Insight into complex rupturing of the immature bending normal fault in the outer slope of the Japan Trench from aftershocks of the 2005 Sanriku earthquake (Mw = 7.0) located by ocean bottom seismometry
 The distribution of aftershocks of a large (Mw = 7.0) normal faulting earthquake beneath the outer slope of the Japan Trench in 2005, measured in 2007 using ocean bottom seismographs, indicates that the earthquake was involved with a set of conjugate normal faults. Although the faults reach to the upper mantle, the estimated Vp and Vp/Vs show no remarkable changes that can be related to extensive hydration in the crust or uppermost mantle. The absence of horst-graben topographic structure in the rupture area suggests that immaturity of the bending fault system is responsible for the relatively unhydrated lithosphere. Several earthquakes below the aftershock zone may belong to the lower plane seismicity of the shallow double seismic zone. Because no earthquakes were recorded in the area for more than 80 years before the 2005 event, shallow extensional and deep compressive earthquakes may be activated concurrently in the focal area of this earthquake.
 On the seaward side of a trench axis—the trench outer slope—many shallow normal faulting earthquakes occur to accommodate near-surface extension of the bending oceanic lithosphere [Forsyth, 1982; Seno and Yamanaka, 1996]. Recent results of seismic surveys made in trench outer slope region in the middle to southern America suggest that this faulting process is closely related to the hydration of the oceanic lithosphere. Ranero et al.  found that normal faults cut through to the oceanic lithospheric mantle and suggested that these faults can be pathways for fluids to enter the lower crust and mantle. Reduced seismic velocity in the uppermost mantle beneath normal faults is considered evidence of hydration caused by downward water transportation though the tensile fault [e.g., Ivandic et al., 2008]. It is essential for understanding how deep water infiltrates into the oceanic lithosphere to clarify focal depth distribution of the bending earthquakes. Most of the previous studies were based on waveform analyses of teleseismic records and precise focal depths have been given to large events [e.g., Seno and Gonzalez, 1987] but few studies on microseismicity representing depth extent of seismogenic layer have been made [e.g., Hirata et al., 1985; Lefeldt et al., 2009].
 Dehydration of descending oceanic lithosphere is considered to play an important role in generating intermediate to deep focus earthquakes within the downgoing slab [e.g., Kirby et al., 1996; Seno and Yamanaka, 1996; Hacker et al., 2003]. Intraslab earthquakes can reactivate existing faults as water released by dehydration reactions lowers effective normal stress on the faults. Normal faults developed in the outer trench slope may act later as rupture planes of intraslab earthquakes [Ranero et al., 2005]. Warren et al. [2007, 2008], however, claimed that observed orientation of rupture planes of intermediate- and deep-focus earthquakes in the Tonga-Kermadec and Middle America Trench subduction zones cannot be explained by reactivation of the normal faults at depth.
 An earthquake of Mw = 7.0 occurred on 14 November 2005 in the outer rise of the Japan trench (Figure 1), where the Mesozoic Pacific plate is subducted at a rate of 8 cm/a beneath the North American plate [DeMets et al., 1994]. The focal mechanism was normal faulting with the T axis in trench-normal direction (GCMT, http://www.globalcmt.org). This earthquake was the largest in the region since the 1933 great Sanriku earthquake (Mw = 8.4) [Kanamori, 1971]. Previous studies based on far-field observations have found it difficult to locate rupture planes of large bending earthquakes. Aftershock distributions determined by ocean bottom seismographs (OBSs) deployed in the focal area promise to significantly improve our understanding about the faulting process of shallow bending earthquakes. We report precise aftershock locations for the 2005 earthquake as well as the seismic velocity structure in the focal area, based on OBS data.
