We carried out a reflection/refraction seismic survey across the southern Izu-Ogasawara arc–back arc system, covering three arcs with different crustal ages. The oldest Eocene arc has middle and lower crust with high velocities of 6.4–6.6 and 6.8–7.4 km/s, respectively, suggesting denser crustal materials. The current volcanic arc has middle and lower crust with lower velocities of 5.7–6.5 and 6.7–7.1 km/s, suggesting advanced crustal differentiation. The crust-mantle transition layer, with a velocity of 7.5–8.0 km/s, is distributed beneath the current volcanic arc and the rear arc, suggesting a pool of dense materials emanating from the crust through the crustal growth. These structural differences between the Eocene arc and current arc indicate a difference of crustal growth based on basaltic and andesitic magmas according to known petrologic studies. Commonly, rifted crusts have lower crusts with high velocities of over 7.0 km/s, and the arc–back arc transition zone also has a thinner more reflective crust that may have been affected by postrifting magmatism.
 Oceanic island arcs, developed by the subduction of a basaltic oceanic plate beneath another oceanic plate, are one of the best locations in which to study the crustal evolution that produces continents. The Izu-Ogasawara arc is located in the western Pacific. Although it is known that the northern Izu-Ogasawara arc (NIA) has middle crust with a P wave velocity (Vp) of 6 km/s composed of continental felsic/intermediate materials [Suyehiro et al., 1996], the process that yields this felsic/intermediate middle crust could prove to be one of the keys to understanding the crustal evolution from oceanic crusts to continents.
 Structural characteristics of oceanic island arcs are not always similar to each other. Some of the major characteristics for the NIA crust are as follows: middle crust has a Vp of 6 km/s, thin upper part of the lower crust has a Vp of 6.8–6.9 km/s, and thick lower crust has a Vp of 7.1–7.3 km/s [Takahashi et al., 1998]. Kodaira et al. [2007a] modeled the velocity structure along the volcanic front in the NIA and the southern Izu-Ogasawara arc (SIA) and found that a layer with an average Vp of 7.4 km/s underlies lower crust. Sato et al.  showed that the Vp of this layer are 7.4–7.7 km/s and suggested that it is composed of a mixture of crust/mantle materials below the lower crust. The above mentioned structural characteristics, which are components of the middle crust (Vp = 6 km/s) and the thick lower crust, were also detected in other island arcs, which are the Mariana arc, West Mariana Ridge [Calvert et al., 2008; Takahashi et al., 2008] and Tonga arc [Crawford et al., 2003]. However, the Aleutian arc has different structural characteristics. The central Aleutian arc has thin upper crust, thin middle crust (Vp = 6.5–6.8 km/s), and thick lower crust (Vp = 7.0–7.2 km/s) [Holbrook et al., 1999]. According to the Aleutian active source seismic studies [Holbrook et al., 1999; Fliedner and Klemperer, 1999, 2000; Lizarralde et al., 2002; Shillington et al., 2004], the characteristics of the Aleutian arc are an upper crust with a Vp of approximately 6 km/s, a middle crust with a Vp of over 6.5 km/s, and a lower crust with a high Vp of over 7 km/s.
 The reason for these structural differences is still unknown. According to suggestions that arc evolution assists in directing dense, and mafic crustal materials toward the mantle [Kay and Kay, 1993; Jull and Kelemen, 2001; Tatsumi et al., 2008], the degree of arc evolution could be one of the candidates for inducing velocity variations. Tatsumi et al.  estimated the mineral compositions of the arcs produced using two different scenarios: the mantle-derived basalt and mantle-derived andesite models. The former causes by underplating basaltic primary magma, and the latter uses andesitic primary magmas. Primary andesitic magmas have been suggested to be produced under high-temperature environments [Tatsumi, 2000]. Tatsumi et al.  concluded that the mantle-derived andesite model causes the high Vp in the middle and lower crust. The SIA is one of the best targets for investigating the degree of arc crustal evolution because it has three arcs with different crustal ages: the extinct Ogasawara Ridge produced during the Eocene (OR), the current volcanic arc (CVA), and the extinct rear arc produced in the Oligocene and Miocene (RA) (Figure 1).
 Structural changes are also brought about by crustal rifting. In this study area, there are two back-arc basins formed by rifting: the Ogasawara Trough (OT) lies between the OR and the CVA, and the Nishinoshima Trough (NT) is between the CVA and the RA. The OT is bounded by normal faults and was formed by rifting during the Eocene [Ishizuka et al., 2006a] and the NT was also formed by rifting [Yuasa and Murakami, 1985]. It is important to find structural commonalities in these rifted crusts in order to understand the structural change brought by the rifting. The Shikoku Basin (SB), which is a back-arc basin formed through arc rifting, is located to the west of the RA. The SB margins might have clues to understanding the arc evolution brought about by crustal rifting.
 Although the SIA has structures important for understanding arc crustal growth, there have been few controlled seismic experiments to understand the velocity structure across the entire SIA. Therefore, we carried out a seismic experiment across the SIA using 110 ocean bottom seismographs (OBSs) and an air gun array with a total capacity of 12,000 cubic inches (197 L) to obtain the velocity structure. The final goal of this paper is to answer why there are crustal variations among oceanic island arcs and what arc rifting brings inside the crust. In this paper, we show that Eocene arc is composed of denser materials than other two arcs and that high-velocity lower crust identified beneath rifted crust commonly is constructed by magmatic underplating rather than serpentinized mantle.
