We present the three-dimensional structures of the P wave velocity (VP), S wave velocity (VS) as well as the P wave to S wave velocity ratio (VP/VS) beneath Mount Fuji and the South Fossa Magna, Japan, using arrival time data collected from 2002 to 2005 by a dense seismograph array. The high-resolution data set and improved methodology reveal not only several velocity features that are consistent with previous studies but also important new details that clarify the velocity structures associated with volcanic processes beneath Mount Fuji and the collision tectonics of the South Fossa Magna. One such particular feature is a low-VP, low-VS and low-VP/VS anomaly at depths of 7–17 km beneath Mount Fuji that corresponds with the locations of deep low-frequency (DLF) earthquakes. The coincidence of the velocity anomaly and the DLF locations suggests that supercritical volatile fluid, such as H2O and CO2, may be abundant in the low-VP/VS region and may play an important role in generating DLF earthquakes. This anomaly overlies a deeper low-VP, low-VS and high-VP/VS anomaly at depths of 15–25 km that may represent a zone of basaltic partial melt. A low-VP, low-VS and low-VP/VS anomaly is seen at depths of 6–14 km beneath Mount Hakone. Isovelocity surfaces (VP = 6.0 km/s and VS = 3.5 km/s) corresponding to the upper limit of hypocenter distribution below Mount Fuji may define the upper surface of the Philippine Sea plate whose existence in a seismic gap beneath Mount Fuji has been controversial.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The “Fossa Magna,” a tectonic zone traversing the central part of the main island (Honshu) of Japan from the Japan Sea coast to the Pacific in a NNW–SSE direction, contains a zone of Quaternary volcanoes. Mount Fuji (Figure 1a) is the most active volcano in the Fossa Magna with an eruption rate that is an order of magnitude higher than those of most other island arc volcanoes [Fujii, 2001]. Mount Fuji has erupted mostly basalt magma with subordinate basaltic andesite composition (50–52 wt % SiO2) and is relatively young (11 ka) among island arc type volcanoes [Tsuya, 1968, 1971]. These characteristics may be due to the unique tectonic setting of Mount Fuji.
 The South Fossa Magna region is the complex focus of the collision of the Philippine Sea (PHS) plate with the Eurasian (EUR) plate in the north and with the North American (NAM) plate in the northeast (Figure 1). This situation is further complicated by the concurrent westward subduction of the Pacific (PAC) plate in the east, providing back-arc volcanism, which penetrates the PHS plate and large portions of Honshu. Mount Fuji is located at the triple junction of the PHS, EUR, and NAM plates called the Fuji triple junction [Nakamura, 1983]. The Tanzawa Mountains, located 20–40 km northeast of the summit of Mount Fuji, were formed in the Izu collision zone, where the Izu Peninsula (part of the PHS plate), moving NW, collides with and is accreted to the Honshu block (EUR and NAM plates) [e.g., Matsuda, 1978].
 A depth contour map of the upper surface of the PHS plate is shown in Figure 1a. A double seismic zone associated with the westward subducting PAC plate is located at depths of 100–200 km beneath the South Fossa Magna. Above the subducting PAC plate, extensive seismicity is observed in the PHS plate dipping eastward along Sagami Bay. The seismicity in the west along Suruga Bay generally ranges from a 50 km depth in the west to depths of 70–80 km in the east [Ishibashi and Ishida, 1989; Ishida, 1992]. A conspicuous fan-shaped seismic gap in the seismic PHS slab exists between the slabs beneath the Kanto and the Tokai districts, in the north and northwest of the Izu Peninsula, respectively (Figure 1a). Ishida  suggested that the seismic gap in the descending PHS slab is due to the relative warmth of the young subducted PHS plate itself and that of the asthenosphere underneath it. Ishibashi  suggested an opposing idea that the PHS plate does not exist in the area of the seismic gap based on the fact that the western edge of the seismic PHS slab beneath the Kanto district is sharp and on the fact that this edge is located 40–70 km seaward of the volcanic front. He proposed that the crust of the Izu-Bonin arc had repeated multiple underplating beneath the Kanto and Tanzawa mountains and that its mantle was delaminated. Consequently, he concluded that the seismogenic crustal part of the PHS plate descending northward in the asthenosphere was actually nonexistent north of the Izu Peninsula. Although the existence of the PHS plate beneath Mount Fuji has not been clearly evidenced due to the scarcity of tectonic earthquakes occurring around Mount Fuji, the expected depth of the upper bound of the PHS plate there is about 10 km (Figure 1a).
 While there are few tectonic earthquakes around Mount Fuji, deep low-frequency (DLF) earthquakes occur at depths of 11–16 km beneath Mount Fuji [Nakamichi et al., 2004a; Ukawa, 2005]. DLF earthquakes have been observed in various volcanic regions and are thought to be related to deep magmatic activity in the uppermost mantle and the midcrust. Therefore the occurrence of DLF earthquakes beneath Mount Fuji may be a manifestation of Mount Fuji's deep magmatic activity.
