Microseismicity at the seaward updip limit of the western Nankai Trough seismogenic zone

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 108, NO. B10, 2459, doi: /2002jb002370, 2003 Microseismicity at the seaward updip limit of the western Nankai Trough seismogenic zone Koichiro Obana, Shuichi Kodaira, and Yoshiyuki Kaneda Institute for Frontier Research on Earth Evolution, Japan Marine Science and Technology Center, Yokohama, Japan Kimihiro Mochizuki and Masanao Shinohara Earthquake Research Institute, University of Tokyo, Tokyo, Japan Kiyoshi Suyehiro Deep Sea Research Department, Japan Marine Science and Technology Center, Yokosuka, Japan Received 25 December 2002; revised 17 June 2003; accepted 2 July 2003; published 4 October [1] The seismogenic zone at subduction zones does not normally extend to the trench axis. The shallowest part of the plate interface is considered to be seismically decoupled. The location of the seaward updip limit of the seismogenic zone and its relation to the crustal structure are important for understanding the transition process from aseismic slip to the seismic rupture at the plate interface. The Nankai Trough seismogenic zone, southwestern Japan, is one of the most well-studied subduction seismogenic zones in the world. However, the offshore seismicity around the Nankai Trough is very low, and hypocenters are not determined accurately by the on-land seismic network. We performed microseismicity observations using ocean bottom seismographs off Cape Muroto. Hypocenters were determined by using a three-dimensional Vp and Vs structure based on seismic survey results. The observed microseismicity seems to be classifiable into two types. The first is earthquakes that occur around the plate interface; the second is a group of earthquakes that occur in the uppermost mantle of the subducting oceanic plate. The seismicity around the plate interface forms several clusters. The seaward limit of the clusters is characterized by earthquakes with a similar waveform. These similar waveform earthquakes are considered to occur at small asperities in the aseismic-seismogenic transition zone of the plate interface. The seismicity in the uppermost mantle may be related to dehydration embrittlement of subducting serpentinized mantle. INDEX TERMS: 7230 Seismology: Seismicity and seismotectonics; 8150 Tectonophysics: Plate boundary general (3040); 9320 Information Related to Geographic Region: Asia; KEYWORDS: Nankai Trough, seismogenic zone, subduction, microseismicity, ocean bottom seismograph, similar earthquakes Citation: Obana, K., S. Kodaira, Y. Kaneda, K. Mochizuki, M. Shinohara, and K. Suyehiro, Microseismicity at the seaward updip limit of the western Nankai Trough seismogenic zone, J. Geophys. Res., 108(B10), 2459, doi: /2002jb002370, Introduction [2] The plate interface along subduction zones causes great earthquakes through thrust faulting between the subducting oceanic plate and the overriding plate. The seismogenic zone at a subduction zone does not normally extend to the trench axis and the shallowest part of the plate interface is considered to be seismically decoupled [e.g., Byrne et al., 1988]. The seaward updip limit of the seismogenic zone is limited by a transition from an aseismic to a seismogenic plate interface. Byrne et al. [1988] considered the aseismic zone at the shallowest part of the plate interface to be caused by the presence of unconsolidated sediments and that consolidation of the sediments is required for seismogenesis. Hyndman et al. [1995, 1997] and Oleskevich et al. Copyright 2003 by the American Geophysical Union /03/2002JB002370$09.00 [1999] suggested that the updip limit of the seismogenic zone is limited by the temperature at the top of the subducting oceanic crust. They showed that the updip limit of great subduction thrust earthquake rupture zones coincides with the C isotherm at the top of the subducting oceanic crust. Stable-sliding clays, such as smectite, in the sediments on the subducting oceanic crust dehydrate at this temperature and cause stick-slip seismogenic behavior. In order to understand the transition from an aseismic to a seismogenic plate interface, it is necessary to understand the crustal structure and seismic rupture along the subduction zone. [3] The Nankai Trough seismogenic zone, southwestern Japan, is one of the most well-studied subduction seismogenic zones in the world. Off Cape Muroto, the Philippine sea plate is subducting beneath the overriding Eurasian plate with a convergence rate of about 4.6 cm/yr [Seno et al., 1993] (Figure 1). Great earthquakes caused by the subduc- ESE 2-1

2 ESE 2-2 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH Figure 1. Map of the Nankai Trough region. The contour interval of water depth is 1000 m. Ocean Bottom Seismographs (OBSs) were deployed off Cape Muroto. Open circles show OBS positions. Rectangular areas labeled A to D are the coseismic rupture areas of great earthquakes [Ando, 1975]. The solid star indicates the epicenter of the 1946 Nankai earthquake [Kanamori, 1972]. Area A and B were ruptured in the 1946 Nankai earthquake. Solid lines, KR9810, KR9704, KY9903, and KR9806, show profiles of OBS-airgun seismic structure surveys [Kodaira et al., 2000a, 2002; Nakanishi et al., 2002b; Takahashi et al., 2002]. The convergence rate of the Philippine sea plate under the Eurasian plate is 4.6 cm/yr off Cape Muroto [Seno et al., 1993]. tion of the Philippine sea plate have occurred along the Nankai Trough repeatedly. Records of great earthquakes can be seen in historical documents from the seventh century and the recurrence interval is about years [Ando, 1975]. The fault region of historical great earthquakes along the Nankai Trough is divided into four segments (Figure 1) [Ando, 1975]. The rupture of the 1946 Nankai earthquake started off Cape Shiono-misaki [Kanamori, 1972]. The coseismic rupture was estimated to occur on the two western segments, A and B, on the basis of tsunami and geodetic data [Ando, 1975]. Recent tsunami analyses show the detailed coseismic slip distribution on the fault plane [Tanioka and Satake, 2001; Baba, 2003]. Tsunami analysis is effective in estimating the coseismic slip on the offshore