We show contrasting subduction structures within the Philippine Sea plate inferred from active-source wide-angle reflection data. Previous studies showed that large-amplitude reflections from the slab are observed in southwest Japan and indicated that a thin low-velocity layer with a high fluid content is formed along the top of the subducting oceanic crust. On the contrary, we found that the slab reflections have smaller amplitudes in the Izu collision zone, central Japan, where the Izu-Bonin volcanic arc has been colliding/subducting, suggesting that such a low-velocity layer does not exist beneath the collision zone. This structural difference is also supported by P-wave and S-wave velocity anomalies by passive-source tomography and electrical conductivity, and correlates with the regional distribution of deep tremors and intraslab earthquakes, both of which are induced by dehydration processes within the downgoing slab. Based on these comparisons, we suggest that the original structure of the incoming plate controls the contrasting subducting systems: typical oceanic plate absorbs water by hydrothermal circulation at spreading centers and/or seawater infiltration at outer rises, whereas volcanic arc crust consumes a large amount of hydrous minerals for melt production and metamorphoses to more stable, anhydrous forms before subduction.
The mechanical friction associated with plate subduction is a fundamental mechanism for earthquake generation. Some of the largest earthquakes occur at subduction zones, where strain is accumulated and released at the contact between the two converging plates. Recent studies showed that an incoming plate mainly controls overall subduction processes [e.g., Hyndman and Peacock, 2003, Yamasaki and Seno, 2003], and that especially progressive dehydration reactions within the slab play a significant role in seismic activities at the plate interface [e.g., Kato et al., 2010]: high pore pressures due to fluids expelled from hydrous minerals probably affect frictional properties along the plate interface and within the downgoing slab [Kodaira et al., 2004; Liu and Rice, 2007; Moore and Lockner, 2007]. Other seismogenic behavior in subduction systems includes various types of seismic activities, such as slow earthquake [Shelly et al., 2006] and intraslab earthquake [Hacker et al., 2003a; Yamasaki and Seno, 2003]. These dehydration-related processes should depend on pressure, temperature, and distribution of water in the slab. However, there are few studies that focus on their regional variation and little is documented about principle causes that make the difference between each subduction zone.
Subduction of the Philippine Sea plate beneath Japan is a great target to improve our understanding of relationship between release and transport of fluids, crustal rheology, and seismogenic processes in subduction zones since fine-scale subduction structures and associated seismicity are well defined by dense seismic observations. For example, nonvolcanic deep tremors [Obara, 2002] are observed along the Nankai subduction zone where the young, warm oceanic crust within the Philippine Sea plate is subducting beneath southwest Japan (Figure 1). Recent seismic studies revealed that other seismic activities with different dominant frequencies, such as short-term and long-term slow slips [Hirose and Obara, 2005, 2006], low-frequency earthquakes [Shelly et al., 2006], and very low frequency earthquakes [Ito et al., 2007], also occur in the Nankai subduction zone. Studies of velocity structures [Kodaira et al., 2004; Matsubara et al., 2009; Kato et al., 2010] showed that these events are confined in a zone with low velocities and high Vp/Vs ratios along the subducting plate and suggested that high fluid pressure plays an important role in generating such phenomena.
It is intriguing that the nonvolcanic deep tremors are almost continuously distributed beneath southwest Japan, but they suddenly disappear in the Izu collision zone, central Japan, where the Izu-Bonin volcanic arc has been colliding with the overriding plate (Figure 1) [Obara, 2002; Obara and Hirose, 2006]. In addition, intermediate-depth earthquakes (at ∼20–50 km depths within the subducting Philippine Sea plate) show a similar pattern (Figure 2a). Since these zoned patterns of seismicity are likely related to metamorphic dehydration processes within the slab, Seno and Yamasaki  postulated that, unlike the Nankai subduction zone, little water is released from the subducting Philippine Sea plate beneath the Izu collision zone. However, structural differences corresponding to this seismogenic pattern are poorly documented and further evidence is necessary to understand the regional seismotectonics associated with the subduction of the Philippine Sea plate.