2. Tectonic Setting
 The entire Izu-Ogasawara arc was formed by intraoceanic subduction [Karig and Moore, 1975; Hall et al., 1995], which began when the direction of the Pacific Plate changed from NW to WNW at 50 Ma [Sharp and Clague, 2006]. The current speed of the Pacific Plate subduction is approximately 45–60 mm/a [Seno et al., 1993]. The infant island arc, including the OR and part of the Mariana frontal arc, was produced by boninitic volcanism beginning in the Eocene [Pearce et al., 1992; Ishizuka et al., 2006a]. The old oceanic crust before the production of the Eocene arc was accumulated around the inner trench slope of the Izu-Ogasawara trench [DeBari et al., 1999]. Volcanism in the Eocene arc continued until the Oligocene and decreased at 27 Ma [e.g., Taylor, 1992]. Although the distribution of the Eocene and Oligocene volcanism have not yet been specified, Kodaira et al.  suggested that the Oligocene arc is located beneath the current rear arc on the basis of correlations between the crustal structures of the CVA and the RA. Volcanogenic sediments collected in the NIA fore arc by ocean drilling indicate that volcanic ash does not exist at 23–17 Ma [Taylor, 1992]. This period of inactive volcanism corresponds to that of the back-arc opening, which separates the Kyushu Palau Ridge from the NIA. Okino et al.  indicated that the period of the SB opening was 30–15 Ma. The Parece Vela Basin also opened around the same time at ∼29–15 Ma [Okino et al., 1998]. Arc volcanism activated again at 17 Ma in the Miocene [Taylor, 1992]. The volcanism, active from the Miocene to the Pliocene, extended as far as 120 km west of the current volcanic front, i.e., from the RA to the CVA [Ishizuka et al., 2006b]. Thereafter, the most recent volcanism in the period between the Pliocene and the Quaternary occurred within 40 km of the volcanic front [Ishizuka et al., 2006b].
 The local tectonic setting for the SIA is as follows: the OR with a boninitic materials was constructed between 48 Ma (Eocene) and 30 Ma (Oligocene) [Ishizuka et al., 2006a]; the RA, including the extinct Tenpo Seamounts, formed during the Miocene until the SB opening ceased [Ikari and Nishimura, 1991]; and the current arc including the active volcanic front was active during the Pliocene-Quaternary period [Ishizuka et al., 2006b]. Yuasa  suggested that the Eocene volcanics was distributed widely at not only the current OR but also the volcanic front, and Stern and Bloomer  indicated that the crustal production rate during the Eocene was several times greater than the current rate. The distribution for the entire Eocene volcanics, however, has not yet been specified. The OT was produced by rifting at 41–45 Ma [Ishizuka et al., 2006a]. After the Oligocene volcanism, the SB opening propagated from the north to the south during the rifting period for the SIA, then, the Parece Vela Basin opening propagated from south [Okino et al., 1994; Sdrolias et al., 2004]. Yuasa  suggested that the Parece Vela Basin opening cut into the SIA and the northern edge corresponds to the NT. Kodaira et al.  suggested that the NT activated as part of the Oligocene rift. The age of the NT sediments identified, however, is after the Pliocene [Higuchi et al., 2007], and the origin of the NT is not yet understood.
3. Seismic Experiment and Data Characterization
 To detect the structural variations of the SIA crust, we carried out a seismic experiment across the entire SIA using the R/V Kaiyo of the Japan Agency for Marine-Science and Technology (JAMSTEC) [Kaiho et al., 2005]. In order to understand the entire crustal evolution for the region, we ran a long offset seismic line from the trench to the back arc. This enabled us to construct a velocity model of the whole arc transecting down to the mantle. The seismic line runs from the inner trench slope to the western SB through the OR, OT, CVA, NT, and the RA (Figure 1). We deployed 110 OBSs on the line at intervals of 5 km and employed an air gun array with a total capacity of 12,000 cubic inches (197 L). The shooting interval was 200 m.
 To obtain entire crustal structures achieving our scientific goal, observing distinct reflections and refractions from the long offsets was required. Continuous reflections are useful to identify extent of layers with different compositions. Identifying structural characteristics is difficult in thin rifted crust and crusts with large interval velocity contrasts, because the tomographic inversion smoothes these abrupt structural changes. Constructing the layered structure using the reflection distribution is useful to overcome the smoothing effect of the tomographic inversion. Our data quality is sufficient for modeling the arc structure, and almost all of the OBS record sections have long first arrivals to an offset of 100–250 km. Here, we show nine vertical sections of the OBSs to introduce the data characterization representative for each ridge and basin (Figure 2). Site 6, site 17, site 33, site 44, site 53, site 74, site 87 and site 108 were deployed on the inner trench slope, the OR, the OT, the CVA, the eastern edge of the NT, the RA, the eastern SB, and the western SB respectively. These record sections have clear refractions as first phases and some reflections within the crust. We modeled a crustal structure mainly to satisfy these traveltimes for refractions and reflections.
