Pure appl. geophys. 154 (1999) 433–456 0033–4553/99/040433–24 $ 1.50+0.20/0

Comparison of Depth Dependent Fault Zone Properties in the Japan Trench and Middle America Trench SUSAN L. BILEK1 and THORNE LAY1

Abstract—Spatial variations in mechanical properties of the interplate thrust faults along the Japan and Middle America subduction zones are examined using teleseismic broadband earthquake recordings. Moment-normalized source duration is used to probe rigidity variations along the interface. We invert body waves to estimate source depth and source duration for 40 events in the Japan subduction zone and 38 events in the Middle America subduction zone. For both areas, there is a systematic decrease in source duration with increasing depth along the subduction zone interface. This is most likely a result of variation in properties of sediments on the plate contact. Variations in source duration are greatly reduced at depths greater than 18 km in both regions. Enhanced spatial heterogeneity at shallow depth may reflect variations in plate roughness, sediment distribution, permeability of the fault zone, and stress. Key words: Fault zone properties, subduction zones, Japan trench, Middle America trench.

Introduction Earthquakes in subduction zones have generated extensive interest because of their spatial and temporal variability. Common questions addressed concern the nature of along-strike variations in the earthquake rupture processes in various subduction zones. This has prompted numerous studies, including many focusing on the failure characteristics of the Japan subduction zone (e.g., SCHWARTZ and RUFF, 1985; RUFF and KANAMORI, 1980; LAY et al., 1982; RUFF, 1992; TANIOKA et al., 1996). This work has provided insight into rupture processes of subduction events, especially in terms of the role of asperities in the rupture process, whether large earthquakes rerupture the same asperities, and variations in rupture directions during an event. Other studies have focused on unusual subduction zone earthquakes, particularly events that produce large tsunamis. KANAMORI (1972) defines tsunami

1 Institute of Tectonics and Earth Sciences Department, University of California, Santa Cruz, CA 95064, USA. Fax: 408-459-3074, E-mail: [email protected]

434

S. L. Bilek and T. Lay

Pure appl. geophys.,

earthquakes as those earthquakes that generate a significant tsunami and have a discrepancy between the surface-wave magnitude and a larger moment magnitude. FUKAO (1979) suggests that low rigidity material in the source region is needed to generate tsunami earthquakes, such as the 1963 Kurile earthquake and the 1975 Hokkaido earthquake. KANAMORI and KIKUCHI (1993) define two types of tsunami earthquakes based on the tectonic environment. Along non-accreting margins where most incoming sediments are being subducted, earthquakes cause slip along the weak sediment-filled interface, breaking to the surface to generate a tsunami. Conversely, along margins where an accretionary wedge is well formed, earthquakes occurring in the wedge can cause slumping and ocean floor motions which in turn can generate tsunamis. Several tsunami earthquakes have occurred historically, most notably the 1896 Sanriku event offshore northern Honshu, and the 1994 event offshore Nicaragua. These events have been studied using both tsunami and seismic records, and have in common anomalously long source durations resulting from slow rupture processes and occurrence in areas where sediments are being subducted in the subduction zone (e.g., SATAKE, 1994; KANAMORI and KIKUCHI, 1993; IDE et al., 1993; VELASCO et al., 1994; KIKUCHI and KANAMORI, 1995; IHMLE´, 1996; TANIOKA and SATAKE, 1996). In the Japan subduction zone, other recent earthquakes have been found to cause slow afterslip over time scales of up to a year. GPS and extensometer measurements suggest that earthquakes offshore northern Honshu in 1992 and 1994 both had significantly large amounts of slow slip, occurring up to one year following the event (KAWASAKI et al., 1995; HEKI et al., 1997; HEKI and TAMURA, 1997). These events, along with the occurrence of both tsunamigenic and nontsunamigenic events in close proximity, indicate variability in subduction zone earthquake processes that must be related to variations in mechanical properties. However, because most work has focused on along strike variations and the study of unusual earthquakes, only a few studies have tried to describe how earthquake rupture processes change with depth along the subduction zone. Depth variations are likely to exist given thermal effects, dehydration and induration of sediments, variations in overthrust plate properties, phase transitions, and changes in frictional behavior. ZHANG and SCHWARTZ (1992) examine how moment release during large interplate earthquakes changes as a function of depth in subduction zones around the world, finding large regional variations in the depth distribution of interplate moment release. TICHELAAR and RUFF (1993) use intermediate size earthquakes to estimate maximum interplate seismogenic zone depths, which range from 20 – 55 km globally. EKSTRO¨M and ENGDAHL (1989) determine earthquake parameters for intermediate depth events to examine the stress distribution in the central Aleutian Islands. VIDALE and HOUSTON (1993) and BOS et al. (1998) examine the relationship between source duration and depth for events between 100 and 650 km depth. These studies provide some insight into the depth dependence of subduction zone earthquake properties, however, there has not yet been a system-

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

435

atic investigation of how earthquake rupture processes and fault zone properties such as rigidity vary as a function of depth in the seismogenic zone. In order to constrain the spatial variation of rigidity on a megathrust, a measurable proxy is needed. Source rupture duration, the duration of seismic energy release during an earthquake, is one parameter that can be used. The rupture duration depends on the rupture velocity of the particular event, which in turn is typically found to scale with shear velocity (KANAMORI and ANDERSON, 1975). Because the shear velocity is defined as the square root of the rigidity divided by the material density, the source duration can be used to study depth dependence of rigidity along subduction zone interfaces. Some caution needs to be taken with the use of source duration, as source duration will also be affected by earthquake rupture area and type of rupture. Thus, we need to scale the measured source durations for total seismic moments, and ideally determine the actual rupture area of each event. The latter is very difficult to do for all but the largest events, thus a trade-off exists between rupture velocity and rupture area if stress drop is not constant. Our objective for this study is to examine subduction zone interface properties for any depth dependence by using scaled source duration as a proxy for rigidity. The northern Japan subduction zone is the site of our earlier study (BILEK and LAY, 1998). Here we will review that data set and compare it with data from the Middle America subduction zone to look for any common trends for these areas.