2. Data and Analyses
 We deployed six OBSs in the rupture area of the 2005 earthquake (Figure 2a) that provided continuous three-component waveform records from 27 April to 21 June 2007. In designing the array, we used main shock and aftershock epicenters relocated by E. R. Engdahl (personal communication, 2006) (Figure 2a) because the locations determined using seismic network data from the Japan Meteorological Agency (JMA) can be biased by structural heterogeneities, such as the high-velocity slab. For our event detections, we used a ratio of short-term average to long-term average (STA/LTA) of squared amplitude of horizontal component waveform records and picked arrival times of P and S waves for the detected events. More than 900 events were detected, of which 309 events with more than three P and three S readings were analyzed. We did not include events with S–P times larger than 30 s. In the arrival time picking, we also read polarities of P first arrivals from vertical component traces with enough emergent onsets.
 We used the code tomoDD [Zhang and Thurber, 2003] to determine hypocenters and simultaneously the depth variation of Vp and Vs by putting a vertical column of velocity nodes at the center of the aftershock area. We expect that additional use of double-difference travel time data will improve constraints on Vp and Vs structures. For initial hypocenters, we used a conventional determination assuming a 1-D Vp structure similar to the result of the active source experiment made in this area [Azuma et al., 2008]. In the initial hypocenter determination, Vp/Vs was assumed to be 1.80 after the results of Shinohara et al. . We also used the 1-D Vp structure as the starting model in the tomoDD analysis using two values of Vp/Vs, 1.75 and 2.00, as initial values.
 In our hypocenter determination and our inversion for seismic velocity using OBS data, we made station corrections to compensate for low seismic velocities in the sedimentary layer. We assumed a structure in which the sedimentary layer is replaced by a crustal layer, then applied the resulting delay times for P and S arrivals at each of the OBS stations. We estimated the sediment thickness from arrival time differences between the P wave and the PS wave converted at the base of the sedimentary layer. The PS arrivals were clear and easily picked on the horizontal component seismograms (Figure 3a).
 We estimated sizes of the observed earthquakes by using radiated seismic energy relative to a reference earthquake for which magnitude was determined by JMA. Relative seismic energy can be measured by taking the ratio of squared amplitude of S coda envelope of the target to that of the reference event as
Here, Ws is radiated energy of S wave and ∣jS coda(tc)∣2 is squared amplitudes of S coda envelope at absolute lapse time tc calculated from jth component velocity seismogram (Figure 3b). Ref denotes the reference earthquake and i is an index of slave events for which magnitude is to be determined. Brackets mean temporal average. Using a relation between seismic energy and moment magnitude [Stein and Wysession, 2003]
magnitude of ith slave event (Mi) can be calculated from magnitude of the reference event (MRef) as
In the calculation, we used the latter part (after twice the direct S travel times) of the coda amplitude to avoid the effects of the radiation pattern [Sato and Fehler, 1998].
 The epicenters of the events in our OBS data form a cluster elongated in a NNE to SSW direction with a length of about 40 km and width of 20 km (Figures 2a and 4a). This cluster has the same dimensions as the aftershock distribution defined by JMA data, but its location is shifted ∼20 km to the east. Direct comparison of epicenter locations by both JMA and us for the same set of 10 aftershocks showed the same degree and direction of eastward shift. As no temporal expansions are evident in the epicenter distribution since the occurrence of the main shock (Figures 2b and 2c), we are confident that the cluster of earthquakes located by our OBS observations in 2007 corresponds to the aftershock activity of the 2005 earthquake. Note that epicenters of aftershocks determined by E. R. Engdahl are located well within the epicenter distribution constrained by our OBS data (Figure 2a).
 Most of the hypocenters were located from 5 to 15 km beneath the seafloor (Figures 4b and 4c). As the estimated thickness of the oceanic crust is 7 km [Azuma et al., 2008], the aftershock activity was present in both the oceanic crust and the uppermost part of the mantle. There is an almost aseismic layer about 4 km thick at the top of the crust. Because the sedimentary layer is only ∼1 km thick, the upper part of the oceanic crust, possibly corresponding to the oceanic layer 2, is also aseismic.