fault zone because tsunamis are generated by seafloor deformation. Their results show that coseismic slip (>1m) extended to the trench axis in the eastern half of the source region off Kii peninsula. However, in the western half of the source region off Shikoku, significant slip only occurred at the landward deeper portion of the fault zone. Although these coseismic slip distributions may be related to variations in the seaward extent of the seismogenic zone, the rupture propagation during the 1946 Nankai earthquake at least, may have been controlled by a subducting seamount off Cape Muroto [Kodaira et al., 2002]. [4] Seismicity during the interseismic period may indicate the location of the updip limit of the seismogenic zone along the plate interface. The updip limit of the seismicity during the interseismic period is comparable with the aftershock area of large interplate earthquakes [Byrne et al., 1988]. The seismogenic zone along the subduction zone is generally located offshore. Ocean bottom seismograph (OBS) observation is useful for observing earthquakes below the seafloor and accurately determining their hypocenters. In the central Aleutians, OBS observations combined with the island network showed that the focal mechanisms of shallow earthquakes in the thrust zone had P axes parallel to the direction of the plate convergence [Frohlich et al., 1982]. Recent seismicity studies using OBSs and ocean bottom hydrophones (OBHs) show the location of the updip limit

3 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH ESE 2-3 Figure 2. Locations of OBSs in each observation phase. Triangles indicate deployed positions of freefall and pop-up type OBSs. Diamonds indicate seismometers of a cable link ocean bottom observation system. OBSs which were used for the analyses are indicated by solid symbols and the others by open symbols. The contour interval of water depth is 1000 m. of seismicity along subduction zones from accurately located hypocenters [e.g., Husen et al., 1999; Newman et al., 2002]. In this article, we report the seismicity around the updip limit of the Nankai Trough seismogenic zone as revealed by OBS observations off Cape Muroto (Figure 1). Along the Nankai trough, many seismic surveys have revealed the structure of the subduction seismogenic zone. Offshore seismicity along the Nankai Trough is not abundant [e.g., Nakamura et al., 1997] and hypocenters are not thought to be determined accurately by on-land seismic stations. Thus the updip limit of seismicity is not constrained well. The purpose of this study is to examine the aseismic-seismogenic transition zone by (1) determining the location of the updip limit of seismicity along the plate boundary, which would indicate how close to the trench axis the plate boundary causes seismogenic behavior and (2) understanding the mechanisms that control the aseismic to seismogenic transition through a comparison between the seismicity, crustal structures, and other geophysical studies. 2. Observation [5] We began microseismicity observations off Cape Muroto in Two free-fall and pop-up types of digital recording OBSs were used for the observations. One type was designed by Shinohara et al. [1993] and is a digital recording version of the original designed by Kanazawa and Shiobara [1994]. The other type of OBS was designed for long-term observations [Mochizuki et al., 1997]. Each OBS has either a three-component short-period seismometer that is designed to maintain the vertical and horizontal directions, or a three-component broadband seismometer produced by PMD Scientific Inc. Some OBSs also contain a hydrophone. Each OBS stores data continuously on a digital audio tape (DAT) or a hard disk drive (HDD) at 100 or 128 Hz sampling frequency. [6] Seismicity off Cape Muroto is not abundant and a long observation period is necessary to detect many earthquakes. However, the maximum observation period of our OBS is limited to about 2 or 3 months. This limitation is a result of the power consumption of the OBS and capacity of the record media. We repeated deployments of OBSs five times from 1998 until 2000 to increase the observation period (Figure 2). The total observation period was about nine months. A cable-linked ocean bottom observation system was installed by the Japan Marine Science and Technology Center (JAMSTEC) off Cape Muroto in 1997 [Momma et al., 1997]. This system consists of 2 ocean bottom seismometers. Data recorded by this system were also used.

4 ESE 2-4 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH Figure 3. A 3D structure boundary model off Cape Muroto based on four airgun-obs seismic surveys along the Nankai Trough. Solid circles indicate locations of OBSs. Two solid lines, KY9903 and KR9704, are seismic survey lines used for constructing the model. We used two other seismic surveys, KR9810 and KR9806, for the construction. Those lines are indicated in Figure 1. Each surface indicates structural boundaries. The rectangular area projected on the map view indicates the horizontal extent of the 3D velocity model. [7] For the first observation in 1998, we deployed five pop-up type OBSs around the seaward limit of the coseismic slip area of the 1946 Nankai earthquake [Obana et al., 2001]. The aseismic-seismogenic transition along the plate boundary is expected to occur in this region based on the coseismic slip distribution and thermal modeling results [Hyndman et al., 1995]. We divided the observation period into two phases and continued observations for about three months. The first phase was from 5 July to 18 August and the second phase was from 20 August to 9 October. At the end of the first phase, we retrieved the OBSs and deployed them soon thereafter. OBSs were deployed in almost the same places for the two phases. [8] In late September 1999, we deployed 6 OBSs in the same region as the first observation. One additional OBS was deployed at the center of the OBS array to construct a Figure 4. An example of velocity structure along seismic survey line KR9704. P wave velocity (Vp) at the top and bottom of each layer and Vp/Vs ratio in each layer are defined for the structure model. Vp at the bottom of the model is assumed to be 8.2 km/s at 50 km depth.

