m is the seismic moment for an Mw = 6.0 event.

3.2 Event Relocations

[11] Each earthquake is relocated in order to provide an improved spatial comparison than are possible with catalog locations alone. We use the method of Pesicek et al. [2010] to identify depth phases (pP, pwP, sP) and improve phase onset times. The additional depth phases are incorporated with existing phase catalogs and relocated using the Engdahl, van der Hilst, and Buland (EHB) teleseismic location approach [Engdahl et al., 1998]. For each event, we first calculate the power spectral density function at 1 Hz for vertical component velocity and displacement waveforms. The resulting time series reflects the magnitude of the discrete Fourier transform at that frequency. We next find the gradient of the time series by calculating the derivative, apply minimum smoothing, and normalize to 1.0. To identify changes in the gradient time series, points less than the mean are set to zero. Abrupt changes in the gradient occurring on both time series are considered triggers. Triggers that occur within a few samples on both time series are associated with P, pP, or sP. We report onset times from the velocity time series. Scatter in the fit to depth phases using the EHB approach can be largely attributed to the effects of complexities in structure across the region on the arrival times of depth phases that have been modeled using the ak135 model [Kennett et al., 1995] overlain by water depth estimates at their bounce points.

[12] Additional depth phases allow improvement of initial EHB solutions that were poorly constrained in depth, improved EHB solutions, and ultimately more accurate determination of rupture parameters. We have provided ~31,000 additional phases including over 10,000 depth phases for inclusion in the EHB catalog at the subduction zones reported on in this study. The mean change in hypocentral location was 4.7 km, with individual subduction zone means ranging from 3.4 (Alaska) to 5.7 km (Peru). On average, the change in depth (~3.6 km) was larger than the change in epicenter. The mean epicenter and depth differences following inclusion of new phases were within the EHB uncertainties, further supporting the overall quality of the EHB teleseismic catalog. The benefit of the new phases lies in better constraining depth for the small percent of outlier events. The maximum depth change ranged from 7.8 (Peru) to 23.3 km (Tonga).

3.3 Depth Comparisons

[13] Once both depth estimates are complete, we find that deconvolution depth estimates are generally shallower than the EHB depths (Figure 3) probably due to the half-space velocity model we use (Vp = 6.7 km/s and Vp/Vs = 1.7) is simpler and slower than the ak135 velocity model used for the EHB relocation. Additionally, depth phases identified for some of the larger and potentially more complex events may be related to large amplitude subevents, leading to relocated depth inaccuracies. In cases where the depth estimates differ by >10 km, both waveform deconvolution results and relocation phase picks are manually reviewed in order to reach a preferred solution where the depth difference is <10 km. In some of these cases, we find a second minimum in the waveform misfit curve that is within 10 km of the relocated EHB depth and use the corresponding STF. In other cases, arrival time picks are reassessed, and the location is recomputed with manually revised picks. We manually reviewed ~25% of the earthquakes used in this study.

Figure 3.

Comparison between relocated and deconvolution depth estimates, with linear regression fit suggesting deconvolution depths are shallower than the EHB-based relocation depths.

Figure 4.

(a) Normalized rupture duration as a function of depth for all regions, including known tsunami earthquakes. Solid vertical line at 26 km marks division between shallow and deep events. Solid horizontal lines show average NRD of shallow and deep events, dashed horizontal lines indicate minimum NRD for classification as very long rupture duration events within each depth group. Long events of interest fall in the gray shaded area. Crosshairs are error in depth estimation (horizontal) and normalized STF error (vertical). (b) Rupture velocity as a function of depth for all regions combined. Black circles show the rupture velocities (0.67–2.51 km/s) for the events classified as long NRD events.

3.4 Parameter Tests

[14] We test the effect of varying focal mechanism and velocity model on depth and normalized rupture duration (NRD) estimates to quantify effects on the waveform modeling based STF and depth estimates (Figure A1 in the Supporting Information). Using two events, one at 10 km from the Tonga subduction zone (to represent shallow and long) and one at 32 km depth in Izu (to represent deep and long), we vary the fault geometry and the seismic velocities (Vp, Vs, and Vp/Vs ratio) by ±10% and compute STF duration and depth using the new parameters. We find that most sources of error tend to decrease source duration.

[15] We take the fault geometry from the Global CMT catalog, with solutions based on a wide frequency band of data. Estimates of likely CMT geometry errors due to unmodeled Earth structure tend to be a few degrees, although dip tends to be larger, up to ~20% variation [e.g., Frolich and Davis, 1999; Ekström et al., 2005; Hjörleifsdóttir and Ekström, 2010]. For both events, we find very minor effects on the depth, with ±1 km variations observed with ±10% variations in geometry (Figure A1a), ±3 km variations for ±20% in dip. These variations produced a 1 s decrease in the duration estimates for the shallow event and no change for the deep event.