 The depth variation of the Vp and Vp/Vs ratio (Figure 4d) was almost independent of the choice of initial values of Vp/Vs ratio; the plotted Vp and Vp/Vs curves were derived with a starting value of Vp/Vs = 1.75. The errors shown in Figure 4d were estimated by using a bootstrap method with grid points set at depths of 5, 7, 9, 11, 13, 15, and 45 km. The 30-km grid spacing in the mantle was necessary to achieve stable solutions given the wide spacing of hypocenters at this depth and the velocity contrasts between crust and mantle.
 The topmost part of the oceanic crust was characterized by a steep vertical gradient in Vp, which is characteristic of oceanic layer 2. Few hypocenters were located in this depth range, in keeping with the aseismic nature of layer 2. In the lower part of the oceanic crust, layer 3, Vp was estimated to be around 7 km/s. Estimated Vp/Vs ratios were high, approximately 1.8, in the oceanic crust. Another steep rise in Vp in the depth range from 13 to 15 km reflects the velocity jump at the Moho. Our inversion did not assume this velocity jump explicitly, and it is difficult to distinguish Vp and Vp/Vs values for the bottom of the crust and the topmost mantle. We used the Vp and Vp/Vs estimated for the grid deeper than 15 km, ∼8.0 km/s and 1.70 to 1.78, respectively, as mantle values.
 Detailed examination of the depth distribution of the aftershocks suggested two characteristics: a slight deepening along the NNE–SSW axis of the distribution (Figure 4b) and a concave distribution in the WSW–ENE cross section (Figure 4c). Although no clear planar structures are evident on the vertical cross section, when they are projected on a plane dipping 77° to the NNW, the aftershock hypocenters are found to lie on two nearly orthogonal planes, one plane dipping 55° east and another dipping 35° west (Figure 5a). The two planes meet at the bottom of the aftershock cluster, and a V-shaped aftershock distribution is apparent.
 The orientations of the two aftershock distribution planes are shown in Figure 5b on a lower hemisphere focal sphere projection along with the P first motions of the aftershocks, defined as the earthquakes with focal depths <20 km. The two planes of the aftershock distribution correspond well to the nodal planes separating dilatant and compressive quadrants of the P wave motions. Furthermore, the polarity distribution is well explained by the focal mechanism solution of the main shock. These observations indicate that the aftershock distribution reflects the geometry of the fault plane of the normal faulting main shock and that most of the aftershocks occurred on or near fault planes matching the main shock focal mechanism solution.
 The earthquakes we detected and located in OBS observations had a magnitude range from about 1 to 4 (Figure 5c). The size distributions of the aftershocks differed between the two planes. Aftershocks with larger magnitudes tended to be located along the east dipping plane, and their b value was lower (∼0.7) than that of the aftershocks along the west dipping plane (∼1.0). Since the number of earthquakes used in this statistics is rather small, b values may not be estimated stably in absolute sense. But we regard it significant that large aftershocks tend to be diminished along the trenchward dipping plane compared to those along the oceanward plane. Short-wavelength component might be more dominant in heterogeneity along the trenchward dipping plane with larger b value.
4. Discussion and Conclusions
 The aftershocks of the 2005 earthquake lay along two conjugate planes whose orientation corresponds to that of the nodal planes of the main shock focal mechanism. There can be two explanations for this complex aftershock distribution. It is possible that the main shock ruptured both of two conjugate fault planes almost simultaneously [e.g., Robinson et al., 2001; Kato et al., 2008]. Another interpretation is that the main shock was a rupture along a single plane but its conjugate fault was eventually activated in the aftershock activity [e.g., Sakai et al., 2005]. Even if the main shock of the 2005 outer rise earthquake was a rupture of single fault plane, it is quite difficult to distinguish the actual fault plane from these two nodal planes using only far-field waveform data [Yamanaka, 2005]. Our results showed that large aftershocks were more abundant along the east dipping plane. If this pattern has persisted throughout the aftershock activity, our observations may indicate that the east dipping plane was the rupture plane of the main shock.