5 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH ESE 2-5 Figure 5. Epicentral distribution obtained by OBS observations. Open circles indicate epicenters observed by OBS. The radius of each circle is taken to be proportional to magnitude. Open diamonds indicate the locations of the OBSs. Open and shaded rectangles indicate the coseismic slip during the 1946 Nankai earthquake estimated from tsunami [Baba, 2003]. The dotted line is the 150 C isotherm contour on the top of the subducting oceanic crust [Hyndman et al., 1995]. The contour interval of the water depth is 1000 m. Two solid lines, KY9903 and KR9704, are Airgun-OBS seismic survey lines. Cross sections along these lines are shown in Figure 6. denser network and obtain more accurate hypocenters. These OBSs continued observations until late November. In mid November, 7 OBSs were deployed for the second phase of the year. All but one were retrieved in mid January [9] The existence of a subducting seamount in the deployment area is evident from magnetic data and seismic surveys [Yamazaki and Okamura, 1989; Kodaira et al., 2000b]. A landward indentation of water depth contours and resulting deformation of the overriding plate are a result of the seamount subduction [Yamazaki and Okamura, 1989; Dominguez et al., 1998]. The observation area was influenced by the subduction of this seamount, which prompted the expansion of the observation area westward. The westward expansion took place in June, with the deployment of 9 OBSs. These were later retrieved in mid August. The aim of the expansion of the observation area was to understand the nature of seismicity around the updip limit of the seismogenic zone by comparing seismicity in areas where the effects of the seamount subduction are different. [10] Some OBSs were not used in the analyses because they were not recovered or could not be corrected for the internal clock drift of the OBS during the observation

6 ESE 2-6 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH Figure 6. Cross sections along two airgun-obs seismic survey lines, KY9903 and KR9704 [Kodaira et al., 2000a, 2002]. Open circles indicate projected location of hypocenters observed by OBSs. Only hypocenters within 25 km of either side of the survey line have been projected. The radius of each circle is taken to be proportional to the magnitude. Isovelocity contours of P wave velocity are drawn. Contour interval is 0.5 km/s. Coseismic slip during the 1946 Nankai earthquake [Baba, 2003] and interplate locked zone [Hyndman et al., 1995] are indicated at the top of both panels. period. The internal clock drift of the OBS was estimated by linear interpolation of the time differences relative to the standard time just before and soon after the observation. A clock based on GPS was used as the standard time to measure the time differences. [11] Seismic events were detected from continuous records and separate event files were created. We picked P and S arrival times and estimated the picking error with changing amplitude and time scaling using the win system [Urabe and Tsukada, 1992]. The P and S arrivals were picked with errors of and s, respectively. 3. Hypocenter Determination in a 3D Velocity Model D P and S Wave Velocity Structure [12] Generally, seismic velocity structure has a large lateral variation along a subduction zone. In this study, we used 3D P and S wave velocity (Vp and Vs) structure models based on seismic surveys to locate earthquakes. Many of the seismic surveys along the Nankai Trough show along-strike variations in crustal structure, such as variations in the subducting plate angle and differences in the shape of the accretionary prism [e.g., Kodaira et al., 2002; Nakanishi et al., 2002b; Takahashi et al., 2002]. Because of the threedimensional (3D) structural variations, a 1D seismic velocity structure is not a good assumption for obtaining accurate hypocenters. Inhomogeneous structure due to the subduction of the Philippine sea plate was taken into account in the 3D structure models. [13] The calculated depths of hypocenters off Cape Muroto in the Nankai Trough strongly depend on pthe ffiffi Vp/Vs ratio [Obana et al., 2001]. A fixed value, such as 3 (Poisson s ratio = 0.25), is often used as the Vp/Vs ratio for hypocenter determinations. However, a higher Vp/Vs ratio in the accretionary prism and sedimentary layer were confirmed by a survey off Cape Ashizuri [Takahashi et al., 2002]. Takahashi et al. [2002] found that the Poisson s ratio is higher than 0.25 in the accretionary prism and

7 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH ESE 2-7 Figure 7. The projections of the estimated 68% confidence ellipsoids of the hypocenters. (a) Map view of the confidence ellipsoids. The rectangle indicates the projected area of the vertical cross sections along the line KY9903. (b, c) Two vertical cross sections along the line KY9903. Hypocenters shallower and deeper than 10 km are projected on Figures 7b and 7c, respectively. sedimentary layer. A realistic Vp/Vs structure is necessary to obtain accurate hypocenters. [14] The P and S wave velocity structure models for the hypocenter determinations were constructed from four airgun-obs seismic survey results along the Nankai Trough (Figure 1) [Kodaira et al., 2000a, 2002; Nakanishi et al., 2002b; Takahashi et al., 2002]. For each survey line, we picked the depths of structural boundaries, such as the lower boundary of the sedimentary layer and accretionary prism, the boundary between upper and lower parts of the island arc crust, the boundary between oceanic crust layer 2 and 3, and the oceanic Moho discontinuity. The upper and lower parts of the island arc crust are widely observed along the Nankai Trough on the basis of their velocities [e.g., Kodaira et al., 2000a; Nakanishi et al., 2002b]. We refer to the upper part as island arc upper crust and the lower part as island arc lower crust, following previous seismic structure studies along the Nankai Trough. The island arc upper crust is considered to correspond to the old accretionary complex [Kodaira et al., 2000a]. A three-dimensional shape for each boundary was produced by interpolating the four seismic lines (Figure 3). [15] The thickness of the sedimentary layer just below each OBS was estimated. We used this estimated thickness in addition to the seismic survey results to produce the sedimentary layer model. Most events showed a P to S conversion phase at the base of the soft sediments. The thickness of the sedimentary layer was calculated from time difference between the PS conversion phase and the P phase, assuming a specific Vp and Vp/Vs ratio for the sedimentary layer. We used the Vp and Vp/Vs ratio estimated from the seismic surveys. The estimated thicknesses are comparable to that of the top sedimentary layer imaged by the seismic surveys at OBSs along the line KY9903. Since two seismic lines exist in our OBS observation area off Cape Muroto, the 3D boundary structure is considered to be modeled well. [16] The P wave velocity of the model was defined at the top and bottom of each layer (Figure 4). The velocity in the layer was defined by a linear interpolation of the velocity at the top and bottom of the layer. The velocities were based on the airgun-obs seismic survey results. Vp in the seawater was fixed to be 1.5 km/s. Vp in the mantle was assumed to be 8.2 km/s at 50 km depth. A low-vp zone beneath the subducted seamount was imaged off Cape Muroto along the KY9903 seismic survey line [Kodaira et al., 2002]. The extent of the low-velocity zone was not well constrained along the survey line. Therefore this lowvelocity zone in the uppermost mantle was not included in our 3D model.

