We performed 3-D seismic tomography in the forearc region of the northeastern Japan subduction zone using both onshore and offshore seismic station data. We obtained the Vp, Vs, and Vp/Vs structures around the plate boundary with high spatial resolution. The position of the plate boundary as defined by relocated hypocenters coincides with the sharp velocity boundary between the oceanic crust and the mantle wedge. The mantle wedge above the coseismic slip area of the 1978 and 2005 off-Miyagi interplate earthquakes (M > 7) is characterized by high Vp and Vs, but low Vp/Vs. Off Fukushima, however, where large earthquakes rarely occur, we found a high Vp/Vs anomaly at the tip of the mantle wedge. The spatial distribution of serpentinized mantle wedge limits the spatial extent of the strongly coupled area on the plate boundary, and thus can explain the difference in seismic activity between the off-Miyagi and off-Fukushima regions.
 The Pacific plate is subducting beneath the North American plate at the Japan Trench (Figure 1). Many large earthquakes occur along the plate boundary in the NE Japan forearc region [e.g., Yamanaka and Kikuchi, 2004]. In this region, the downdip limit of the thrust type earthquakes is about 60 km in depth [Igarashi et al., 2001], whereas the average thickness of the island arc is 30 km [e.g., Miura et al., 2003]. GPS studies have shown that there is spatial heterogeneity in the back-slip distribution on the plate boundary, and that an area of strong interplate coupling corresponds to the distribution of asperities associated with large earthquakes [e.g., Iinuma et al., 2007].
 In the off-Miyagi region, middle of the NE Japan forearc, interplate earthquakes of M ∼7.5 occur repeatedly at intervals of about 40 years (Headquarters for Earthquake Research Promotion, List of long-term evaluation (in Japanese), 2008, http://www.jishin.go.jp/main/choukihyoka/kaikou.htm). The most recent earthquake occurred in 1978 (M 7.4). Seno et al.  proposed that this earthquake was comprised of three rupture areas. A thrust-type earthquake (M 7.2) on 2005 off Miyagi is considered as the re-rupturing of one of the three asperities of the 1978 earthquake [Okada et al., 2005].
 South of the off-Miyagi region (the off-Fukushima region), large earthquakes have rarely occurred. The 1938 Shioya-oki earthquake sequence, comprising four earthquakes with Ms ∼7.8 [Abe, 1977], is the only known major interplate activity in more than 800 years. Iinuma et al.  showed that the area of strong interplate coupling is narrower off Fukushima than off Miyagi (Figure 1).
 Several seismic tomographic studies have been carried out in the NE Japan forearc region. Zhao et al.  found that an area with a history of no large earthquakes corresponded spatially to a part of the mantle wedge with low velocity and a high Poisson ratio, but the spatial resolution of their study was too limited to characterize individual asperities associated with large earthquakes. Yamamoto et al.  showed that an area of the mantle wedge with high Vp corresponded to the asperity of an off-Miyagi earthquake, but they could not resolve Vs structure. To discuss the cause of the correlation between seismic structure and seismogenesis along the plate boundary, both the Vs and the Vp/Vs structures must be known.
 To clarify heterogeneities in the mantle wedge corresponding to the differences in interplate coupling between the off-Miyagi and off-Fukushima regions, we estimated the 3-D seismic velocity structure by seismic tomography. We used travel-time data from an onshore seismic network and ocean-bottom observations obtained from 2004 to 2005 in the forearc region off Miyagi and off Fukushima.
2. Data and Method
 We used ocean-bottom observation data from free-fall/pop-up type ocean bottom seismographs (OBSs) equipped with three component sensors. We compiled data obtained during four deployments in 2004 and 2005: (1) 15 OBSs in the off-Miyagi, observations from May to October 2005; (2) 15 OBSs additional to (1), observations of aftershocks of the 2005 off-Miyagi earthquake [Hino et al., 2007]; (3) temporary network of 20 OBSs off Fukushima from June to November 2005; and (4) OBSs deployed for an active-source survey in August 2004 [Watanabe, 2005], used as passive seismic stations. We used data from a total of 89 OBS stations in this study. We also used data from a seismic network operated by the National Institute for Earth Sciences and Disaster Prevention, Japan Meteorological Agency (JMA), and national universities comprising 141 on-land stations and four OBSs (three in the northern and one in the southern study area).
 We selected the target earthquakes for our analysis from the JMA catalogue according to the following criteria: (1) M > 2, and (2) epicentral distance to the nearest station < 30 km. We included intermediate-depth earthquakes among the target events to improve the resolution of the tomographic imaging of the downgoing slab.