 It is often assumed that fault planes of bending earthquakes in trench outer slopes dip toward trenches [e.g., Ranero et al., 2005], while Jiao et al.  assumed coexistence of trenchward and oceanward dipping faults. Our study proves that both of the trenchward and oceanward dipping planes can be ruptured by bending normal faulting earthquakes beneath the trench outer slope. In Japan trench, existence of normal faults with opposite dipping directions is well demonstrated by seafloor togography [Kobayashi et al., 1998] and seismic reflection data [Tsuru et al., 2000]. Fromm et al.  found a tensional outer rise earthquake of Mw = 6.8 occurred off Chile had aftershock distribution forming a pair of conjugate planes conforming to the main shock focal mechanism.
 If intraslab earthquakes are caused by reactivation of existing faults, we may expect earthquakes rupturing the faults generated by outer rise bending earthquakes such as the 2005 event. The rupture plane of a large (Mw = 7.0) downdip compression intraslab earthquake in 2003 (Figure 1a) was estimated from its aftershock distribution to be almost vertical. The fault cut through the oceanic crust to the upper mantle [Okada and Hasegawa, 2003]. Given that the slab dips by about 30°, this rupture plane corresponds to a plane in the incoming oceanic lithosphere dipping about 60° landward, corresponding to one of the fault planes associated with the 2005 outer rise earthquake. Kita et al.  precisely relocated the hypocenters of earthquakes in the deep double seismic zone beneath the northeastern Japan arc [Hasegawa et al., 1978] and showed the existence of planar clusters of seismicity extending in vertical and horizontal directions in the upper plane. One of the horizontal planes was found to be attached to the fault of the 2003 earthquake, forming a conjugate system. These observations can be interpreted as evidence of the reactivation of complex fault systems that first developed in the outer rise area.
Warren et al. [2007, 2008] studied orientation of fault planes of intermediate- and deep-focus intraslab earthquakes and successfully identified earthquakes with subhorizontal fault planes. They argued that the subhorizontal ruptures are not consistent with the idea of reactivation of the normal faults created beneath trench outer slope, assuming that all the bending faults are trenchward dipping. However, the subhorizontal faults imaged by the microseismicity in the Pacific slab beneath the northeastern Japan is consistent with the orientation of the oceanward dipping plane of the aftershock distribution of the 2005 earthquake. Therefore, existence of subhorizontal fractures in the deep slab does not rule out plausibility of reactivation hypothesis explaining intraslab earthquake generation, at least in the Japan trench subduction system.
 Seafloor of the outer slope of the Japan trench shows well developed horst and graben structure with normal faults (Figure 1). Tsuru et al.  showed that throws of the normal faults increase gradually toward the trench axis. Since amount of fault throw represents accumulated fault displacements, normal faults near the trench axis with larger throws are more matured than distant faults with small offsets. In the focal area of the 2005 earthquake, about 80 km east from the axis of the Japan trench, no significant fault scarps can be recognized on the seafloor (Figures 1a and 2a). However, the fault ruptured by the 2005 earthquake seems to belong to the series of normal faults progressively developing toward the trench since the strike of the rupture plane is almost parallel to that of prominent fault scarps observed near the trench. Therefore, the earthquake can be interpreted as a rupture of relatively immature bending faults.
 The estimated Vp and Vp/Vs values in the rupture area of the 2005 earthquake are not anomalous as compared to observations from Central and South America [e.g., Ivandic et al., 2008; Ranero and Sallares, 2004]. The values are similar to those in the center of the northwest Pacific basin [Shinohara et al., 2008], where no significant tectonic processes modifying seismic structure are expected. Therefore, we conclude that the velocity reduction due to faulting and water percolation into the oceanic lithosphere in the 2005 earthquake area was lower than the detection level of our study.