8 ESE 2-8 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH Figure 8. Example seismograms of similar waveform earthquakes. These seismograms were recorded at S04 (Figure 9). Each trace is a vertical component seismogram and was band-pass filtered from 2 to 8 Hz. P and S phase picks are shown as P and S. [17] S wave velocity structure was estimated from the Vp model using an assumed Vp/Vs ratio. Vp/Vs in the sedimentary layer and accretionary prism were assumed to be 3.32 and 2.14, respectively. In other parts of the p modeled structure, Vp/Vs was assumed to be 1.73 (= ffiffi 3 ). These values are based on the results of the seismic survey KR9810 off Cape Ashizuri (Figure 1) [Takahashi et al., 2002]. Except for this survey, there is no other seismic survey which estimates S wave velocities along the Nankai Trough. We assumed that there is no significant difference in the Poisson s ratio for each structural unit along the Nankai Trough. After the hypocenter determination described below, we compared hypocenters obtained only from P wave arrivals to those obtained from both P and S wave arrivals. There are no systematic changes between them. We therefore consider that the assumed Vp/Vs ratios used in this model were appropriate. Although S waves can not propagate in seawater, Vp/Vs in seawater was taken to be the same as the sedimentary layer for convenience in the calculations. Finally, we constructed 3D Vp and Vs models that extend to 200 km in both horizontal directions and 50 km in depth, with a 1 km grid interval Hypocenter Determination [18] Hypocenters were obtained by an inversion in the 3D velocity model. For the first step of the hypocenter determination procedure, P and S wave travel times between OBSs and 1 km spacing grids in the 3D model were calculated. The calculation was done by solving the eikonal equation using finite differences [Zelt and Barton, 1998]. In the second step, hypocenters were obtained by using the P and S wave travel time tables calculated in the first step. Hypocenters were relocated iteratively from an initial location by a linearized inversion to minimize the sum of the square of travel time residuals. During the inversion, residuals for each phase (P and S wave arrivals) were weighted by the inverse of their phase picking error. When errors were less than 0.10 and 0.50 s for P and S wave arrival, respectively, we adopted these values as a minimum picking error for each phase. [19] We examined several initial locations for each event. Because our 3D model includes large velocity contrasts, there would be several local minima of travel time residuals. Examined initial positions were distributed in a space of 100 km in the horizontal direction with a 10 km interval, and from 5 km to 30 km depth with a 5 km interval. Final hypocenter locations were determined at the location with the minimum sum of the square of the weighted residual from all inversion results. We also determined the magnitude of each event based on the maximum amplitude of vertical components [Watanabe, 1971]. [20] Finally, we estimated hypocenter location uncertainties using a probabilistic earthquake location program with nonlinear global search methods. We used the software package NonLinLoc [Lomax et al., 2000], which follows the probabilistic formulation of inversion presented by Tarantora and Valette [1982]. The posterior probability density function (PDF) represents a probabilistic location of the hypocenter and uncertainty due to the errors of picking and travel time calculation, the network-event geometry, and incompatibility of the picks [Lomax et al., 2000]. The NonLinLoc also determines traditional Gaussian or normal estimators, such as expectation of the hypocenter location and the covariance matrix, in addition to the

9 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH ESE 2-9 Figure 9. Open circles are earthquakes which show similar waveforms. Crosses are double difference locations of similar waveform earthquakes. Solid circles indicate epicenters of other earthquakes shallower than 10 km in depth. Each similar earthquake cluster is enclosed by dotted or solid line ellipses. Seismograms of earthquakes included in a cluster enclosed by the solid line ellipse is shown in Figure 8. Solid lines, KY9903 and MS105, indicate MCS profile lines. The dashed line indicates a 150 C isotherm contour on the top of subducting oceanic plate [Hyndman et al., 1995]. Open and shaded rectangles indicate the coseismic slip during the 1946 Nankai earthquake estimated from tsunami [Baba, 2003]. Open diamonds indicate locations of OBSs. The OBS where the seismograms shown in Figure 8 were recorded is indicated with label S04. maximum likelihood hypocenter. The 68% confidence ellipsoid is obtained from the singular value decomposition of the covariance matrix [Lomax et al., 2000]. We used this confidence ellipsoid as the hypocenter location uncertainties when the PDF was not ill conditioned and was comparable to the confidence ellipsoid. 