This paper investigates physical properties of the subducting Izu-Bonin arc crust and compares them with the structures in the Nankai subduction zone from seismic refraction/wide-angle reflection data and intends to provide direct evidence for structural differences dominating seismogenic features along this subduction zone. Seismic reflectivity is a powerful tool for this purpose because recent studies in subduction zones revealed a good correlation between amplitude variations of wide-angle reflections and frictional properties on the plate interface: large-amplitude reflections are found from within aseismic regions that have weak plate coupling probably due to a high fluid content, while a weakly reflective interface corresponds to the coseismic rupture zone of large megathrust earthquakes [Nakanishi et al., 2004; Mochizuki et al., 2005; Azuma et al., 2012]. Together with other evidences from passive-source seismic tomography [Nakamichi et al., 2007; Arai, 2011] and electrical conductivity [Yamaguchi et al., 2009; Aizawa et al., 2004], we show contrasting subduction systems between hydrous oceanic crust beneath southwest Japan and anhydrous volcanic arc crust beneath the Izu collision zone, which supports the hypothesis by Seno and Yamasaki  that dehydration is inactive beneath the Izu collision zone. The results provide a new insight into the relationship between subduction properties and original structure of an incoming plate, which may also be applicable to other collision/subduction settings.
2 Structure of the Philippine Sea Plate
The Izu-Bonin arc, the intraoceanic volcanic arc located south of Japan, has developed on the eastern flank of the Philippine Sea plate since the Pacific plate started to subduct at ∼43 Ma [Stern et al., 2003]. The Izu-Bonin arc began to separate from the Kyushu-Palau ridge at 30 Ma, and a subsequent backarc spreading formed deep basins named Shikoku basin (Figure 1) and more southerly Parece Vela basin [Kobayashi et al., 1995; Okino et al., 1998, 1999; Ohara et al., 2001]. At almost the same time that the backarc spreading ceased (∼15 Ma), the Philippine Sea plate started to northwardly subduct along the Nankai trough, and the Izu-Bonin arc started to collide against the Honshu arc in central Japan [Amano, 1991; Takahashi and Saito, 1997; Aoike, 1999]. Studies of reconstruction modeling indicate that there was a change in a convergence direction between the Philippine Sea plate and Eurasian plate over its subduction history: the plate motion direction of the Philippine Sea plate had been more northwardly than the present northwestward direction until ∼4 Ma and approximately in the strike direction of the Izu-Bonin arc [Seno et al., 1988, 1993, 1996], resulting in collision tectonics in a relatively small area over time. This ideal tectonic setting enables us to study direct impacts of structural heterogeneities of an incoming plate on resulting subduction systems.
The aforementioned tectonic framework of the Philippine Sea plate can be also understood from the viewpoint of seafloor bathymetry: there is a significant along-trench variation in present seafloor topography, or correspondingly total crustal thickness (Figure 1): Shikoku basin is characterized by deep seafloor (>3500 m). Underlying the basin, the young oceanic crust has 5–10 km thickness with typical seismic structure; the upper crust with Vp of 2.0–6.8 km/s, the lower crust with Vp of 6.8–7.0 km/s, and the peridotite (and partially serpentinized) mantle [Kodaira et al., 2006; Nishizawa et al., 2006, 2007]. In contrast, the Izu-Bonin arc has shallower seafloor depth (<3000 m) and 20–35 km thick crust [Suyehiro et al., 1996; Nishizawa et al., 2006; Kodaira et al., 2007, 2008]. Seismic studies indicate that the Izu-Bonin arc has compositionally differentiated crust consisting of three major crustal layers: the upper volcanic layer with Vp < 6.0 km/s, the middle crust with felsic/intermediate rock composition (Vp = 6.0–6.8 km/s), and the mafic/ultramafic lower crust with Vp > 6.8 km/s [Kodaira et al., 2007; Takahashi et al., 2008; Tatsumi et al., 2008].