Tectonic Setting The Japan and Middle Americas subduction zones have been well studied using drilling, seismic reflection, and earthquake studies. Additionally, both areas have abundant seismicity as well as earthquakes with unusual rupture processes. Japan Subduction Zone This region is characterized by the Pacific plate subducting beneath the Okhotsk plate at a rate of approximately 8 cm/yr (SENO et al., 1996) (Fig. 1). Using seismic reflection studies and the presence of only a small accretionary wedge, it has been determined that most sediments are being subducted to at least a depth of 12 km (VON HUENE and CULOTTA, 1989; VON HUENE et al., 1994; SUYEHIRO and NISHIZAWA, 1994). The subduction zone geometry changes along strike of the trench, and there is a change in the morphology of the subducting Pacific plate, as horst and graben structures are being subducted between 38.6° and 39.7° (TANIOKA et al., 1997). Several earthquakes in this region have had unusual rupture processes, and there seems to be both along-strike and downdip variability in their occurrence.

436

S. L. Bilek and T. Lay

Pure appl. geophys.,

Rupture areas of the well-documented unusual earthquakes are shown in Figure 1. The 1896 Ms 7.2 earthquake generated a very large tsunami, much larger than expected for that magnitude (ABE, 1981; ABE, 1994), characteristic of tsunami

Figure 1 Map of Japan subduction zone study area, modified from BILEK and LAY (1998). Japan trench location from bathymetry maps of the region. Inset shows regional geography. Rupture areas of previous large earthquakes, the 1968 Tokachi-Oki, 1994 Sanriku-Oki, and the 1896 Sanriku tsunami earthquake, are estimated by the dashed lines. Convergence rate of Pacific plate from SENO et al. (1996).

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

437

earthquakes. Modeling of the tsunami records indicates that this event had a shallow, slow rupture through subducted sediments (TANIOKA and SATAKE, 1996). Other events occurring in this region failed to produce large tsunamis, even though they were larger magnitude events. The 1968 Mw 8.2 and 1994 Mw 7.7 events did not produce anomalously large tsunamis, but both did have unusual rupture characteristics. Both exhibited variability in their updip and downdip rupture directions and showed evidence for a slow initiation phase along the updip section of the rupture zone before the main energy pulse (SCHWARTZ and RUFF, 1985; TANIOKA et al., 1996; HARTOG and SCHWARTZ, 1996). TANIOKA et al. (1996) and HARTOG and SCHWARTZ (1996) conclude that these two events ruptured different asperities. The 1994 event was unusual with respect to post-seismic slip as well. Post-seismic ‘‘slow’’ slip has been measured using various GPS techniques for time periods spanning a few days to one year. This post-seismic slip has generated more moment release than the mainshock event (KAWASAKI et al., 1995; HEKI et al., 1997; HEKI and TAMURA, 1997). Additional complexity arises around the actual rupture dynamics of this event, as NAKAYAMA and TAKEO (1997) estimate a spatially varying rupture velocity, finding an average velocity of 1.8 km/s shallower than about 20 km depth and 3.0 km/s at greater depths. These unusual earthquake rupture characteristics, along with the tsunamigenic events, suggest changing fault zone properties in this region. Middle America Subduction Zone In this region, the Cocos plate is subducting beneath the Caribbean plate at the Middle America trench at a rate ranging from 7 cm/yr offshore Guatemala to 9 cm/yr offshore Costa Rica (PROTTI and SCHWARTZ, 1994) (Fig. 2). Two spreading ridges are located at the edges of the Middle America trench, Tehuantepec Ridge to the north of Guatemala and the Cocos Ridge to the south of Costa Rica, where the Middle America trench terminates. Early work suggested that there is a change in sediment accretion around the Tehuantepec Ridge, with an accreting margin to the north of the ridge and a non-accreting margin to the south (AUBOUIN et al., 1982). Off the coast of Nicaragua, neither a large accretionary wedge nor a large amount of sediment has been observed, suggesting a non-accreting margin in that area (VON HUENE and SCHOLL, 1991). Recent ODP drilling off the Nicoya Peninsula also provides evidence for a non-accreting margin offshore Costa Rica. Data from ODP Leg 170 (KIMURA et al., 1997) indicate that the margin wedge, once proposed to be formed by accreted sediment (SHIPLEY et al., 1992), is not accreted sediment from the incoming plate but deformed sedimentary material from the overlying plate. This suggests that most of the approximately 400 m of sediment is being subducted with the downgoing plate (SILVER et al., in preparation, 1999).

438

S. L. Bilek and T. Lay

Pure appl. geophys.,

Figure 2 Map of Central America study area, showing Middle America trench as located on bathymetry maps. Ellipse indicates the estimated rupture area of the September 1992 tsunamigenic earthquake occurring offshore Nicaragua (VELASCO et al., 1994). Nicoya Peninsula is shown by letters N.P. Convergence rate of Cocos plate from DEMETS et al. (1990). Inset gives regional geography with study area enclosed in dashed box.

Other features in this subduction zone environment include varying topography on the subducting Cocos plate as well as changes in dip angle. Horst and graben structures have been imaged on the Cocos plate both near Guatemala (AUBOUIN et al., 1982) and offshore Costa Rica to at least a depth of 10 km (HINZ et al., 1996). Additionally, many seamounts have been imaged on the Cocos plate north of the Cocos Ridge (HINZ et al., 1996). The dip angle of the subducting plate also changes due to a tear in the plate north of the Cocos Ridge, with the northwestern portion of the plate dipping 20° more steeply than the southeastern portion (PROTTI et al., 1996). This region is similar to the Japan subduction zone area in terms of unusual earthquake ruptures. The 1992 Mw 7.62 earthquake offshore Nicaragua generated a large tsunami and had a larger tsunami magnitude (Mt 7.9) than surface wave magnitude (Ms 7.2), similar to the 1896 Sanriku earthquake (IDE et al., 1993) (Fig.

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

439

2). Because of the recent occurrence, the 1992 event provided an opportunity to study a tsunami earthquake using modern recordings. Several studies examined this earthquake to determine source parameters, finding a range of unusually small rupture velocities between 0.6 km/s–2.3 km/s, with a changing rupture velocity with rupture direction (IDE et al., 1993; VELASCO et al., 1994). Anomalously low static stress drops were also calculated for this event, ranging between 1–7 bars (IDE et al., 1993; VELASCO et al., 1994). SATAKE (1994) models the tsunami records in order to explain the discrepancy between the Ms and Mw for this event. He finds that by changing the rigidity of the fault zone materials from a normal 3×1010 N/ m2 to a smaller 1 ×1010 N/m2, he can estimate similar moment magnitudes as measured with other methods, as well as generate a tsunami by having the earthquake rupture to the ocean bottom through low rigidity sediments, with a very narrow fault plane. This mechanism is consistent with the tsunami mechanism proposed by KANAMORI and KIKUCHI (1993). Other events in this same region have not produced tsunamigenic earthquakes, similar to the variability offshore Honshu, which again suggests variability in fault zone properties.