 The surface geology of the South Fossa Magna is dominated by volcanic deposits, marine sediments [Niitsuma, 1989; Shimazu, 1989; Takahashi, 1989], and the Quaternary volcanoes including Mount Fuji, Mount Hakone and volcanoes on the Izu Peninsula. North of Mount Fuji and Mount Hakone, along the Tanzawa Mountains, and in the southern portion of the Izu Peninsula, Neogene volcanoes are mapped with several pockets of exposed granitic intrusions. A submarine eruption occurred in this region in 1989, forming the Teishi Knoll (Figure 1a), and the eruption was modeled as a dike intrusion based on crustal deformation data [Okada and Yamamoto, 1991]. The Itoigawa-Shizuoka Tectonic Line (ISTL), which is a major boundary between the EUR and NAM plates, trends northward through Honshu toward the Japan Sea. There is an extensively faulted zone of clastic marine sediments between ISTL and Mount Fuji [Takahashi, 1989].
 The structures of Mount Fuji and the South Fossa Magna have been investigated using several geophysical methods. Seismic tomographic studies done by Sekiguchi  and by Lees and Ukawa  found a low-velocity anomaly at depths more than 20 km beneath Mount Fuji. A seismic refraction survey using controlled source along a WSW–ENE line passing through the summit of Mount Fuji revealed a high-velocity body swelling just beneath the summit [Oikawa et al., 2004]. The profile obtained from a magnetotelluric survey showed that a conductive body is located at a depth of greater than 20 km just beneath Mount Fuji [Aizawa et al., 2004]. These studies suggest that magmatic bodies exist at depths more than 20 km and that a shallow solidified magma plumbing system exists beneath Mount Fuji. However, the relationship between DLF earthquakes and the structure of Mount Fuji is still unknown.
 Seismic observations around Mount Fuji have been conducted since the 1980s by the Earthquake Research Institute of the University of Tokyo (ERI), the National Research Institute for Earth Science and Disaster Prevention (NIED) and the Japan Meteorological Agency (JMA). DLF earthquakes beneath Mount Fuji have been detected frequently by these seismic networks since the early 1980s [Kanjo et al., 1984; Ukawa and Ohtake, 1984; Shimozuru et al., 1986]. The most intense DLF earthquake activity beneath Mount Fuji occurred in two periods during September–December 2000 and April–May 2001 [Ukawa, 2005]. However, no unusual shallow seismic activity or crustal deformation around Mount Fuji accompanied or followed the DLF earthquake activity [Ukawa, 2005].
 After the onset of intense DLF earthquake activity in 2000 and 2001, a dense passive seismic experiment was conducted in and around Mount Fuji and the South Fossa Magna [Nakamichi et al., 2004b] to elucidate the seismic structures of this region and the source mechanism of the DLF earthquake in the period from 2002 to 2005. In this paper, we determine the three-dimensional velocity structures of Mount Fuji and the South Fossa Magna using first arrival time data obtained in this experiment. This study combines a high-quality subset of P and S wave recordings of local earthquakes with the recordings of several explosions obtained by both temporary and permanent stations. We then compare the tomographic results with surface geological features, gravity anomalies and earthquake hypocenters around this region.
Figure 2 shows the distribution of the seismic stations used in this study. Scientists at national universities in Japan began deploying 28 temporary seismic stations in September 2002 to cover regions with few permanent stations in and around Mount Fuji and the South Fossa Magna [Nakamichi et al., 2004b]. This temporary network was maintained until April 2005. Each temporary station was equipped with a three-component short-period (1 s) seismometer or a broadband seismometer (40–360 s). Data from the temporary stations were transmitted by VSAT satellite, radio and line telemeters using IP protocols and 24 bit A/D converters with a sampling frequency of 100 Hz. This observation also included 138 permanent short-period stations, which are operated by several institutes: ERI, NIED, JMA, the Hot Springs Research Institute of Kanagawa Prefecture (HSIK), and Nagoya University. Half of them are Hi-net stations of NIED [Okada et al., 2004]. A network around Mount Hakone is operated by HSIK.
 All the P and S wave arrival time picks in the period from October 2002 to November 2004 were manually picked. S waves were picked only on the horizontal components of seismograms. The P wave picking error was estimated to be on the order of a few samples (±0.02 s). S waves were generally contaminated by the P wave coda and had picking errors of approximately ±0.05 s. Data selection was based on the following criteria: (1) the number of stations having clear P or S arrival was greater than or equal to 25 for each event; (2) residuals of P and S wave arrival times were less than or equal to 0.3 s and 0.6 s, respectively; (3) the ranges of latitude, longitude and depth were N34.8°–36.05°, E137.85°–139.4°, and −1 to 50 km, respectively; and (4) magnitudes, M > = 0.8. This resulted in 1066 earthquakes with 66,038 P and 62,265 S arrival time data. Figure 3 shows the geographical distribution of the selected hypocenters and the configuration of grid points in the final step of the inversion, as described later. Five timed explosions [Oikawa et al., 2004], indicated by stars in Figure 2, were also used along with natural earthquakes. The locations and origin times of the explosions were fixed as inputs to the tomographic inversions.