 The western side of site 6 has clear refractions until an offset of about 200 km and clear reflections at offsets of 15–30, −40 to −20, and −100 to −40 km (Figure 2a). On the eastern side, reflections are traced until an offset of 40 km. In particular, the reflections identified at offsets of 10–30 km have relatively large amplitudes. At the far side of an offset, we traced large amplitude reflections with no refractions. Site 17 also indicates clear refractions until an offset of 120 km (Figure 2b). On the eastern side, refractions are traced until an offset of 80 km, and clear reflections with large amplitudes are identified at offsets of 70–80 km and a traveltime of 6 s. The western side of site 33 has first phase undulations corresponding to the severe topography (Figure 2c). Clear reflections are observed at offsets of 25–50 and 90–110 km on the eastern side and offsets of −75 to −40 km on the western side. Site 44 also has clear refractions on both sides until an offset of 100 km and reflections with large amplitudes at offsets of 40–50, 50–80, and −60 to −25 km (Figure 2d). At site 53, refractions are observed as first phases until offsets of 80 and −100 km on both sides (Figure 2e). Clear reflections are identified at offsets 40–75, and 65–80 km on the eastern side. Site 74 indicates refractions as the first phases until far offsets of over 100 km on both sides, and the apparent Vp of refractions in the western side becomes 8 km/s near the offset of 25 km (Figure 2f). Clear reflections are observed at eastern offsets of 50–75 km and −30 to −20 km. Site 87 has very short offsets with an apparent Vp of less than 8 km/s (Figure 2g). Site 108 has clear refractions as the first phases until offsets of over 100 km, with apparent velocities indicating severe undulation at offsets of 10–20 km and corresponding to the Kinan Seamount Chain.
4. Data Analysis
 To obtain a velocity structure with high resolution, we employed a tomographic inversion technique using first arrivals [Zhang et al., 1998] and a diffraction stack migration approach using traced reflections [Fujie et al., 2006]. Forward modeling using a layered model [Zelt and Smith, 1992; Zelt and Ellis, 1988] was then used to fit the traveltimes and amplitudes of the reflections. An elastic modeling using finite difference wave propagation code (E3D) [Larsen and Grieger, 1998] was employed to confirm the amplitudes of the refractions and reflections.
 First, we carried out the tomographic inversion using a simple initial model to detect lateral variations of the crustal structure (Figure 3). On the basis of the seismic structure obtained from previous studies [Suyehiro et al., 1996; Takahashi et al., 2007], the initial model with sedimentary layers citing velocity tables obtained using reflection data was composed of a homogeneous thin crust (∼6 km thick) beneath the back arc and a thick crust (∼20 km thick) beneath the arc region (Figure 3a). This initial tomographic inversion comprised two steps: the first step was carried out for the shallow section using limited first arrivals within an offset of 20 km. Using the results of the first step, we carried out the inversion in the second step using all of the first arrivals (Figure 3b). We repeated the iteration until the RMS residuals reached minimum values of less than 100 ms. We then mapped the reflection arrivals on the velocity image using the diffraction stack migration approach [Fujie et al., 2006] (Figure 3c). We constructed a new initial model, shown in Figure 3d, with variations of the crustal thickness based upon the image shown in Figure 3c. We included the subducting oceanic crust in this model. We carried out the tomographic inversion again and obtained the final result from the second inversion (Figure 3e). We then obtained a velocity image with reflectors as mentioned above (Figure 3f). The grid size for these tomographic inversions was 500 m × 500 m. We estimated the picking error for each trace depending upon the signal-to-noise ratio, with a maximum of 200 ms. Figure 4 shows the final tomographic image obtained from the result of a checkerboard test used to identify the reliable region and its interpretations. In the checkerboard test, we adopted a Vp perturbation of 5% and two perturbation grid sizes of 10 km × 5 km for regions shallower than 10 km and 20 km × 10 km for regions deeper (Figure 4b). We regard portions having a clear positive-negative perturbation pattern as the resolved areas. Most of the traveltime residuals fall within ±100 ms (Figure 5).
 Although the above tomographic inversion is advantageous for constructing lateral heterogeneous velocity images, it is impossible to use for modeling the reflection traveltimes. Therefore, we constructed a layered model with a velocity contrast by performing conventional forward modeling for each layer [Zelt and Smith, 1992; Zelt and Ellis, 1988] using the traveltimes of the identified reflections and refractions in order to distinguish continuous layers (Figure 6). We selected four layers as the targets to be distinguished using continuous reflectors (A–E in Figures 4c and 6c–6g). The forward modeling result was confirmed by calculations of the RMS traveltime residuals and chi-square values which denote the RMS traveltime residuals normalized by the picking error and by comparing the observed and calculated traveltimes (Figure 7). The node interval for the forward modeling is 5 km, which is similar to the OBS interval. Therefore, the velocity structure from forward modeling has a resolution for lateral variation of about 5 km. After the construction of the velocity model, we verified an error range of Vp, indicating the minima of the RMS residuals and chi-square values by using the simplified Vp value to confirm the specified structural characteristics (Figure 8).
 To model the amplitudes of deep reflections from interface E beneath the CVA, we calculated the amplitudes of the reflections using E3D software [Larsen and Grieger, 1998] and determined the velocity contrast by trial and error. Such reflections are observed on some OBSs that were deployed around the current active arc, from site 33 to site 54. Although it is difficult to quantitatively evaluate the consistency between the synthetic and observed seismograms, we formulated the synthetic seismograms by changing the velocity contrast with intervals of 0.1 km/s (Figure 9).
5. Seismic Structure of the SIA
 We obtained two velocity models using the abovementioned analysis flow: one is a tomographic image and the other is a layered structure. These two velocity models are very similar to each other and have common structural characteristics. We identified seven regions on the basis of the obtained structural characteristics: the OR, OT, CVA, NT, RA, the arc–back arc transition zone (ABT) between the RA and the SB, and the SB (Figure 10). We extracted the crustal layers of each area on the basis of the continuous interface distribution and the Vp, as described in section 4.1. Although there are velocity variations within each layer, we differentiate them as the upper, middle, and lower crust, and a crust-mantle transition layer (CMTL) (Figure 10a). In this section, we summarize the characteristics and reliability of the velocity model with a layered structure (Figures 10b and 10c).