Data We used the Harvard CMT (Centroid Moment Tensor) catalog in order to find those events consistent with interplate thrusting along the subduction zone interfaces. The major criteria for choosing events included a source depth less than 80 km, a fault plane approximately parallel to the local trench axis (allowing for a 20° variation in this strike), dip of the fault plane less than or equal to 35°, and a rake consistent with a thrust mechanism (909 35°). These criteria are similar to those in other studies (ZHANG and SCHWARTZ, 1992; TICHELAAR and RUFF, 1993). These criteria discriminate most interplate events in the earthquake catalog, however, we realize that a few non-interplate events could be incorrectly classified as interplate events. Thus, we consider improved depth estimates to better determine whether the events actually occurred on the interface. A catalog search for the Japan trench study area used a latitude range of 37°– 43° and a longitude range of 140°–145°. This produced a list of 70 events of Mw =5.0 – 7.7 between 1989 – 1995. The magnitude bounds were chosen because of signal-to-noise ratio concerns, and the time window is defined so that enough broadband digital stations existed to provide good azimuthal coverage. An initial analysis of the corresponding catalog earthquake duration and depth estimates revealed large scatter, and we decided to perform detailed wave-form analysis in order to improve the depth and duration estimates. Thus, we chose the 40 events for which seismograms had a good signal-to-noise ratio and good azimuthal distribution for further study. A similar search was performed for events in the Central America region, using latitude bounds of 6°–16° and longitude bounds of

440

S. L. Bilek and T. Lay

Pure appl. geophys.,

Table 1

Longitude

Mw

Depth (km)

Normalized source duration (s)

143.27 143.38 143.32 143.39 143.35 143.43 143.90 143.69 143.63 142.46 142.24 143.50 143.59 143.68 142.74 142.14 143.75 142.09 142.58 143.49 142.37 143.54 143.02 143.06 143.29 143.38 143.25 142.04 142.05 142.55 143.36 143.45 142.84 143.33 143.42 143.70 143.52 142.14 144.34

5.63 6.31 5.50 5.87 5.94 5.63 5.17 6.15 5.41 5.54 6.95 5.33 6.41 6.21 6.14 5.87 6.38 5.35 5.47 5.96 5.67 6.18 5.92 5.98 5.59 6.89 5.89 5.38 6.16 6.38 5.65 5.85 5.20 6.44 5.83 6.33 5.95 5.26 6.14

21 14 12 31 6 12 7 8 8 43 37 4 9 6 6 33 5 44 41 30 32 5 13 18 12 7 14 41 47 50 15 17 28 14 24 7 12 44 9

1.49 2.74 5.20 3.39 6.28 4.47 2.52 6.51 3.83 1.66 1.96 10.59 4.23 4.95 3.81 3.41 4.42 4.10 1.80 3.06 2.84 6.35 4.28 6.00 4.66 3.49 4.44 0.99 0.81 0.63 4.36 4.62 1.84 2.34 1.19 2.65 3.08 1.14 7.47

(b) Events in the Middle America subduction zone 97/12/22 10:03:51.5 13.76 −90.39 96/12/17 20:20:33.3 12.47 −88.92 96/12/14 21:15:35.5 12.73 −88.79 96/11/17 19:17:41.9 11.08 −86.05 96/09/04 19:06:49.8 9.36 −84.27 96/08/28 17:16:17.1 9.38 −84.31

6.10 5.60 5.71 5.92 6.18 5.87

25 10 13 27 20 21

4.36 4.63 4.07 3.20 3.16 3.38

Date

Origin time

Latitude

(a) Events in the Japan subduction zone 95/12/30 20:45:16.0 41.15 95/12/30 12:11:07:4 40.79 95/11/23 14:13:13:5 40.24 95/09/26 07:14:37.7 41.79 95/07/08 05:42:56.9 39.64 95/05/25 09:11:35.3 40.21 95/04/25 13:32:52.1 40.17 95/02/23 05:01:25.9 39.66 95/01/21 06:56:33.4 40.56 95/01/11 07:48:20.4 41.97 95/01/06 22:37:37.9 40.23 95/01/06 16:41:32.7 39.15 95/01/01 06:59:54.8 40.63 94/12/28 22:37:49.2 40.25 94/12/28 20:52:28.9 40.03 94/08/14 09:06:52.4 38.75 94/04/08 01:10:41.6 40.55 93/10/28 01:52:06.4 41.60 93/09/11 04:55:33.1 42.01 93/03/25 07:08:18.8 41.79 93/02/25 09:27:43.0 40.24 92/07/29 04:30:49.3 39.44 92/07/25 02:53:29.7 38.65 92/07/18 21:18:56.8 39.29 92/07/18 20:55:13.3 39.60 92/07/18 08:36:59.1 39.36 92/07/16 00:00:40.8 39.33 92/07/13 14:19:05.8 41.49 92/07/12 11:08:56.0 41.44 91/11/26 19:40:48.2 42.02 90/01/10 03:09:15.8 39.69 89/11/04 20:12:04.9 39.08 89/11/01 18:25:35:9 39.80 89/10/29 05:25:41.1 39.55 89/10/29 03:09:13.5 39.58 89/10/27 01:45:58.2 39.74 89/10/26 17:06:44.6 39.79 89/08/22 07:56:09.7 41.45 89/01/22 22:20:16.8 41.79

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

441

Table 1 (Continued)

Date

Origin time

Latitude

Longitude

Mw

Depth (km)

Normalized source duration (s)

96/08/27 96/04/01 96/03/03 95/09/06 95/01/20 94/06/29 94/05/01 94/03/12 93/09/30 93/09/19 93/09/14 93/09/13 93/09/11 93/09/10 93/09/10 93/09/10 93/09/03 93/09/03 93/07/21 93/07/16 93/05/20 92/09/05 92/08/11 92/08/10 92/05/30 92/04/05 92/02/21 91/06/21 91/03/16 90/04/28 90/03/25