3. Method and Tomography Procedure
 We estimated the three-dimensional velocity structure using the double-difference (DD) tomography method developed by Zhang and Thurber . The DD tomography method simultaneously inverts absolute and differential arrival time data for velocities. The DD algorithm can remove model errors in hypocenter locations by using differential data [Waldhauser and Ellsworth, 2000]. This method uses the pseudobending (PB) ray-tracing algorithm [Um and Thurber, 1987] to determine rays and traveltimes between events and stations. The PB method has been frequently used for exploring volcanoes where large velocity perturbations are expected to exist [e.g., Chiarabba et al., 2000; Tanaka et al., 2002; Yamawaki et al., 2004].
 We used 146,441 P and 137,304 S wave arrival time differences, computed directly from the absolute P and S wave arrival picks. Although the accuracy of the differential data was not higher than that of the absolute arrival data from which the differential data were produced, their use improved the inversion result [Zhang and Thurber, 2003]. We selected the catalog arrival time differences so that any event would be linked to a maximum of 10 neighboring events by at least eight pair-wise observations.
 We obtained a one-dimensional velocity model by inverting the P and S wave arrival time data, which served as our starting velocity structure for the three-dimensional inversion. We followed a “graded inversion” approach. The number of velocity nodes was progressively increased to allow the velocity of poorly sampled volumes at the periphery of the target region to be adjusted to the best values obtained in the early inversions. As a first step, we took 9 × 9 horizontal grid nodes with a coarse nodal spacing of 20 km for the latitudinal and longitudinal directions and 5 vertical grid nodes at 10 km intervals. The vertical grid extended from −5 km below sea level (bsl) (i.e., 5 km above sea level) down to 45 km bsl. The initial three-dimensional model was produced from the one-dimensional model obtained by inverting the arrival time data. As a second step, the resulting coarse three-dimensional model was interpolated to provide a starting model for an inversion of 16 × 16 grid nodes with nodal spacing of 10 km in the horizontal direction. The vertical grids were the same as in the first step. As a third step, the resulting three-dimensional model in the second step was interpolated to provide a starting model of the same horizontal grids and vertical grids at −5, 0, 5, 10, 15, 20, 25, 35 and 45 km. In the last step, the horizontal interval was reduced to 5 km except for the grids near the periphery. The final grid configuration with 6561 nodes is shown in Figure 3.
 We applied both damping and smoothing to the inversion to make the solution more stable. Smoothing weights of 0, 5, 10, 20 and 30 were tested. The main features of these models were similar in the regions with good ray coverage. In other regions, the models using weights of 0 and 5 showed some oscillations, which were reduced by using larger weights. Considering the trade-off between the roughness and the stabilization of the model [Zhang and Thurber, 2003], we chose weights 20 and 10 in the horizontal and vertical directions, respectively. We chose damping parameters of 150 for hypocenter relocation and 300 for simultaneous inversion of the hypocenters and velocities. The hypocenter relocations and simultaneous inversions were iterated 16 times.
4. Resolution Tests
 We examined a well-imaged region of the three-dimensional velocity model using a checkerboard test [Humphreys and Clayton, 1988]. We constructed checkerboards for P and S wave velocity structures by assigning ±10% of positive and negative velocity perturbations alternately to each node of the final grid. The background velocity model used in the checkerboard test was the one-dimensional velocity model, the same as that used for the velocity inversion for the actual data. The synthetic P and S wave arrival times were computed using the source-receiver geometry of the observed P and S wave data sets and the checkerboard velocity model. We added Gaussian random noise with zero mean and standard deviations of 0.05 s and 0.08 s to the entire synthetic P and S wave arrival times, respectively. We also added a constant noise term having a uniform distribution between −0.12 s and 0.12 s to the arrival times at each station. This simulated the case in which the systematic errors (model errors and pick biases) associated with the arrival times are larger than the random ones. We then conducted an inversion of the simulated data with the same damping and smoothing factors as used for the real data set.
Figure 4 shows the synthetic P wave reconstructions of the checkerboard velocity model. The resolution is very good for areas around Fujikawa, the Misaka Mountains, Mount Fuji, the Tanzawa Mountains, Mount Hakone and the northern part of the Izu Peninsula at depths of 5, 10 and 15 km. The resolution is fairly good at a depth of 0 km beneath Mount Fuji and Mount Hakone. The pattern is well resolved at depths of 20 and 25 km for areas around Fujikawa, Mount Fuji and the Tanzawa Mountains. Although the S wave resolution pattern is similar to the P wave pattern, the area with good S wave resolution is somewhat smaller than that for P waves (Figure S1 in the auxiliary material).
