Seismic structure of western end of the Nankai trough seismogenic zone

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 107, NO. B10, 2212, doi:10.1029/2000jb000121, 2002 Seismic structure of western end of the Nankai trough seismogenic zone Narumi Takahashi, Shuichi Kodaira, Ayako Nakanishi, Jin-Oh Park, Seiichi Miura, Tetsuro Tsuru, Yoshiyuki Kaneda, Kiyoshi Suyehiro, and Hajimu Kinoshita Japan Marine Science and Technology Center, Yokosuka, Japan Naoshi Hirata and Takaya Iwasaki Earthquake Research Institute, University of Tokyo, Tokyo, Japan Received 20 December 2000; revised 21 February 2002; accepted 26 February 2002; published 5 October 2002. [1] In 1998 a seismic experiment along a 320-km-long profile using an air gun array and ocean bottom seismographs (OBSs) was collected in the on shore-offshore, forearc region of the western Nankai trough seismogenic zone, southwestern Japan. This area is outside the coseismic rupture area of the 1946 Nankaido earthquake (M s = 8.2). The main objectives of this paper are to clarify the P wave and the Poisson s ratio profile around the forearc region of southwestern Japan, and estimate differences between this area and the Nankaido coseismic rupture area. The P wave velocity model estimated by two-dimensional (2-D) ray tracing and travel time inversion indicates three units with distinctive velocity patterns. The lens-shaped unit has a maximum thickness of 5 km and velocity of 3.4 4.0 km/s. The underlying unit has velocity of 5.4 5.6 km/s, and its P wave velocity increases on land (V p = 6.2 6.4 km/s). The relative homogeneous unit with P wave velocity of 6.6 6.9 km/s exists just beneath secondary unit with P velocity of 5.4 5.6 km/s. The angle of subduction is 7. We also obtained a Poisson s ratio of the crustal structure of the forearc region. The Poisson s ratio of the southern part of secondary unit is larger than that of the northern part. We conclude that these data indicate clay-rich metamorphic rocks plus oceanic basaltic rocks, rather than high-porosity materials, comprise the majority of the layer. We suggest that the region of high Poisson s ratio probably corresponds to underplating and the metamorphism of the clay-rich sediments and possibly the oceanic basalt just below the decollement. INDEX TERMS: 3025 Marine Geology and Geophysics: Marine seismics (0935); 8105 Tectonophysics: Continental margins and sedimentary basins; 8150 Tectonophysics: Evolution of the Earth: Plate boundary general (3040); 7205 Seismology: Continental crust (1242); 7220 Seismology: Oceanic crust; KEYWORDS: Nankai trough, seismogenic zone, velocity structure, Poisson ratio, ocean bottom seismographs (OBSs) Citation: Takahashi, N., et al., Seismic structure of western end of the Nankai trough seismogenic zone, J. Geophys. Res., 107(B10), 2212, doi:10.1029/2000jb000121, 2002. 1. Introduction [2] The Nankai trough is an active convergence boundary formed by subduction of the Philippine Sea plate beneath the southwestern Japan arc (Figure 1). Great earthquakes have occurred repeatedly around the Nankai trough and are well documented [e.g., Aida, 1981]. The reoccurrence interval is estimated to be between 100 and 200 years, and six rupture zones associated with large earthquakes are denoted Z, A, B, C, D, and E [e.g., Ando, 1975; Awata and Sugiyama, 1989]. The last great earthquake, Nankaido earthquake (M s = 8.2), occurred in 1946 with an epicenter located at the eastern part of area B and had a low-angle thrust-type source mechanism [e.g., Kanamori, 1972]. The rupture area covered areas A and B; however, the rupture Copyright 2002 by the American Geophysical Union. 0148-0227/02/2000JB000121$09.00 pattern is different for their two areas (Figure 1). Ando [1975] proposed that the area B was broken by brittle rupture and that area A was deformed by slow slip. Area Z (Figure 1), located off Cape Ashizuri, was not broken during the 1946 Nankaido earthquake according to geodetic and tsunami data [Kanamori, 1972; Ando, 1975; Ando, 1982; Aida, 1981; Kato, 1983; Sagiya and Thatcher, 1999]. In other studies, the rupture area does not include area Z, or the slip movement of area Z is small [Fitch and Scholz, 1971; Iwasaki, 1981; Satake, 1993]. According to Kato [1983] and Sagiya and Thatcher [1999], the slip movement at the plate boundary becomes small near Cape Ashizuri, although a splay fault located off Cape Ashizuri produced the coseismic slip. Cummins and Kaneda [2000] also proposed splay faults within area A during coseismic slip. Thus the slip movement at the plate boundary by the 1946 Nankaido earthquake around the Z area seems to be very small compared to that of area A. The rupture area of ESE 2-1