02:12:40.2 06:10:52.9 16:37:31.5 22:48:52.6 11:29:08.3 12:00:57.9 21:26:05.3 23:00:29.2 18:27:53.3 14:10:59.5 03:59:28.0 22:58:13.0 19:29:42.7 19:12:54.8 18:58:52.9 17:28:05.7 12:35:01.5 03:16:06.7 22:06:32.8 12:55:36.2 04:55:55.5 21:48:11.9 05:23:48.7 06:09:14.5 16:30:02.5 14:13:39.9 22:09:52.7 06:27:37.5 06:02:10.6 01:23:12.1 13:22:54.9

13.51 14.55 11.90 14.98 13.96 11.40 10.74 11.93 15.51 14.56 14.30 11.18 14.35 14.73 14.44 14.57 14.55 14.40 13.28 10.74 13.81 11.97 11.74 11.72 14.51 11.19 13.91 13.30 10.19 8.86 9.89

−91.28 −93.53 −86.77 −94.25 −91.75 −86.31 −85.93 −87.02 −94.71 −93.28 −93.06 −86.56 −92.31 −92.68 −92.72 −92.73 −92.82 −92.72 −90.86 −86.31 −91.19 −87.34 −87.47 −87.33 −92.90 −86.61 −92.28 −89.79 −85.15 −83.56 −84.89

5.68 5.74 6.67 5.78 5.65 5.60 5.78 5.60 6.45 6.40 5.56 5.75 5.72 7.21 6.01 5.74 6.72 5.45 5.59 5.28 5.74 5.97 5.47 5.67 6.30 5.76 5.74 5.80 6.27 6.35 7.29

11 7 16 5 16 24 17 14 13 13 7 13 20 13 7 7 11 22 10 19 20 13 8 10 22 15 6 21 10 10 11

4.24 5.22 9.03 7.49 4.36 6.16 4.99 4.64 4.63 3.70 6.45 5.18 2.70 2.89 7.68 9.23 4.27 3.64 4.66 6.70 5.23 7.08 7.19 4.26 4.16 5.14 7.84 4.91 6.40 4.56 2.41

−83° – −95°. Thirty-eight events were selected for study. Teleseismic vertical component broadband records were used for all inversions, and we manually picked the P-wave onset times. Table 1 lists the events used for both study areas.

Methodology We perform point source wave-form inversions with constrained fault geometries to estimate the two parameters, source duration and depth, needed for this study. Because depth estimates are constrained by the time differences between the

442

S. L. Bilek and T. Lay

Pure appl. geophys.,

P wave and surface reflections, teleseismic recordings of these phases are used in these inversions. The wave-form inversion is based on the seismogram represented in terms of the convolution of a point source Green’s function and the moment rate, or source time function. We use the best double-couple of the moment tensor solution determined by the Harvard CMT catalog to compute the P-wave Green’s function and deconvolve this from the seismogram to get the source time function. The Green’s function is dependent on source depth, and thus performing the inversion for a range of depths allows us to find the optimal depth for each event based on minimizing misfit between the data and synthetic seismograms. We performed inversions for a range of 12–25 depths, usually starting the inversion at the Harvard catalog depth and changing the depth in later iterations in order to converge on the best depth. We could also invert for focal mechanism, however our data sets are usually too small to do this stably, and the Harvard CMT solutions are relatively consistent, tectonically sensible, and presumably realistic if the events are indeed interplate thrust events. The velocity model used in computing the Green’s functions is important, as there are trade-offs between the source depth and the velocity. We use a P-wave velocity of 6.0 km/s as an average P-wave velocity from the hypocenter to the water layer interface. Water depth above the source is estimated from bathymetry data. Rather than inverting single station records, we chose instead to invert multiple azimuthally distributed records simultaneously, as this reduces the trade-offs between the source depth and the source duration (TICHELAAR and RUFF, 1991; CHRISTENSEN and RUFF, 1985). For further details of the inversion process, see RUFF (1989); RUFF and MILLER (1994); TICHELAAR and RUFF (1991). Once the optimal depth for the event was determined, the source duration is estimated from the corresponding source time function. We measure source duration as the time extent of the first large pulse of moment release, from the onset to the return to a baseline level. The limited bandwidth of the data often causes some overshoot in the source time function which complicates identification of the rupture duration, but we strive to be consistent in our measurements. These source duration estimates need to be corrected for earthquake size, as we expect about cube-root scaling with seismic moment (KANAMORI and ANDERSON, 1975). We correct for moment by dividing the measured source duration by the cube root of the Harvard CMT determined seismic moment (M0 ), normalized to a moment magnitude 6.0 event. This scaling differs from that in our earlier study (BILEK and LAY, 1998); however, there is no qualitative difference in the results. The cube-root scaling law is more widely used (e.g., KANAMORI and ANDERSON, 1975; EKSTRO¨M and ENGDAHL, 1989; VIDALE and HOUSTON, 1993; BOS et al., 1998), thus allowing our data to be compared with other studies. We are confident that this normalization scheme removes the average effects of magnitude, as there is no relationship between our normalized durations and log M0 for our data.

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

443

In6ersion Results Japan Inversions provide source duration and depth estimates for forty events seaward of Honshu. Figure 3 shows inversion applications for six representative events. These plots show the variability in the data quality as well as a few of the complications in the inversion process. As seen in Figures 3a,b,c, there can be a range of depths where misfit is low. For these cases, we add error bars to the depth estimates to account for these low misfit ranges. We measure source duration at the optimal depth and take into account the range of duration over the depth range, assigning corresponding error bars for source duration. As it is sometimes difficult to pick the exact termination of the energy pulse, as in Figure 3c, we also use the character of the source time function to assign relative errors due to the difficulty of picking the termination points. Wave propagation effects can also cause complications in determining the source time function for the shallowest events, as the closer arrivals of P, pP, and sP along with the complex structure of the margin wedges could cause the measurement of longer source time functions for these events. Another complexity arises in estimating the actual values for source duration and source depth. In a few cases, inverting the data for a range of depths results in a double minimum of misfit (Figs. 3d,e,f). To determine the optimal depth for these events, we examine the resulting source time functions for the depths of low misfit, choosing the simplest source time function which has the most moment release in a pulse near the beginning of the source time function. We also look at the estimated moment produced in the inversion, which should be comparable to catalog listed values. Because catalog moments are calculated with many stations at different frequencies, our inversion moments do not match exactly the catalog values, but they are generally similar. This subset of inversion results is representative of the entire data population. There is a large range in number of input seismograms and data quality, and consequently, variation in misfit values. Those events with higher misfits had fairly low signal-to-noise ratios and/or few stations with poor azimuthal coverage. Figure 4 shows results for the Japan subduction zone data set. Normalized source duration and event depth are plotted with respect to event distance from the Japan trench axis. The trench axis was determined from bathymetry data. The data generally define the dipping plate interface, with about a 10-km scatter in the interface surface at shallow depth and 5-km scatter at greater depth. Some scatter is expected simply due to variations in plate dip along the subduction zone and variations in position of the trench axis due to sedimentary wedge structure. Those events lying off of this ‘‘plane’’ have been examined for the possibility of a double minimum in misfit, which might allow for a change in depth, but none of the five