5.1. Reliability of the Velocity Model
 In this study, velocities down to the upper mantle are generally resolved along the entire profile on the basis of the results of the checkerboard test. The maximum depths of the resolved area from beneath the SB to the ABT, the RA to the NT, and the CVA to the OR are approximately 15, 20, and 25 km, respectively (Figure 4b).
 For forward modeling on the layered model, we confirmed the reliability of the velocity model as shown in section 4, and modeled the offset ranges and amplitudes for observed refractions changing the velocity gradients of each layer. The error for each Vp was confirmed by using the RMS traveltime residuals and is estimated to be ±0.1 km/s. The RMS traveltime residuals of the refractions through the upper, middle, and lower crust, the CMTL, and the upper mantle are 88, 92, 105, 130, and 172 ms, respectively. The chi-square values of these refractions are 3.509, 1.694, 1.687, 1.284, and 1.429.
 As shown by Figure 8, we estimated the Vp ranges for the middle crusts beneath the RA (Figure 8a), the CVA (Figure 8b), and the OR (Figure 8c), and the lower crusts beneath the ABT (Figure 8d), the NT (Figure 8e), and the OT (Figure 8f). In this test, we estimated the Vp error using each time difference between the minimum peaks of the profiles for the RMS traveltime residuals and chi-square values. By adding the time difference to Vp step used in this test (0.1 km/s), the Vp errors for the middle crusts beneath the RA, CVA, and OR are 0.1, 0.2, and 0.2 km/s, respectively. The Vp errors for the lower crusts beneath the ABT, NT, and OT are 0.4, 0.2, and 0.3 km/s, respectively. On the basis of the test results, we concluded that the Vp for the middle crust of the OR is larger than for the others, and that the Vp of the lower crust for the three trough regions are higher than for the typical oceanic crust beneath the Pacific seafloor with a depth of over 2.5 km (6.81 ± 0.16 km/s [Shor et al., 1970]).
 In order to model the amplitudes of deep reflections from interface E, we constructed some models with a velocity contrast ranging from 0.1 to 0.3 km/s. Figure 9 shows an example of this modeling for site 49. Assuming the velocity contrast of interface E to be 0.3 km/s, the amplitudes of the modeled reflections are consistent with the observed ones (Figures 9a and 9b). However, a velocity contrast of 0.1 km/s yields relatively small amplitudes of reflections at an offset of 75–85 km (Figure 9c). The error in the Vp determined by modeling the reflection amplitudes is estimated to be approximately 0.2 km/s. On the other hand, a velocity contrast of 0.5 km/s at interface D yields reasonable reflections with similar amplitudes to those observed (Figure 9).
5.2. Ogasawara Ridge
 The OR is located at a distance of 465–580 km on Figure 10 on the basis of the shape of the Moho and the distribution of the middle crust. The maximum crustal thickness is approximately 18 km. Under the thin sedimentary layer, we confirmed four layers: the basement layer and the upper, middle, and lower crust. Their velocities are shown in Table 1.
Table 1. Comparisons of Velocity, Crustal Volume, and Proportion Among Three Arcs and Aleutians
Current Volcanic Arc
Average Vp (km/s)
 On this ridge, the upper crust has a Vp of approximately 6 km/s; toward the west it thickens and the Vp decreases. The middle crust is pinched out at both ends and the maximum thickness is approximately 6 km. Although the lower crust has a constant thickness of approximately 9 km except for the ends, the Vp parting the center position is slightly faster than the surrounding region. As observed by the continuous reflectors, the tops of the upper, middle, and lower crust and the Moho are likely to have velocity contrast. The Vp of the easternmost mantle wedge is 7.6 km/s, which is lower than Vp of mantle materials (∼7.8 km/s [Christensen, 1996]).
 Although no refractions beneath the subducting plate were observed, we observed reflections from the top and inside of the plate. Assuming the velocities of oceanic layers 2 and 3 to be 5.0–6.2 and 6.7–6.9 km/s, respectively, we modeled the Pacific Plate subducting beneath the OR. The subducting oceanic crust is tangential to this ridge at a range of 530–600 km and a depth of 18 km.
5.3. Ogasawara Trough
 The OT is located at a distance of 395–465 km on Figure 10c. The crustal thickness is generally 12–13 km and increases toward both sides. Underlying a thick sedimentary layer and the basement, we identified two layers: the upper (Vp = 4.3–6.1 km/s) and lower crust (Vp = 6.7–7.4 km/s). There is no middle crust with a Vp of 6.4–6.6 km/s identified beneath the OR. The mantle Vp is constant at 7.8 km/s and lateral variation was not detected.
 The western half of the upper crust is thicker than on the eastern side and its Vp (4.3–5.6 km/s) is lower than that on the eastern side (5.7–6.1 km/s). At the center of the trough, we identified a sharp-pointed structure (like an intrusive body) with a relatively high Vp of approximately 6 km/s. The thickness of the lower crust increases westward from 6 to 8 km. Its Vp is relatively high with a maximum value of 7.4 km/s. The Vp of the lower crust decreases to 6.7–7.1 km/s at the western boundary with the CVA. A reflector inside the lower crust, which is discontinuous toward the OR and the CVA, coincides with a contour line of 7.2 km/s (Figure 4c).