[16] Another source of uncertainty in our depth and duration estimates is the velocity model. For the deconvolution, we assume a 1-D homogenous half-space velocity model, which is a simplistic model given the complex velocity structure of a subduction zone. Variations in subduction zone seismic velocity have been documented to range between 5% and 12% along the interface and across lithologic boundaries [e.g., Mitronovas and Isacks, 1971; Snoke et al., 1978; Matsuzawa et al., 1987; Helffrich and Stein, 1993]. We explore the effects of a range of velocity models on our depth and duration estimates, including varying Vp, Vs, and Vp/Vs by ±10% (Figure A1b). For the Izu event, we see ±3 km range from the original solution of 32 km. The effect on duration is also larger, with up to a 4 s decrease in duration when the depth becomes deeper by 3 km. For the shallow Tonga event, the depth variation is ±2 km from the original solution, and up to a 3 s decrease in duration results. Future efforts could involve refining the velocity model used for this analysis; however, for consistency with past studies and ease of comparison between regions, we choose to use a consistent velocity model for all regions in the study. Consistency in depth between the two methods described here also provides confidence in the depth and duration estimates. As an additional check on the depths, we also find reasonably good fit between the depths here and the slab 1.0 plate interface models [Hayes et al., 2012] for the Alaska and Chile margins (Figure A2).

4 Results

[17] We focus on eight subduction zones: Peru, Chile, Alaska-Aleutian, Tonga-Kermadec, Kuril-Kamchatka, Izu, Java, and Sumatra subduction zones (Figure 1, Table 1). We compare our source parameters to locations of other slow-slip occurrences and other geologic and tectonic parameters for each zone (Figures 5-12, Table 2). One target comparison is in and around the location of past tsunami earthquakes. Within our study areas, these include the 1907 Mw = 7.6 Sumatra [Kanamori et al., 2010], the 1946 Mw = 7.4 Aleutian [Kanamori, 1972; Abe, 1979; Johnson and Satake, 1997], the 1960 Mw = 7.6 Peru [Pelayo and Wiens, 1990; 1992], the 1963 Mw = 7.8 and 1975 Mw = 7.5 Kuril [Abe, 1979; Pelayo and Wiens, 1992], the 1982 Mw = 7.7 Tonga [Lundgren et al., 1989], the 1994 Mw = 7.8 Java [Abe, 1979; Polet and Kanamori, 2000; Abercrombie et al., 2001], the 1996 Mw = 7.5 Peru [Polet and Kanamori, 2000], the 2006 Mw = 7.7 Java [Ammon et al., 2006; Bilek and Engdahl, 2007], and the 2010 Mw = 7.8 Sumatra [Bilek et al., 2011; Lay et al., 2011; Newman et al., 2011; Singh et al., 2011]. Other events have occurred in other regions not covered in this paper (e.g., 1992 Nicaragua and the 2011 and 1896 Japan, discussed in Bilek et al., [2012]), and still others are debated, such as the 2010 Mw = 8 Tonga event [Beaven et al., 2010; Lay et al., 2010]. Here we also briefly describe the tectonic setting, sediment type and thickness, bathymetric structures, significant earthquakes associated with these bathymetric features, the occurrence of tsunami earthquakes and afterslip for each subduction zone, and finally the source parameters for the events in each region. Margin type and variation of slab dip, age, convergence rate, and estimates of sediment thickness are listed in Table 1 and shown in the respective figures for each region.

Figure 5.

Alaska subduction zone showing convergence rate, direction (arrow), and bathymetric features on the incoming plate. Circles and inverted triangles denote shallow and deep earthquakes, respectively, color indicating NRD (see color scale and legend) yellow star indicates tsunami earthquake. Symbols with thick black outline are long NRD events. Sediment thickness estimates indicated by white lines show sediment extent and thickness as determined by Kelemen et al. [2003] and Singer et al. [2007]. Size of clusters A, B and C are determined according to the size of the geographic sector divisions by Dobson et al. [1996], size of cluster D determined according to the extent of thickest sedimentary layer and size of cluster E determined according to rupture extent of the 1946 tsunami earthquake [Johnson and Satake, 1997]. K-BS, Kodiak-Bowie Seamount Chain; DS, Derikson Seamount; AFZ, Amlia Fracture Zone; ADZ, Adak Fracture Zone; RFZ, Rat Fracture Zone; TS, Terrigenous sediments.

Figure 6.

Tonga-Kermadec subduction zone showing convergence rate, direction (arrow) and bathymetric features on the incoming plate. Data plotted as in Figure 4. Symbols with thick outline are long NRD events. Sediment thicknesses are estimates from Divins [2006]. CT, Capricorn Tablemount; LR, Louisville Ridge.

Figure 7.