 We introduce a quantity called the “recovery rate,” which is unity when the initial velocity perturbations are completely recovered and is zero when the resulting velocity perturbations are zero or have signs opposite the initial ones. Overall, the patterns of recovery rates for P and S waves are the same, but the area with good recovery rates for S waves (Figure S3) is slightly smaller than that for P waves (Figure S2). Recovery rates greater than 0.6 occur beneath Fujikawa, Mount Fuji, the Tanzawa Mountains and Mount Hakone at depths of 5–15 km for P and S waves.
 Solution quality is also described by a weighted hit count called the “derivative weight sum (DWS)”, an estimate of ray density near a grid point scaled by ray node separation and raypath length near a node [Evans et al., 1994]. Larger DWS values indicate better model resolution. We checked the distributions of the DWS values. Recovery rates of 0.4 and 0.6 approximately are equal to DWS values of 200 and 500, respectively.
 We also conducted a restoring resolution test [Zhao et al., 1992] on the obtained structure including the low VP/VS ratio to determine whether or not the main trend or some special structures in the real result were actually restored, and if they were restored, to determine how well. The method of the inversion was the same as in the checkerboard resolution test. The velocity anomalies were clearly imaged (Figure S4).
 The PB method might not be sufficient for exploring volcanic regions where very strong velocity heterogeneities may exist. Our model shows strong velocity perturbations of up to 20–30%. Ray-tracing methods other than the PB method, such as the graph theory (GT) and finite difference (FD) methods, have sometimes been used for tomographic studies in volcanic regions [e.g., Nishi, 2001; Benz et al., 1994]. The GT and FD methods, however, require much more time to calculate traveltimes for large target volumes such as those used in this study. These methods cannot be incorporated with small station intervals relative to the grid intervals of the methods without consuming enormous computational resources. We therefore used the PB method because the advantage of its computational efficiency exceeds the limitations of its ray-tracing ability. We show in the next section that the perturbations are within ±10% over most of the target volume, and thus the results obtained with PB are reliable.
5. P and S Wave Tomography Results
 After the four steps of the graded inversion, the weighted root-mean-square (RMS) arrival time residual was reduced by 76% from 0.180 s to 0.044 s, which is within the range of a priori picking uncertainty for S phases.
 The P wave velocity (VP) structure (Figure 5) at depths of 0–10 km appears to be a heterogeneous mixture of low- and high-velocity anomalies related to the complex juxtaposition of sedimentary and volcanic features in the South Fossa Magna region. At a depth of 0 km, high-velocity anomalies (5% or more relative to the initial one-dimensional (1-D) reference model; 5 km/s or more) exist beneath Mount Fuji, Mount Hakone, Misaka and the Tanzawa Mountains. In contrast, the western, southern and eastern bases of Mount Fuji are underlain by low-velocity (−20% or less relative to the initial 1-D reference model; 4 km/s or less) anomalies. At depths of 5 and 10 km, a linear north–south (NNW and SE) trend of low-velocity anomalies (−5 to −15% relative to the initial 1-D reference model; 4.5–5.5 km/s) coincides with those of Fujikawa and the Itoigawa-Shizuoka Tectonic Line (ISTL). In the northern part of Sagami Bay, there is a broad low-velocity anomaly (−5 to −15% relative to the initial 1-D reference model; 4.5–5.5 km/s), extending on shore in the northerly direction at depths of 5 and 10 km. At 20 and 25 km depths, a low-velocity zone is observed from 5 to15 km east from the summit of Mount Fuji.
 The S wave velocity (VS) structure (Figure S5) shares many of the prominent features observed in the VP structure. For example, in the ISTL and Fujikawa areas, a north–south trending low-VS anomaly (−10 to −20% relative to the initial 1-D reference model; 2.8–3.2 km/s) is also observed at depths of 5 and 10 km (Figure 5). A low-VS region (−10 to −15% relative to the initial 1-D reference model; 3.2–3.6 km/s) also exists beneath Mount Fuji and extends to 10 km northeast at 20 km depth.
Figure 6 displays VP, VS and VP/VS structures along the cross section A-A′ shown in Figure 3. The VP/VS model is derived from the direct division of VP by VS. This approach can be so unstable that splotchy anomaly patterns can be generated, as seen in Figure 6c. Therefore, in the interpretation, we use only the sign of the VP/VS change from the initial value of 1.73.