 We relocated the JMA-catalogued hypocenters of the earthquakes in the study area by adding the OBS arrival-time data to onshore data, assuming a 1-D velocity structure model [Hino et al., 2007], to obtain the initial hypocenters for the tomographic inversion. We also used data of 1658 earthquakes that occurred on land and were recorded only by the onshore stations because we needed to determine the detailed seismic structure of the land area as well as of the offshore region.
 To obtain the detailed structure of the seismogenic zone around the plate boundary, we applied the double-difference tomography method “tomoFDD” [Zhang and Thurber, 2006]. We obtained 144,878 P-wave arrivals and 117,501 S-wave arrivals from 3293 local earthquakes by manually picking. We used the picked arrival times whose accuracies were less than 0.1 and 0.2 s for P-waves and S-waves, respectively. From these arrival data, we obtained 398,548 double-differences for P and 334,523 for S [Waldhauser and Ellsworth, 2000] from event pairs separated by no more than 10 km.
 We established a grid with nodes spaced at intervals of 25 km (15 nodes) in the trench-normal, 30 km (13 nodes) in the trench-parallel, and 5 km (31 nodes) in the vertical direction. Vp and Vs in the initial model were taken from the 1-D model used in the hypocenter determination. To remove the dependence on the distribution of the grid nodes from tomography results, after the first inversion, we carried out another set of inversions by shifting the grid nodes horizontally by half a grid interval in each horizontal direction. We calculated the mean velocities of the two sets of inversion results to obtain the final velocity model.
3. Results and Discussion
 We performed a checkerboard resolution test (CRT) [Spakman and Nolet, 1988] and a restoring resolution test (RRT) [Zhao et al., 1992] to assess the resolution of our results. In these test, we added the noise of 0.1 s for P-wave and 0.2 s for S-wave to each synthetic data. From the CRT results (Figures S1a–S1d in the auxiliary material), we concluded that we could evaluate velocity heterogeneity at a resolution of about 10 km in the vertical direction and 1 grid interval in each horizontal direction. On the basis of the CRT results, we regarded the estimated seismic velocities to be well constrained in the area with derivative weight sum (DWS) values [Thurber and Eberhart-Phillips, 1999] > 2000 for Vp, and > 1000 for Vs. In these areas, the inverted velocity distribution obtained by the RRT agreed within ±1% difference with the inversion results (Figures S1e–Sh). The mean accuracy of the hypocenter location was 0.39 km in the horizontal directions and 0.34 km in the vertical direction. The root mean square of the travel-time residuals was reduced from 0.64 s in the initial model to 0.15 s in the final model obtained after 20 iterations.
 Because most of the relocated earthquakes were distributed along the landward-dipping layer, we estimated the position of the plate boundary from the relocated hypocenters. The locations of thrust-type earthquakes and characteristic repeating earthquakes (CREs) [Uchida et al., 2006] in particular were used to define the boundary. We estimated the focal mechanisms by using the P-wave first-motion polarity and the S-wave/P-wave amplitude ratio, using the FOCMEC software package [Snoke, 2003]. The plate boundary determined from the hypocenter distribution corresponds to a sharp boundary in the obtained Vp and Vs distributions. We identified a high-velocity zone above the plate boundary, and a landward-dipping layer with low velocities below the plate boundary at depth down to about 60 km (Figures 2 and S2). We inferred that these high- and low-velocity layers correspond to the mantle wedge and oceanic crust, respectively.
 Since it is probable that the mantle wedge just above the plate boundary strongly affects the frictional properties of the plate interface, we examined spatial heterogeneities in the mantle wedge and their relation to spatial variation in the interplate coupling. Because the vertical resolution of our velocity model was about 10 km, we used the mean vertical velocity in the 10-km-thick layer above the plate boundary to represent the velocity just above the plate boundary. We set the bottom of this layer at 2 km above the boundary to avoid smearing from the oceanic crust (Figures 2a and S2a, orange line). We also excluded values from shallower than 25 km from the calculation to avoid smearing from the continental crust.
Figures 3a–3c show the Vp, Vs and Vp/Vs structures of the mantle wedge just above the boundary. The mantle wedge above the asperities of the 1978 off-Miyagi earthquake [Yamanaka and Kikuchi, 2004] is characterized by high Vp, Vs and low Vp/Vs. Vp decreases northward in the mantle wedge, and this is consistent with Yamamoto et al. . On the other hand, a relatively high Vp/Vs (>1.8) zone is present at the eastern, updip end of the mantle wedge in the off-Fukushima forearc region. In the mantle wedge, Vp/Vs decreases to less than about 1.75 and Vp increases to 8 km/s toward the west, showing structural variation in the downdip direction to a plate boundary depth of about 40 km or more. These patterns of velocity variation in the mantle wedge are consistent with the results of previous active-source experiments conducted in the off-Miyagi and the off-Fukushima regions [Watanabe, 2005; Miura et al., 2003].