Izu-Bonin subduction zone showing convergence rate, direction (arrow), and bathymetric features on the incoming plate. Data plotted as in Figure 4. Symbols with thick outline are long NRD events. Open black square shows location of event with longest NRD in deep group. Sediment thicknesses are estimates from Divins [2006]. SMT, Seamounts; MS, Mogi Seamount; SR, Shatsky Rise; KS, Kashima Seamount; PSP, Philippine Sea Plate.

Figure 8.

Kuril subduction zone showing convergence rate, direction (arrow) and bathymetric features on the incoming plate. Data plotted as in Figure 4. Sediment thickness estimates indicated by line are obtained from Bailey [1996]. Symbols with thick outline are long NRD events. Size of cluster A determined by location of subducting seamounts, cluster B covers rupture area of 1969 Mw = 8.2, 1958 Mw = 8.3, 1963 Mw = 8.5 [Schwartz and Ruff, 1987], and 2006 Mw = 8.3 [Lay et al., 2009], cluster C by rupture area of 1952 Mw = 8.3 [Kelleher and Savino, 1975]. Black open circle shows location of event with shortest NRD in deep group. ES, Erimo Seamount; ZR, Zenkevich Rise; MS, Morozko Seamount; KT, Krusenstern Trough; MJS, Meiji Seamount.

Figure 9.

Java subduction zone showing convergence direction, rate, and bathymetric features on the incoming plate. Data plotted as in Figure 4. Symbols with thick outline are long NRD events. Open black square shows location of event with longest NRD in shallow group. Square symbols indicate normal-faulting events color-coded with NRD. Sediment thicknesses from Masson et al. [1990]. RR, Roo Rise; UG SMT, Umbgrove Seamount; SMT, Seamounts; FZ, Fracture zone.

Figure 10.

Sumatra subduction zone showing convergence direction, rate, and bathymetric features on the incoming plate. Data plotted as in Figure 4. Symbols with thick outline are long NRD events. Open black circle shows location of event with shortest NRD in shallow group. Size of cluster A is approximated by the extent of thickest sedimentary layer (reaching up to 2.5 km) along the trench [e.g., Dean et al., 2010], size of cluster B covers the area of the aftershocks of 2010 Mw = 7.8 tsunami earthquake [Bilek et al., 2011]. 90°ER, 90°East Ridge; WB, Wharton Basin; IFZ, Investigator Fracture Zone.

Figure 11.

Peru subduction zone showing convergence rate, direction (white arrow), and bathymetric features on incoming plate. Data plotted as in Figure 4. Sediment extent and thickness from Schweller et al. [1981]. NR, Nazca Ridge; MFZ, Mendana Fracture Zone; SR, Sarmiento Ridge; AR, Alvarado Ridge; GR, Grijalva Ridge; CS, Carnegie Seamount.

Figure 12.

Chile subduction zone showing convergence rate, direction (white arrow), and bathymetric features on incoming plate. Data plotted as in Figure 4. Sediment thickness estimates indicated by line show sediment extent and thickness [Schweller et al., 1981; Bangs and Cande, 1997] along the trench. JFR, Juan Fernandez Ridge; IR, Iqiuque Ridge; NR, Nazca Ridge.

Table 2. Number of Events in Long and Deep, Long and Shallow, Short and Deep, and Short and Shallow Groups in Each of Our Study Regions
RegionsNos. of eventsNos. of long and deep eventsNos. of long and shallow eventsNos. of tsunami earthquakesNos. of events near tsunami earthquakesNos. of slow-slip eventsNos. of events in slow slipNos. of events near subducting bathymetric features
Long and deepLong and shallowLong and deepLong and shallowLong and deepLong and shallow
  1. NF, Normal-faulting event.

Alaska9241012000007
Tonga; Kermadec6437202 (near 1982 event)00034
Izu315400000051
Kuril1209423136331
Java132120000021
Java (NF)5101 0100001
Sumatra46123 (near 1907 event) 1 (near 2010 event)034151
Peru140020020000
Chile380000030000

[18] We find a general trend of decreasing normalized rupture durations (NRD) with increasing depth for all the study regions, consistent with previous studies (Figure 4) [e.g., Bilek and Lay 1998; 1999; Bilek et al., 2004]. In order to focus on the spatial distribution of the longest duration events, we group the events into separate depth and duration classes based on mean values of the dataset. The mean depth for the entire dataset is 26 ± 3 km; thus, we consider events ≤26 km as shallow and those >26 km as deep events. Within the shallow group, the average NRD is 5.1 s with a standard deviation of 2.5 s; for this group, we classify events with NRD ≥ 7.5 s (1 standard deviation above mean) as long duration in subsequent sections. Within the shallow group, the shortest NRD (0.7 s) event occurred in Sumatra at 13 km depth (Figure 10), and the longest NRD (14.8 s) occurred in Java at 12 km depth (Figure 9). Within the deep group, the average NRD is 4.0 s, and the standard deviation is 1.6 s; thus, long-duration events in this group have NRD ≥ 5.6 s. In the deep group, the shortest NRD (0.8 s) is at 45 km depth in Kuril (Figure 8), and the longest NRD (9.8 s) at 29 km depth is in Izu (Figure 7).