 The VP = 5.0 km/s and VS = 2.8 km/s isovelocity contours shallow under the summit of Mount Fuji. The depth of the VP = 6.0 km/s and VS = 3.5 km/s isovelocity contours are very close to the ground surface beneath the Tanzawa Mountains, and the depths of the isovelocity contours gradually increase toward the southwest. Beneath the Fujikawa area, these depths are between 5 and 10 km. A clear low-VP anomaly (6.0 km/s or less) is located between 7 and 17 km in depth beneath Mount Fuji. We also observe a broad low-VP region at depths greater than 20 km beneath Mount Fuji. This low-VP region shallows to a depth of 15 km and extends to 10 km northeastward. A clear low-VS region (3.5 km/s or less) is observed beneath Mount Fuji beginning at a depth of 8 km and extending to 25 km, where the low-VS region becomes broad and dispersed. At a depth of 20 km, this low-VS region also extends northeast and reaches to a depth of 15 km. The resolution of the velocity structures is so poor at depths greater than 25 km beneath Mount Fuji that we could not detect any velocity anomalies there. The VP/VS model shows a large low-VP/VS anomaly between 7 and 17 km depth beneath Mount Fuji (Figure 6c). High values of VP/VS occur broadly at depths of 20 and 25 km beneath Mount Fuji, probably due to the low-VS in the same region (Figure 6b).
 The vertical cross sections of the velocity structures along B-B′ (location in Figure 3) are shown in Figure 7. The VP = 6.0 km/s and VS = 3.5 km/s isovelocity contours approach the ground surface under the Tanzawa Mountains and Mount Hakone. A low-velocity wedge with VP = 5.0–6.0 km/s and VS = 2.4–3.5 km/s is located between the Tanzawa Mountains and Mount Hakone. This wedge is located above the region of the earthquake swarm beneath the Tanzawa Mountains. A clear low-VP (5.0–6.0 km/s) region is located between 6 and 14 km in depth just beneath Mount Hakone. Since the low-VP region corresponds to slightly low-VS (3.5 km/s or less), it results in significantly low-VP/VS.
6.1. Comparison With Other Geophysical and Geological Studies
 We interpret the results described in the last section primarily in terms of the major volcano tectonic features in the target volume, which includes the northern boundary of the subducting PHS plate, the major tectonic line ISTL, Mount Fuji and Mount Hakone. While the results appear to be complicated at first glance, several velocity anomalies correspond to the geologically observed features.
 The obtained three-dimensional velocity structure correlates well with surface geologic features and is consistent with earlier interpretations of the velocity structures derived from a controlled source refraction profiling survey [Oikawa et al., 2004] and a tomographic inversion [Lees and Ukawa, 1992]. Our results, however, yield a higher-resolution image of the three-dimensional structural details than do those of the previous studies. For example, the shallower low-velocity anomaly at depths of 7–17 km seen in our model is not seen in the previous model of Lees and Ukawa  probably due to a lack of spatial sampling.
 The deeper low-velocity anomaly around the 15–25 km depth interval beneath Mount Fuji in our model is in good agreement with that of Lees and Ukawa . This anomaly also corresponds to the conductive body at depths greater than 20 km just beneath Mount Fuji obtained from a magnetotelluric survey [Aizawa et al., 2004]. These coincidences further confirm the existence of the magmatic body at depths greater than 20 km beneath Mount Fuji.
 While Lees and Ukawa  detected a low-velocity zone at 10 km beneath the eastern shore of the Izu Peninsula, we did not observe any low-velocity anomalies there (Figure 7). Since the region has sufficient resolution (recovery rate of 0.4) in our models, the absence of a low-velocity anomaly is not due to poor resolution. Lees and Ukawa  used data from the earthquake swarm associated with the 1989 submarine eruption off the eastern shore of the Izu Peninsula. Although a dike intrusion fed by a large magma body occurred soon after the 1989 submarine eruption [Okada and Yamamoto, 1991; Ukawa and Tsukahara, 1996], it is likely that the dike had cooled enough by the time of our observation that we could not detect any low-velocity bodies.
 We observe a clear, relatively low-velocity anomaly around a depth of 10 km beneath Mount Hakone (Figures 5 and 7), while Lees and Ukawa  did not find the low-velocity anomaly. The reason for this discrepancy is that we have a much greater station density around Mount Hakone and thus we have better resolution there. The low-velocity anomaly is also found at depths of 5–10 km beneath Mount Hakone using a VP tomography with fine grids and local seismic network data [Tanada, 1999].
 The velocity structures at a depth of 0 km show low-velocity anomalies at the eastern and southeastern flanks of Mount Fuji (Figure 5). These anomalies correspond to deposits of volcanic mudflows [Miyaji, 1988] and avalanche deposits [Miyaji et al., 2004]. A low-velocity zone at a depth of 0 km also exists at the western and southwestern feet of Mount Fuji. This low-velocity zone correlates well with a region of thick pyroclastic deposits [Tsuya, 1968, 1971]. The velocity structure at shallow depths of −3 to 2 km confirms the existence of the high-velocity (VP = 4.5–5.0 km/s) region that shallows just beneath the summit of Mount Fuji (Figure 6). The VP of the high-velocity body beneath the summit obtained by the refraction survey is in the range of 4.5–5.0 km/s [Oikawa et al., 2004], which is identical to that of the high-velocity body found in our result. Shallow high-velocity bodies have been found at other volcanoes by explosion seismic experiments [e.g., Tanaka et al., 2002; Yamawaki et al., 2004]. The high-velocity body directly beneath the summit is considered to be the formation edifice and the manifestation of a near-solid magma system beneath Mount Fuji.