ESE 2-2 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 1. Map of this experiment area. Large solid circles and thick line show the OBS locations and the seismic line in this study, respectively. Numerals are OBS number. The small solid circles and thin line are the OBS location and the seismic line cross the 1946 Nankaido earthquake coseismic rupture zone, respectively [Kodaira et al., 2000]. Dot lines show the tectonic lines, and MTL, BTL, and ATL are the median tectonic line, the Butsuzo tectonic line, and the Aki tectonic line, respectively. RB, SB, CB, NSB, and SSB are the Ryoke metamorphic belt, the Sanbagawa metamorphic belt, the Chichibu metamorphic belt, the northern Shimanto belt, and the southern Shimanto belt, respectively. Rectangles are 1946 Nankaido earthquake coseismic rupture zone (a, Kanamori [1972]; b, Ando [1982]; c, Ando [1975]; d, Kato [1983]; e, Sagiya and Thatcher [1999]). The A, B, and Z notation follows Awata and Sugiyama [1989]. CA and CM are cape Ashizuri and cape Muroto. the 1707 Hoei-Nankai earthquake, however, appears to correspond to areas Z, A, and B [Aida, 1981]. Thus area Z has not been a region of large deformation by large earthquakes since 1707. [3] Seismic experiments have been carried out around area Z and the western part of area A in the past [Nishizawa and Suyehiro, 1989; Moore et al., 1990; Stoffa et al., 1992]. Nishizawa and Suyehiro [1989] constructed a velocity model at the island arc slope of the western part of area A by air gun-ocean bottom seismograph (OBS) data. Moore et al. [1990] clearly imaged the decollement of the subducting oceanic crust using multichannel seismic reflection (MCS) data, and Stoffa et al. [1992] estimated a high-resolution velocity structure model using expanding spread and split spread seismic data. The velocity model shows that a lowvelocity zone, related to out-of-sequence thrusts, has developed on the island arc slope and that the decollement zone does not appear to exhibit low velocity. The results of these previous studies, although helpful, were limited to near the trough axis. [4] Velocity models around the eastern part of area A were constructed [Yoshii et al., 1973; Kinoshita and Matsuda, 1989; Kodaira et al., 2000], and Kodaira et al. [2000] estimated the crustal structure across area A using controlled source seismic data and the rupture area of the 1946 Nankaido earthquake proposed by Ando [1975]. The crustal structure is characterized by a gently sloping subducting oceanic crust and a thick overlying unit of thick lowvelocity sediments (9 km thick at a distance of 70 km from trough axis). [5] Many land-based geological investigations have also been performed on the Shikoku metamorphic belts [e.g., Taira et al., 1989; Omori et al., 1997]. According to these studies the Shikoku Island is divided by four tectonic lines

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-3 Figure 2. MCS profile along the main seismic line [after Park et al., 2002]. Amplitude corrections, deconvolution, band-pass filtering, normal move out corrections, stacking, multiple suppression, and time migration are applied. that strike ENE-WSW; these are (from south to north): the southern Shimanto belt, the northern Shimanto belt, the Chichibu metamorphic belt, the Sanbagawa metamorphic belt, and the Ryoke metamorphic belt. The southern and northern Shimanto belts and the Chichibu metamorphic belt consist of Neogene, Cretaceous, and Paleozoic and/or Mesozoic sediments, respectively. [6] The crustal structure of Shikoku, estimated using an explosive seismic experiment [Ito et al., 1982], consists of a low-velocity layer, and two relative higher-velocity layers, an upper layer (V p = 6.1 km/s) and a lower (V p = 6.7 km/s). The depth of the island arc Moho is 35 km beneath the central Shikoku. These two distinguishable layers are characteristics of velocity structure deduced by explosive seismic experiments. [7] Nakamura et al. [1997] clarified the distribution of background seismicity around the outer zone of the Nankai trough and suggested that the descending slab appears to form three sequences and directions. Kimura and Okano [1994] imaged the change in focal depth distribution of central and western Shikoku using mantle earthquakes along the strike of the geological tectonic lines of Shikoku, which drops gradually toward west. However, because landbased reflection data are limited and because of very low background seismicity and the significant errors of hypocenters attributed to the exclusive use of a land station network with no offshore stations, it has been impossible to identify exactly the subduction of the oceanic crust across area Z. [8] We carried out an air gun-obs seismic experiment, including several land stations, across area Z in 1998 and estimated the P wave velocity model of the seismogenic zone as well as the Poisson s ratio distribution in the forearc. Our objectives in this paper are to construct the seismic velocity model across area Z of the Nankai seismogenic zone and clarify the differences of the velocity models across areas Z and A, which is the western limit of the rupture area of the 1946 Nankaido earthquake. 