 One reason for the minimal structural modification of the velocity structure in this area may be the immaturity of normal faulting in this area. In the central American subduction zone, significantly reduced Vp is observed in the oceanic crust and uppermost mantle beneath the outer trench slope but not beneath the outer rise without horst and graben structure [Ivandic et al., 2008]. Seismic section presented by Tsuru et al.  shows several buried normal faults in the vicinity of the epicenter of the 2005 earthquakes, but their throws were too small to reach the seafloor. Because the sedimentary layer is thought to be impermeable, it might be difficult for seawater to enter into the lithosphere without faults reaching the seafloor surface.
 Apart from the aftershock activity, there were scattered earthquakes around 40 km depth, as shown in Figures 1 and 4. These earthquakes can be distinguished from the shallow aftershocks by S–P times at the OBSs. Although we could not determine focal mechanism solutions for the deep earthquakes, several earthquakes had P wave motions consistent with reverse faulting but not normal faulting, similar to the main shock. Existence of a thrust type earthquake with focal depth of about 40 km has been reported by Seno and Gonzalez . Although the epicenter of the earthquake was located at the axis of the Japan trench and about 100 km apart from the focal area of the 2005 earthquake, its focal depth is in good agreement with the deeper seismicity which we found by the OBS observation. Gamage et al.  documented a double-plane structure of the shallow seismicity near the Japan Trench from analyses of a depth phase recorded by the local land seismic network. There, thrust earthquakes are dominant in the lower plane seismicity, whose depth is about 40 km. From these similarities we suggest that the deep seismicity we observed is caused under compressional stress of the lower portion of the bending lithosphere, as proposed by Seno and Gonzalez  and Gamage et al. .
 No previous earthquakes were observed around the epicenter of the 2005 earthquake. This may indicate that shallow and deep seismicity were inactive prior to 2005. We think it improbable that all earthquakes in this area were smaller than the detection threshold of the land network for more than 80 years. The deep thrust earthquakes could be activated concurrently with shallow normal faulting or triggered by the occurrence of shallow large earthquakes.
 The focal area of the 2005 earthquake is adjacent to the rupture area of the 1933 Sarniku earthquake, a shallow normal faulting earthquake. But Kanamori  suggested that it ruptured entire thickness of the lithosphere and was different from ordinary earthquakes representing the bending stress along the surface of oceanic lithosphere. In the rupture area of the 1933 earthquake defined by Kanamori , there is currently active microseismicity. Gamage et al.  showed that the seismicity, including normal faulting events, can be explained as the aftershock activity of the 1933 earthquake. Depth range of the aftershock distribution is 10–23 km, almost the same as that of the aftershock activity of the 2005 earthquake (Figure 4). The correspondence indicates that tensional stress due to the lithosperic bending prevails down to this depth. It is also notable that the coseismic slip of the 2007 Kuril outer rise earthquake (Mw = 8.1) was concentrated in the upper 25 km of the oceanic lithosphere [Ammon et al., 2008]. Ordinary shallow bending earthquakes can rupture down to 20–25 km but may not extend any deeper than that in the old Pacific lithosphere, except for the 1933 earthquake which could be a very unusual earthquake. Lack of deep focus aftershocks in the current activity may indicate that the present stress state is different from that prior to the occurrence of the 1933 earthquake; temporal change of stress state might be required to cause an unusually large normal faulting earthquake.
 OBS deployment and retrieval were conducted by R/V Ryofu-maru and R/V Kofu-maru, JMA. We thank P. van Keken, the editor of G-cubed, and an associate editor for their valuable comments. Careful review by S. Bilek and an anonymous reviewer is cordially acknowledged. Discussion with S. Kirby, H. Fujimoto, and A. Nishizawa was helpful. We used J-EGG500 bathymetric data provided by Japan Oceanographic Data Center. We also thank E. R. Engdahl and T. Yamada for providing their epicenter data. Figures 1–5 were drawn by using the Generic Mapping Tool developed by P. Wessel and W. H. F. Smith. This study was partly supported by JSPS KAKENHI (19540436) and 21st COE program of Tohoku University.