5.4. Current Volcanic Arc
 The CVA is located at a distance of 305–395 km on Figure 10c. The crustal thickness has a maximum of approximately 20 km and thins toward both ends. The crust is composed of five layers, which are similar to the OR. Underneath the sedimentary layer, we identified the basement layer, and the upper, middle, and lower crust. Their velocities are shown in Table 1. In addition, there is the CMTL with a Vp of 7.5–7.6 km/s between the lower crust and upper mantle.
 The model exhibits markedly strong lateral variations in the middle and lower crust. The middle crust has a maximum thickness of approximately 5 km immediately beneath the volcanic front. The Vp beneath the volcanic front and beneath the next western topographic high (distance = 320 km) are 6.0–6.5 and 5.7–6.2 km/s, respectively. The Vp of the lower crust immediately beneath the volcanic front is 6.7–7.1 km/s, which is lower than for the OT. A reflector inside the lower crust is identified in the eastern half and discontinuous to the western part (distance = 380 km; Figure 4c).
 The CMTL with a Vp of 7.5–7.6 km/s is distributed beneath the CVA. A velocity contrast on interface E (Figure 4c) between this transition layer and the upper mantle reaches a maximum of 0.3 km/s immediately beneath the volcanic front. This layer thins beneath the OT and is obscured toward the west. The upper mantle Vp beneath this transitional layer is 7.9 km/s, which is close to velocities of mantle materials (∼7.8 km/s [Christensen, 1996]). Beneath interface E, a few reflectors also exist in the upper mantle; however, they are laterally discontinuous.
5.5. Nishinoshima Trough
 The NT is located at a distance of 260–305 km on Figure 10b. The crustal thickness is approximately 10 km. The mantle Vp is 7.9 km/s. The crust is composed of four layers beneath the sedimentary layer, the basement layer and the upper (Vp = 4.6–5.6 km/s), middle (Vp = 6.0–6.3 km/s), and lower crust (Vp = 7.0–7.4 km/s).
 The upper crust with a maximum thickness of approximately 2 km underlying the basement has a Vp of 4.5–5.5 km/s. The thickness beneath the eastern side is higher than that beneath the western side. The middle crust is thin beneath this trough and is indicated by the continuous distribution of the reflections from the middle and lower crust (Figures 6d and 6e). The lower crust also becomes thin toward the center of the trough and the minimum thickness is approximately 5 km.
5.6. Rear Arc
 The topographic high located in a range of 160–260 km on Figure 10b is named the Tenpo Seamount, which is one of the seamounts in the RA. The crustal thickness has a maximum of 16 km and decreases toward both ends. The crustal composition is similar to the CVA. The velocities of the upper, middle and lower crust are shown in Table 1, and the Vp of the lower crust is faster than that of the CVA.
 The thickness of the upper crust is lowest in the central part of the RA and thickens toward both flanks. The middle crust is the thickest in the central part and decreases toward both flanks. The lowest Vp of 6.6 km/s within the lower crust is distributed at the center of this seamount. However, the Vp increases westward and is over 7.3 km/s beneath the boundary between the RA and the ABT. Beneath the lower crust, the CMTL has a maximum Vp of 7.6 km/s. This Vp is significantly lower than the typical mantle Vp (8.0 km/s) and is similar to that beneath the CVA. However, it increases toward both of the ridge ends, becoming 7.8–8.1 km/s. Furthermore, this low Vp does not extend to the deep reflector at the depth of approximately 27 km (Figures 4c and 10). The reflector is discontinuous next to both the ABT and the NT.
5.7. Arc–Back Arc Transition Zone Beneath the Eastern Shikoku Basin
 We see a high-velocity lower crust (HVLC) with a Vp of over 7.0 km/s at a distance of 75–160 km on Figure 10b, which is greater than that of a typical oceanic crust [White et al., 1992]. In this paper, the region with the HVLC has been named as the ABT. The Koka Seamount is composed of a basement layer and an upper, middle, and lower crust beneath the sedimentary layer. For the other region except the Koka Seamount, we identified two major layers: the upper and lower crust. Although the error for the high Vp within the lower crust is relatively large, it is still probably faster than that the one bordering the west side, as described in section 5.1.
 The velocity model of the ABT is characterized by a middle crust with a Vp of 6.1–6.5 km/s only beneath the Koka Seamount and the HVLC. The lower crust with a high Vp thickens beneath the Koka Seamount and the Moho probably deepens underneath. The Vp of the lower crust in the eastern half (distance: 115–160 km) is over 7.2 km/s, and it is greater than that in the western half (distance: 75–115 km), which is in the range of 7.0–7.2 km/s. There is high reflectivity in the lower crust (distance: 115–170 km). As shown in Figure 4c, regions with high reflectivity are located within the areas having a Vp of over 7.0 km/s. The upper mantle is almost homogeneous with a Vp of 7.9–8.1 km/s.
5.8. Shikoku Basin
 The SB is one of the back-arc basins between the Izu-Ogasawara arc and the Kyushu-Palau Ridge and broadly occupies the northern Philippine Sea. The area, except for the region including the HVLC, is located at a distance of 0–75 km on Figure 10b. The crustal thickness is approximately 6 km. The mantle Vp is 7.9–8.1 km/s and similar to that of a typical mantle (8.0 km/s) [Christensen and Mooney, 1995]. The crust is composed of three layers beneath the sedimentary layer: a basement layer and the upper (Vp = 5.6–5.9 km/s) and the lower crust (Vp = 6.5–6.7 km/s). The lower crust is homogeneous with a thickness of 2–3 km and a constant Vp.