444

S. L. Bilek and T. Lay

Pure appl. geophys.,

exhibit this feature. Thus, these events could possibly be mislocated, be intraplate events with a mechanism fortuitously similar to that of an interplate thrust, or, in the case of the two events above the defined interface, could be thrust events in the

Figure 3 Inversion results for six events in the Japan data set that span a range of depth and source durations. Top graph shows misfit as a function of depth for each inversion result. Bottom plot shows the best event depth, the misfit, the source time function for the best depth (left), source time function with uncertainty bounds (middle) and data in solid lines, with superimposed synthetic seismograms in dashed lines (right).

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

445

Figure 4 Normalized source duration and depth as a function of distance from the Japan trench, modified from BILEK and LAY (1998). Error bars show range in depths from the range of misfit from the inversion. Source duration has been corrected for moment and normalized to a Mw 6.0 event. Dashed box indicates those events which define the plate interface. In general, longer duration events occur at depths less than 20 km, with shorter duration events occurring at greater depths along the interface.

accretionary wedge. Another feature evident in Figure 4 is the systematic change in scaled source duration with depth. Longer duration events all occur at depths less than about 20 km, and duration decreases with increasing depth. There is significant scatter in duration at the shallow depths, which may represent heterogeneities in the shallow portion of the subduction zone. Figure 5a shows the same data set plotting the scaled source durations as a function of source depth. The trend of decreasing source duration with increasing depth is apparent, however, there is a rather complex relationship. It is possible to characterize the behavior as a linear decrease in duration with depth with large scatter in the shallow portion, or as a step-like change in the relationship around 20-km depth, or a combination of the two. In either case, if we assume uniform scaled rupture areas for all of these durations, these results indicate rupture velocity and thus rigidity increases with depth along the plate interface, as well as the presence of heightened spatial heterogeneity at shallow depths. If instead we assume that all events have an equal rupture velocity, but have varying rupture area, we can look at static stress drop variations with depth. The stress drop, being related to fault displacement divided by fault length or width, can be related to the seismic moment and thus the rupture velocity and source duration. Using a circular crack model and assuming that the rupture velocity is equal to the shear velocity, 6s , we can use the relationship Ds = 7P*M0 /16*6 3s *t 3

(1)

446

S. L. Bilek and T. Lay

Pure appl. geophys.,

to find the static stress drops (Ds) for these events, where t is the estimated source duration. These stress drop-depth results are shown in Figure 5b, calculated using the determined source durations and a constant 6s of 3.5 km/s. Of course, the shear

Figure 5 (a) Normalized source duration as a function of depth for the Japan data set, modified from BILEK and LAY (1998). Open symbols indicate those events which do not lie on the plate interface. Error bars indicate range in duration and depth as determined by range of misfit from the inversion. Shallow events generally have longer source durations with duration decreasing with depth. (b) Stress drop as a function of depth in the Japan data set, with stress drop calculated from the determined source durations using Equation (1). Stress drop increases with depth.

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

447

velocity may vary with depth due to rigidity variations, which might then cause rupture velocity variations, but we assume that it is constant because we lack constraints on this. Again we see a variation with depth, in this case with general increase in stress drop as depth increases. However, we emphasize that we have not independently constrained the fault area for each event, as is needed to resolve the trade-offs between rupture area and rupture velocity (VIDALE and HOUSTON, 1993). Thus Figure 5 gives two end member characterizations of the variation in properties. The key point is that clear, readily detected variations with depth exist, and this forms the basis for investigating fault zone properties that have, up to now, been ignored for lack of measurements. We also examined the possibility of along-strike variations in this durationdepth relationship, possibly caused by changes in the subduction geometry in this region. Figure 6a shows focal mechanisms of the thirty-five events located on the interface, scaled in size by source duration. Most longer duration events occur near the trench axis, not a surprising finding given Figure 4. There does seem to be a change in the northern region of the study area, with shorter events occurring above 41° latitude. This could be related to the change in dip of the subducting Pacific plate, the bend in the trench axis, or the limited number of events in this data set. More events are needed in order to fully examine these along-strike variations, but this will require use of older events with fewer or no digital recordings, or examination of smaller events that have poor teleseismic signals.

Middle America Trench We used the same procedure to examine thirty-eight events from the Middle America subduction zone. The same inversion complexities seen in the Japan data set also arise with this data set. Figure 7 shows a representative subset of the event population for comparison to the Japan data set, again having some events with a double minimum in misfit (Figs. 7b,e) as well as variation in overall optimal misfit (Figs. 7c,f). Figure 8 shows the normalized source duration and depth with respect to the distance from the Middle America trench. The events vaguely define a dipping interface; however, there is greater scatter than for Japan. Much of this may result from lateral variation in the trench morphology and sedimentary wedge structure, with a much larger region being sampled than for Japan. Again, most of the slower or longer duration events are shallower than about 20 km, but there are several short to intermediate source duration events intermingled at shallow depths. The event population has longer source durations on average than those from the Japan data set. Another difference is the thrust events only go down to about 30 km depth. This is consistent with an earlier study of seismic

448

S. L. Bilek and T. Lay

Pure appl. geophys.,

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

449

Figure 7 Inversion results for six events in the Central America data set. The format follows that of Figure 3. Figure 6 (a) Analyzed events plotted in map view for the Japan data set. Only events that lie on the plate interface are plotted. Symbols indicate focal mechanism of the Harvard CMT catalog plotted at the NEIC (National Earthquake Information Center) epicentral locations. These symbols are scaled to represent the normalized source durations estimated in this study. Most long source duration events are located close to the trench. There is a possibility that the events furthest to the north have a different character than those to the south due to the change in subduction geometry. (b) Analyzed events for the Central America data set plotted in map view. Symbols indicate focal mechanism of the Harvard CMT catalog plotted at the NEIC epicentral locations. These symbols are scaled to represent the normalized source durations estimated in this study. There does not appear to be a spatial pattern in the location of the long duration versus short duration events.