4. Results 4.1. Hypocenter Distribution [21] We attempted to locate 582 events through our OBS observation period and 301 events converged within the 3D model. RMS travel time residuals of all converged events are 0.06 and 0.27 s for P and S arrivals, respectively. When hypocenters were determined in a considerable one-dimensional structure, RMS residuals were 0.18 and 0.34 s for P and S arrivals, respectively. The uncertainties of these events were estimated from the PDF and the confidence ellipsoid. Finally we selected 80 events with a maximum length of the 68% confidence ellipsoid semiaxis of less than 10 km. Figure 5 shows the epicentral distribution and Figure 6 shows two vertical cross sections along the seismic survey lines, KY9903 and KR9704. These events have at least 5 readings of phase arrivals. The largest magnitude of these events is 2.6 and many of events are less than 2.0. Since these events are too small to observe at on-land seismic stations, we used only the data from the OBSs. Seismicity extends seaward to the 4000 m isobath, which is parallel to the 150 C isotherm contour on the top of the subducting oceanic crust [Hyndman et al., 1995]. [22] Cross sections along two airgun-obs seismic survey lines, KY9903 and KR9704, with projected hypocenters

10 ESE 2-10 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH Table 1. Earthquakes With Similar Waveforms Date Time, UT Latitude, deg N Longitude, deg E Depth, km Mag. ErrMax, a km Double-Difference Location, b m X Y Z ErrX ErrY ErrZ Cluster 1 10 Oct c Nov Cluster 2 10 Nov Nov Cluster 3 27 June 2000 c June Cluster 4 21 July 2000 c July July 2000 c July 2000 c July 2000 c July Aug c Aug c a ErrMax is maximum length of the semi-axis of confidence ellipsoid. b Double-Difference locations (X, Y, Z, ErrX, ErrY, and ErrZ) were obtained by hypodd [Waldhauser and Ellsworth, 2000]. Hypocenter location (X, Y, and Z) is east-west, north-south, and depth location relative to cluster centroid, respectively. ErrX, ErrY, and ErrZ are east-west, north-south, and depth error, respectively. c Those earthquakes having a large confidence ellipsoid and not indicated in Figures 5 and 6. show two groups of seismicity (Figure 6). One is a group of shallower earthquakes that occurred around the top of the subducting oceanic crust. The other is a group of deeper earthquakes that occurred in the uppermost mantle of the subducting oceanic plate. These two groups characterize the seismicity during the whole observation period. Shallow earthquakes are located around the boundary between the subducting and overriding plates. These earthquakes show several clusters on both seismic survey lines. Almost all the shallow hypocenters are shallower than 10 km depth. The confidence ellipsoids projected on the vertical cross section indicate that these shallow earthquakes occurred at depths shallower than the subducting oceanic Moho (Figure 7b). Most of the seismicity in the uppermost mantle is located between 20 km and 30 km in depth. Confidence ellipsoids show that these earthquakes occurred below the Moho of the subducting oceanic crust (Figure 7c). The hypocenters in the uppermost mantle show a scattered distribution. Although more earthquakes appear along line KY9903, this is artificial since the observation period in the eastern part of the OBS array was longer Shallow Earthquake Clusters [23] Earthquakes of the shallow group show some clustering around the seaward limit of the coseismic rupture area of the 1946 Nankai earthquake (Figure 6). Some of the earthquakes in the shallow group were characterized by very similar waveforms (Figure 8). We calculated crosscorrelation coefficients of vertical seismograms between pairs of events. Cross-correlation coefficients were calculated from 1 s before the P arrival to at least 3 s after the S arrival. The seismograms were band-pass filtered from 2 to 8 Hz before the cross-correlation calculation to increase their signal-to-noise ratio. Pairs of events with correlation coefficients larger than 0.95 at more than 2 OBSs are treated as similar earthquakes. [24] We found five similar earthquake clusters from all of located earthquakes, including earthquakes with large hypocenter uncertainties (Figure 9). Each cluster consists of two to four earthquakes. The variation in magnitude within each cluster ranges from 0.0 to 0.5. Although a northern cluster on line KY9903 (enclosed by a solid line ellipse in Figure 9), includes two clusters of similar earthquakes, there is no significant difference in the seismograms. Crosscorrelation coefficients between these two clusters are larger than 0.93 at two or more OBSs. We treated these two groups as the same cluster. The final number of clusters is four in total. All the similar earthquakes are listed in Table 1. [25] The similar earthquake clusters were located at the seaward limit of the shallow earthquakes, which occurred at the top of the subducting oceanic crust. These clusters also coincide with the 150 C isotherm at the top of the subducting oceanic crust [Hyndman et al., 1995]. The apparent dimensions of the similar earthquake clusters based on the absolute location of the hypocenters are less than 10 km in horizontal dimension. We tried to estimate the actual size of similar earthquake clusters based on the relative location of the hypocenters using the double-difference algorithm Table 2. One-Dimensional Velocity Model Used for the Double Difference Locations a P Wave Velocity, km/s Layer Thickness, km a Vp/Vs ratio is assumed to be 1.93.