 As described in the introduction, there are differences of the structural characteristics of island arcs, such as the Izu-Ogasawara-Mariana arc and the Aleutian arc. The processes controlling arc growth may be the key to understanding the different crustal structures in these arcs. Here, we discuss how the structural variations are brought about in view of the crustal growth process. In order to achieve our goal, we take following steps:
 1. We note the structural variations among the different arcs: the oldest OR from the Eocene, the current active CVA, and the extinct RA from the Miocene and summarize their structural commonalities and differences.
 2. We discuss the possibility that such variations can be brought about by crustal growth based upon two different petrologic scenarios.
 3. We compare the velocity model of the Izu-Ogasawara-Mariana arc with that of the Aleutian arc. Because both arcs include boninitic materials, finding common characteristics between them is significant.
 4. We note the structural commonalities among rifted crusts as another candidate for causing crustal alterations. Although the crustal growth of the Izu-Ogasawara-Mariana arc was accompanied by back-arc opening, the Aleutian arc has no back-arc basins. The influences of the tectonic differences for crustal growth are discussed.
6.1. Structural Variation of Arcs
 As mentioned above, the three arcs in this region (the OR, the CVA, and the RA) have three layers beneath the basement in common: the upper, middle, and lower crust. Here, we compare the Vp volume profiles for these arcs on the basis of tomographic images and the layered model (Figure 11). The profiles of the three arcs using both the tomographic image and the layered model are roughly similar, although the effects of smoothing on the tomographic image may be slightly excessive when identifying the characteristics of the shallow crust. The OR has two major peaks at velocities of 6.6 and 7.1 km/s (Figure 11a). The CVA has three peaks at velocities of 6.3, 6.9, and 7.6 km/s (Figure 11b). The RA has two major peaks at velocities of 6.3 and 7.3 km/s (Figure 11c).
 We compared these profiles to identify commonalities and differences in the structural characteristics of these arcs (Figure 12a). The most remarkable characteristic on the OR profile is the peak at a Vp of over 6.5 km/s corresponding to the middle crust. On the other hand, the peaks beneath the CVA and the RA are both located at Vp of less than 6.5 km/s. For the lower crust, the OR and the RA have a peak between 7.0 and 7.5 km/s, whereas the peak for the CVA is located between 6.8 and 7.0 km/s. As shown in Table 1, the velocities beneath the OR are the highest for all crustal layers. The CMTL with a Vp peak of 7.6 km/s is distributed beneath the CVA and the RA. Sato et al.  indicated that a CMTL with a Vp of 7.4–7.7 km/s observed beneath the entire current volcanic front is composed of a mixture of dense crustal materials and mantle materials. A CMTL with a Vp of less than 8.0 km/s also exists beneath the RA; however, the low Vp does not extend to the interface beneath the Moho. The relatively slow mantle Vp was also detected beneath the OR; however, it is difficult to evaluate whether the origin is the CMTL brought about by crustal growth or mantle materials serpentinized by subduction. Therefore, we do not discuss the slow mantle Vp beneath the OR separately from the discussion of the CMTL.
 Approximately 50% of the total crustal volume of all three arcs is made up of the lower crust. The upper and middle crusts occupy 11–18% and 17–26% of these arcs, respectively. These values are consistent with those for the Mariana arc proposed by Takahashi et al. . Although the values of these crustal layers are roughly similar to each other, the bulk Vp of the OR crust (6.71 km) is the highest (Table 1). The Vp of the CVA and the RA are 6.44 and 6.61 km/s, respectively, and the CVA has the lowest Vp.
6.2. Crustal Growth Processes for Three Arcs
 Although the proportions of the crustal layers are similar to each other, Vp variations in the middle and the lower crust can be seen. According to Behn and Kelemen , one of the origins for the Vp variations may be the SiO2 and MgO content. The process of one of crustal growth models could be regarded as removal of dense and mafic materials [Tatsumi et al., 2008]. Therefore, possible origins to bring about these Vp variations might be (1) different models for the crustal growth or (2) the degree of the crustal growth. Here, we discuss which origin is possible comparing the CVA with the OR and the RA.
 The structural differences between the CVA and the OR may have been brought about by different scenarios of crustal growth. According to Tatsumi et al.  introduced in section 1, their Vp profiles for the mantle-derived basalt model and the mantle-derived andesite model are consistent with those obtained beneath the CVA and the OR in this study, respectively. For example, the petrologic Vp profile [Tatsumi et al., 2008] indicates that the velocities of the middle crusts estimated using the mantle-derived basalt and andesite models are 6.0–6.5 and 6.5–7.0 km/s, respectively. This suggests that the crustal growth processes producing the middle crusts for the CVA and the OR are quite different. Because the OR is known as the boninitic arc [Ishizuka et al., 2006a], this result is also consistent with the background geology. The Vp of the lower crust beneath the OR is also faster than that beneath the CVA. In other words, an area with a Vp of less than 7 km/s beneath the OR, corresponding to the upper part of the lower crust studied by Tatsumi et al. , is narrower and limited mainly within a region near the CVA. This suggests that the OR crust remains dense and is composed of mafic materials.