[19] Fifty-seven events meet the long-duration classification, and these make up a small subset (13%) of the overall dataset (Table 2). The range is high, however. Peru and Chile have no long-duration events, while nearly 30% of the Izu dataset are long duration. Although a general trend of decreasing NRD with depth exists for all regions, Alaska, Java, Peru, and Chile show the strongest trend of an overall decrease in NRD with increasing depth (Figure 13). Subsections below outline along-strike variations in each zone; for comparisons of NRD to specific areas, we count the number of events within ±1° around defined areas of afterslip, tsunami earthquakes, and specific topographic features (Table 2).

Figure 13.

Normalized rupture duration as a function of depth for individual regions showing best fit-line and equation.

[20] These values can also be used to provide a rough estimate of rupture velocity (Vr) variations with depth. This assumes that the duration variations are a result of rupture velocity variations and not due to a difference in fault size (thus stress drop). Using a constant fault length of 10 km (the characteristic length of a fault producing a Mw = 6 event), we compute Vr of 0.7–2.5 km/s for the longest duration events in the dataset. These values fall within the range of values estimated for tsunami earthquakes of 1–1.5 km/s [e.g., Polet and Kanamori, 2000; Lay and Bilek, 2007; Ammon et al., 2006; Lay et al., 2011], suggesting similar processes between tsunami earthquakes and these long-duration events.

4.1 Alaska-Aleutian

[21] The Alaska subduction zone marks the boundary where the Pacific plate is subducting under the North American Plate (Figure 5). At the eastern end of the study region, seamounts are present on the ocean floor, while towards the western side there exist several fracture zones and trough structures that intersect the trench. Here convergence is oblique and increases in obliquity towards the west [Scholl et al., 1982] where the relative plate motion becomes nearly trench-parallel at the westernmost end of the arc [Ryan and Scholl, 1993] (Figure 5). Sediments in the Aleutian trench are mainly composed of terrigenous material eroded from the Alaska Range and older sediments from the Chugach, Wrangell, and St. Elias Mountains [Kelemen et al., 2003].

[22] There have been several observations of afterslip and creeping motion in this subduction zone, but none has been observed in our study area. A number of Mw > 7.0 subduction earthquakes ruptured most of the length of the subduction zone since the mid-1900s [e.g., Plafker, 1969; Sykes, 1971; Johnson et al., 1994; Ichinose et al., 2007]; some are thought to be related to seamount subduction [Estabrook et al., 1994]. The Aleutian trench has only one known tsunami earthquake, the 1946 Mw = 7.4 [Kanamori, 1972], located at the eastern edge of the region with thick terrigenous sediments.

[23] Within Alaska, 14 events are classified as long duration, 10 in the shallow group and four in the deep group (Figure 5, Table 2). All these events occur within regions where fairly thick (>1 km) terrigenous sediment is subducted [Kelemen et al. 2003; Singer et al. 2007] and accretion efficiency is low (7%) [Clift and Vannucchi, 2004]. Six events fall within the 1946 rupture zone boundaries [Johnson and Satake, 1997], but only two are long duration. Seven of the long shallow events occur within a region of either seamount or fracture zone subduction. In total, only nine of the 14 long-duration events spatially correlate with our parameters of interest (Figure 5, Table 2). No slow-slip events have been document along Alaska, so this comparison cannot be made.

4.2 Tonga-Kermadec

[24] The Tonga-Kermadec trench is an intra-oceanic subduction zone where the Pacific plate is subducting beneath the Australian plate [e.g., Turner and Hawkesworth, 1997; Bonnardot et al., 2009] (Figure 6). The incoming plate has several subducting features such as the Capricorn Seamounts [Scholz and Small, 1997] and the Louisville Ridge [Ballance et al., 1989], which divides the trench into the Tonga (in north) and the Kermadec (in south) subduction zone [Christensen and Lay, 1988]. The Tonga trench is a sediment-starved trench [Ballance et al., 1989] mainly containing pelagic/hemipelagic sediments [Underwood, 2007]. Five large earthquakes (Mw > 7.0) have occurred near the Louisville Ridge [Christensen and Lay, 1988] including the 1982 Mw = 7.5 tsunami earthquake in addition to the more recent possible 2009 Mw = 8 tsunami earthquake [Beavan et al., 2010; Lay et al., 2010] near the Capricorn Seamounts.