450

S. L. Bilek and T. Lay

Pure appl. geophys.,

Figure 8 Normalized source duration and depth as a function of distance from the Middle America trench. Slower duration events are generally located in the shallowest 20 km, however, there is a lot of scatter in this data set. It is difficult to determine any spatial pattern for those events with shorter source durations, possibly due to variations in trench structure.

coupling depths, which finds a shallow depth extent of coupling (20–30 km) in the northern portion of this study area (TICHELAAR and RUFF, 1993). In Figure 9a, normalized source duration is plotted as a function of depth. We assume all events are in fact on the thrust interface, but of course, some may be intraplate events (open circles in Fig. 9). Generally, source duration decreases as depth increases. This relationship is very similar to that in the upper 30-km depth range of the Japan data set in Figure 5. Below 18 km the durations are much more tightly clustered. The data gap between 15–20 km depth could be important in defining the nature of the relationship. The range of source durations for this data set span about 6 seconds, which is comparable to the corresponding 0–30 km depth range in the Japan data set, but with about a 1 s baseline shift to longer scaled durations. Figure 9b shows static stress drop as a function of depth, calculated as for Japan, with an increase in stress drop as depth increases. These results indicate a change in fault zone properties as a function of depth, similar to that seen in the Japan data set. We examined the possibility of along-strike variations in this relationship by plotting the focal mechanisms scaled in size by source duration (Fig. 6b). There does not appear to be much of a lateral pattern in the estimated source durations, as is the case near the Japan trench. Expanding the data set in localized regions is needed to further address this issue.

Discussion The observed variations of scaled source duration as a function of depth in the Japan and Middle America subduction zones indicate changes in physical proper-

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

451

ties along the megathrust interfaces. The two regions exhibit similar trends of decreasing source duration with increasing depth, suggesting the mechanism that causes this change is common to both areas. Also, both areas have experienced variability in large earthquake rupture processes that vary with depth, which further strengthens the argument of variable interface properties with depth. However, while the general trend is similar for each area, there are differences in detail between the two areas. So while the basic mechanism might be the same, the magnitude of the processes may vary in each region, possibly because of the differences in details of the tectonic setting. There are several possible causes of the general decrease in source duration with increasing depth. Figure 10 is a schematic of the Japan subduction zone interface

Figure 9 (a) Normalized source duration as a function of depth for Central America data set. Similar to Figure 5a, source duration decreases with depth. This trend seems more linear than that for the Japan data set, and source durations are on average longer for this data set. (b) Stress drop as a function of depth in the Central America data set. There is a general trend of increasing stress drop with depth.

452

S. L. Bilek and T. Lay

Pure appl. geophys.,

Figure 10 Summary of results shown in a schematic of the subduction zone interface in the Japan trench. The seismogenic zone extends to about 50 km depth along the interface. Events with slow source duration occurred on the interface between 5–20 km depth, while events with faster source durations occurred deeper. There is a change in the duration-depth relationship around 25 km, which may reflect in duration and dehydration or offscraping of sediments on the interface.

indicating the relationship between these physical properties and depth. Sediments that are being subducted are less rigid than the rest of the subducting plate, likely causing a longer source duration for an earthquake rupturing through these sediments. Most tsunami earthquakes are postulated as being shallow ruptures through these interface sediments (FUKAO, 1979; PELAYO and WIENS, 1990, 1992; IDE et al., 1993; KANAMORI and KIKUCHI, 1993), and both of these study areas have had at least one tsunamigenic event historically. Subduction of sediments to at least 12-km depth appears to occur near Japan (SUYEHIRO and NISHIZAWA, 1994; VON HUENE and CULOTTA, 1989; VON HUENE et al., 1994), while along the Middle America trench, there is a variable amount of sediment being subducted. For example, offshore Nicaragua and further south, most sediments transported on the oceanic plate are being subducted, as evidenced by the lack of accretionary prism (e.g., AUBOUIN et al., 1982; VON HUENE and SCHOLL, 1991). Other important factors are phase transitions occurring both within the subducting sediment as well as in the subducting plate, particularly the dewatering and dehydration reactions which add water to the interface system. One such transition is the smectite to illite transition. Smectite has low rigidity and retains water within its structure, thus providing a zone of weakness along the interface. When exposed to pressure and temperature increases as experienced with increasing depth along the subduction zone interface, smectite will transform to the stronger mineral illite, which may influence seismicity with depth (VROLIJK, 1990). In the Japan study area, VON HUENE and LALLEMAND (1990) estimate 409 20% smectite present in the subducting sediments.

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

453

Water along the interface is also important in determining fault zone conditions in subduction zones. Shear stresses and friction will be reduced between the subducting and overriding plates when there is high pore pressure. When water is trapped in sediments as they are subducted or when dehydration reactions occur in the subducting materials, pore pressures will increase, thus producing lower rigidity fault zone materials. However, further subduction will produce compaction of most of this material, creating higher rigidity materials and more frictional resistance along the interface. This change from water-filled sediments to more indurated materials along the interface should correspond to a change from longer source durations to shorter ones. Depending on the permeability conditions in the fault zone itself, this effect of changing source durations may be gradual or abrupt, and it may be splotchy, with lateral variations, particularly at shallow depths. Other factors likely play a role in the variability of fault zone properties downdip in the subduction zone. Plate roughness combined with sediment supply influence fault zone properties. In the Japan trench area, TANIOKA et al. (1997) show that large underthrusting earthquakes occur where the subducting Pacific plate has a smooth surface of sediment, while unusual events, such as normal faulting outer rise earthquakes and the 1896 tsunami event, occurred where horst and graben structures are present on the surface of the Pacific plate. Using smooth and rough designations for areas along the Pacific plate from TANIOKA et al. (1997), we see only shallow, longer duration events in the areas where horst and graben structures are being subducted, while events in the smooth sections fit with our general trend of longer duration events at shallow depths progressing to shorter durations events at deeper depths. While this classification does not take into account the change in geometry of the Pacific plate to the north, these results do suggest that plate roughness can play a role in the variability of these properties. It is likely that along the Middle America trench plate roughness also affects the interface properties, especially in areas where seamounts and horst and graben structures are being subducted on the Cocos plate. Other factors are likely important, such as thermal structure of the interface, rate of plate convergence, plate age, and changes in subduction geometry with depth. Further modeling is needed to constrain how these factors interact with those discussed above to influence interface properties. But our observation of a trend of decreasing source duration with increasing depth in two subduction zones is an initial step towards probing these variable interface properties. The next steps are to examine other subduction zones around the world to see if this trend is general, and to pursue more in-depth modeling in hopes of constraining which factors are most important in defining rigidity and frictional properties with depth.