 The RA has lower crust with a high Vp and higher bulk crustal Vp compared to the lower crust of the CVA. Because the structural characteristics of the RA (the Vp of the middle crust, the Vp of less than 7 km/s in the lower crust and the CMTL) are common to those beneath the CVA, the crustal compositions are probably similar. According to Tatsumi et al. , the CVA was produced with the mantle-derived basalt model. On the other hand, the deep reflector beneath the CVA has velocity contrast between 7.6 and 7.9 km/s, whereas the RA has no deep reflectors with such a velocity contrast. The average Vp of the CMTL beneath the RA is higher than that beneath the CVA. These characteristics suggest that the transformation of the dense crustal materials to the CMTL, as discussed by Tatsumi et al.  and Takahashi et al. . Sato et al.  suggests that the velocity variation in the CMTL is induced by the variation in the proportions of the dense crustal materials and mantle materials. Because the transformation of the dense crustal materials to the mantle needs a heat supply accompanied by materials rising from the slab [Tatsumi et al., 2008], it has been suggested that the RA underwent a shorter crustal growth process compared to the CVA. Indeed, the RA along this seismic line corresponds with the Tempo Seamount, which is an extinct seamount formed at 15 Ma just before the end of the SB opening [Ikari, 1991]. However, undifferentiated mafic materials may not be the only reason for the HVLC because the HVLC beneath the RA is distributed continuously until the ABT. Because such a HVLC is sometimes identified beneath continental margins, the rifting process, discussed in section 6.4, may have contributed to the development of the layer.
 As one of candidates brought above velocity differences of each layer, a difference among thermal structures of arcs is possible, because the CVA is currently active and the OR is inactive. However, we consider that the influence of thermal structure is limited. Kitamura et al.  modeled the velocity structure of the Izu-Ogasawara arc on the basis of the velocity measurement and adopted a temperature derivative of −4 × 10−4 km/s °C [Rudnick and Fountain, 1995] and adopted the maximum thermal gradient of 40°C/km according to heat flow measurements for this arc [Yamazaki, 1992] and a study of Japan arc [Nakajima and Arima, 1998]. Furukawa  used 25°C/km as a thermal gradient of typical arcs. The high temperature such as 1000°C at a bottom of the crust bring the melt in the crust according to Tatsumi et al. , however, such anomalous structures are not identified in this study. Therefore, above temperature derivative and thermal gradient are reasonable. According to Christensen and Mooney , the influence of temperature on seismic velocity has received only limited attention. In addition, it may be difficult to explain the differences of crustal proportions difference using gradual change of the temperature. We believe that a difference of the crustal composition bring the structural difference.
6.3. Comparison of the Crustal Structure With the Aleutian Arc
 We compared the velocity structures between the CVA, OR, and the Aleutian arc and found that the Aleutian arc has similar structural characteristics to the OR. From the many studies on the Aleutian arc [e.g., Holbrook et al., 1999; Fliedner and Klemperer, 1999, 2000; Lizarralde et al., 2002; Shillington et al., 2004], we selected a velocity structure across the arc summarized by Holbrook et al.  because the experimental geometry (including the OBSs) of the seismic survey was similar to that used in this study. Because their seismic line is far away from western Alaska, we believe that their velocity model is not influenced by the structure of the Alaska continental crust.
 The Aleutian arc may have similar crustal growth process as the OR. The Aleutian velocity model has three crustal layers; the upper (Vp = 5.2–6.0 km/s), the middle (Vp = 6.5–6.8 km/s), and the lower crust (Vp = 7.0–7.2 km/s). These structural characteristics of the central Aleutian arc are common to the OR. Moreover, their background geologies are also similar. The profile of the MgO content along the volcanoes of the Aleutian arc summarized by Kelemen et al. , and the MgO content beneath the velocity model of Holbrook et al.  shows a relatively larger MgO content compared to the OR with a maximum of 13.5 wt%, although there are variations of up to 4.0 wt%. Therefore, this suggests that the Aleutian arc experienced boninitic volcanism like the OR and that boninitic materials are included within the arc.
Shillington et al.  obtained the structural variation along the Aleutian arc and suggested that the crust was produced under conditions with higher heat flow and lower pressure. Studies around the Izu-Ogasawara-Mariana arc also suggest that boninites were produced under conditions with high heat flow. Pearce et al.  proposed the subduction of an actively spreading ridge beneath the young lithosphere. Ishizuka et al. [2006a] proposed that the boninitic arc crustal growth accompanied the sinking of the Pacific Plate with a trench retreat and adiabatic upwelling of the asthenosphere at 46–48 Ma. The adiabatic upwelling of the asthenosphere may have induced higher heat flow in the mantle wedge beneath the spreading fore-arc and boninitic volcanism in that region. In view of the conditions to produce boninites, the abovementioned studies show commonalities between the Izu-Ogasawara arc and Aleutian arc.
 We suggest that the structural differences identified between the Izu-Ogasawara arc and the Aleutian arc originate from differences in the crustal growth scenario as proposed by Tatsumi et al. . The crust beneath the CVA developed from basaltic primary magma. On the other hand, the crusts beneath the OR and the part of the Aleutian arc shown by Holbrook et al.  may have developed from andesitic primary magma instead.
6.4. Commonalities and Differences in the Rifted Crust
 As described in section 5, the main survey line crosses the rifted region at the HVLC, the OT, NT, and the ABT. These structures are characterized by the high Vp of the lower crust at over 7.0–7.5 km/s (Figure 10). The Vp volume functions clearly show that the Vp peaks in the rifted regions are 7.1 or 7.3 km/s, which is significantly higher than for the SB (Figure 12b). The function for the SB region has a peak at a Vp of 6.7 km/s and no peaks at Vp of 7.0–7.5 km/s (Figure 12b).