Along-dip variations in subduction structure and their relation to seismogenic processes are well documented in the Tokai district where the Philippine Sea plate is subducting through the Suruga trough (the eastern end of the Nankai subduction zone) (Figure 2a). In this area, megathrust earthquakes with magnitude of ∼8 have repeatedly occurred at intervals of approximately 150 years [Ando, 1975]. More recently, deep tremors and short-term and long-term slow slips were found to occur at the transitional domain from the coseismic (locked) zone to the deeper aseismic (conditionally stable) zone [Obara, 2002; Hirose and Obara, 2006; Suito and Ozawa, 2009]. Using active-source seismic data, Kodaira et al.  and Iidaka et al.  discovered a highly reflective region at the plate boundary. Kodaira et al.  correlated this reflective zone with the distribution of slow slip and suggested that high-pressure fluids supplied from hydrous minerals within the slab effectively extend a region of stable slips and consequently generate the slow slip. The interpretation by Kodaira et al.  was supported by further evidence from reflectivity analysis of wide-angle reflection data by Iidaka et al.  (Figure 3) and dense seismic tomography by Kato et al. . Similar highly reflective portions in the deeper extension of the coseismic zone are also observed beneath the Shikoku region [Kurashimo et al., 2001; Kodaira et al., 2002] (Figure 3) and Kii peninsula [Ito et al., 2006; Kurashimo et al., 2013]. Quantitative evaluations of the velocity structure in the Nankai subduction zone consistently show that there exists a thin low-velocity layer along the top of the subducting oceanic crust [Kodaira et al., 2002; Iidaka et al., 2004].
In comparison, structure and physical properties of the Izu-Bonin arc subducting beneath the Izu collision zone are poorly understood. Previous studies using active-source data revealed that the crustal blocks derived from the Izu-Bonin arc (Koma, Misaka, Tanzawa, and Izu in Figure 2a) have been obducted/accreted onto the Honshu arc in the process of collision while the lower part of the crust has been subducting into the mantle [Sato et al., 2005, 2006; Arai et al., 2009, 2013]. In this paper, we expand this view by evaluating the velocity structure atop the subduction portion of the Izu-Bonin arc.
3 Active-Source Data and Modeling Method
To determine crustal structure associated with collision/subduction of the Izu-Bonin arc, multiple active-source seismic experiments have been carried out in the Izu collision zone (Figure 2b); the eastern profile in 2003 [Sato et al., 2005; Arai et al., 2009] and the western profile in 2006 [Sato et al., 2006; Arai et al., 2013]. These surveys provided the densest seismic reflection/refraction data sets in this region: the eastern profile consists of a seismic line extending approximately 130 km on which 2518 receivers were deployed at 50 m intervals. As seismic sources, 15 large-energy (dynamite or vibroseis) shots were fired for the refraction/wide-angle reflection experiment. For the western line, 1642 receivers were deployed on a 75 km long profile and 18 shots were used. See additional details concerning the experiments in Sato et al.  and Arai et al.  for the eastern line, and in Arai et al.  for the western line.
Figure 4 shows seismic record sections focusing on deep wide-angle reflections. Using these phases, previous studies imaged northward dipping boundaries that were interpreted to be the top of the subducting portion of the Izu-Bonin arc [Sato et al., 2005, 2006; Arai et al., 2009]. This interpretation is also consistent with the boundary imaged by converted waves [Iidaka et al., 1990; Tsumura et al., 1993]. An important consideration is that these slab reflections are detectable by enhancing signal/noise ratio, but they are much weaker than the first arrivals on both (eastern and western) survey lines, implying that the velocity jump across the slab top is extremely small.
In order to explain these weak reflections, synthetic seismograms were calculated using the code SEIS83, which is based on asymptotic ray theory [Červený and Pšenčík, 1983]. For the eastern profile, we used P-wave velocity model by Arai et al.  for the overriding crust. For the western profile, results of refraction tomography [Arai et al., 2013] and passive-source tomography [Nakamichi et al., 2007; Arai, 2011] were used to create a starting model. Next, we used models with and without a low-velocity layer on the top of the subducting slab, and compared the synthetic seismograms with the observations. Location and a dip angle of the plate boundary were obtained from previous seismic reflection studies [Sato et al., 2005, 2006] and refraction/wide-angle reflection analyses [Arai et al., 2009] and fixed in our new analysis. In this study, we only focused on overall characteristics by visually comparing amplitude variations of reflections and first arrivals since this approach is robust and efficient to clarify first-order structural features [e.g., Kodaira et al., 2002]. Although finer-scale and more quantitative studies may be possible, such analyses are sensitive to data noise and assumed initial models, and often show large uncertainties [e.g., Mjelde et al., 1997]. This paper is the first attempt to investigate subduction properties beneath the Izu collision zone using dense active-source data and puts more interests in possible causes for the systematic difference from the Nankai subduction zone.