2. Data Acquisition [9] In 1998 the Japan Marine Science and Technology Center (JAMSTEC) and Earthquake Research Institute (ERI), University of Tokyo, carried out a collaborative seismic refraction-reflection experiment in the forearc-arc region of southwestern Japan using R/V Kairei of JAM- STEC. A single seismic line was run perpendicular to the trough axis measuring 320 km from the southern end of marine receivers to the northern limit of land-based stations. Fifteen ocean bottom seismographs (OBSs) were deployed on this line (14 OBS records are available) and three temporary land stations (ERI1, ERI2, and ERI3) on Shikoku Island also recorded the air gun signals. Furthermore, the records of the permanent land station MSK1, located at the western end of Shikoku, were also used because the station is in line with the rest of the seismic profile. The R/ V Kairei towed a 120-channel hydrophone streamer and multichannel reflection data (MCS) were obtained. Unfortunately, the MCS data were only obtained from the southern end of the seismic line, to OBS 5, because of the many fishing boats north of this location. Total capacity of the air gun array was 4000 cubic inches (68 L). The shot interval for the MCS survey was 50 m, and that for the wide-angle survey was 200 m, in order to reduce the noise of previous shots. The profile was designed with reference to the rupture area of the 1946 Nankaido earthquake to compare the differences between velocity structures of ruptured and nonruptured areas.

ESE 2-4 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 3. Comparison of the observed record section of the vertical component recorded by OBS 15 and the synthetic seismograms. All travel times are reduced by 6 km/s. (a) Record section. Band-pass filter (5 15 Hz) and geometry amplitude correction are applied. P2, P3, Pn, and PmP are the refraction arrivals from the oceanic layer 2, the oceanic layer 3, the oceanic upper mantle, and the reflection arrivals from the oceanic Moho, respectively. (b) Synthetic seismograms. [10] The OBSs were equipped with a hydrophone sensor and three-component geophones using gimbal-leveling mechanisms; natural frequency of the geophones was 4.5 Hz. The digital recorder used a 16-bit A/D converter and stored data on digital audiotape at 100 Hz, sampling continuously for 17 days [Shinohara et al., 1993]. [11] The seafloor position of our free-fall instruments differed from their deployment location because of the strong Kuroshio current. We estimated the exact OBS positions using arrivals of direct waves of air gun shots to within 5 km from the deployment position of each OBS. 3. Modeling Procedure 3.1. Construction of P Wave Velocity Model [12] To make the initial 2-D velocity model, we estimated the one dimensional velocity structure of the shallow section beneath each OBS using t-p mapping [Stoffa et al., 1981] and the t-sum inversion method [Diebold and Stoffa, 1981; Shinohara et al., 1994]. We compared these 1- D velocity results with the MCS profile and confirmed that the structures roughly agreed with the MCS profile. [13] A 2-D velocity model was constructed using forward modeling, trial and error, 2-D ray tracing [Zelt and Ellis, 1988] and travel time inversion [Zelt and Smith, 1992]. We picked travel times of reflection and refraction phases with uncertainties of 0.1 0.3 s depending on the signal-to-noise ratio. Most of the synthetic travel times fell within the error bars. Any model parameter with the resolution value greater than 0.5 was generally defined as well resolved. To demonstrate uniqueness and reliability, a root-mean-square of misfit between calculated and observed travel times (T rms ) and misfit normalized by uncertainty of the observed travel times (c 2 ) were used [Zelt and Smith, 1992]. Synthetic seismograms were also calculated and compared with the observed seismograms to estimate detailed structure, which cannot be determined from travel times alone. In particular, we considered the whole characteristics of the amplitude variation, and in constructing the velocity model we paid attention to the range of offset distances across which the refraction phases were observed. 3.2. Construction of Poisson s Ratio Profile [14] Poisson s ratio is influenced strongly by porosity and clay content, which is controlled by material [e.g., Eastwood and Castagna, 1986]. In order to obtain other physical information of the structure related to the seismogenic zone, we tried to construct a Poisson s ratio profile beneath the forearc region from OBS 3 to OBS 9. [15] Our OBS instruments have two horizontal components perpendicular to each other, but we cannot measure the azimuth of each horizontal sensor of the OBS because there is no device to measure the absolute (e.g., magnetic)