454

S. L. Bilek and T. Lay

Pure appl. geophys.,

Conclusions We model forty earthquakes from the Japan subduction zone and thirty-eight events from the Middle America subduction zone to examine the relationship between source duration and depth. We find in both cases a decrease in source duration with increasing depth and an increase in stress drop with increasing depth, suggesting a change in fault zone properties along the megathrusts in these two subduction zones. This may be either a continuous variation with depth, or more of a stepwise change near a depth of about 20 km, with the shallow region having more scatter and on average lower rigidity material (or lower stress drop events). While this general trend is seen in both areas, there is somewhat more heterogeneity and longer average source durations in the Middle America data set, indicating possible differences in the mechanisms causing these variations in interface properties. This method of using scaled source duration as a probe of fault zone variations appears promising for examining large-scale variations, but further data collection and modeling will be necessary to address the details of the changes as well as mechanism of these variations.

Acknowledgments We used data provided through the IRIS DMC. GMT software was used for figure preparation. We wish to thank R. Dmowska, J. E. Vidale, H. Houston, and an anonymous reviewer for careful reviews and helpful comments. Also thanks to E. A. Silver for providing a preprint. This work was supported by grant NSF EAR-9418643. This is contribution number 355 of the Institute of Tectonics, University of California, Santa Cruz.

REFERENCES ABE, K. (1981), Physical Size of Tsunamigenic Earthquakes of the Northwestern Pacific, Phys. Earth Planet. Int. 27, 194–205. ABE, K. (1994), Instrumental Magnitudes of Historical Earthquakes, 1892 – 1898, Bull. Seismol. Soc. Am. 84, 415–425. AUBOUIN, J., VON HUENE, R., BALTUCK, M., ARNOTT, R., BOURGOIS, J., FILEWICZ, M., KVENVOLDEN, K., LEINERT, B., MCDONALD, T., MCDOUGALL, K., OGAWA, Y., TAYLOR, E., and WINSBOROUGH, B. (1982), Leg 84 of the Deep Sea Drilling Project: Subduction Without Accretion: Middle America Trench off Guatemala, Nature 297, 458 – 460. BILEK, S. L., and LAY, T. (1998), Variation of Interplate Fault Zone Properties with Depth in the Japan Subduction Zone, Science 281, 1175–1178. BOS, A. G., NOLET, G., RUBIN, A., HOUSTON, H., and VIDALE, J. E. (1998), Duration of Deep Earthquakes Determined by Stacking of Global Seismograph Network Seismograms, J. Geophys. Res. 103, 21,059–21,065.

Vol. 154, 1999

Fault Zone Properties in Japan and Middle America Trenches

455

CHRISTENSEN, D. H., and RUFF, L. J. (1985), Analysis of the Trade-off Between Hypocentral Depth and Source Time Function, Bull. Seismol. Soc. Am. 75, 1637 – 1656. DEMETS, C., GORDON, R. G., ARGUS, D. F., and STEIN, S. (1990), Current Plate Motions, Geophys. J. Int. 101, 425–478. EKSTRO¨M, G., and ENGDAHL, E. R. (1989), Earthquake Source Parameters and Stress Distribution in the Adak Island Region of the Central Aleutian Islands, Alaska, J. Geophys. Res. 94, 15,499 – 15,519. FUKAO, Y. (1979), Tsunami Earthquakes and Subduction Processes Near Deep Sea Trenches, J. Geophys. Res. 84, 2303–2314. HARTOG, J. R., and SCHWARTZ, S. Y. (1996), Directi6ity Analysis of the December 28, 1994 Sanriku-Oki Earthquake (Mw = 7.7), Japan, Geophys. Res. Lett. 23, 2037 – 2040. HEKI, K., MIYAZAKI, S., and TSUJI, H. (1997), Silent Fault Slip Following and Interplate Thrust Earthquake at the Japan Trench, Nature 386, 595 – 598. HEKI, K., and TAMURA, Y. (1997), Short Term Afterslip in the 1994 Sanriki-Haruka-Oki Earthquake, Geophys. Res. Lett. 24, 3285–3288. HINZ, K., VON HUENE, R., RANERO, C. R., and PACOMAR WORKING GROUP (1996), Tectonic Structure of the Con6ergent Pacific Margin Offshore Costa Rica from Multichannel Seismic Data, Tectonics 15, 54–66. IDE, S., IMAMURA, F., YOSHIDA, Y., and ABE, K. (1993), Source Characteristics of the Nicaraguan Tsunami Earthquake of September 2, 1992, Geophys. Res. Lett. 20, 863 – 866. IHMLE´, P. F. (1996), Frequency-dependent Relocation of the 1992 Nicaragua Slow Earthquake: An Empirical Green’s Function Approach, Geophys. J. Int. 127, 75 – 85. KANAMORI, H. (1972), Mechanism of Tsunami Earthquakes, Phys. Earth Planet. Int. 6, 246 – 259. KANAMORI, H., and ANDERSON, D. L. (1975), Theoretical Basis of Some Empirical Relations in Seismology, Bull. Seismol. Soc. Am. 65, 1073 – 1095. KANAMORI, H., and KIKUCHI, M. (1993), The 1992 Nicaragua Earthquake: A Slow Tsunami Earthquake Associated with Subducted Sediments, Nature 361, 714 – 716. KAWASAKI, I., ASAI, Y., TAMURA, Y., SAGIYA, T., MIKAMI, N., OKADA, Y., SAKATA, M., and KASAHARA, M. (1995), The 1992 Sanriku-Oki, Japan, Ultra-slow Earthquake, J. Phys. Earth 43, 105–116. KIKUCHI, M., and KANAMORI, H. (1995), Source Characteristics of the 1992 Nicaragua Tsunami Earthquake Inferred from Teleseismic Body Wa6es, Pure appl. geophys. 144, 441 – 453. KIMURA, G., SILVER, E., BLUM, P. et al. (1997), Proceedings ODP, Initial Reports, 170: College Station, TX (Ocean Drilling Program), 1–458. LAY, T., KANAMORI, H., and RUFF, L. (1982), The Asperity Model and the Nature of Large Subduction Zone Earthquakes, Earthquake Pred. Res. 1, 3 – 71. NAKAYAMA, W., and TAKEO, M. (1997), Slip History of the 1994 Sanriku-Haruka-Oki, Japan, Earthquake Deduced from Strong-motion Data, Bull. Seismol. Soc. Am. 87, 918 – 931. PELAYO, A.M., and WIENS, D.A. (1990), The No6ember 20, 1960, Peru Tsunami Earthquake — Source Mechanism of a Slow E6ent, Geophys. Res. Lett. 17, 661 – 664. PELAYO, A. M., and WIENS, D. A. (1992), Tsunami Earthquakes: Slow Thrust-faulting E6ents in the Accretionary Wedge, J. Geophys. Res. 97, 15,321 – 15,337. PROTTI, M., and SCHWARTZ, S. Y. (1994), Mechanics of Back Arc Deformation in Costa Rica: E6idence from an Aftershock Study of the April 22, 1991, Valle de la Estrella, Costa Rica Earthquake (Mw = 7.7), Tectonics 13, 1093–1107. PROTTI, M., SCHWARTZ, S. Y., and ZANDT, G. (1996), Simultaneous In6ersion for Earthquake Location Beneath Central Costa Rica, Bull. Seismol. Soc. Am. 86, 19 – 31. RUFF, L. J. (1989), Multi-trace Decon6olution with Unknown Trace Scale Factors: Omnilinear In6ersion of P and S wa6es for Source Time Functions, Geophys. Res. Lett. 16, 1043 – 1046. RUFF, L. J. (1992), Asperity Distributions and Large Earthquake Occurrence in Subduction Zones, Tectonophysics 211, 61–83. RUFF, L., and KANAMORI, H. (1980), Seismicity and the Subduction Process, Phys. Earth Planet. Int. 23, 240–252. RUFF, L. J., and MILLER, A. D. (1994), Rupture Process of Large Earthquakes in the Northern Mexico Subduction Zone, Pure appl. geophys. 142, 101 – 171.