 The isovelocity (VP = 6.0 km/s) surface at shallow depths beneath the Tanzawa Mountains gradually deepens toward the southwest and is at a 5–10 km depth beneath Fujikawa (Figure 6). The same trend is also found in the refraction profile [Oikawa et al., 2004]. The depth of a layer with VP = 5.7 km/s at its top is located at 2 km beneath the Tanzawa Mountains. The depth gradually increases toward the southwest beneath Mount Fuji, and reaches 5 km beneath Fujikawa. Mount Fuji is located slightly inland from the position where the PHS plate subducts beneath the EUR and NAM plates in association with the collision with the Izu Peninsula. Matsuda  suggested that Mount Fuji lies on the same uplifted body as the Tanzawa Mountains in the east. This uplifted body is formed by the plate subduction and collision with the Izu Peninsula, and its influence is believed to extend to a significant depth. This tectonic background is generally considered to be the reason for the change in the geologic structure beneath Mount Fuji from west to east.
 Our interpretation of the velocity structure is supported by the distribution of gravity anomalies. Figure 8 shows the Bouguer gravity anomaly distribution of the South Fossa Magna obtained by the Geological Survey of Japan . The pattern of high- and low-magnitude positive Bouguer anomalies resembles that of the velocity anomalies at depths of 0–10 km (Figure 5). The shallow high-velocity region around Mount Fuji and the Misaka Mountains is in good agreement with Bouguer gravity highs of up to about 30 mGal. The high-velocity anomaly beneath the Tanzawa Mountains corresponds to the high-Bouguer gravity anomaly. We have already noted a prominent low-velocity zone, trending northward east of the ISTL and extending to a 10 km depth on both the VP and VS results (Figures 5 and S5). This anomaly correlates well with a low-Bouguer gravity anomaly, which extends northward from Suruga Bay surrounding the west side of Mount Fuji. This low-Bouguer gravity anomaly can be explained by the existence of the belt of a low-density sediment with a thickness of 2–3 km [Satomura and Anma, 1986], or by the extensive fracturing of rocks due to fault activity [Nishimura et al., 1986].
 The shallow low-velocity anomaly at a depth of 0 km (Figure 5) correlates well with the slightly low-Bouguer anomaly of −20 to 0 mGal at the eastern, southern and western feet of Mount Fuji. A high-Bouguer gravity of up to 80 mGal is seen around Mount Hakone. A high-velocity anomaly at depths of 0 and 5 km and a low-velocity anomaly at a depth of 10 km exist beneath Mount Hakone (Figure 5). The spatial correlation between high velocities and positive Bouguer anomalies has been observed beneath other volcanoes [e.g., Villaseńor et al., 1998; Yamawaki et al., 2004], where the high-velocity and high-Bouguer anomaly zones are interpreted to correspond to regions of rocks filled with magma that intruded from a greater depth and then slowly solidified. Such material is expected to be denser and to have a higher seismic wave velocity than the surrounding deposited material.
 The present discussion is limited to qualitative relationships between seismic velocity and gravity anomalies. The quantitative relationship between seismic velocity and gravity anomalies, however, should be discussed in the future as was done by Lees and VanDecar . When the relationship between velocity and density is determined in a future study, a joint tomographic inversion of traveltime and gravity anomaly data will give us more reliable constraints on the shallow velocity structure [e.g., Onizawa et al., 2002].
6.2. Structure and Tectonics of the Collision Zone of the South Fossa Magna
 The seismic swarm activity beneath the Tanzawa Mountains (Figures 1, 5, 6, and 7) is commonly associated with the collision of the EUR and PHS plates. There is a relatively high-velocity perturbation in both the VP and VS results (Figures 5, 6, and 7). Most of the earthquakes beneath the Tanzawa Mountains are distributed in a narrow zone with a thickness of about 10 km and dipping toward the northwest (Figure 7). No earthquake was located in the northwestern or downward extension of this zone (Figure 7). Tsumura et al.  found that S-to-P converted waves beneath the Tanzawa Mountains were reflected from a plane at the upper limit of the narrow zone of high seismicity beneath the Tanzawa Mountains (thick white lines in Figure 7). A low-velocity wedge between the Tanzawa Mountains and Mount Hakone (Figure 7) corresponds to trough-filled deposits (sandstone and conglomerate), where the Izu Peninsula collides and accretes with the Honshu block [Taira et al., 1998]. At the Izu collision zone toward the north, there is a plutonic body of tonalite (Tanzawa Tonalite) within an accreted crustal slice of the PHS plate [Taira et al., 1998]. The plutonic body of the Tanzawa Tonalite is seen as the high-velocity zone (VP = 6.0–6.5 km/s) just beneath the Tanzawa Mountains (Figure 7). The seismic reflecting plane coincides with the boundary between the Tanzawa Tonalite and a tonalite block below Mount Hakone. It also coincides with the downward extended portion of the low-velocity wedge corresponding to the trough-filled deposits. Thus our tomographic result clearly images the collision tectonics of Izu and Honshu.