[25] Our Tonga subduction zone catalog contains 10 long-duration events, seven shallow, and three deep (Figure 6, Table 2). Roughly constant thin layers (<400 m) of pelagic sediment subduct along this erosive margin [Turner and Hawkesworth, 1997; Scholl and von Huene, 1991; Clift and Vannucchi, 2004]. Two (out of eight events within that rupture zone) occur in the rupture zone of the possible 2008 tsunami earthquake [Beavan et al., 2010; Lay et al., 2010], and no events are recorded in the region of the possible 1982 tsunami earthquake [Lundgren et al., 1989]. The majority (seven) of long-duration events occur within a region of seamount subduction. Slow-slip events have been documented along the very southern end of the margin along New Zealand [e.g., Wallace and Beavan, 2010] but outside of our study area.

4.3 Izu

[26] The Izu subduction zone is a segment of the larger Izu-Bonin-Mariana (IBM) system that marks the plate boundary where the Pacific plate subducts beneath the Philippine Sea Plate (Figure 7). The Izu segment has generally smooth subducting bathymetry; the incoming plate increases in roughness to the south [Staudigel et al., 2010] but outside of our study area. In the IBM trench, all of the oceanic crust, sediments, and seamounts are being subducted [Staudigel et al., 2010]. The IBM is an intra-oceanic subduction zone and is far from the continent, a main source of sediment supply, and therefore has a thin sedimentary cover mainly composed of a mixture of clayey and siliceous, pelagic/hemipelagic, biogenic, and calcareous deposits [Underwood, 2007; Chauvel et al., 2009].

[27] Nine long-duration events occur along the erosive Izu margin, four shallow and five deep (Figure 7, Table 2). The sediment cover in the area of long-duration events is very thin (<200 m) pelagic material [Plank et al., 2007; Underwood, 2007]. No tsunami or slow-slip events have been recorded in this region. The majority (six) of long-duration events occur within regions of seamount subduction, including the event with longest NRD in the deep group (9.8 s at 29 km depth) in the region where the Kashima Seamount chain enters the trench (Figure 7). Interestingly, the Izu margin has the largest percentage of long-duration events in this study.

4.4 Kuril-Kamchatka

[28] Our study region for the Kuril-Kamchatka margin extends from the southern portion of the Kamchatka trench to the northern portion of the Izu Trench (Figure 8). Here, the Pacific plate is subducting under the Eurasian plate. There are a few bathymetric features on the subducting plate entering the trench, such as the Erimo Seamount [Lallemand and Le Pichon, 1987], the Meiji Seamount, and Kruzenstern fracture zone [Gorbatov et al., 1997]. The Kurile-Kamchatka trench is covered with thin pelagic sediments [Bailey, 1996].

[29] In Kuril, afterslip has been observed and attributed to viscoelastic relaxation [Kogan et al., 2011] and other factors such as plate roughness, negative buoyancy of the plate, sediments, and pore fluid pressure [Baba et al., 2006]. Several Mw > 8 have occurred along the Kuril-Kamchatka trench, with the 1923 Mw = 8.5, 1952 Mw = 9, and 1969 Mw = 8.2 having occurred near the subducting features. The 1963 and 1975 tsunami earthquakes also occurred along this segment [Pelayo and Wiens, 1992].

[30] This margin contains 13 long-duration events, four shallow and nine deep (Figure 8, Table 2). The pelagic sediment thickness here ranges between 200 and 600 m [Langseth and Shipboard Scientific Party, 1980; Scholl and von Huene, 1991; Bailey, 1996; Clift and Vannucchi, 2004], with most events occurring in areas of 500–600 m sediment thickness. Within the rupture zones of the two tsunami earthquakes, only one (of 28) long-duration event occurred near the 1963 event, and one (of 20) occurred within the 1975 rupture zone. Only five of the long-duration events occur within regions of topographic feature subduction; two are aligned with Erimo Seamount, one with the Zenkevich Rise, one with the Morozko Seamount, and one at the northern end of our study area is aligned by the Krusenstern Trough. Nine of the long-duration events occurred within regions defined by slow slip or afterslip (Figure 14a) [Burgmann et al., 2001; Miyazaki et al., 2004; Kogan et al., 2011].

Figure 14.