11 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH ESE 2-11 Figure 10. Poststack depth migrated section of a MCS profile along KY9903 (upper panel) and its interpretation with projected hypocenters (lower panel) [after Park et al., 1999]. Coseismic slip during the 1946 Nankai earthquake [Baba, 2003] and interplate locked zone [Hyndman et al., 1995] are indicated at the top of the figure. Only hypocenters within 5 km of either side of the survey line have been projected. Open circles are earthquakes which show similar waveforms. Crosses are double difference locations of similar waveform earthquakes. Solid circles indicates epicenters of other earthquakes. Each similar earthquake cluster is enclosed by dotted line ellipses. [Waldhauser and Ellsworth, 2000]. The software package hypodd was used for this calculation. The relative locations of the hypocenters were estimated from the picked travel times and cross-correlation differential travel times of P and S arrivals. A one-dimensional layered model with a fixed Vp/Vs ratio (=1.93) was used for the calculation (Table 2). The size of clusters including the errors, are estimated to be about a few hundred meters for most of the clusters (Table 1). The calculated sizes are almost the same even when considerably different velocity models with Vp/Vs ratios ranging from 1.73 to 2.14 were used. 5. Discussion 5.1. Similar Earthquakes as Small Asperities in the Transition Zone of the Plate Interface [26] Although there are large uncertainties in the earthquake locations, similar waveform earthquakes are located around the top of the subducting oceanic crust (Figures 10 and 11). The similar waveforms imply that the events were caused by repeated ruptures at the same asperity [Geller and Mueller, 1980]. These earthquakes could be explained by repeated ruptures of small asperities in the aseismicseismogenic transition zone of the plate interface (Figure 12). [27] From the macroscopic point of view, coupling of the plate interface between the subducting and the overriding plates changes from aseismic stable sliding to seismogenic locked in the transition zone. The decollement steps down to the subducting oceanic basement and duplex structures are formed around the similar earthquake clusters (Figure 10). The stepping down of the decollement results in underplating of subducting sedimentary materials. Thus friction between the subducting and overriding plates may increase. Landward of the similar earthquake clusters, the water depth of the seafloor rapidly becomes shallower. Where the friction at the bottom boundary of the accretionary prism increases, the prism taper becomes steeper [Davis et al., 1983]. Large coseismic slip (>2m) occurs where the crustal block immediately landward of the sedimentary wedge contacts the subducting oceanic crust [Nakanishi et al., 2002a]. That crustal block, referred to as island arc upper crust in this article, is interpreted to be old accreted sediments and forms the backstop described by Byrne et al. [1988]. The backstop is located landward of the similar earthquake clusters. Coseismic rupture during large interplate earthquakes can propagate to the shallower transition zone. Along line MS105, the similar earthquakes were located where the out-of-sequence thrusts (OST) converge to the decollement (Figure 11). These OSTs that cut through the accretionary prism are interpreted as seismic thrusts during large interplate earthquakes, including the 1946 Nankai earthquake [Park et al., 2000]. The relatively small coseismic slip during the large thrust earthquake around the similar earthquake clusters indicates weak coupling at the plate boundary during the interseismic period. On the basis of geodetic data analysis, the minimum seaward extent of the

12 ESE 2-12 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH Figure 11. Poststack depth migrated section of MCS profile along MS105 (upper panel) and interpretation with projected hypocenters (lower panel) [after Park et al., 2000]. Only hypocenters within 5 km of either side of the survey line have been projected. Symbols are the same as in Figure 10. OST1 and OST2 are the out-of-sequence thrust faults. fully locked plate interface is at 15 km depth along the plate boundary [Mazzotti et al., 2000], deeper than the similar earthquake clusters. [28] On a smaller scale, there may be small locked patches in the transition zone (Figure 12). Double-difference locations of similar earthquakes show that each cluster of them could be limited to within an area of a few hundred meters radius. The similarity between the earthquakes is closest at frequencies lower than 8 Hz, and deteriorates at higher frequencies. This frequency-dependent character can be explained by earthquakes in the same cluster occurring within a radius of one quarter wavelength [Geller and Mueller, 1980]. Assuming the S wave velocity to be 2.7 km/s, which corresponds to the velocity at the top of the subducting oceanic crust, one quarter of the wavelength is 84 m for a frequency of 8 Hz. The size of the similar earthquake clusters is concluded to be smaller than a few hundred meters. [29] These locally locked patches could be caused by spatial variations in the progress of the transition process. Similar waveform earthquakes are located at the seaward limit of the seismicity along the plate boundary and coincide with the 150 C isotherm on the top of subducting oceanic crust [Hyndman et al., 1995]. Recent seismicity studies using OBSs and OBHs in Costa Rica and Chile show the possibility that the updip limit of seismicity is controlled by the temperature of the subducting crust [Husen et al., 1999; Newman et al., 2002]. Aseismic-seismogenic transition at this temperature can be explained by smectite-to-illite dehydration reactions [Hyndman et al., 1995] or by a suite of diagenetic to low-grade metamorphic processes [Moore and Saffer, 2001]. [30] The steady slip in the surrounding area causes repeated rupture at the asperities. A study of microearthquakes at Parkfield along the San Andreas fault showed seismicity clusters characterized by similar waveform earthquakes [Nadeau et al., 1995]. The earthquakes that occurred in individual clusters had a periodic recurrence interval and were interpreted to be repeated slip on a locally strong fault driven by tectonic loading [Nadeau and Johnson, 1998]. Similar earthquakes were observed in the subduction seis- Figure 12. A schematic image of a transition zone in interplate coupling between subducting and overriding plates at the subduction zone.

13 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH ESE 2-13 Figure 13. Cross section along the seismic survey KY9903 off Cape Muroto with projection of hypocenters. Solid circles are hypocenters determined by OBS observations. Open circles are hypocenters determined by Japan Meteorological Agency (JMA) from October 1998 to September Only hypocenters within 25 km of either side of the survey line have been projected. Projected position of the Cape Muroto and deepest part of the Nankai Trough is indicated by arrows in the upper part of the panel. The shaded area is a low-vp zone in the uppermost mantle [Kodaira et al., 2002]. Contours indicate P wave velocities. The isovelocity contour interval is 0.5 km/s. mogenic zone along the Japan Trench, northeastern Japan [Igarashi et al., 2000]. Many of the similar earthquakes were located where the interplate coupling ratio derived from GPS observation was relatively low during the interseismic period. The slip rate estimated from the similar earthquakes using the scaling relationship along the San Andreas Fault [Nadeau and Johnson, 1998], was consistent with the slip rate on the plate boundary obtained from GPS data [Igarashi et al., 2001]. For similar waveform earthquakes off Cape Muroto, time intervals between the earthquakes included in individual clusters are not regular (Table 1). There are also variations in the magnitude of earthquakes within the same cluster. However, these earthquakes should occur within a few hundred meters. The similar waveform earthquakes off Cape Muroto are considered to occur at the same asperity, which is a locally locked patch in the aseismic-seismogenic transition zone. [31] Focal mechanisms of the similar earthquakes have not been determined because the magnitudes of the events were small and the number of OBS was insufficient to determine the focal mechanism from the polarities of the first motions. However, observed first motions of these earthquakes can be explained by P axes parallel to the direction of the plate convergence. All of the observed first motions cannot be explained by a low-angle thrust fault along the plate boundary. However, the size of the clusters are small enough that the observed first motions can be explained by dip variations on the fault, even along the plate boundary. [32] Existence of local asperities around duplex structures is supported by pseudotachylyte, a type of fault rock, observed in old accretionary complexes. Pseudotachylyte is formed by frictional heating due to rapid slip on a fault, such as an earthquake. Pseudotachylyte was found in the Okitsu Melange in the Shimanto accretionary complex, southwest Japan [Ikesawa et al., 2003]. In the Okitsu Melange, pseudotachylyte was considered to have formed on the roof thrust of a duplex structure (A. Sakaguchi et al., Tectonic setting of seismogenic fault in the ancient subduction zone, Okitsu Melange, Shimanto accretionary complex, SW Japan, submitted to Tectonics, 2003). The horizontal continuity of the pseudotachylyte is less than 1 km, which is comparable in size to the asperities identified by the similar waveform earthquakes Earthquakes in the Uppermost Mantle of the Subducting Slab [33] Earthquakes occurred in the uppermost mantle below the subducting oceanic crust near the trench axis and also beneath Shikoku Island (Figure 13). These earthquakes appear both in our OBS observations and in earthquake catalogues of the Japan Meteorological Agency (JMA). The hypocenters of the JMA catalogues are mainly determined using the on-land seismic network. Although the depths of offshore earthquakes are not considered to be determined accurately by the on-land seismic network, the hypocenters beneath Shikoku Island are reliable. [34] Intermediate-depth and deep earthquakes are observed as a double seismic zone in some subducting slabs. These earthquakes may be explained by stresses due to unbending [Engdahl and Scholz, 1977] or sagging [Sleep, 1979], or by thermal stresses [House and Jacob, 1982]. Dehydration of the subducting slab most likely controls the occurrence of the intermediate-depth intraslab earthquakes [e.g., Kirby et al., 1996]. Beneath southwestern Japan, modeled temperatures indicate that dehydration from the subducting oceanic crust could occur at shallow depths, which correspond to the upper plane seismicity of the double seismic zone along the northeastern Japan [Peacock and Wang, 1999]. The lower plane earthquakes can be

14 ESE 2-14 OBANA ET AL.: MICROSEISMICITY AT THE NANKAI TROUGH explained by dehydration of serpentine in the subducting oceanic mantle [Seno et al., 2001; Peacock, 2001]. Serpentinization of the mantle may be caused by the infiltration of seawater through faulting in the trench-outer rise region [Peacock, 2001]. However, the trench-outer rise, and earthquakes in that area, are not distinct along the Nankai Trough. Seno et al. [2001] suggested that dehydration embrittlement is an additional mechanism that can explain intermediate-depth earthquakes. They considered that earthquakes in the mantle beneath part of southwestern Japan were caused by dehydration embrittlement of serpentinized mantle associated with the back arc igneous activity in the Izu-Bonin arc. [35] The seismicity in the uppermost mantle off Cape Muroto could be explained by the subduction of serpentinized mantle beneath the Kinan seamount chain undergoing dehydration embrittlement [Kodaira et al., 2002]. Serpentinized mantle was imaged as a low P wave velocity zone (=7.5 km/s) beneath the subducted seamount along the seismic survey line KY9903. This serpentinized mantle is associated with past plume activity along the Kinan seamount chain. Kodaira et al. [2002] explained the seismicity in the uppermost mantle beneath Shikoku Island by dehydration of serpentinized mantle. Dehydration of serpentinized mantle is supported by a possible trapped water layer on the subducting oceanic crust above the high seismicity in the upper most mantle [Kodaira et al., 2002]. The trapped water layer was thought to be produced by dehydration of the uppermost mantle. A low-resistive layer at the top of the subducting lithosphere revealed by a magnetotelluric (MT) study [Yamaguchi et al., 1999] supports this hypothesis. [36] If the mantle portion of the subducting slab has been hydrated at the trench, dehydration may begin at a depth of about 20 km, based on thermal modeling and dehydration loci of serpentine [Wang et al., 1995; Seno et al., 2001]. Earthquakes in the mantle observed by our OBS networks were deeper than 20 km depth. Although the exact area of the serpentinized mantle could not be estimated, the seismicity beneath Shikoku Island implies that the hydrated mantle extends about 100 km from the seamount along the survey line. The seismicity in the uppermost mantle is active only in the area along the line KY9903 that is a northern continuation of the Kinan seamount chain [Kodaira et al., 2000a, 2002]. This observation supports the contention that serpentinized mantle along the Kinan seamounts chain relates to the uppermost mantle seismicity. The seismicity in the uppermost mantle continues from beneath the Shikoku Island to the OBS array. The dehydration of serpentinized mantle and resulting embrittlement may relate to the earthquakes in the uppermost mantle beneath the OBS array. 6. Conclusions [37] We observed seismicity off Cape Muroto around the updip limit of the seismogenic zone using OBSs. The obtained hypocenters based on a 3D velocity structure may be classified into two types. The first is seismicity around the top of the subducting oceanic crust, and the second is in the uppermost mantle of the subducting Philippine Sea plate. Earthquakes around the top of the subducting oceanic crust form several clusters characterized by pairs of earthquakes with very similar waveforms. These earthquake clusters are considered to be located in the transition zone of the interplate coupling, where the interplate coupling changes from aseismic stable sliding to seismogenic locked. We suggest that these similar earthquakes occurred at small asperities in the transition zone. Stress is concentrated on these asperities by steady slip in the surrounding area where the interplate coupling is relatively weak. Deeper earthquakes in the uppermost mantle could be explained by dehydration embrittlement of serpentinized mantle, with initial hydration of the mantle provided by past plume activity along the Kinan seamount chain. [38] A part of the seismicity around the updip limit of the Nankai Trough seismogenic zone was made clear by repeated OBS observations for nine months in total. The seismicity described in this article is, however, only a snap shot. Seismicity would change in time and space during the interseismic periods between large thrust earthquakes, and over longer timescales. Focal mechanisms are still unknown as the magnitudes of the events were small and the number of OBSs was insufficient to determined the focal mechanism from the polarities of first motions. These are subjects for future studies. [39] Acknowledgments. A part of this study has been done as JAMSTEC Frontier Research Program for Subduction Dynamics. We would like to thank the captains, crew members, and shipboard scientists of the R/V Kairei, R/V Kaiyo, R/V Yokosuka (Japan Marine Science and Technology Center), R/V Tansei-maru (Ocean Research Institute, Univ. of Tokyo), M/V Shintatsu-maru (Shin Nippon Kaiji Co., Ltd.), M/T Choyo-maru, M/V Asean-maru (Dokai Tug Boat Co., Ltd.), and M/V Kaiko-maru No. 5 (Offshore Operation Co., Ltd.) for OBS deployment and recovering. We acknowledge N. Takahashi, E. Araki, S. Yoneshima, Y. Nakamura, T. Higashikata, A. Nakanishi, T. Kanazawa, and marine technicians of Nippon Marine Enterprises, Ltd. for their support. A. Smith and M. Handler are greatly acknowledged for careful corrections of the manuscript. Some figures were drawn by Generic Mapping Tools [Wessel and Smith, 1995]. We thank G. Abers, S. Hussen and an anonymous reviewer for their constructive comments. References Ando, M., Source mechanisms and tectonic significance of historical earthquakes along the Nankai Trough, Japan, Tectonophysics, 27, , Baba, T., Slip distributions of the 1944 Tonankai and 1946 Nankai earthquakes including the horizontal movement effect on tsunami generation, in Frontier Research on Earth Evolution, vol. 1, pp , IFREE Jpn. Mar. Sci. and Technol. Cent., Yokosuka, Japan, Byrne, D. E., D. M. Davis, and L. R. Sykes, Loci and maximum size of thrust earthquakes and the mechanics of the shallow region of subduction zones, Tectonics, 7, , Davis, D., J. Suppe, and F. A. Dahlen, Mechanics of fold-and-thrust belts and accretionary wedges, J. Geophys. Res., 88, , Dominguez, S., S. E. Lallemand, J. Malavieille, and R. von Huene, Upper plate deformation associated with seamount subduction, Tectonophysics, 293, , Engdahl, E. R., and C. H. Scholz, A double Benioff zone beneath the central Aleutians: An unbending of the lithosphere, Geophys. Res. Lett., 4, , Frohlich, C., S. Billington, E. R. Engdahl, and A. Malahoff, Detection and location of earthquakes in the central Aleutian subduction zone using island and ocean bottom seismograph stations, J. Geophys. Res., 87, , Geller, R. J., and C. S. Mueller, Four similar earthquakes in central California, Geophys. Res. Lett., 7, , House, L. S., and K. H. Jacob, Thermal stresses in subducting lithosphere can explain double seismic zone, Nature, 295, , Husen, S., E. Kissling, E. Flueh, and G. Asch, Accurate hypocenter determination in the seismogenic zone of the subducting Nazca Plate in northern Chile using a combined on-/offshore network, Geophys. J. Int., 138, , Hyndman, R. D., K. Wang, and M. Yamano, Thermal constraints on the seismogenic portion of the southwestern Japan subduction thrust, J. Geophys. Res., 100, 15,373 15,392, 1995.