 There is a seismic gap beneath the northern extent of Mount Fuji and the Izu collision zone where an extension of the PHS plate is expected (Figure 1). There are two interpretations of this seismic gap, as explained in the introduction. One is that the absence of the intraslab seismicity north of Mount Fuji and Izu does not necessarily mean that there is no slab. The absence of seismicity can be explained by the relatively high temperature of the slab itself and asthenosphere underneath. If there is no slab, the difference in slab length between the Izu collision zone and the Chubu or Kanto region amounts to 200 km (Figure 1), and it is difficult to account for this difference by the deformation of the PHS plate north of the collision zone. The other interpretation is that the seismic plate is absent in the seismic gap due to delamination of the PHS plate by the mantle wedge.
 Previous seismic tomography studies have revealed a slab-like structure with a high-VP around the aseismic region, which is north of Mount Fuji at a depth of 70–80 km [Sekiguchi, 2001]. The hypocenters of high-frequency earthquakes beneath Mount Fuji are located at depths of 7–20 km in our study (Figure 6). These depths correspond to the expected depth (10 km) of the upper surface of the PHS plate around Mount Fuji [Ishibashi and Ishida, 1989; Ishida, 1992; Noguchi and Sekiguchi, 2001]. These earthquakes have not been well located in previous studies because there were few seismic stations around Mount Fuji. This is the reason that the previous studies did not clearly define the upper surface of the PHS plate beneath Mount Fuji, resulting in a controversy over whether there is no slab in the seismic gap. The VP = 6.0 km/s and VS = 3.5 km/s isovelocity surfaces correspond to the upper limit of the hypocenter distribution beneath Mount Fuji and may define the upper surface of the PHS plate. Our results strongly support the former idea that the descending PHS plate exists in the apparent seismic gap.
 The results of the petrologic studies also suggest the existence of the PHS plate beneath Mount Fuji. Mount Fuji has produced large amounts of basaltic magma [Tsuya, 1968, 1971; Fujii, 2001], while volcanoes in the Izu-Bonin arc have produced not only basaltic magma but also andesitic and dacitic magma. The basaltic magma of Mount Fuji has been much more strongly differentiated than that from oceanic ridges. The composition of the basaltic magma of Mount Fuji is neither compatible with nor in equilibrium with mantle peridotite [Fujii, 2004]. It is completely different from the composition of magma that would be produced by the partial melting of the decompressed warm mantle rising to fill a gap in the proposed splitting PHS plate. This petrologic evidence also supports our conclusion that there is no gap in the PHS plate beneath Mount Fuji.
 The magma differentiation trend of Mount Fuji is different from that of volcanoes of the Izu-Bonin arc. This difference is attributable to the depths at which magma differentiates. Differentiation of basaltic magma beneath Mount Fuji occurs with an overburden pressure of 0.1–0.2 GPa, which is greater than that beneath the volcanoes in the Izu-Bonin arc [Fujii, 2004]. The reason for the large pressure difference is that the EUR (or NAM) plate overlaps the PHS plate beneath Mount Fuji [Fujii, 2004]. Because our results do not clearly image the lower bound of the PHS plate beneath Mount Fuji due to poor resolution at depths greater than 25 km, future studies should reveal this lower bound by exploring at greater depths.
6.3. Low-VP, Low-VS and Low-VP/VS Anomalies and Deep Low-Frequency Earthquakes of Mount Fuji
 Seismic velocities in crustal rocks depend on factors such as temperature, pressure, composition, crack density, and fluid content. A few general properties have emerged from laboratory and in situ velocity experiments and from theoretical studies, though the relationships are not completely understood for all rock types. For example, it has been suggested that low-VP, low-VS and a high-VP/VS ratio are associated with partial molten rock and/or a magma reservoir [e.g., Takei, 2002]. A low-VP, low-VS and high-VP/VS anomaly at depths of 15–25 km beneath Mount Fuji is consistent with the volume of partial melt of basaltic magma.