456

S. L. Bilek and T. Lay

Pure appl. geophys.,

SATAKE, K. (1994), Mechanism of the 1992 Nicaragua Tsunami Earthquake, Geophys. Res. Lett. 21, 2519–2522. SCHWARTZ, S. Y., and RUFF, L. J. (1985), The 1968 Tokachi-Oki and the 1969 Kurile Islands Earthquakes: Variability in the Rupture Process, J. Geophys. Res. 90, 8613 – 8626. SENO, T., SAKURAI, T., and STEIN, S. (1996), Can the Okhotsk Plate be Discriminated from the North American Plate, J. Geophys. Res. 101, 11,305 – 11,315. SHIPLEY, T. H., MCINTOSH, K. D., SILVER, E. A., and STOFFER, P. L. (1992), Three-dimensional Seismic Imaging of the Costa Rica Accretionary Prism: Structural Di6ersity in a Small Volume of the Lower Slope, J. Geophys. Res. 97, 4439 – 4459. SILVER, E. A., MCINTOSH, K., SAITO, S., SAFFER, D., and LEG 170 SCIENTIFIC PARTY. (1999), in preparation. SUYEHIRO, K., and NISHIZAWA, A. (1994), Crustal Structure and Seismicity Beneath the Forearc off Northeastern Japan, J. Geophys. Res. 99, 22,331 – 22,347. TANIOKA, Y., RUFF, L., and SATAKE, K. (1996), The Sanriku-Oki, Japan, Earthquake of December 28, 1994 (Mw 7.7): Rupture of a Different Asperity from a Pre6ious Earthquake, Geophys. Res. Lett. 23, 1465–1468. TANIOKA, Y., RUFF, L., and SATAKE, K. (1997), What Controls the Lateral Variation of Large Earthquake Occurrence Along the Japan Trench?, Island Arc 6, 261 – 266. TANIOKA, Y., and SATAKE, K. (1996), Fault Parameters of the 1986 Sanriku Tsunami Earthquake Estimated from Tsunami Numerical Modeling, Geophys. Res. Lett. 23, 1549 – 1552. TICHELAAR, B. W., and RUFF, L. J. (1991), Seismic Coupling Along the Chilean Subduction Zone, J. Geophys. Res. 96, 11,997–12,022. TICHELAAR, B. W., and RUFF, L. J. (1993), Depth of Seismic Coupling Along Subduction Zones, J. Geophys. Res. 98, 2017–2037. VELASCO, A. A., AMMON, C. J., LAY, T., and ZHANG, J. (1994), Imaging a Slow Bilateral Rupture with Broadband Seismic Wa6es: The September 2, 1992 Nicaraguan Tsunami Earthquake, Geophys. Res. Lett. 21, 2629–2632. VIDALE, J. E., and HOUSTON, H. (1993), The Depth Dependence of Earthquake Duration and Implications for Rupture Mechanics, Nature 35, 45– 47. VON HUENE, R., and LALLEMAND, S. (1990), Tectonic Erosion Along the Japan and Peru Con6ergent Margins, GSA Bulletin 102, 704–720. VON HUENE, R., and SCHOLL, D. W. (1991), Obser6ations at Con6ergent Margins Concerning Sediment Subduction, Subduction Erosion, and the Growth of Continental Crust, Rev. Geophys. 29, 279 – 316. VON HUENE, R., KLAESCHEN, D., and CROPP, B. (1994), Tectonic Structure Across the Accretionary and Erosional Parts of the Japan Trench Margin, J. Geophys. Res. 99, 22,349 – 22,361. VON HUENE, R., and CULOTTA, R. (1989), Tectonic Erosion at the Front of the Japan Trench Con6ergent Margin, Tectonophysics 160, 75–90. VROLIJK, P. (1990), On the Mechanical Role of Smectite in Subduction Zones, Geology 18, 703 – 707. ZHANG, Z., and SCHWARTZ, S. Y. (1992), Depth Distribution of Moment Release in Underthrusting Earthquakes at Subduction Zones, J. Geophys. Res. 97, 537 – 544. (Received April 27, 1998, revised November 5, 1998, accepted January 12, 1999)

.