Locations of afterslip in (a) Kuril, (b) Sumatra, (c) Peru, and (d) Chile. In Figure 14a, dashed black box, area of afterslip model following the 2003 Tokachi-Oki earthquake [Baba et al., 2006]; white dashed box, area of afterslip following the 2006–2007 Kuril earthquake sequence [Kogan et al., 2011]; red dashed box, area of afterslip following the 1997 event [Burgmann et al., 2001]. In Figure 14b, area of afterslip following the 2004 Mw = 9.1 Sumatra earthquake shown in dashed red ellipses [Chlieh et al., 2007] and 2005 Mw 8.7 Nias earthquake dashed black box from Hsu et al. [2006], dashed white box from Kreemer et al. [2006]. In Kuril and Sumatra, long NRD events fall within or close to areas with afterslip. In Figure 14c, dashed black box, Peru afterslip model following the 2007 Mw = 8 Pisco earthquake [Perfettini et al., 2010]. In Figure 14d, dashed black box. Chile afterslip model following the 2010 Mw = 8.8 Maule earthquake [Vingy et al., 2010], white dashed box. area of afterslip following the 1995 Mw = 8.1 northern Chile earthquake [Pritchard and Simons, 2006]; red dashed box. area of afterslip following the 2007 Mw = 7.7 Tocopilla earthquake [Bejar-Pizarro et al., 2010]. No long NRD events in Peru and Chile, also areas of observed afterslip in Peru are far from location of tsunami earthquakes. Symbols and labels same as in Figures 5-12.

4.5 Java

[31] The Java trench is a segment of the Sunda Arc stretching from the Banda Sea to the Sunda Strait where the Indo-Australian plate subducts beneath the Sunda plate (Figure 9). Along Java, there are a number of subducting or completely subducted seamounts along the trench [Masson et al., 1990] (Figure 9). The central to eastern segment of the Java trench is generally devoid of sediments except for local ponds of pelagic sediments derived from erosion of seamounts due to collision with the margin [Masson et al., 1990]. Earthquakes along the Java trench have small to moderate (Mw < 8) magnitudes with two similar tsunami earthquakes in 1994 (Mw 7.8) and 2006 (Mw 7.7) [Polet and Kanamori, 2000; Ammon et al., 2006].

[32] Four events had long duration along this margin, two shallow and two deep (Figure 9, Table 2). Pelagic sediment thickness ranges from ~500 m [Masson et al., 1990] to 1.5 km [e.g., von Huene and Scholl, 1991; Clift and Vannucchi, 2004] in the region of long-duration events. One normal-faulting event (of four total in the rupture zone) occurred nearby the 1994 tsunami earthquake [Abercrombie et al., 2001]. No long-duration events occurred within the region of the 2006 tsunami earthquake. The remaining events appear to be located in the area of the Roo Rise and seamounts [Masson et al., 1990] or the fracture zones identified by Newcomb and McCann [1987]. The longest NRD (14.8 s at 12 km depth) is located in the region between the 1994 and 2006 tsunami earthquakes (Figure 9), which also corresponds to the location of a 500 m thick pelagic patch of sediments [Masson et al., 1990]. No afterslip or slow slip has been observed in this region.

4.6 Sumatra

[33] The Sumatra trench is the segment of the Sunda Arc that continues from the Sunda Strait to the Andaman Sea marking the subduction of the Indo-Australian plate under the Sunda plate (Figure 10). Along Sumatra, two main bathymetric highs, the Investigator Fracture Zone (IFZ) and the 90°E Ridge, separated by the Wharton Basin, are prominent features (Figure 10). The area west of the 90°E ridge contains several E-W trending reverse and NW-SE strike-slip active faults responsible for the seismicity around and along the ridge. The area to the east of the 90°E ridge is cut by the N-S striking Investigator Fracture Zone and other fossil transform faults [Newcomb and McCann, 1987] (Figure 10). Convergence obliquity increases and becomes nearly trench-parallel along the Andaman Islands [Prawirodirdjo et al., 1997].

[34] Prawirodirdjo et al. [1997] suggested that these thick subducting sediments (northern segment) containing excess pore-fluids elevate pore pressure and reduce the coupling (ratio of seismic slip to total slip) along the thrust interface and therefore enhance aseismic slip. Afterslip has been observed in this region following the 2005 Mw = 8.5 Nias earthquake and was attributed to the presence of thick subducting sedimentary layers [e.g., Hsu et al., 2006; Kreemer et al., 2006; Chlieh et al., 2007]. Sumatra has been the site of several recent Mw > 8.0 earthquakes including the 2004 Mw 9.2 and 2005 Mw 8.7 events. Two tsunami earthquakes have also been identified along this margin, one in 1907 (Mw 7.6) [Kanamori et al., 2010] and another in 2010 [e.g., Lay et al., 2011; Newman et al., 2011; Bilek et al., 2011].

[35] During the time period of this catalog, seven events had long durations, one shallow and six deep events (Figure 10, Table 2). There is a mix of sediment type along this accretionary margin, but it is all relatively thick at 1.5–2.5 km [e.g., von Huene and Scholl, 1991; Clift and Vannucchi, 2004; Dean et al., 2010] in the zones of long-duration events. Three events (out of 20) occurred within the estimated rupture zone of the 1907 tsunami earthquake [Kanamori, 2010], which interestingly lies between the 2004 and 2005 great Sumatra earthquake rupture zones. Aftershocks of the 2010 Southern Sumatra tsunami earthquake are not included here, but several of these also had long duration [Bilek et al., 2011]. One long-duration event occurred near the Investigator Fracture Zone, and five others occurred within the Wharton Basin. The majority of the long-duration events (five of seven) occurred within regions of observed afterslip (Figure 14b, Table 2) [Hsu et al., 2006; Kreemer et al., 2006; Chlieh et al., 2007].