4 Subduction Structure Beneath the Izu Collision Zone
Figure 5 shows an observed seismic record of the eastern experiment and its comparison with synthetic seismograms. Although slab reflections can be recognized by enhancing signal/noise ratios with use of filtering, observed reflections have even smaller amplitude than the first arrivals (as indicated by an arrow in Figure 5a). An increase in seismic velocity from 6.2 to 6.4 km/s in the middle crust to 7.1–7.2 km/s in the lower crust satisfactorily explains the observed amplitude behavior (Figures 5b and 5c) [Arai et al., 2009]. Based on a trial and error approach (assuming a lower and higher velocity below the reflector), we estimated that the velocity uncertainty on the subducting portion is smaller than 1.0 km/s. Although this range of errors cannot fully exclude a possibility of a low-velocity layer itself, the estimated minimum value (Vp = ∼5.5 km/s) is nonetheless high compared to those reported in other subduction zones [Kodaira et al., 2002; Nakanishi et al., 2004; Iidaka et al., 2004] and can be explained by middle/lower crustal lithologies of the Izu-Bonin arc [Kodaira et al., 2007].
Another example from the western profile is shown in Figure 6. As in the case of the eastern profile, the amplitude of the slab reflections is even smaller than first arrivals (Figure 6a), and we confirmed that a velocity model without a low-velocity layer, in which Vp = 6.4–6.9 km/s for the overriding plate and Vp = 7.0–7.1 km/s for the subducting part are assumed, successfully reproduced weak reflections consistent with the observed data (Figure 6c). On the other hand, synthetic seismograms from a velocity model with a thin low-velocity (Vp = 5.5 km/s) layer show reflections with amplitude clearly larger than the first arrivals (Figure 6d), which contradicts the observation. A key point is that this assumed velocity of 5.5 km for the low-velocity layer is still higher than that proposed in the Nankai subduction zone [e.g., 3.0 km/s in Kodaira et al., 2002]. Even in this case, the calculated amplitudes are too large to explain the observation. It is also noted that similar weak reflections are observed from other multiple shots in the eastern and western profiles (Figure 4). Therefore, the existence of a low-velocity layer with a high water content, as in the Nankai subduction zone, is less plausible in the Izu collision zone.
In the western profile, our preferred velocity model is consistent with the interpretation by petrological studies. One of the prominent geological features in the Izu collision zone is intrusive granitic rocks exposed at the surface (pink-colored regions in Figure 2b): the Miocene Tanzawa plutonic complex, characterized by tonalite and minor gabbro, is exposed at the center of the Tanzawa Mountains [Kawate and Arima, 1998], and the Kofu granitic complex, which mainly comprises granite and tonalite, occurs in the northwestern part of the Izu collision zone [SaIto et al., 2007]. Petrological studies based on zircon U-Pb ages showed that these granitic rocks formed simultaneously with the collision/subduction of the Izu-Bonin arc [SaIto et al., 2007; Tani et al., 2010]. In the process of subduction, the middle crust of the Izu-Bonin arc partially melted and produced a magmatic body at a subMoho level beneath the Honshu arc [Tamura et al., 2010], from which the felsic magma ascended and intruded into the Honshu arc and the Tanzawa block to form the Kofu granitic complex and Tanzawa plutonic complex, respectively [Arai et al., 2013; Arai and Iwasaki, 2014]. Tamura et al.  suggested that the lower crust with dense mafic rock compositions has been descending deep into the mantle without being accreted. Therefore, the overriding crust with Vp = 6.7–6.9 km/s in our velocity model probably corresponds to the middle crust of the Izu-Bonin arc that accreted to the bottom of the Honshu arc crust by tectonic/magmatic underplating, and the subducting portion with Vp = 7.0–7.1 km/s corresponds to the lower crust of the Izu-Bonin arc.
A possible alternative for the low-amplitude reflections is a strong effect of seismic attenuation. Figure 6e shows synthetic seismograms from a model with a low-velocity layer of Vp = 5.5 km/s in which seismic attenuation is also taken into account. To investigate the most effective case of seismic attenuation, we used the model consisting of the upper 12 km part with extremely low attenuation (Qp = 5000) and the lower part (deeper than 12 km) with extremely high attenuation (Qp = 100) where only reflected waves from the slab propagate. In other words, refracted first arrivals propagate through a region without any effect of seismic attenuation, while reflected waves undergo a significant effect of attenuation when traveling through the lower part of the model. This test shows that the seismic attenuation causes a slight reduction of the reflected wave amplitudes but its effect is insufficient to produce the observed amplitude. Thus, the small impedance contrast across the slab is probably the main cause for the small-amplitude reflections.