 The HVLC can effectively be seen beneath the continental margins. Although the continental margins are divided into two types, volcanic continental margins and nonvolcanic margins [e.g., White et al., 2003], there exist two corresponding scenarios producing the HVLC. One candidate is the dense crustal materials confirmed to exist beneath the volcanic continental margin [Holbrook et al., 1994]. The other candidate is serpentinized peridotites not crustal materials, which have been confirmed beneath part of the nonvolcanic continental margin [Dean et al., 2000]. The former is characterized by flood basalts on the seafloor and a large volume of the HVLC. The latter is characterized by detachment faults cutting through the entire crust and a small volume of the HVLC. However, the reflection profiles around three rifted regions have neither seaward dipping reflectors from the flood basalts nor large detachment faults (Figure 13). We consider the former as the cause for the HVLC because of the following discussion. The volume of the dense crustal materials can be small in an environment with mantle temperatures of less than 1350°C [Parkin and White, 2008] and unaccompanied by seaward dipping reflectors. The latter case often accompanies weak Moho reflections, according to Dean et al.  and Chian et al. . However, we observed clear reflections from the Moho beneath the three rifted regions (Figures 2f and 2g). Stern et al.  suggested magmatic underplating beneath the rifted crust in the Mariana Trough from petrologic studies, and Takahashi et al.  suggested that the HVLC was produced by magmatic underplating during the initial rifting periods on the basis of the velocity structure of the Mariana arc–back arc system. A similar process could have occurred beneath these three rifted regions.
 On the other hand, there are structural differences among these rifted crusts, although these crusts commonly include the HVLC. The crusts of the OT and the NT have similar structural characteristics to each other: a crustal thickness of 12–13 km and lower reflectivity of the lower crust (Figure 4c). The ABT has thin crustal thickness and high reflectivity. The thickness, except near the Koka Seamount, is approximately 7–8 km, and is almost the same as a typical oceanic crust as defined by White et al. .
 Recently, Ishizuka et al.  reported that the geochemical characteristics of the rock samples obtained from the seamounts in the ABT reveal arc-like characteristics, suggesting melts rising from the slab, as opposed to back-arc basin-like structures, and produced by mantle decompression. Considering that McBride et al.  suggested that high reflectivity in the lower crust is produced by igneous intrusion into the preexisting crust, the high reflectivity within the lower crust beneath the ABT identified in this study suggests that the ABT experienced the intrusion of igneous materials from the slab. The area deformed by the intrusive materials has clear breaks in the acoustic basement (Figures 13b and 13c). The area with the HVLC and high reflectivity extends from the western part of the RA to near the Kinan Escapement. The western end of the HVLC may be on the eastern edge of the magnetic lineament [Okino et al., 1999]. Although the age of the seamounts located at the ABT are known to be 12–15 Ma [Ishizuka et al., 2009], it is difficult to focus on one origin for the HVLC. It is possible that the crust at the initial rifting stage may have also been contaminated by the old arc crustal materials and magmatic materials underplated by mantle decompression and intrusive materials from the slab, as pointed out by Stern et al. .
 We obtained structural commonalities and differences for the OR, the CVA and the RA. Although they each have three layers (the upper, middle, and lower crust) characterized by their Vp, each structure is characterized by the velocities of each layer. The OR has the highest bulk crustal velocity, and the major characterizations are the middle crust with a high Vp of 6.4–6.6 km/s and the lower crust with a high Vp of 6.8–7.4 km/s. The CVA has the lowest bulk crustal velocity, and the major characterizations are the middle crust with a Vp of 5.7–6.5 km/s and the lower crust with a Vp of 6.7–7.1 km/s. The structural characteristics of the RA are similar to that of the CVA. These characteristics suggest that the OR is composed of much denser crustal materials than the other two. These suggest that the crustal growth beneath the CVA is more advanced the other two.
 When comparing these structural characteristics to the petrologic models developed by Tatsumi et al. , the crustal growth process CVA and the OR might follow the mantle-derived basalt and mantle-derived andesite models, respectively. The structural differences between them are expressed by the Vp of the middle crust. The velocity model of the Aleutian arc [Holbrook et al., 1999] has similar characteristics to the OR. The apparent structural difference between the Izu-Ogasawara-Mariana arc and the Aleutian arc may have originated from the different crustal growth scenarios.
 The rifted regions, which are located beneath the OT, NT, and the ABT, commonly have thin crust and the HVLC, suggesting the underplating of dense crustal materials. However, the ABT has structural differences compared to the other two trough areas: a thin crust similar to that of a typical oceanic crust and a high-reflectivity signature. The ABT also has intrusive materials on its reflection profile and the intrusive seamounts have arc-like geochemical characteristics [Ishizuka et al., 2007]. Therefore, we suggest that the lower crust of ABT developed through magmatic underplating on the thin rifted crust.
 This study was funded by the Institute for Research on Earth Evolution, Japan Agency for Marine-Earth Science and Technology, and partially supported by the Grant-in-Aid for Creative Scientific Research (19GS0211) and Grant-in-Aid for Scientific Research (B) (20340122). We are grateful to Donna J. Shillington and Simon L. Klemperer for providing their velocity models of the Aleutian arc and fruitful discussions and to Mitsuhiro Toriumi for his encouragement and discussions about the seismic structures. We thank Peter van Keken, Sara H. Pozgay, Adrienne Oakley, and one anonymous referee for their helpful comments on improving the manuscript. We received data acquisition support from Gou Fujie.