4.7 Peru

[36] We study the segment between 1°S and 15°S of the Peru-Chile margin (Figure 11) where the Nazca Plate subducts beneath the South American plate. The Nazca Plate contains several fracture zones that intersect the trench. Except for the Nazca Ridge, the subducting plate has an overall thin cover [Schweller et al., 1981; Clift and Vannucchi, 2004; Scholl and von Huene, 2007; Syracuse et al., 2010] of pelagic sediments [e.g., Mix et al., 2003; Gamage et al., 2005] (Figure 11; Table 1). The Peru trench ruptured in several large earthquakes (Mw > 7) and two tsunami earthquakes (Figure 11). Some of these large events are associated with subducting ridges or fractures that intersect the trench; for example, the Nazca Ridge was the site for the 1942 Mw = 8.2 and the 2001 Mw = 8.4 events [e.g., Swenson and Beck, 1996; Robinson et al., 2006; Bilek, 2009], and the region of high slip in the 2007 Mw = 8.4 event [Bilek, 2009] and the 1996 Mw = 7.5 tsunami earthquake occurred near the subducting Mendana Fracture [Swenson and Beck, 1999; Bilek, 2009]. Perfettini et al. [2010] identified regions of afterslip that they relate to the subduction of the Nazca Ridge.

[37] No events along the Peru or Chile margins are classified as long duration (Figures 11 and 12) based on our definition of NRD being 1 standard deviation above the mean. Both erosive margins have exhibited either tsunami earthquakes or slow-slip events [Pritchard and Simons, 2006; Bejar-Pizarro et al., 2010; Vingy et al., 2011] in the past and are characterized by thin sediment covers [Schweller et al., 1981; Clift and Vannucchi, 2004; Scholl and von Huene, 2007]. The longest duration Peru event observed occurs within ~100 km of the 1960 tsunami earthquake [Pelayo and Wiens, 1990; 1992], but we do not see any long-duration events in the afterslip region [Polet and Kanamori, 2000] (Figure 14c). The longest duration events in Chile fall at the edge of observed afterslip (Figure 14d).

4.8 Chile

[38] Our Chile study region extends from 18°S to 36°S (Figure 12). The Chile trench is an extension of the Peru-Chile margin described above. The Nazca plate along this segment contains numerous features such as the Nazca Ridge, Iquique Ridge, and the Juan Fernandez Ridge (JFR) that collide with the margin. The JFR intersects the Chile trench at 33°S, forming a topographic high that separates the trench into two distinct sedimentary domains [Völker et al., 2006]. North of the JFR, the trench contains no or little sediments (pelagic) [Mix et al., 2003]. South of the JFR, the trench becomes partially or completely filled with turbidite sediments [Schweller et al., 1981; Bangs and Cande, 1997].

[39] Studies of postseismic slip [Pritchard and Simons, 2006; Bejar-Pizarro et al., 2010; Vingy et al., 2011] attributed the occurrence of afterslip to the presence of sediments along the Chile trench. The intersection of the Iquique Ridge and the JFR with the trench plays an important role in the earthquake processes. For example, the Iquique Ridge was identified to be the location of high slip for the 2007 Mw = 7.7 Chile event [Bilek, 2009].

4.9 Comparison With Other Studies

[40] Convers and Newman (CN) [2011] examine a set of global subduction zone earthquakes to determine which events have a low energy to seismic moment ratio (inline image), another characteristic of tsunami earthquakes. They find a global value for thrust earthquakes of inline image = −4.74 and a specific value for Java of inline image = −4.91. Because long-duration and tsunami earthquakes should both have low inline image, we compare their low inline image events with our catalog (Figure 15). There are 38 events in common for both catalogs. Sixteen of these common events are not low inline image or long duration in either catalog; 16 of these are classified as low inline image in the CN catalog but do not fall within our long-duration classification. Of these; six events have NRD above the mean but within 1 standard deviation so fall just outside of our long classification. Five of the common events are classified as low inline image and long duration in both catalogs, and one event is long duration but not low inline image (see Supporting Information Table S1). The difference in number and location of events identified as having long NRD in our catalog and those with low inline image in the CN catalog is probably due to the different threshold and method used to identify tsunami-like events and rupture duration, respectively. While we use a threshold based on the mean duration + 1 standard deviation to define long NRD events, Convers and Newman [2011] based their threshold on the global mean of the energy to seismic moment ratio of the events. We obtain the durations from the resulting STF obtained through the deconvolution, while they estimate rupture duration from the cross-over time. It is probable that the differences in the method used to approximate the rupture duration of the events is leading to different values of NRD and inline image, thus the dissimilarity in number and location of long events in both catalogs.