5.1 Other Geophysical Evidences
Besides active-source seismic data, the spatial distribution of seismicity is another useful tool to refine the nature of the subducting structure in the Izu collision zone. The study area is located on the western flank of the source region of the great Kanto earthquake with magnitude of 7.9 that occurred in 1923 at the plate boundary between the subducting Philippine Sea plate and the overriding crust [Matsu'ura et al., 1980; Wald and Somerville, 1995]. A deep reflector constrained by our active-source data occurs at the deeper extension of the source fault of the Kanto earthquakes (Figure 5d) [Sato et al., 2005; Arai et al., 2009]. This coincidence supports our interpretation of the reflector as a plate boundary, not instead of a simple midcrustal reflector. The interesting feature is that although the spatial pattern of reflectivity (a shallow coseismic region with low reflectivity and a deeper aseismic region with relatively high reflectivity) is similar to the Nankai subduction zone, the degree of reflectivity in the aseismic zone is significantly different.
Another insight in the seismogenic processes is obtained from focal mechanisms. Intensive seismicity occurs beneath the Tanzawa Mountains (Figure 2a), and these events have the maximum compressive axis (P axis) in the northwest-southeast direction [Yukutake et al., 2012], which is consistent with the present motion of the Philippine Sea plate [Seno et al., 1993]. Seismicity induced by dehydration reactions often shows a normal fault-type mechanism due to a tensional stress field that is caused by volume reduction [Kita et al., 2006]. However, most events beneath the Tanzawa Mountains are reverse fault, strike slip or a combination of the two, and a normal fault mechanism is observed in very few events [Yukutake et al., 2012; Arai, 2011].
A number of tomographic studies have succeeded in imaging seismic structure of the subducting Philippine Sea plate: the results consistently show that the oceanic crust forms a zone with low P-wave and S-wave velocities and high Vp/Vs ratios in the Nankai subduction zone, indicating high water contents along the plate interface and/or within the slab [Hirose et al., 2008; Matsubara et al., 2009; Kato et al., 2010]. In contrast, the subducting portion beneath the Izu collision zone is imaged as a “high” velocity anomaly [Matsubara et al., 2009; Nakajima et al., 2009], which is interpreted to be the lower crustal/uppermost mantle materials of the Izu-Bonin arc with mafic compositions [Arai et al., 2009; Tamura et al., 2010]. In addition, local tomographic studies show that any velocity reduction cannot be seen in the seismogenic zone beneath the Tanzawa Mountains (Figure 2a) [Nakamichi et al., 2007]: intermediate values of Vp/Vs ratio there are consistent with the values of gabbroic rocks, a major component of the middle/lower crust of the Izu-Bonin arc, in a dry condition [Nishimoto et al., 2008; Arai, 2011].
A similar structural difference between the Izu collision zone and Nankai subduction zone is also reported by studies of electrical resistivity. In the Nankai subduction zone, a highly conductive zone is imaged above and along the subducting Philippine Sea slab where deep low-frequency tremors repeatedly occur (Figure 7) [Yamaguchi et al., 2009]. This zone is also coincident with a high Vp/Vs region inferred from seismic tomography, supporting the model that the hydrous oceanic plate releases a large amount of fluid to the plate interface and the overriding plate. On the other hand, the Izu collision zone is characterized by high resistivity [Aizawa et al., 2004]: the seismogenic zone beneath the Tanzawa Mountains is separated from a highly conductive magmatic body beneath Mt. Fuji (Figure 7), which excludes the possibility that the seismicity beneath the Tanzawa Mountains is related to dehydration process or volcanic activity.