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-5 Figure 4. Comparison of the observed record section of the vertical component recorded by OBS 8 and the synthetic seismograms. All travel times are reduced by 6 km/s. (a) Record section. Band-pass filter (5 15 Hz) and geometry amplitude correction are applied. Psed, Puc and Plc are the refraction arrivals from the sedimentary wedge; the 5.4 5.9 km/s layer connects to the island arc upper crust and the island arc lower crust, respectively. (b) Synthetic seismograms. directions of these horizontal components. Therefore we estimated the azimuth of each horizontal sensor by calculating particle motions using two horizontal components for direct waves of air gun signals for locations between 5 and 15 km from each OBS. Because of distances from OBS location at seafloor to the seismic line, using data less than 5 km offsets from OBS location to air gun shot position may enlarge an error of the estimated azimuth. [16] Marine air gun arrays generate only P waves. To determine a Poisson s ratio model, P waves converted to S waves need to be used. PSS phases that convert from P waves to S waves beneath the OBS and that refract from each layer were not observed on radial component records (see section 4). Because of a strong heterogeneity of the velocity structure the PSS phases do not seem to have been generated. Therefore we used PPS phases that converted from P waves to S waves beneath the OBS. [17] The PPS phases on a radial component of each OBS were picked. We assumed that P waves converted to S waves on an interface whose velocity gap between the upper and the lower layers is >1 km/s. We paid special attention to distinguish the PPS arrivals with multiple arrivals. Then a Poisson s ratio profile of the island arc slope was estimated using forward modeling with trial and error [Zelt and Ellis, 1988]. We picked reflection and refraction PPS travel time arrivals with uncertainties of 0.2 0.4s depending on the signal-to-noise ratio, as in the case of the construction of the P wave velocity model. Most of the synthetic travel times lie within the error bars. 4. Seismic Data 4.1. Multichannel Reflection Data [18] Figure 2 shows a time migrated section of the MCS profile analyzed by Park et al. [2002]. The MCS data were obtained from OBS 6 to OBS 15. The section clearly images the subduction of the oceanic crust. Hemipelagic sediments on the oceanic crust can be seen clearly, and a turbidite layer accumulated on the oceanic crust at the trough axis is also imaged. The decollement in very clear between OBS 9 and OBS 11, but it becomes obscure between OBS 8 and OBS 9. The oceanic Moho is also imaged at 9 s in two-way travel time beneath OBS 15 and can be traced to OBS 9. A very clear interface can also be traced between OBS 6 and OBS 8 at 6 s in two-way travel time. Within the accretionary prism, imbricated thrusts are imaged and a splay fault is well defined at 6 s and breaks the seafloor between OBS 7 and OBS 8. within the error bars. 4.2. Vertical Record Section of Wide-Angle Data [19] The air gun signals were recorded at all OBSs except OBS 14, which, because of a failed release mechanism, was not recovered. Data quality of almost all OBSs was good. That of OBS 2, OBS 3, and ERI2 was poor,

ESE 2-6 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 5. Comparison of the observed record section of the vertical component recorded by OBS 4 and the synthetic seismograms. All travel times are reduced by 6 km/s. (a) Record section. Band-pass filter (5 15 Hz) and geometry amplitude correction are applied. Psed2 is the refraction arrivals from the lower sedimentary wedge. (b) Synthetic seismograms. perhaps due to the shallow depth and ocean currents at these locations. [20] We note three patterns in the OBS records: (1) refraction and reflection phases within the oceanic crust; (2) first refraction phases with apparent velocity of 2 3 km/s, corresponding to a sedimentary layer; and (3) first refraction phases with higher apparent velocity than 2 3 km/s. [21] The record sections of OBSs deployed on the Philippine Sea plate (OBS 11 OBS 15) have four main phases, which we interpret (Figure 3) as (1) refractions from the oceanic layer 2 (P2); (2) refractions from the oceanic layer 3 (P3); (3) reflections from the oceanic Moho (PmP); and (4) refractions from the oceanic mantle (Pn). The Pn phases have high amplitude at offsets between 30 80 km; however, the amplitude becomes small at distances >80 km. This characteristic of the Pn phases probably means that the velocity structure of the uppermost mantle is stratified. [22] The record sections of OBSs deployed on the accretionary prism (OBS 4 OBS 10) have refraction phases with very low apparent velocities (Psed) and relatively small amplitude (Puc and Plc). Figure 4 shows the record section of OBS 8. The characteristics of the record section of OBS 4 are similar to those of OBS 8 (Figure 5). The difference between the record sections of OBS 8 and OBS 4 is adding another refraction phase (Psed2) whose apparent velocity is faster than that of Psed on the record section of OBS 4. [23] Figure 6 shows the record section of a land station (ERI1) with high-amplitude refraction phases from the island arc upper crust and low-amplitude refraction phases from the island arc lower crust. The reflection phases from the top of the subducting oceanic crust (Ptop) can be seen. High-amplitude reflection phases (PmP) and low-amplitude refraction phases (Pn) from the oceanic Moho can be modeled by synthetic seismograms. 4.3. Radial Record Section of Wide-Angle Data [24] The record section of the radial component calculated from two horizontal components perpendicular to each other (OBS 5) and synthetic travel times of PPS arrivals are shown in Figure 7. Deep refraction phases with an S wave apparent velocity were not observed. However, many refraction and reflection phases whose apparent velocities equal the corresponding P wave velocities could be traced and regarded as PPS phases converted at an interface with a large velocity contrast just beneath the OBS. For example, the large velocity contrast beneath OBS 5 exists on units A, B, and C. Figures 7a, 7b, and 7c show the comparisons of the observed arrivals with the estimated arrivals of S waves converted at the tops of unit A (Psed-2, Puc-2, Ptop-2, and PmP-2), unit B (Puc-4 and PmP-4), and unit C (PmP-5), respectively. 5. Crustal Model and Discussion [25] A P wave velocity model along the wide-angle OBS profile is shown in Figure 8. Figure 9 shows comparisons

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-7 Figure 6. Comparison of the observed record section of the vertical component recorded by ERI1 and the synthetic seismograms. All travel times are reduced by 6 km/s. (a) Record section. Band-pass filter (5 15 Hz) and geometry amplitude correction are applied. Ptop is the reflection arrivals from the top of the subducting oceanic crust. (b) Synthetic seismograms. between calculated and observed travel times. Figures 10 and 11 are ray diagrams of the refraction and reflection arrivals from each layer, and the resolution values of the interface and the P wave velocity nodes calculated by the 2-D travel time inversion, respectively. This dimensionless resolution value is not directly related to the absolute parameter uncertainty, however, a higher resolution is usually associated with lower uncertainty [Zelt and Smith, 1992; Kodaira et al., 2000]. We regard resolution values greater than 0.5 as significant. By the resolution values we tried to estimate for the subduction angle using refraction and reflection phases (Figure 12). [26] A Poisson s ratio distribution was also modeled using the PPS data recorded on six OBSs deployed on the forearc region. Figures 13 and 14 show the Poisson s ratio profile of the forearc region and comparisons between the calculated PPS and the observed travel times. 5.1. P Wave Velocity Model 5.1.1. Oceanic Crust and Uppermost Mantle [27] The P wave velocity structure of the oceanic crust is near that of normal oceanic crust [Spudich and Orcutt, 1980; White et al., 1992]. The P wave velocities of two layers of the shallow part are 1.7 1.9 km/s and 1.9 2.5 km/s, respectively. Beneath these are a 4.5 5.3 km/s layer and a 6.8 7.0 km/s layer that probably correspond to oceanic layer 2 and oceanic layer 3, respectively. Oceanic layer 2 thickens beneath the deformation front. The depth of the oceanic Moho is 11 km, and the crustal thickness is 5 km. The P wave velocity of the oceanic mantle is 8.0 km/s. [28] Most of the synthetic travel times lie within the error bars (0.1 0.3 s) of the picked values (Figure 9). The depth of the oceanic layer 2 was determined by comparing the MCS profile with reflections from the top of the oceanic layer 2 recorded on OBSs. Almost resolution values of interface nodes from the top of the oceanic layer 3 and the Moho are both >0.5 due to good ray coverage (Figures 10c and 11a). The resolution values for velocity nodes of the oceanic layer 2, the layer 3, and the uppermost mantle, except the bottom of the oceanic layer 2, are also greater than 0.5 due to good ray coverage (Figures 10a, 10b, 10d, and 11b). [29] The Pn phases recorded by OBS 15 at ranges of 30 km to 80 km have relatively high amplitude compared with the Pn phases from 80 to 200 km. Such changes of the amplitudes of the Pn phases seem to be due to the structure of the uppermost oceanic mantle. Consequently, we assumed the presence of a layer with high vertical velocity gradient of 0.06 s 1 and thickness of 1 kmjust above the oceanic uppermost mantle to explain the amplitude changes. Thus the interface between the high-velocity gradient layer and the oceanic upper mantle does not have an appreciable velocity contrast. The P wave velocity at the top of the oceanic mantle is 8.0 km/s. 5.1.2. Forearc Region of the Island Arc [30] The velocity structure beneath the forearc region of the island arc consists of three layers: units A, B, and C.