Figure 15.

Comparisons of location of long-duration events found in this study and by Convers and Newman [2011]. Red open diamonds show long-duration events as found by Convers and Newman [2011]. In all regions (except Peru and Chile), we see a similarity in location of long NRD events found in this study and in Convers and Newman [2011]. Symbols and labels same as in Figures 5-12.

5 Discussion

[41] Based on the locations of the long-duration events in our study and comparison with various geologic and tectonic parameters, we suggest that no single feature strongly correlates with the long-duration (slower rupture velocity) earthquakes. Long-duration and tsunami earthquakes occur at most subduction zones with a range of ages (~40–130 Ma), sediment thicknesses (<200 m to >2 km), sediment type (pelagic, terrigenous, and combinations of these), and rates of subducted sediment volumes (5850–81,000 km3/Myr, Table 1). This last point is important for reconciling models for tsunami earthquakes, as many studies suggest rupture through low strength sediments as important in triggering tsunami earthquakes or long-duration earthquakes. This idea may be correct, but we suggest that a very small thickness of sediment within the system is all that is required.

[42] The majority of the long-duration events occur in erosive margins (65%) rather than accretionary margins (35%), but this is likely due to the fact that four of our margins with long-duration events are erosive and only two are accretionary. There is a small difference in the rupture duration between erosive and accretionary margins. The mean NRD are similar (4.8 s in erosive margins, 4.6 s in accretionary), but the maximum NRD is longer in erosive margins (14.8 s) compared to 11.9 s in accretionary margins.

[43] Spatial correlation between the long-duration events and location of tsunami earthquake rupture or slow-slip/afterslip processes is limited. Approximately 19% of the long-duration events occur within or at the edges of known tsunami earthquake rupture zones, but long-duration events are not the only type of earthquakes within these rupture zones. The other long-duration events occur in regions without historical records of tsunami earthquakes. It is possible that we have simply not recorded long enough to capture a larger magnitude tsunami earthquake in these regions. Another explanation is that the patches of the fault that are able to rupture with slow velocities are highly variable, with the size of these patches limiting the size of the earthquake. Thus, some of these regions may only be able to produce smaller magnitude (M < 7) slow events. Approximately 25% of the long-duration events occur in areas of defined afterslip. It is possible that the conditions required for afterslip and slow slip are similar to those required to produce long-duration and/or tsunami earthquakes, but given the weak correlation, these are likely not linked.

[44] There is limited spatial correlation between the locations of the long-duration events and subduction of topographic features. Roughly 60% of the long-duration events occur in or near topographic features. Fluids may play a role, with subducting topography such as fractures, ridges, and seamount acting as permeable zones that allow seawater to penetrate deeply into the crust [e.g., Peacock, 1990; Rupke et al., 2004]. Fluids within the crust can escape with increasing temperature and pressure through various dehydration reactions and/or flow through permeable features [e.g., Moore and Saffer, 2001]. These features can also cause deformation of the upper plate [e.g., Wang and Bilek, 2011], thus disrupting the plate interface contact and likely reducing seismic and rupture velocities. Our dataset cannot distinguish the causes for this link between long-duration events and subduction of topographic features, but we suggest future study of the role of these features is warranted.

[45] Other long-duration events of interest are those within our deep classification (>26 km). In a few cases, deep long-duration events occur within past tsunami earthquake rupture zones (one in Sumatra, two in Alaska) or in regions of other shallow long-duration events (three in Tonga, two in Izu). Also, four deep long-duration events, in both Sumatra and Kuril, are in regions of afterslip. These connections suggest that conditions important on the shallow interface are present deeper on the interface within limited geographic regions. Thus, heterogeneous patches of fault conditions, such as variable subducting features and sediment, can extend in both the along-strike and down-dip directions down to ~40 km in the subduction zone.

6 Conclusions

[46] We estimate rupture durations and focal depths for 427 relocated subduction zone megathrust earthquakes for comparison with along-strike variations in geologic and tectonic parameters. We focus on those events with NRD that are 1 standard deviation above the mean, classifying these as our long-duration events for interpretation. We find long-duration events at all depths within the seismogenic zone, not just at the shallow end. Our long-duration events are associated with various sediment thickness and/or type as we find no correlation with incoming sediment thickness or type; therefore, we suggest that sediment thickness and type are not critical factors for long-duration events. There is limited spatial correlation with regions of past tsunami earthquakes, regions of observed afterslip, and subducting bathymetric features. The relatively stronger correlation between long-duration events and subducting bathymetric features suggests a role of fluids in producing long-duration events.