 A striking feature of our new result is that a low-VP anomaly and a low-VP/VS anomaly are found at depths of 7–17 km beneath Mount Fuji. Since a high-VP/VS ratio is a physical property of magma, the signature of low-VP/VS excludes the presence of molten rocks at these depths. Theoretical modeling shows that randomly distributed pores with large aspect ratios (∼0.1) filled with supercritical fluid in the rock matrix will decrease VP and VS as well as VP/VS [Takei, 2002]. The presence of the supercritical fluid of volatile components may explain the relatively low-VP/VS at depths of 7–17 km. If the low-VP/VS is caused by the rock matrix filled with the supercritical fluid, then abundant volatiles at these depths are required. In order to explain the abundant volatiles, these volatiles must have a very low solubility and exist in relatively large quantities within basaltic magma. CO2 is one of the leading candidates for such a volatile because it has by far the lowest solubility and exists in large quantities among volatiles commonly present in basaltic magmas [Bottinga and Javoy, 1990]. H2O is also one of the candidates for crack-filling supercritical fluid because H2O is easily generated from the dehydration of the partial melting materials [e.g., Stern, 2002]. Nakajima et al.  suggest the presence of H2O at midcrustal depths beneath a volcanic front in a subduction zone based on the low-VP, low-VS and low-VP/VS anomalies seen in their tomography results.
 CO2 emissions from the flanks of Mammoth Mountain in eastern California, together with the continued occurrence of DLF earthquakes, indicate the presence of basaltic magma in the upper 20 km of the crust beneath Mammoth Mountain [Hill, 1996]. The DLF earthquakes beneath Mammoth Mountain are likely associated with CO2-rich basaltic magma distributed within a plexus of dikes and sills at midcrustal depths of 10–25 km [Hill and Prejean, 2005].
 Our data show that DLF earthquake locations beneath Mount Fuji are distributed mostly within the low-VP, low-VS and low-VP/VS anomaly, where fluid-filled pores are expected to exist. At Mount Fuji, the source mechanism of a DLF earthquake includes large nondouble-couple components, suggesting that DLF earthquakes result from fluid flow [Nakamichi et al., 2004a]. These items of circumstantial evidence strongly suggest that CO2 and/or H2O play an important role in DLF earthquake excitation in Mount Fuji as in the Mammoth Mountain case.
 A seismic survey around Mount Fuji using a dense array was conducted during 2002–2005. VP and VS tomography, using both local earthquakes and controlled seismic sources, elucidated important details about the crustal structures of the South Fossa Magna and Mount Fuji with greater resolution than previous geophysical studies. Important results include the following:
 1. The upper surface of the PHS plate may be defined by the upper limit of the hypocenter distribution and the isovelocity (VP = 6.0 km/s and VS = 3.5 km/s) contours beneath Mount Fuji.
 2. A low-VP, low-VS and low-VP/VS anomaly is seen at depths of 7–17 km beneath Mount Fuji, corresponding to the DLF earthquake locations. Volatile fluid may be abundant and would facilitate the generation of the DLF earthquakes.
 3. A low-VP, low-VS and high-VP/VS anomaly at depths of 15–25 km beneath Mount Fuji is consistent with the presence of partial melt of basaltic magma.
 4. VP of 6.0–6.5 km/s in the Izu collision zone beneath the Tanzawa Mountains is interpreted as a plutonic body of tonalite within an accreted crustal slice of the Izu-Bonin arc.
 5. A low-velocity anomaly trends northward in the west along Fujikawa, from Suruga Bay to northwest of Mount Fuji. This feature well correlates with Bouguer gravity anomalies of the South Fossa Magna.
 6. A low-VP, low-VS and low-VP/VS anomaly is seen at depths of 6–14 km beneath Mount Hakone. Thus the results of the present tomographic imaging study clarify the volcanic process beneath Mount Fuji and the collision tectonics of the South Fossa Magna.
 We are grateful to T. Kagiyama, J. Oikawa, M. Hagiwara, T. Takeda, E. Koyama, N. Osada, H. Tsuji, and T. Kobayashi of the University of Tokyo; H. Ohsima, H. Aoyama, T. Maekawa, and A. Suzuki of Hokkaido University; S. Tanaka, T. Nishimura, K. Nida, and J. Yamazaki of Tohoku University; K. Yamaoka and M. Yamada of Nagoya University; Y. Sudo, M. Iguchi, and A. Ohkura of Kyoto University; H. Shimizu, T. Matsushima, and K. Uehira of Kyushu University; and H. Yakiwara of Kagoshima University for their participation in the field experiment. We are also grateful to HSIK, JMA and NIED for providing us with the waveform data. This study was partially done at the U.S. Geological Survey, Menlo Park, during H.N.'s stay as a visiting postdoctoral fellow. We used the tomoDD program [Zhang and Thurber, 2003] and would like to thank the authors. We appreciate the discussions the first author had with David Hill, Bruce Julian, Gillian Foulger, Margaret Mangan, and Gregory Waite. We thank Jonathan Lees, Michael West, Thomas Wright, and the Associate Editor of JGR for their constructive comments and for helping us to refine our English expression. H.N. was supported by JSPS Postdoctoral Research Fellowships. This work was supported by a Grant-in-Aid for JSPS Fellows to H.N. The seismic observation was financially supported by the National Project for Prediction of Volcanic Eruptions and by a Special Coordination Fund for Promoting Science and Technology from the Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan.