Seismic Structure of the Izu Arc‐Backarc System

Arc‐backarc systems are inherently shaped by subduction, representing an essential window into processes acting in the Earth's interior such as the recycling of subducted slabs. Furthermore, they are setting where new crust is formed and are believed to be sites where juvenile continental crust emerges. We present a seismic refraction and wide‐angle velocity model across the Izu arc‐backarc system, and use its characteristic features to constrain geochemically and petrologically different compartments, revealing processes governing crustal formation overlying subduction zones. Our result delineates the Izu arc with a maximum thickness of ∼20 km and the Shikoku Basin with thicknesses of ∼7 to 11 km. In the volcanic arc, the middle crust of the felsic to intermediate tonalitic layer (6.0–6.5 km/s) is remarkably thicker beneath the basalt‐dominated area than in the rhyolite‐dominated area, indicating that basaltic volcanism is indispensable in the transformation process from arc to continental crust. However, rhyolitic volcanism may relate to the juvenile stage of arc evolution or the remelting of middle crust due to the insufficient supply of basaltic magma from the mantle. The mafic restite and cumulates, which used to be part of the arc crustal material, are delaminated and foundered into the mantle, forming extremely low mantle velocities (<7.5 km/s). In the Shikoku Basin, our result supports a fertile mantle source with passive upwelling and normal temperature during the opening process, but the lack of high velocity in the lower crust rules out hydrous melts entrained from the subducting slab or anomalous mantle trapped during subduction zone reconfiguration.

• A long seismic refraction and wide-angle profile presents the seismic structure across the Izu arc and Shikoku Basin • The transformation from arc to continental crust is closely associated with basaltic volcanism from the rear arc to volcanic front • Passive melting of a fertile mantle source under normal temperature governs the opening of the Shikoku Basin Supporting Information: Supporting Information may be found in the online version of this article.
One of the basic and still open questions is the formation and evolution of Earth's continental crust, which is unique in our solar system in having a buoyant andesitic composition (Tatsumi et al., 2015).The occurrence of continental crust might be an important and vital feature in supporting the conditions for the evolution of life and ecosystems.Taylor and McLennan (1995) proposed that post-Archean growth of continental crust is a result of island arc evolution.However, the paradox between the predominant basaltic volcanic rocks in oceanic arcs and the andesitic composition of continental crust makes the formation mechanism of the continental crust enigmatic (e.g., Rudnick, 1995).
To solve this problem, a large number of wide-angle seismic and petrological surveys were conducted in intraoceanic volcanic arcs (e.g., Ishizuka et al., 1998;Kodaira et al., 2007aKodaira et al., , 2007b) ) because the contamination of preexisting continental crust can be avoided here and the construction of continental crust based on oceanic crust can be examined.A petrological modeling study tested different theoretical concepts for arc lava formation, including the mantle-derived basalt model and the mantle-derived andesite model, delineating the evolution of arc crust to continental crust in the Izu-Bonin-Mariana (IBM) arc (Tatsumi et al., 2008).The mantle-derived andesite model suggests that primary mantle-derived andesitic melts are generated from slab melting (Kelemen et al., 1993;Pearce et al., 1992).Yet, the crustal anatexis pattern in the mantle-derived basalt model suggests that the intermediate to felsic melts are generated via partial melting of basaltic crustal compositions (e.g., Kawamoto, 1996), while the magma mixing mechanism proposes that the bulk composition of continental crust is generated by the mixing of basaltic magma from mantle melting induced by slab dehydration with a felsic magma produced by partial melting of an initial basaltic arc crust (Tatsumi & Kogiso, 2003).The results of petrological modeling show that the calculated velocity values for each layer of the IBM crust are consistent with the observed values on the basis of the basaltic model, whereas the andesite model cannot account for the velocity structure (Tatsumi et al., 2008).Moreover, seismic refraction and wide-angle seismic studies running along the magmatic arc demonstrated that an intermediate composition layer (V p = ∼6.5 km/s) that is close to the bulk continent is observed beneath the basaltic volcanoes, while the rhyolitic volcanism may relate to a more juvenile stage of crustal evolution, or remelting of preexisting continental crust (Kodaira et al., 2007a(Kodaira et al., , 2007b)).Nevertheless, the correlation between the arc crustal structure and volcanism type is poorly defined in the cross-arc direction.Diverse forms of volcanism characterize from the volcanic arc to backarc region, reflecting variation in the depth of dehydration melting, temperature variability, and mantle composition.Therefore, additional efforts are required to reveal which conceptual model for the construction of continental crust can be best applied from forearc to rear arc, or if the basaltic model can only apply to the active volcanic front.
In the IBM arc-backarc system, a number of wide-angle seismic surveys were acquired to explore not only arc growth and formation of continental crust, but also rifting, breakup, and backarc seafloor spreading (Grevemeyer et al., 2021;Suyehiro et al., 1996;Takahashi et al., 1998Takahashi et al., , 2007Takahashi et al., , 2008Takahashi et al., , 2009)).Similar to passive continental margins, breakup and seafloor spreading in backarc regions are related to extensional stresses and are usually characterized by high P-wave velocities in the lower crust of the arc-backarc transition zone (e.g., Grevemeyer et al., 2021;Takahashi et al., 2008).A high-velocity lower crust at volcanic continental margins is generally related to Mg-rich mafic/ultramafic material formed by extensive partial melting of mantle peridotites at unusual high mantle temperatures (e.g., White & McKenzie, 1989).However, the structural diversity of backarc basins, including high-velocity lower crust, may differ from normal mid-ocean ridge type melting, which was suggested to be controlled by entrainment of hydrous melts (Arai & Dunn, 2014;Dunn & Martinez, 2011), or cold and depleted mantle invading the backarc while a subduction zone reconfigures (Grevemeyer et al., 2021).The interplay between arc magmatism and backarc seafloor spreading magmatism is another principal focus of our study.
In this study, we present the tomographic result of a long seismic refraction and wide-angle transect across the Izu arc-backarc system.Structural constraints from our seismic results indicate the nature of crustal domains across the Izu arc and neighboring regions.Along with geochemistry and petrological evidence from previous drilling sites by the Ocean Drilling Program (ODP) and International Ocean Discovery Program (IODP) expeditions and dredge samples, we analyzed the correlation between the seismic structure and different types of arc magmatism, elucidating the arc growth process and the transformation from arc crust to continental crust.In addition, we discussed the impact of hydrous melting or heterogeneity of mantle source on the structure of the Shikoku Basin by establishing the mantle melting model and comparing different backarc basins.

Geological Setting
The IBM arc is an intraoceanic volcanic arc located near the eastern boundary of the Philippine Sea plate (Figure 1).Since ∼50 Ma the Pacific plate subducted beneath the Philippine Sea plate and hence resulted in the 10.1029/2023JB027213 3 of 20 formation of the IBM arc, which extends over 2,800 km from Sagami Bay (Japan) in the north to Guam in the South, and its backarc basins including the Shikoku Basin and the Parece Vela Basin (e.g., Bloomer et al., 1995;Stern et al., 2003;Taylor, 1992).Based on tectonic characteristics, the IBM arc can be further divided into the Izu arc in the north, the Bonin arc and Mariana arc in the south.The Izu arc is bounded by the collision zone with southern Honshu in the north and the Sofugan tectonic line in the south (Figure 1).
The controlling factor that triggered subduction initiation might be a change in the direction of plate motion of the Pacific plate, which changes from NW to NWW at ∼50 Ma (e.g., Ishizuka et al., 2011;Reagan et al., 2013).In the initial stage of subduction in the Eocene, the subducting plate sunk into the mantle, leading to the onset of seafloor spreading, forming forearc basalts.
Thereafter, hydrous fluids from the subducting slab aided the formation of boninites (Christeson et al., 2016;Ishizuka et al., 2006;Reagan et al., 2010).With ongoing subduction, the volcanism in the IBM arc gradually transferred to andesitic, basaltic, or rhyolitic.Until ∼27 Ma, the volcanism was significantly diminished and was feeblest between ∼23 and ∼17 Ma according to the IODP drilling result (Taylor, 1992).At the same time, seafloor spreading of the Shikoku Basin in the direction of NW-SE to E-W started from the north at ∼26 Ma and propagated southward (Okino et al., 1994).In the South, the Parece Vela Basin opened at ∼26 Ma and connected with the Shikoku Basin at ∼23 Ma (Okino et al., 1998).As a result, the conjugated Kyushu-Palau Ridge, which is a remnant arc, separated from the IBM arc (Figure 1).Backarc spreading of the Shikoku Basin and Parece Vela Basin ceased at ∼15 Ma, while arc volcanism was reactivated in the Izu-Bonin regions (Okino et al., 1998;Taylor, 1992).Initially, basaltic to rhyolitic magmatism occurred in the rear arc at ∼17 Ma.Volcanic activity migrated in north-northeast direction until ∼3 Ma, forming en echelon ridges (Ishizuka, Uto, & Yuasa, 2003).Farther east, small knolls were formed by the eruption of basalt and minor felsic rocks between 2.8 and 1 Ma (Ishizuka et al., 1998;Ishizuka, Uto, & Yuasa, 2003).The most recent back rifting occurred right behind the volcanic front at ∼1.5 Ma, forming a narrow backarc rift belt in a zigzag pattern (Taylor, 1992).Accompanied by rifting, bimodal magmatism was active in this rift belt.The overall eastward migration of Neogene and Quaternary magmatism is perhaps controlled by the rapid retreat of the Philippine Sea plate and steepening of the subducting Pacific plate (Ishizuka, Uto, & Yuasa, 2003).
The Bonin arc in the south suffered from vigorous rifting in the Oligocene, forming the Ogasawara trough that separates the Bonin ridge and the Bonin arc (Figure 1).In the Mariana region, a new backarc basin-the Mariana Trough-started rifting at ∼10 Ma, and the onset of seafloor spreading occurred at 3-4 Ma (Yamazaki & Stern, 1997).As a result, the Mariana arc is split into a remnant arc-the West Mariana Ridge, and its current active arc front.
At present, along the volcanic front of the Izu arc, a prominent feature is that basaltic-dominated volcanism usually forms relatively high bathymetry features, such as conical volcanoes and islands, while the rhyolitic-dominated volcanism is mostly located in depressions and calderas (e.g., Kodaira et al., 2007b;Tamura et al., 2009).Kodaira et al. (2007aKodaira et al. ( , 2007b) ) discussed the relationship between the crustal structure and the type of volcanism.They proposed that the bulk of juvenile continental crust is predominantly generated below the basaltic volcanoes, while the rhyolitic volcanism might relate to an early stage of arc evolution or the remelting of the intermediate composition middle crust.

Seismic Data Collection and Processing Procedure
From May to June 2008, the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) acquired a 585-km-long wide-angle seismic refraction line IBr5 perpendicular to the Izu arc, spanning from the serpen- tine forearc seamount ridge to the middle of the Shikoku Basin (Figure 1).A total of 110 Ocean-Bottom-Seismometers (OBSs) were deployed at approximately 5 km intervals along the profile.All the instruments were successfully recovered and recorded data to be used in geophysical inversion (Supporting Information S1).An array of eight BOLT air-guns with a total volume of 12,000 cubic inches fired 2,926 shots.The air-gun shooting was carried out at 200 m spacing with a ship's speed of ∼3.5 to 4.5 knots.The raw OBS data were corrected for clock drift and data were cut into seismic segy-format.Butterworth band-pass filter and gain processing using balancing trace by mean were applied before displaying seismic record sections.Figure 2 shows some examples of the seismic record sections, exhibiting the excellent quality of the data set.First-arrival seismic refraction phases with apparent velocity of <7.0 and ∼8.0 km/s were defined as crustal refraction phases Pg and mantle refraction phases Pn, respectively.Pg were usually observed to offset of ∼30 km at instruments in the Shikoku Basin and offset of ∼100 km at instruments in the Izu arc.Prominent seismic wide-angle reflections, representing the crust-mantle boundary or Moho reflection PmP, were observed extensively in our data set.In addition, at far-offsets, Pn occurs in most of the OBS data, and secondary deep mantle reflection phases PdP were observed in the OBS data located on the Izu arc and the forearc region (e.g., Figures 2e-2j).Generally, the intersection of PmP and refraction phases is located between Pg and Pn, while the PdP appears later than PmP and the intersection with refraction phases is located after Pn emerges.

Seismic Tomography
Seismic tomographic inversion software Tomo2D (Korenaga et al., 2000) was used to obtain the P-wave velocity structure of the Izu arc and the Shikoku Basin along line IBr5, conducting a joint travel time inversion of first arrivals and onset times of secondary wide-angle reflections.This method employs a hybrid ray-tracing scheme combining the graph method with further refinements utilizing ray bending with the conjugate gradient method.To better constrain the sedimentary layer of the initial model, the geometry and depth information of the sedimentary basement were extracted from the near congruent multi-channel seismic (MCS) line IBr5 (Figure 1).As the MCS and wide-angle profiles were not exactly coincident, we use forward raytracing RayInvr software (Zelt & Smith, 1992) to further refine the sedimentary velocity and basement depth by modeling the delay of near-offset Pg arrivals.Consequently, the initial model for the tomographic inversion contains a reasonably precise structure of the sedimentary blanket, minimizing bias introduced to the crustal and upper mantle inversion for seismic velocity structure of the igneous base rocks.
In total, we picked 49,675 Pg arrivals, 13,693 PmP arrivals, 10,721 Pn arrivals, and 2,942 PdP arrivals for the tomographic inversion (Supporting Information S1 and Figure S1 in Supporting Information S2).The picking errors were set to 0.05-0.07,0.07, 0.08, and 0.08 s for the Pg, PmP, Pn, and PdP arrivals, respectively (Table 1).In a layer-stripping strategy, the seismic tomographic inversion includes seven steps that yield the velocity structure from top to bottom of the model (Figure S2 in Supporting Information S2).The fitting parameters root-mean-square (RMS) value and χ 2 of each step are shown in Table 1, indicating good travel-time fit of the inverted models.Figure 3 shows the final P-wave velocity structure, which is constrained by dense ray coverage benefitted from the dense OBS deployment.The travel-time residuals of each OBS are summarized in Figure 4, showing good fitting except for ray-paths from the edge of the model and some far-offset arrivals in a

Uncertainty Tests
To evaluate the validity of the velocity model, a nonlinear Monte-Carlo error analysis was used to derive model uncertainties (e.g., Korenaga et al., 2000).Random noise of ±40 ms was added to the travel-time data set to verify the stability.One-hundred 1D randomly perturbed initial models were established for the Monte-Carlo tests, in which the velocity perturbation is ±8% at the top of the crust, changing to ±2% at the bottom of the model.The depth of initial Moho reflector was perturbed ranging from 11 to 19 km.The corresponding 100 reference models were used to estimate the uncertainties of the inversions by the mean standard deviation of all solutions (Tarantola, 1987).Figure 3e shows the mean standard deviation of the crustal velocity model.The smallest standard deviation of <0.05 km/s occurs in the upper to middle crust of the Izu arc region from basement top to ∼12 km depth, the middle to lower crust of the forearc area, and the lower crust beneath the seamounts in the Shikoku Basin.In the other areas of our model, the velocity uncertainties are mostly ranging from ±0.05 to ±0.1 km/s.The Moho depth variation is approximately stable in the Shikoku Basin (model distance 30-260 km) and in the forearc area (model distance 430-510 km), ranging from 0 to ±0.6 km.In the model distance 320-430 km, Moho uncertainties are between ±0.6 and ±1.0 km.The poorest stability of the Moho reflector, with maximum uncertainties of ±2 km, is beneath the rear arc and Enpo-Manji Basin (model distance 270-320 km).In summary, the velocity and Moho depth standard deviations are small, showing well-controlled tomographic results owing to the dense receiver spacing along line IBr5.
Similarly, the Monte-Carlo analysis was applied to test the uncertainty of the velocity structure in the mantle.The crustal velocity was taken from the final result of Pg + PmP inversion.One-hundred random 1D initial velocity models were added beneath the Moho, ranging from 7.6 to 8.2 km/s at the top of the mantle to 8.2-8.8 km/s at the bottom of the model.The results show that the best control of the mantle velocity is located beneath the forearc at model distance of >450 km, which only has standard deviation values of <0.1 km/s in ∼7 to ∼10 km depth below Moho (Figure 3f).To the arc and backarc region (model distance of <450 km), the uncertainty gradually increases from ±0.15 km/s within ∼2 km below Moho to ±0.2 km/s in the deeper mantle.

Resolution Tests
Resolution of the velocity model is critical in the research of volcanic arc and backarc areas since the robustness in the discrimination of different crustal layers and variation between different tectonic domains hugely impact the geological interpretation.In this study, we applied checkerboard tests using synthetic data in perturbed cells in width and height of 15 × 4 km and 10 × 4 km, with velocity perturbation of ±5% (Figures 5a and 5b).The synthetic travel-times were perturbed by random noise of ±40 ms during the forward calculation.The velocity perturbations were recovered remarkably well in the Shikoku Basin (Figures 5c and 5d).The upper and middle crust of the arc to forearc area shows a satisfactory recovery, indicating that the detailed velocity structure is well-resolved, even at small scales.In the lower crust, some perturbation of 15 × 4 km can still be recovered from the volcanic front to the forearc basin (model distance 370-500 km), while smaller cells can hardly be recovered in these areas.The resolution tests indicate that our velocity model generally has good robustness and the detailed analysis on the bottom of the crust beneath the arc and forearc areas should be avoided.

Model Description
The velocity structure along line IBr5 shows strong lateral variations across the arc.It depicts a thick crust beneath the Izu arc in the middle of the model, thinning to both sides of the arc in the backarc and forearc regions.The thickest crust occurs beneath the backarc knolls zone and the Myojin rift, reaching ∼20 km (Figure 3).From the rear arc to the current volcanic front, the crustal thickness ranges from ∼15 to ∼20 km.The crustal thickness gradually thins approaching the Shikoku Basin and the forearc region to ∼7 to ∼10.8 km and ∼5 to ∼16.5 km, respectively.
In the Shikoku Basin (model distance 0-260 km), the sedimentary layer gradually thickens from the extinct mid-ocean spreading center to the Izu arc, showing larger depression and higher sediment accumulation on the older oceanic crust adjacent to the arc (Figure 3).At the top of basement, velocities range from ∼4.0 to ∼5.5 km/s, probably reflecting variations of the porosity and crack density.Distinct velocity gradients characterize the upper and lower crust of the Shikoku Basin.From top to bottom, velocities in the crust increase sharply with a velocity gradient of ∼0.46 s −1 , reaching ∼6.7 km/s at ∼3 km depth, while the lower crust has a small velocity gradient of ∼0.07 s −1 and increasing to ∼6.9 to ∼7.2 km/s at the bottom of the crust (Figure 6b).Two submarine volcanoes occurred in the model at 140-190 km, showing crustal thicknesses of 9-10.8 km.The Moho beneath the volcanoes is slightly deeper than that in the adjacent normal oceanic crust, reflecting flexural deformation of the oceanic plate due to the load of the seamounts.Further, low mantle velocities of ∼7.0 to ∼7.8 km/s occur beneath these volcanoes.
Approaching the arc-backarc transition zone (model distance 240-280 km), a prominent ∼5-km-thick high-velocity anomaly (7.0-7.5 km/s) characterizes the lower crust (Figure 3).A similar high-velocity anomaly was found in the same geological setting further south in the Bonin arc region and was interpreted as magmatic intrusions in the thin rifted crust (Takahashi et al., 2009).
In the Izu arc region (model distance 260-410 km), an extremely thin sedimentary blanket covers the igneous basement, calderas, and small knolls, while a thicker sedimentary layer of up to 2 km exists in the backarc rifts (e.g., Enpo-Manji Basin, Myojin rift) (Figure 3).The crustal velocity increased from ∼4.1 to ∼4.8 km/s at the basement top to ∼6.5 km/s at ∼10 km depth.In the lower crust, the velocity gradient becomes smaller and the velocities range from ∼6.5 to ∼7.0 km/s (Figure 6a).The velocities in the lower crust beneath the Enpo-Manji Basin are higher than in the adjacent areas, which may relate to the rifting process.Two prominent mantle velocity anomalies, revealing low velocities of 7.0-7.5 km/s, were observed in model distances 280-310 km and 330-410 km, and showing a complementary relationship with high-velocity anomalies in the lower crust.In addition, a deep mantle reflector was imaged at ∼27 to ∼30 km depth beneath the Izu arc, across the iso-velocity contour from 7.5 to 8.2 km/s.
Farther east in the forearc area (model distance 410-585 km), the sedimentary layer is rather thick in the forearc basin, reaching ∼2 km.From the outer high to the trench, the sedimentary layer gradually thinned and pinched out at the trench axis (Figure 3).The crustal velocities are between ∼4.0 and ∼5.0 km/s at the basement top, increasing to 6 km/s at ∼7 km depth.At greater depth, the 6.5 km/s iso-velocity contour undulates profoundly throughout the entire forearc.Furthermore, velocities also vary strongly from ∼6.5 to ∼7.2 km/s at the bottom of the crust.From model distance 450 to 580 km, low velocities anomalies of <7.5 km/s also occur below the Moho.The deep mantle reflector remains nearly flat beneath the forearc basin, while it changes to sharp inclined at model distance of 520 km and becomes shallower toward the trench.

Crustal and Upper Mantle Layer Divisions and Their Interpretation
From the basement to Moho, the crust is generally divided into different layers, such as upper, middle, and lower crust.In the early period of island arc research, forward modeling results subdivided upper crust (4.5-6.0 km/s), middle crust (6.1-6.5 km/s), lower crust (6.7-7.3 km/s) and they were interpreted as basaltic layer, tonalitic (or granitic) layer, and gabbroic layer, respectively (e.g., Takahashi et al., 1998Takahashi et al., , 2007Takahashi et al., , 2009)).However, in along-strike model, seismic tomographic first arrival travel-time inversion, supplemented with float reflectors imaging obtained from a diffraction stack migration approach (e.g., Fujie et al., 2006) gives a subdivision of upper crust (<5.8 km/s), upper part of the middle crust (6.0-6.5 km/s), lower part of the middle crust (6.5-6.8 km/s), upper part of the lower crust (6.8-7.2 km/s), lower part of the lower crust (7.2-7.6 km/s) (e.g., Kodaira et al., 2007aKodaira et al., , 2008Kodaira et al., , 2010)).The main difference between the two divisions above is the definition of the Moho position, either on ∼7.0 or ∼7.6 km/s.In our data set IBr5, only one strong seismic reflector was observed and imaged continuously along the entire model thus we interpreted it as a reflection from the seismic Moho  (Figure 3).A crucial layer occurring in an arc and forearc setting is a middle crustal layer of 6.0-6.5 km/s, which generally was interpreted as an intermediate tonalitic layer.Such a layer will be nearly absent in the backarc basin, such as the Shikoku Basin, as it indicates a continental rock composition.Therefore, to account for the strong variability of our transect crossing different geological domains, we use a simplified classification, defining the upper crust from the top of the basement to the 6.0 km/s isoline, the middle crust as 6.0-6.5 km/s, and the lower crust from the 6.5 km/s isoline to Moho for the arc-forearc region.
The upper crust is 3.0-5.5 km thick in the Izu arc area, varying strongly to 1.0-6.0km thick in the forearc (Figure 3).Based on petrologic data (Christensen & Mooney, 1995) and rock samples collected from the arc-forearc region (Ishizuka, Uto, & Yuasa, 2003;Tamura & Tatsumi, 2002), the major composition of the upper crust was inferred as basaltic.Huge lateral velocity variations exist in the upper crust.Beneath the rift center, the basement top usually has relatively low velocities of <5 km/s.
In our model, the thickness of the middle crust is quite variable in the arc-forearc region, ranging from ∼1 to ∼6 km.This layer has long been considered as a felsic to intermediate composition layer (e.g., Kodaira et al., 2007a;Takahashi et al., 2007), mimicking the average composition of continental crust.The exposure of a plutonic complex in the Tanzawa mountains, which formed in the collision of the northern Izu arc with Honshu Island in the middle Miocene (Soh et al., 1991), consists of predominantly tonalite-quartz gabbro intrusions with subordinate hornblende gabbroic intrusions (Kitamura et al., 2003).The velocity measurement and estimation under different temperatures of these rock samples have shown that the middle crust likely consists mainly of tonalite containing high SiO 2 (>64 wt% SiO 2 ) (Kitamura et al., 2003;Kodaira et al., 2007a).
The lower crust is thickest beneath the backarc rift (Myojin rift) and volcanic front, thinning toward the rear arc and forearc.The layer with 6.5-6.8 km/s is generally interpreted as plutonic intermediate rocks (54-63 wt% SiO 2 ) consisting mainly of gabbro and tonalitic intrusions (Kitamura et al., 2003;Kodaira et al., 2007a).The high velocities (>7.0 km/s) are mainly located beneath the arc-backarc transition zone and forearc region.Laboratory experiments on rock sample (Kitamura et al., 2003) and global compilations of seismic velocity data (Christensen & Mooney, 1995) suggest that these relatively high velocities may represent more mafic and hence gabbroic plutons.et al., 1999), Lesser Antilles arc (Christeson et al., 2008), Alaska Peninsula (Lizarralde et al., 2002), and Mariana arc (Calvert et al., 2008) are extracted from the volcanic front.The velocity structure of the Bonin arc is extracted at the volcanic front beneath the Suiyo seamount (Kodaira et al., 2007b).The velocity structure of the Tonga arc is extracted from the remnant arc (Crawford et al., 2003).The red error bars show the velocity range of average continental crust (Christensen & Mooney, 1995), while the blue and red envelopes denote the velocity range of the Pacific oceanic crust and magmatic oceanic crust, respectively, which was compiled by Grevemeyer et al. (2018).The gray envelope represents the velocity range of the Izu arc in the model distance of 280-410 km of our model.
Beneath the Moho, extremely low uppermost mantle velocity anomalies (<7.5 km/s) are observed dispersedly (Figure 3), which significantly deviate from normal upper mantle velocities in the range of 8.0-8.2 km/s.These low velocity anomalies are unlikely to be related to mantle anisotropy because either slab-induced mantle flow or olivine alignment caused by backarc seafloor spreading would result in fast mantle speed along the subduction and backarc opening direction (e.g., Kodaira et al., 2014;Stern et al., 2003), which is NW-SE to E-W in the Izu arc-backarc region.Petrologic modeling has shown that these anomalies are mafic restites consisting of plagioclase, garnet, and clinopyroxene, which are generated during the crustal anatexis and differentiation processes (Tatsumi et al., 2008).Crustal delamination or foundering is needed for transfering the restite and cumulates into the mantle to form an overall felsic to intermediate continental crust.Sato et al. (2009) verified that these anomalies are a mixture of mafic residues and olivine cumulates using amplitude modeling.The deep mantle reflector at ∼27 km depth may represent the interface between the restite/cumulate and pure mantle harzburgite layer, which results from the increasing of garnet and decreasing of orthopyroxene and plagioclase (Tatsumi et al., 2008).

The Interpretation of Seismic Velocity Structure
The seismic profile cuts through different geological domains and to appreciate the structural variability we used seismic characteristics to subdivide our model into six domains.For convenience in depicting the geological features of the arc-backarc system, the discussion will be rolled out from the forearc to the backarc.

Serpentinite Seamount Ridge and Inner Trench Wall
At the eastward limit of the seismic model, previous studies (Fryer et al., 1985;Ishii, 1992) identified an along-trench serpentinite seamount ridge to the west of the trench (Figure 1), which resulted from mantle diapirs formed by the release of water from the subducting oceanic slab (Fryer, 1992;Kamimura et al., 2002).Our model shows low crustal velocities beneath this ridge, but the ray-coverage is rather poor (Figure 3).To the north of our line at 32°15′N, Suyehiro et al. (1996) and Takahashi et al. (1998) obtained extremely low velocities of <5 km/s beneath the forearc serpentinite diapir.To the south at ∼31°N, Kamimura et al. (2002) also found low velocities in the vicinity of the serpentinite diapirs.In addition, they also revealed low mantle velocities overlying the subducting plate, most likely illustrating the role of water and mantle hydration in the mantle wedge, issuing serpentinite diapirism.Similarly, our model reveals low mantle velocity (7.0-7.6 km/s) below the forearc and volcanic arc (Figure 3b).The deep mantle reflector shows an inclined boundary in 17°, which is close to the estimation of 22° for the subducting Pacific plate (Jarrard, 1986).Thus, the deep mantle reflector from model distance ∼520 to 570 km might be related to the top of the subducting slab.In this area, the low mantle velocity zone above the subducting slab probably shows the effect of serpentinization produced by the release of water during the dehydration of subducting slab.Kamimura et al. (2002) interpreted it as chrysotile, which may act as a lubricant to reduce the seismic activity along the subducting zone.
At the inner trench wall (model distance 550-572 km), the crust gradually thickens westward from ∼6 to ∼9 km (Figure 3b).This domain beneath the inner trench wall represents the suprasubduction zone crust that formed by decompressional melting soon after subduction initiation (Reagan et al., 2010).IODP Expedition 352 collected samples from the basement located at the northeastern edge of the Bonin Ridge (Figure 7c) which consists of forearc basalt and boninite lavas and dikes (Christeson et al., 2016).The corresponding P-wave sonic log velocities of the top basement are only ∼3.1 to ∼3.4 km/s, which is significantly lower than in normal oceanic crust (Christeson et al., 2016).Low crustal velocities in our model are consistent with the observations from IODP Expedition 352, indicating the extensive fracturing and alteration of the Izu forearc crust beneath the inner trench wall compared with normal oceanic crust.

Forearc Outer High
In the forearc outer high (model distance 500-550 km), crustal thickness ranges from ∼9 to ∼12 km and is associated with slightly higher velocities of ∼7.0 to ∼7.2 km/s at the bottom of the crust (Figure 3b).The 1D velocity-depth structure shows that this domain is close to the lower bound of the envelope of typical oceanic crust, while the thickness is larger than a global average of ∼6 km (Figure 6a).Site 786 conducted by ODP Leg 125 located at ∼12 km north of line IBr5 has revealed the boninitic edifice in the Eocene (Figure 7a; Pearce et al., 1992).The boninitic volcanism represents the initial forearc spreading process that occurred after the sinking of the Pacific plate.Fluids released from the subducting slab entered the upwelling asthenosphere mantle to form boninitic magmas (Ishizuka et al., 2006).The oceanic-like structure may relate to the similar accretion process of seafloor spreading, while the high velocities at the bottom of the crust and thicker crust perhaps represent the amphibole-rich pluton formed by hydrous magma (Tatsumi et al., 2008).

Forearc Basin
The Izu forearc basin occurred at model distance of 410-500 km, which is characterized by a thick sedimentary cover.The crustal thickness increases westward from ∼11 to ∼16.5 km.This domain can be divided into two sub-domains based on their distinct velocity structure (Figure 3b).The eastern part (model distance 460-500 km) has a ∼5.5 km thick middle crust and velocity of only ∼6.8 km/s at the base of the crust, while the western part (model distance 410-460 km) has a thin middle crust (∼2 km) and with high-velocity anomalies ∼7.0 to 7.2 km/s in the lower crust.
Although no drilling or dredge samples were collected in the forearc basin along our seismic line, the adjacent Site 792 and 793 from ODP Leg 126 provide important evidence for revealing the nature of the crust in the forearc basin basement since the tectonic features from the subduction zone to island arc are basically a function of distance from the trench axis.Site 793 is located in the eastern part of the forearc basin, in which the basement consists of volcanic breccias, pillow flows, and massive flows of dominantly high-Mg two-pyroxene basaltic andesites and andesites (Figure 7a; Taylor et al., 1990).The velocity structure of this domain (x = 480 km in Figure 6a) is close to the arc crust formed by basaltic volcanism (x = 357 km in Figure 6a), which may represent a transition from forearc volcanism to arc volcanism.
In contrast, in the western part of the forearc basin, the basement at Site 792 is composed of plagioclase-rich, two-pyroxene andesites and dacites with calc-alkaline affinities (Taylor, 1992).Compare with Site 793, the silicic content in the Site 792 basement samples is higher, which is petrologically between the volcanic front (rhyolitic) and the eastern forearc basin (andesitic) (Figure 7).Seismically, the thinner middle crust and higher velocities at the bottom of the crust indicate that the crust of this domain has higher mafic content (Figures 3b and 7b), similar to the velocity structure of the volcanic front (x = 400 km in Figure 6a).Thus, we suggest that this domain was significantly affected by rhyolitic volcanism, forming different structures between the eastern and western parts of the forearc basin.

Volcanic Front and Myojin Rift
Seismic line IBr5 crosses the volcanic front at model distance of ∼400 km, south of Myojin-sho, which is a large dacite-dominant submarine volcano (Figures 3a and 7; Tamura et al., 2009).The iso-velocity contours show an arched-up pattern beneath the volcanic front.A small backarc rift called Myojin rift is located right on the west of the volcanic front (model distance 362-395 km), which was covered by ∼1 to ∼2 km of thick sediments.
The crust gradually thickens from ∼17 to ∼20 km toward the west and reaches its maximum in the Myojin rift (Figure 3b).However, the middle crust is rather thin with thicknesses of only ∼1.5 to ∼3 km (Figure 7b).The velocities at the bottom of the crust are between ∼6.8 and ∼7.0 km/s, while low mantle velocities of ∼7.0 to ∼7.5 km/s occur beneath the Moho.
Wide-angle seismic data acquired along the volcanic front demonstrates a relationship between the type of volcanism and the thickness of the middle crust (Kodaira et al., 2007a).A thick middle crust was usually observed beneath basaltic volcanoes while a thin middle crust usually occurred beneath rhyolitic volcanoes.Dacitic magma is also a felsic magma, close to but contains less silicon than rhyolitic magma.The results in the dacite-dominated domain agree with previous observations that the middle crust is thin beneath the rhyolitic-dacitic volcano (Kodaira et al., 2007a(Kodaira et al., , 2007b;;Tamura et al., 2009).The 1D velocity-depth profile of the volcanic front in our line (x = 400 km) is between the average continental crust and oceanic crust, deviating much more from the continental crust than the backarc region (x = 300, 357 km in Figure 6a).
To the west, a series of backarc rift basins occur parallel to the volcanic front and initiated from 2.9 to 2.35 Ma, in which the age of volcanism in the Myojin rift is <1.0 Ma (Taylor, 1992).Site MWD40-5 obtained dacite-rhyolite with an age of 0.38 ± 0.16 Ma (Figure 7a; Ishizuka, Uto, & Yuasa, 2003), which is consistent with the adjacent volcanic front in lithology.Coincidently, the middle crustal layer is also thin beneath the Myojin rift (Figures 3b  and 7b), which supports the observations beneath the volcanic front.

Backarc Knolls Zone to Rear Arc
Farther west, seismic line IBr5 crosses the backarc knolls zone, the Enpo-Manji Basin, and the rear arc from model distance 362 to 260 km (Figure 3).The crustal thickness decreases westward from ∼20 to ∼10 km, while the middle crust thickness is ∼3 to ∼6 km.Lower-crustal velocities are ≤7.0 km/s, while up to ∼6.5 km thick low mantle velocities of 7.0-7.5 km/s occur heterogeneously below the Moho.This area was overprinted by a series of NE-SW-trending en echelon ridges (Figure 1).In our model, the seismic structure from the backarc knolls zone to the rear arc is rather consistent and is close to the velocity structure of average continental crust (Figure 6a).Dredging samples are mainly basaltic andesite and andesite on the backarc seamount chain and mainly basalt on the backarc knolls (Figure 7a; Ishizuka et al., 1998;Ishizuka, Taylar, et al., 2003;Ishizuka, Uto, & Yuasa, 2003).
The active volcanism migrated and converged eastward with a rock age of ∼17 Ma in the rear arc, and changing to ∼3 Ma in the backarc knolls zone.Basaltic volcanism in this domain resulted in a rather thick intermediate composition middle crust, which is consistent with features found along the volcanic front (Kodaira et al., 2007a), hinting that the crustal layer of the backarc region was not strongly affected by rifting.This domain is not only similar to continental crust in velocity structure, but also comparable to the composition of continental crust in both SiO 2 /K 2 O content and light rear-earth elements (Tamura et al., 2015).

Arc-Backarc Transition Zone to Shikoku Basin
The crustal thickness decreases sharply from the volcanic arc to the backarc region, and a high-velocity anomaly zone (V p : 7.0-7.5 km/s) occurs in the lower crust beneath the arc-backarc transition zone (Figure 3b).This high-velocity anomaly is larger and faster than the high velocities beneath the rifts on the Izu arc, probably indicating more abundant magmatic intrusions during the break-up of the IBM arc and the Kyushu-Palau ridge.Similar findings were also observed in the arc-backarc transition of the Bonin and Mariana regions (Takahashi et al., 2008(Takahashi et al., , 2009)).Since only one reflector was imaged at the bottom of this anomaly, we interpret it as magmatic intrusion in the crust.This type of magmatic activity is interpreted in terms of adiabatic melting of mantle lithosphere occurring during break-up, showing a strong resemblance to rifted volcanic or intermediate continental margins (e.g., Li et al., 2021;White et al., 2008).Yet, whether hydrous melting entrained from the subducting slab was captured in the break-up event is still unclear.Although geochemistry data reveals that some seamounts in the Eastern Shikoku Basin and the backarc seamount chain have a significant contribution from the hydrous melting and sediment melting, the age of magmatism was determined in the post-spreading stage (<16 Ma; Ishizuka et al., 2009; Figure 7).To the west, the seismic structure of the Shikoku Basin shows a typical two-layered structure of oceanic crust, consisting of a high velocity gradient in the basaltic layer (oceanic layer 2) and a low velocity gradient in the gabbroic layer (oceanic layer 3) (Figure 6b).Detailed discussion about the interplay between arc volcanism and backarc magmatic accretion refers to Section 7.

Formation and Evolution of Tonalitic Layer
The middle crust in the vicinity of the arc is interpreted in terms of a felsic to intermediate tonalitic layer (∼6.0 to ∼6.5 km/s) characterizing volcanic arcs and hence is similar to continental crust both in bulk composition and velocities.Seismic velocity structure along the volcanic front of the Izu-Bonin arc indicates that this layer plays a key role in the growth of the arc crust and the generation of continental crust (Kodaira et al., 2007a(Kodaira et al., , 2007b)).Remarkably, the tonalitic layer was observed predominately beneath the basaltic volcanoes, and thus rhyolitic volcanism was seen as an indicator of a juvenile stage of crustal evolution, or remelting of preexisting continental crust, or both (Kodaira et al., 2007a).In this study, we found that the correlation between the thickness of the tonalitic layer and basaltic/rhyolitic volcanism can also be observed spanning from the forearc to the rear arc, not only at the active volcanic front.We extracted the thickness of the middle crust and calculated the ratio between the lower and middle crust (T l /T m ; Figure 7b).In the Shikoku Basin and forearc outer high formed by seafloor spreading, the middle crust is nearly absent and the T l /T m is larger than four.In contrast, from the backarc knolls zone to the rear arc that is dominated by basaltic volcanism (e.g., Ishizuka et al., 1998;Ishizuka, Uto, & Yuasa, 2003), the middle crust is mostly thicker than 3 km, and the T l /T m ranges between ∼1.5 and ∼3.5, which is closer to the average value of continental crust.In the rhyolite-dominated area spanning from the western forearc basin to the Myojin rift, the middle crust is instead thinner (∼1.5 to ∼3 km) and the T l /T m (∼3 to ∼8.5) deviates from the range of continental crust, in agreement with the findings in the volcanic front (Kodaira et al., 2007a(Kodaira et al., , 2007b)).
The abrupt variation of middle crustal thickness and T l /T m between the oceanic domain, basalt-, and rhyolite-dominated arc domains indicate the role of different types of volcanism in shaping the crustal structure.The primary magma generated in the mantle wedge is considered to be of basaltic composition (e.g., Tatsumi & Eggins, 1995).Petrological modeling has demonstrated that only the basalt model for the arc magma generation can obtain results consistent with the observed velocities, while the andesite model does not (Tatsumi et al., 2008).The compositional difference between the primary magma and the continental crust leads to a paradox in the arc growth process.Thus, the generation of the felsic-intermediate layer in the arc growth processes may require remelting, anatexis, differentiation, and delamination/foundering of mafic material.The initial arc crust may form on the pre-existing oceanic crust emplaced by incipient basaltic arc magmatism, creating mainly basaltic bulk composition.Successive upwelling and underplating of basaltic magma heat the initial crust and lead to partial melting and crustal anatexis, creating partial melts and relative mafic restite.Felsic to intermediate partial melts of low density migrate upward and are aggregated due to larger buoyancy, forming the tonalitic middle crust, whereas mafic restite is left at the bottom of the crust.
In our model, the basalt-dominated domain was governed by an eastward migrated volcanism from ∼17 to ∼3 Ma, while the volcanism in the rhyolite-dominated domain occurred after ∼2.8 Ma (Figure 7a; Ishizuka et al., 1998;Ishizuka, Uto, & Yuasa, 2003).Basaltic volcanism originates from the primary magma generated from the mantle wedge, forming a thick tonalitic layer either at the volcanic front (Kodaira et al., 2007a(Kodaira et al., , 2007b) ) or in the backarc region.However, rhyolitic volcanism in contrast is considered to occur as a result of partial remelting of the initial basaltic crust.Considering the young age of volcanic rocks collected in the rhyolite-dominated domain, a juvenile stage of arc evolution might be responsible for rhyolitic volcanism (Tatsumi & Kogiso, 2003).Another interpretation is that the insufficient supply of basaltic primary magma did not support the growth of the middle crust, but led to the middle crust remelting by the heating of lateral influx of hot basaltic magma at adjacent mantle diapir (Tamura et al., 2009;Tamura & Tatsumi, 2002), leaving a thinner middle crust and thicker lower crust (Figure 7).
Compared with bulk continental crust, the arc composition from our model indicates a more mafic composition (Figure 6a) owing to larger T l /T m .Therefore, further accretion of rising magma is needed for the growth and maturation of the arc crust.The arc crust spanning from the rear arc to the backarc knolls zone (model distance 260-362 km) shows a closer relationship to continental crust than the backarc rift and volcanic front (model distance 362-460 km) when considering its velocity structure (Figure 6a).The velocity structure of global volcanic arcs is comparable with the Izu arc (Figure 6c), exhibiting a more mafic composition compared with continental crust.The Bonin arc is similar to the volcanic front of our model in the top 10 km below the sedimentary basement (Kodaira et al., 2007b), which may present the juvenile stage of arc evolution.In comparison, the Tonga arc (Contreras-Reyes et al., 2011;Crawford et al., 2003), Mariana arc (Calvert et al., 2008), and the Lesser Antilles arc (Christeson et al., 2008) present thicker middle crust, showing a similar structure to the rear arc to volcanic front area of the Izu-Bonin arc and mimicking a middle stage of arc evolution.The velocity structure of the western Alaska Peninsula is close to the upper bound of continental crust and hence represents the type of most mature arc (Lizarralde et al., 2002).However, for the Aleutian arc, a tonalitic layer with velocities between ∼6.0 and ∼6.5 km/s is absent, illustrating the most immature arc crust (e.g., Holbrook et al., 1999;Shillington et al., 2004).

Delamination of Mafic Restites and Cumulates
Low upper-mantle velocities (∼7.0 to ∼7.5 km/s) are observed broadly beneath the Izu arc in our model, which were interpreted as residues consisting of mafic restites and cumulates after crustal differentiation (Figure 3b).
A large amount of this layer is coincidentally located beneath the area that has a thicker middle crust except for the Enpo-Manji Basin which was affected by the backarc rifting.Therefore, it is reasonable to infer that this anti-continent layer was a part of the crust in the juvenile stage of arc evolution.To achieve the delamination or foundering processes, a permeable Moho is required to allow mass transfer (Kodaira et al., 2007a;Tatsumi et al., 2008).In the petrological calculation by mixing a rhyolite and a basalt melt containing 75 and 50 wt% SiO 2 , respectively, the thickness of the restite (T R ) was expressed as a function of the thickness of the middle crust (T M ): T R = 2/5 × T M × 9 (Tatsumi et al., 2015).Accordingly, the thickness of mafic restite can be estimated as ∼5.5 to ∼21.5 km, which is substantially thicker than the features found in our model (Figure 3b).The exposed section of the Kohistan arc has proved that a section with garnet granulite rocks (∼7.5 km/s) embedded between the gabbroic lower crust (<7.0 km/s) and ultramafic rocks (∼8.0 km/s) is denser than the underlying ultramafic rocks, providing unstable conditions to allow foundering of this layer (Jagoutz & Behn, 2013).High pressure experiments and geochemical modeling have demonstrated that the sinking anti-continent is always denser than the surrounding mantle, resulting in foundering without stagnation (Tatsumi et al., 2015).The continuing delamination or foundering of mafic restites and cumulates leads to the harzburgitic sub-arc mantle being replaced, resulting in a seismic structure comparable to the continental crust.
In previous studies of IBM arc, the layer with velocities of ∼7.0 to ∼7.5 km/s was defined as the lower part of the lower crust (e.g., Kodaira et al., 2007aKodaira et al., , 2008) ) or low upper-mantle velocities (e.g., Takahashi et al., 2007Takahashi et al., , 2009) ) based on different definitions of Moho.In the Aleutian arc, the Moho was also placed underneath the layer of ∼7.3 to ∼7.7 km/s (Shillington et al., 2004).However, this layer is not detected in the Lesser Antilles arc and thus the mafic cumulate layer is missing (Christeson et al., 2008;Kopp et al., 2011).The contradiction of the Moho and the layer we defined as "crust-mantle transition" is widely observed in global volcanic arcs, through the interpretation of composition for this layer is consistent as mafic restite and cumulates.We proposed that different evolution stages of volcanic arcs might be responsible for this.In the juvenile stage of arc evolution, the anti-continent materials have not yet separated from the crust, which are supplied by primary basaltic melting (Tatsumi & Eggins, 1995), and therefore prominent Moho reflector appears under ∼7.5 km/s isoline, such as the Aleutian arc and some parts of IBM arc.With sufficient magma supply and ongoing arc evolution, a stronger differentiation occurs in arc crust, forming a thicker middle crust and gradually foundering the mafic restite into the mantle.Higher differentiation degree gradually results in larger composition contrast between the plutonic intermediate rocks (∼6.5 to ∼7.0 km/s) and mafic residuals (∼7.0 to ∼7.5 km/s) and thus a new Moho, which may be the Moho of future continental crust, between these two layers.For the Lesser Antilles arc, low magma production rate may prevent the differentiation of the crust (Christeson et al., 2008).While the Moho temperature cooling below 700°C, the anti-continent will stop foundering and replaced by upper-mantle rocks (Jagoutz & Behn, 2013), enhancing the discontinuity of newly formed Moho.Therefore, the waning of volcanism between the late Oligocene and early Miocene might facilitate the forming of a new Moho beneath the Izu arc in our model.Diverse crustal and deep mantle structures of volcanic arcs suggest the dynamic property of the Moho with the growth of volcanic arc.Nevertheless, it should be noted that the low mantle velocities could also represent partial melt and/or high temperatures in the uppermost mantle (Kelemen et al., 2003), like the case beneath the seamounts in the Shikoku Basin (model distance 140-190 km), which we cannot preclude beneath volcanic arc and need more evidence to discuss this.
At the depth of ∼27 km, a deep mantle reflector occurs across the iso-velocity contours from ∼7.5 to ∼8.0 km/s (Figure 3b), which was previously interpreted as the boundary between the crust-mantle transition and the ultramafic mantle.Amplitude modeling demonstrated that this transition zone is a mixture of mafic restites and olivine cumulates formed as a result of arc crustal growth (Sato et al., 2009).However, this reflector cannot be seen as a distinct boundary between different bulk rock compositions.The variation of velocity contrasts at the top and bottom of this layer and the average velocity in this layer may represent the ratio between the mafic restites and olivine cumulates.

Backarc Seafloor Spreading
Generally, the upper-crustal velocity in oceanic basins increases with age due to the off-axis mineralization of cracks, fissures, and veins at the top of the basement.However, in our model, the older crust (>23 Ma) located on the model distance 190-240 km has upper-crustal velocities of ≤5.4 km/s, lower than the younger crust (<20 Ma) located on the model distance 50-140 km (Figure 3b and Figure S3 in Supporting Information S2).
One explanation is the entrainment of hydrous melt during the initial stage of spreading, comparable with the observations in the Lau Basin (Dunn & Martinez, 2011).The melting contribution from the subducting slab chemically reformed the mantle composition, resulting in a higher differentiated and thicker oceanic crust, which contains more andesitic compositions in the upper crust.As the backarc spreading center moves away and separates from the volcanic arc, mantle downwelling replaces upwelling that happens close to the volcanic arc, and thus the dominated mantle composition transits from hydrous melt to enriched mantle melt rapidly (Dunn & Martinez, 2011).Nevertheless, the crustal thickness is relatively uniform and the lower-crustal velocities do not show prominent distinction in the Shikoku Basin (Figure S3 in Supporting Information S2).Also, the volcanism in the Izu arc ceased before the opening of Shikoku Basin (Taylor, 1992), which does not support the entrainment of hydrous melt.We also notice that the spreading direction of the Shikoku Basin changes from nearly E-W to NE-SW at ∼20 Ma (Figure 7a; Okino et al., 1994).Thus, either the change of geodynamic source or the anisotropy led to the "paradox" feature of the upper crust in the Shikoku Basin.
To further analyze the mantle condition and the melting contribution from the subducting slab in the Shikoku Basin, we extracted the average lower-crustal velocity (V p ) and crustal thickness (H) at an interval of 10 km with a 10-km-wide window (Figure 8).Only the lower crust was used for the H-V p analysis because it contains the first fractionates from the primary melt, while the upper crustal velocity is expected to be hugely affected by large porosity and alteration.Theoretical H-V p curves were constructed to discuss different factors during mantle melting, for example, mantle temperature, active mantle upwelling, mantle source, and crustal/subcrustal fractionation (Holbrook et al., 2001;Korenaga et al., 2002;Sallarès et al., 2005).As a standard model, McKenzie and Bickle (1988) demonstrated that the normal oceanic crust of ∼6-km crust with MORB composition stems from the adiabatic passive melting of a pyrolytic mantle in a temperature of ∼1,280°C.Higher potential temperature deepens the solidus and increases the height of the melting column, resulting in increasing melting and more MgO content thus leading to higher V p and thickness (Figure S4 in Supporting Information S2).In contrast, active upwelling provides an enhanced flux of mantle but does not change the mantle composition, and thus only increases crustal thickness.For a hypothetical fertile mantle comprising 70% depleted pyrolite mantle and 30% MORB, the crust becomes thicker while the crustal velocity becomes lower.Besides, crustal melt fractionation would increase the velocity while subcrustal fractionation has the opposite effect.
Figure 8 shows the H-V p plotting of the Shikoku Basin derived from our model.The velocity data were extracted from the oceanic domain where the clear magnetic isochron occurs and has been corrected to a pressure of 600 MPa and temperature of 400°C following the temperature coefficient of −0.00063 km/s/°C and the pressure coefficient of 0.00022 km/s/MPa, with the geothermal gradient of 11°C/km given by Christensen (1979).The results indicate enormous active upwelling (r > 10) of pyrolite mantle melting in moderate low temperatures of ∼1,200 to 1,250°C following the model of Korenaga et al. (2002) (Figure 8a).However, we noticed that most of the data corresponds to an extreme condition and is located outside of the curves given by Korenaga et al. (2002).Furthermore, such a rigorous active upwelling of mantle melting can hardly be explained in the overlying plate considering the mantle wedge flow monitored by subduction process.In contrast, a fertile mantle source with passive upwelling and normal temperature (∼1,300°C) can fairly explain the H-V p plotting of the Shikoku Basin (Figure 8b).Compared with the normal oceanic crust from slow to fast spreading rate controlled by MORB source, our data reveals that the Shikoku Basin has a thicker crust and lower V p .The recycling of ancient oceanic crust and underlying lithospheric mantle enhance the Fe content in the mantle compared to the MORB source, leading to the thicker crust, such as the Atlantic province (Korenaga & Kelemen, 2000) and the Galápagos hotspot (Sallarès et al., 2005).In the west Pacific, the Pacific plate was subducted beneath the Euro-Asia plate since the Eocene.As a result, the long-term supply of the subducting oceanic crust provided the raw material for altering the mantle composition.
The trend of H-V p with time in the Shikoku Basin mainly reflects the changing of fractionation position during mantle melting process (Figure 8b).The crust with larger thickness of >8.0 km is located near post-spreading volcanoes, or Kinan seamount chain, and thus does not reflect the accurate melting condition during backarc opening.The oceanic domain, which is less affected by the post-spreading magmatism, has rather uniform thickness of ∼7 to ∼8 km, while the average V p changes from ∼6.7 to ∼7.0 km/s (Figure 8b).Overall, the early stage  (Korenaga et al., 2002).Nearly diagonal dashed curves mark different ratios of active mantle upwelling (r), and the thick solid curve denotes the standard passive upwelling (r = 1), while nearly horizontal thin lines show mantle potential temperature in °C.The green star and orange rectangle represent the average value in the 10 km averaging window extracted from model distances 20-140 km and 190-240 km, respectively, while the error bar shows the range of those.The black circle shows the average thickness and lower-crustal velocity of the typical oceanic crust (White et al., 1992).The gray background shows the compilation H-V p range of oceanic crust generated at ultraslow to fast spreading centers or governed by hotspot provinces (Grevemeyer et al., 2018(Grevemeyer et al., , 2021)).For the backarc basin, the red, dark green, and blue ovals represent the H-V p range of Parece-Vela Basin (PVB; Grevemeyer et al., 2021), East Lau Basin (ELB; Arai & Dunn, 2014), and Mariana Trough Basin (MTB;Grevemeyer et al., 2021).All the data were calibrated to a pressure of 600 MPa and a temperature of 400°C.The correction of H-V p data in our model applies the temperature coefficient of −0.00063 km/s/°C and the pressure coefficient of 0.00022 km/s/MPa, with the geothermal gradient of 11°C/km, which are given by Christensen (1979).
of spreading (∼27 to ∼23 Ma, model distance 20-140 km) shows greater crustal fractionation and less subcrustal fractionation than the late stage spreading (∼20 to ∼15 Ma, model distance 190-240 km).The mantle melting condition of the Shikoku Basin may be similar to the Parece-Vela Basin (Grevemeyer et al., 2021), but exhibits a higher mantle temperature (∆T = ∼50°C) (Figure 8b).In contrast to the backarc opening that incorporates the hydrous melt or anomalous mantle from the subducting slab, for example, Mariana Trough Basin (Grevemeyer et al., 2021) and East Lau Basin (Arai & Dunn, 2014), the Shikoku Basin shows comparable thickness but prominent low V p , indicating that the oceanic crust of the Shikoku Basin barely has a melt contribution from the subducting slab (Figure 8b).

Conclusions
We derived the P-wave velocity structure model from a 585-km-long wide-angle seismic transect across the Izu arc-backarc system.The crustal thickness of the forearc region, arc region, and backarc basin (Shikoku Basin) is ∼5 to ∼16.5 km, ∼15 to ∼20 km, and ∼7 to ∼10.8 km, respectively.From top to bottom, the arc crust was divided and interpreted as upper crust (from the basement to 6.0 km/s) consisting of volcanic rocks, middle crust (6.0-6.5 km/s) consisting of felsic to intermediate composition layer, and lower crust (from 6.5 km/s to Moho) consisting of mainly gabbro and tonalitic intrusions.Beneath the arc crust, an extremely low velocity (<7.5 km/s) in the uppermost mantle was interpreted as mafic restite and cumulates.A deep mantle reflector was observed at a depth of ∼27 km beneath the Izu arc, which is considered as the boundary between the contaminated mantle and the unaltered mantle.
From the forearc to the rear arc region of the Izu arc, our seismic velocity correlates well with geochemistry and petrologic data.From the rear arc to the backarc knolls zone, the middle crust is thicker than 3 km and the ratio between the lower crust and middle crust (T l /T m ) is ranging between ∼1.5 and ∼3.5, which is controlled by predominant basaltic volcanism.However, the middle crust at the volcanic front and in the Myojin rift, which belongs to an area of rhyolitic volcanism, is rather thin.This demonstrated that the formation of tonalitic middle crust, which is crucial for transferring arc crust to continental crust, is closely associated with basaltic volcanism.The juvenile stage of arc evolution or remelting of the middle crust, which is controlled by the insufficient supply of basaltic magma from the mantle, may be responsible for the thin middle crust beneath the rhyolitic volcanoes.Another process for the transformation of continental crust is the delamination and foundering of mafic restites and cumulates, which are observed as low-velocity anomalies (∼7.0 to ∼7.5 km/s) in the uppermost mantle.
In the Shikoku Basin, the seismic structure does not support the impact of arc magmatism or any entrainment of hydrous melts during the opening of the backarc basin.On the contrary, a fertile mantle source with passive upwelling and normal temperature (∼1,300°C) can explain the seismic structure of the Shikoku Basin.

Figure 1 .
Figure 1.Bathymetry map of the study area and layout of seismic lines.Inset shows the location of our study area (indicated by red box) in the Izu-Bonin-Mariana (IBM) arc-backarc system.The black and blue lines mark the location of wide-angle and multi-channel seismic profiles of line IBr5, respectively.The red stars show the positions of International Ocean Discovery Program (IODP)/ODP drilling.SB, Shikoku Basin; PVB, Parece-Vela Basin; KPR, Kyushu-Palau Ridge; SFG, Sofugan tectonic line.

Figure 2 .
Figure 2. Original seismic record sections and the identification of seismic phases of (a-b) OBS18, (c-d) OBS49, (e-f) OBS64, (g-h) OBS83, (i-j) OBS100.The yellow error bars show the picking travel-time data and their uncertainty range, while the red and blue dots represent the calculated traveltime of seismic refraction and reflection phases respectively.Pg, crustal refraction phase; PmP, Moho reflection phase; Pn, upper mantle refraction phase; PdP, deep mantle reflection phase.

Figure 3 .
Figure 3. P-wave velocity tomographic result of line IBr5-OBS.(a) The position of Ocean-Bottom-Seismometer (OBS) stations along the seismic line is plotted on the bathymetric map.(b) P-wave velocity model.The thick black, blue, red, and green line represents the interface of the seafloor, sedimentary basement, Moho, and deep mantle reflector, respectively.(c) Derivative Weight Sum (DWS) for the rays traveling in the crustal tomographic inversion.(d) DWS beneath the Moho for the rays traveling in the mantle tomographic inversion.(e) Uncertainty test for the crust.(f) Uncertainty test for the mantle.The error bars show the uncertainty range of Moho depth.The ray-tracing and travel-time fitting figures for the inversion of Pg + PmP and Pg + Pn + PdP can be seen in Supporting Information S3 and S4, respectively.Myn, Myojin knoll; Myj, Myojin-sho; Sc, Sumisu caldera; Sn, Sumisu knoll; Ssc, South Sumisu caldera; KS, Kinan seamount chain; EMB, Enpo-Manji Basin; BK, Backarc knolls zone; MR, Myojin rift; V, Volcanic front; OH, Outer high; ITW, Inner trench wall.

Figure 4 .
Figure 4. (a) Travel-time residuals of Pg + PmP + Pn phases in each Ocean-Bottom-Seismometer (OBS).The vertical axis indicates the OBS station number.The color scale indicates the travel-time residuals for each shot.(b) All travel-time residuals of Pg + PmP + Pn phases along the entire model.The red dots denote the travel-time residual, showing good travel-time fits with mostly <0.15 s residual after inversion.

Figure 5 .
Figure 5. Recovered velocity perturbations from checkerboard tests.The input velocity perturbation anomaly is ±5% (positive in blue and negative in red), which is shown in (a) and (b).The recovered checkerboard patterns with cells of 15 × 4 km and 10 × 4 km are shown in (c) and (d) respectively.

Figure 6 .
Figure 6.1D depth-V p profiles extracted from (a) arc-forearc area and (b) backarc basin area in the V p model of line IBr5.(c) Comparison of in situ island arc velocity profiles.The thick red line in (c) represents the velocity structure of the volcanic front (x = 400 km) in our model.Velocity structures of the Aleutian arc(Holbrook et al., 1999), Lesser Antilles arc(Christeson et al., 2008), Alaska Peninsula(Lizarralde et al., 2002), and Mariana arc(Calvert et al., 2008) are extracted from the volcanic front.The velocity structure of the Bonin arc is extracted at the volcanic front beneath the Suiyo seamount(Kodaira et al., 2007b).The velocity structure of the Tonga arc is extracted from the remnant arc(Crawford et al., 2003).The red error bars show the velocity range of average continental crust(Christensen & Mooney, 1995), while the blue and red envelopes denote the velocity range of the Pacific oceanic crust and magmatic oceanic crust, respectively, which was compiled byGrevemeyer et al. (2018).The gray envelope represents the velocity range of the Izu arc in the model distance of 280-410 km of our model.

Figure 7 .
Figure 7. (a) Geochemistry data of the rock samples and tectonic features in the Izu arc and Shikoku Basin.The geochemistry data are compiled from Fryeret al. (1990),Ishizuka et al. (1998Ishizuka et al. ( , 2009)),Ishizuka, Uto, & Yuasa, 2003, Pearce et al. (1992),Tamura et al. (2009), andTaylor (1992).The italics mark the age of the samples.The normal faults of the backarc rifts are fromTaylor (1992), while the magnetic isochrons in the Shikoku Basin are fromOkino et al. (1994).(b) The variation of the middle crust thickness (T m ) and the thickness ratio between the lower crust and middle crust (T l /T m ) along line IBr5.The blue and red curves represent T m and T l /T m respectively.The pink area marks the range of T l /T m of the average continental crust calculated byChristensen and Mooney (1995).(c) Close-up of the International Ocean Discovery Program (IODP) Expedition 352 drill sites(Christeson et al., 2016) at the northeastern edge of the Bonin Ridge, showing the locations of forearc basalt (FAB) and boninite lavas.AR, Aoga-shima rift; SR, Sumisu rift; Ags, Aoga-shima; EDM, Enriched depression melting.The other abbreviations are consistent with those in Figure3.

Figure 8 .
Figure8.Mantle melting model illustrated by proxy of lower-crustal velocity (V p ) and crustal thickness (H) calculated every 10 km in the Shikoku Basin.The black and purple curves in (a) show the standard theoretical H-V p diagram of igneous crust generated by primary melts beneath mid-ocean ridge based on the method ofKorenaga et al. (2002) andSallarès et al. (2005), respectively.(b) Same as (a) but the mantle composition corresponds to a hypothetical high-Fe source fertile mantle composed of 70% depleted pyrolite mantle and 30% MORB(Korenaga et al., 2002).Nearly diagonal dashed curves mark different ratios of active mantle upwelling (r), and the thick solid curve denotes the standard passive upwelling (r = 1), while nearly horizontal thin lines show mantle potential temperature in °C.The green star and orange rectangle represent the average value in the 10 km averaging window extracted from model distances 20-140 km and 190-240 km, respectively, while the error bar shows the range of those.The black circle shows the average thickness and lower-crustal velocity of the typical oceanic crust(White et al., 1992).The gray background shows the compilation H-V p range of oceanic crust generated at ultraslow to fast spreading centers or governed by hotspot provinces(Grevemeyer et al., 2018(Grevemeyer et al., , 2021)).For the backarc basin, the red, dark green, and blue ovals represent the H-V p range of Parece-Vela Basin (PVB;Grevemeyer et al., 2021), East Lau Basin (ELB;Arai & Dunn, 2014), and Mariana Trough Basin (MTB;Grevemeyer et al., 2021).All the data were calibrated to a pressure of 600 MPa and a temperature of 400°C.The correction of H-V p data in our model applies the temperature coefficient of −0.00063 km/s/°C and the pressure coefficient of 0.00022 km/s/MPa, with the geothermal gradient of 11°C/km, which are given byChristensen (1979).

Table 1 Travel
-Time Picking Information of the Seismic Tomographic Inversion and Root-Mean-Square (RMS) Misfit and χ 2 Values of Each Step Inversion