 Serpentinization in the mantle wedge is one of the causes of decreasing velocity in the mantle wedge. Several seismic tomography studies have found low velocities in the mantle wedge, for example, in Cascadia [Ramachandran et al., 2005] and Costa Rica [DeShon and Schwartz, 2004]. Interplate coupling is believed to be weak along the plate boundary where the oceanic crust contacts the hydrated mantle wedge [e.g., Hyndman and Peacock, 2003]. These low-velocity anomalies are interpreted to indicate serpentinization of mantle wedge, and in these regions, the plate boundary is reported to be almost aseismic in their papers.
 We estimated the volume fraction of serpentinized mineral in the mantle wedge of the NE Japan forearc from our velocity models using the relation proposed by Christensen . The mantle wedge above the rupture area of the off-Miyagi earthquakes is less than 10% serpentinized, whereas at the tip of the mantle wedge off Fukushima serpentinization of 20–40% is inferred. Almost identical values were obtained from the Vp, Vs, and Vp/Vs structures. Escartin et al.  reported that the presence of 10–15% serpentine reduces the strength of altered peridotite to that of pure serpentinite. On the basis of their result, we suggest that the rheological characteristics of the mantle wedge are significantly different between the off-Miyagi region and the updip region in the off-Fukushima. Our velocity model indicates that interplate coupling is strong in the off-Miyagi, whereas the tip of the mantle wedge in the off-Fukushima is an area of weak coupling. This interpretation is consistent with the findings of a back-slip distribution study that showed weak coupling in the off-Fukushima [Iinuma et al., 2007; Matsumoto et al., 2008].
 The high Vp/Vs region seems to occupy the eastern half of the rupture area of the 1938 Shioya-oki earthquake (Figure 3b). However, using GPS data, Suwa  detected an episodic slow-slip event in the off-Fukushima forearc and located this event in the eastern part of the 1938 earthquake rupture zone. Combining this result with ours, we suggest that the shallow part of the fault plane of the 1938 earthquake estimated by Abe  is not a strongly coupled area but that interplate coupling is strong in the deep part.
 Our results suggest that a non-serpentinized mantle wedge exists along the plate boundary between 30 and 60 km depth off Miyagi, making the width of the strongly coupled zone about 70 km in the downdip direction. Off Fukushima, the presence of the greatly serpentinized mantle in the updip end of the mantle wedge means that the strongly coupled zone there is about 40 km wide, much narrower than that in the off-Miyagi. The difference in the width of the less-serpentinized mantle wedge is the most notable difference in structure between the off-Miyagi and off-Fukushima regions. We suggest that the width of the highly serpentinized part of the mantle wedge influences the spatial extent of the interplate seismogenic zone. The difference in the amount of fluid-rich subducted sediments in along-arc direction [Tsuru et al., 2002] may be responsible for the heterogeneous distribution of the serpentinized mantle. Kato and Hirasawa  reported on the basis of a numerical study that episodic creep events become dominant in a subduction zone with a narrow seismogenic zone. Their result well explains the seismic behavior in the off-Fukushima region.
 We performed a 3-D seismic tomography study in the middle to southern part of the NE Japan forearc region to clarify the relationship between spatial heterogeneities in the velocity structure and the spatial variation in interplate coupling. We obtained well-resolved, stable velocity structures for both Vp and Vs. The plate boundary was estimated from the relocated hypocenters of CREs and earthquakes with thrust-type focal mechanisms. This estimated plate boundary coincides with the velocity boundary between the subducting oceanic crust and the mantle wedge.
 We found a spatial correspondence between seismic velocity variation in the mantle wedge just above the plate boundary and interplate seismic coupling. In the off-Miyagi region, where M > 7 interplate earthquakes frequently occur, the mantle wedge shows low Vp/Vs, which we interpreted as indicating a less-serpentinized state. On the other hand, at the updip end the mantle wedge in the off-Fukushima region, where large interplate earthquakes are rare, Vp/Vs was high, possibly because of extensive serpentinization. The spatial distribution of serpentinized mantle thus limits the spatial extent of the interplate seismogenic zone and explains the less-frequent occurrence of large earthquakes off Fukushima.
 We thank F. Florindo, H. R. DeShon and one anonymous reviewer for helpful comments, S. Miura, A. Ito, A. Kuwano, E. Araki, and M. Nishino for useful discussions. N. Uchida and T. Iinuma kindly gave us their latest results. We also thank the captains, crews, and scientists on board R/Vs Shinnichi-Maru, Seifu-Maru, Kofu-Maru, Ryofu-Maru, Hakuho-Maru, Tansei-Maru, and Yokosuka for their kind support. Figures were prepared using Generic Mapping Tools [Wessel and Smith, 1991]. JSPS research fellowships for young scientists and the 21st COE program of Tohoku University supported this study.