5.2 Sources of the Structural Variation
Based on the geophysical evidences presented above, we suggest that the dehydration process is inactive and thus a significant low-velocity layer is not formed atop the slab beneath the Izu collision zone. Figure 8 shows two contrasting structural models of the Izu collision zone and the Nankai subduction zone. The oceanic crust is generally composed of basaltic rocks that contain abundant hydrous minerals, such as blueschists [Hacker et al., 2003b]. Mantle peridotite can also hold water in hydrous minerals (serpentinite) [Hacker et al., 2003b]. Although oceanic plates are thought to hydrate through several mechanisms including hydrothermal circulation near the spreading centers where they are formed [Kirby et al., 1996; Hacker et al., 2003b], recent seismological studies indicate that plate-bending-related normal faulting at outer rises facilitates seawater penetration into a lower crustal level or even down to the uppermost mantle [Contreras-Reyes et al., 2007; Grevemeyer et al., 2007; Ivandic et al., 2008; Fujie et al., 2013]. These hydrated crust and upper mantle progressively release fluids with increasing depth of subduction (correspondingly increasing pressure and temperature), and then the fluids are supplied to the plate boundary to form a low-velocity layer. On the other hand, active volcanic arcs including the Izu-Bonin arc are thought to have consumed a large amount of hydrous minerals for melt production during crustal differentiation [e.g., Johannes and Holtz, 1996] and metamorphosed into more stable, anhydrous forms [Tatsumi et al., 2008]. In addition, seawater infiltration typical at outer rises is unlikely to occur in the Izu-Bonin arc: although the maximum depth of possible hydration for oceanic lithospheres remains unconstrained, the effect is expected to wane as it goes deeper or farther from the seafloor. In the Izu-Bonin arc, the lower crust, which finally leads to subduction without obducting/accreting to the Honshu arc, is covered by 15–20 km thick upper/middle crust that must impede water inflow from the seafloor. Matsubara et al.  pointed out that in southwest Japan deep tremors are relatively inactive where ridges or seamounts are subducting, implying that seafloor topography is closely related to fault developments before subduction and final hydration condition within the subducting oceanic lithosphere.
One may argue that temperature anomalies due to the proximity to active volcanoes are another factor contributing to the aseismic region beneath the Izu collision zone (Figure 1). However, the thermal effect seems less dominant for the regional pattern of the seismogenic zone: first, the width of the aseismic region is several times wider than that of volcanic front itself (Figures 1 and 2). Second, deep seismicity beneath the Izu collision zone concentrates on a few very narrow spots (Figure 2a): seismicity beneath Mt. Fuji and Hakone volcano is probably induced by deep magmatic activities [Nakamichi et al., 2007], and events beneath the Tanzawa Mountains are interpreted to be collision-related activities [Arai, 2011; Yukutake et al., 2012]. In short, there is no clear correlation between the distance from an active volcano and the distribution of the seismicity. In order to demonstrate this hypothesis, further studies including geodynamic modeling of the thermal state of the subduction zone will be necessary in the future.
We comparatively investigate the subduction structures of the Izu-Bonin volcanic arc using active-source seismic data. Large-amplitude reflections from the slab are observed in southwest Japan, indicating that a low-velocity layer with a high fluid content is formed along the plate boundary due to dehydration reactions of hydrous minerals within the subducting oceanic lithosphere. On the contrary, the slab reflections have smaller amplitude in the Izu collision zone where the Izu-Bonin volcanic arc has been colliding/subducting. Together with other geophysical evidences, we suggest that such a low-velocity layer does not exist atop the plate boundary, which consistently supports the original hypothesis by Seno and Yamasaki  that dehydration reaction is inactive beneath the Izu collision zone. Based on these comparisons, we propose two contrasting subduction models of the Philippine Sea plate: in the Nankai subduction zone, the oceanic crust and upper mantle, enriched in water by hydrothermal circulation at a mid-ocean ridge and/or seawater inflow at an outer rise, releases fluids when subducted. On the other hand, the Izu-Bonin arc crust has consumed a large amount of hydrous minerals for melt production and metamorphosed to more stable, anhydrous forms before subduction, leading to an inactive dehydration reaction beneath the collision zone. This systematic difference is probably a major cause for the variation in seismic activities associated with the subduction of the Philippine Sea plate.
We are grateful to T. Yoshida, S. Gao, and H. Van Avendonk for helpful comments on this study. We also thank Japan Meteorological Agency for the hypocentral data. This study was supported by the Special Project for Earthquake Disaster Mitigation in Urban Areas and the Special Project for Earthquake Disaster Mitigation in Tokyo Metropolitan Area of the Ministry of Education, Culture, Sports, Science, and Technology of Japan, and JSPS KAKENHI (09J09214). We used the GMT software [Wessel and Smith, 1998] to draw figures.