ESE 2-8 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 7. Comparison of the observed record section of the radial component recorded by OBS 5 and the synthetic arrivals. All travel times are reduced by 6 km/s. Band-pass filter (5 15 Hz) and geometry amplitude correction are applied. Insets represent a schematic ray path down to the island arc lower crust. (a) Comparison of the record section with S wave synthetic arrivals converted from P wave on a top of the sedimentary layer. (b) Comparison of the record section with S wave synthetic arrivals converted from P wave on a top of the 5.4 5.6 km/s layer. (c) Comparison of the record section with S wave synthetic arrivals converted from P wave on top of the island arc lower crust.

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-9 Figure 8. P wave velocity model. The horizontal and the vertical axes are the distance from the southern end of the line (Figure 1) and the depth from the sea level, respectively. Numerals are P wave velocity (km/s). The contour interval is 0.1 km/s. Open circles show OBS locations. Unit A is defined as a lens-shaped layer with velocities in the range of 2.9 4.0 km/s. Unit B underlies unit A and is defined as a wedge-shaped layer characterized by a velocity structure that grades horizontal from 5.4 km/s on its seaward edge to 6.2 6.5 km/s on its landward edge. Unit C underlies unit B and is defined as a second wedge-shaped layer that has a more homogeneous velocity structure as well as a higher average velocity (6.6 6.9 km/s). [31] The uppermost unit A can be subdivided into three parts. The shallowest part of the wedge has P wave velocity of 1.7 2.7 km/s and appears to thicken seaward. The main body of the sedimentary wedge has velocities of 2.9 4.0 km/ s and has maximum thickness of 5 km at 60 80 km north of the trench axis. The base of the wedge has P wave velocities of 4.2 4.5 km/s with a vertical velocity gradient that beneath the northern part of the forearc region is larger (0.2 s 1 ) than that of the main body. Refraction arrivals (Psed2) from this third layer are observed in OBS 4 (Figure 5). [32] The middle unit B underlying in velocity with 5.4 5.9 km/s exhibits lateral differences. The southern area (between OBS 9 and OBS 5) has displaying velocities of 5.4 5.6 km/s and lying depth of 7.7 8.0 km below sea level (bsl). The northern part of this unit (between OBS 5 and ERI1) shows slightly faster velocities (e.g., 5.4 6.2 km/s), and the unit as whole thickens and shoals landward. Furthermore, the vertical velocity gradient is larger (maximum 0.13 s 1 ) than that of the southern part (0.04 s 1 ). [33] The lowermost unit C has P wave velocities of 6.6 6.9 km/s and has average flat top at a depth of 13 km bsl. Its P wave velocity does not change laterally, and this unit is relatively homogeneous compared to unit B. The lower boundary of this unit C corresponds to the top of the subducting oceanic crust. [34] Most of the synthetic travel times of each OBS deployed beneath the forearc region also lie within the error bars (0.1 0.3 s). The resolution values of the interface nodes of the tops of unit B and C are greater than 0.5 due to good ray coverage (Figures 10g and 11a). The resolution values of the velocity nodes in unit A, B and C except the bottoms of unit A and C are also greater than 0.5 due to good ray coverage of each refraction phase (Figures 10e, 10f, 10h, and 11b). 5.1.3. Crustal Structure Beneath Shikoku [35] Unit A thins toward the shoreline and almost disappears on land. The crustal structure beneath Shikoku appears to consist of two layers: a 6.2 6.4 km/s layer, and a 6.6 6.9 km/s layer. This velocity structure is similar to that of Ito et al. [1982], who measured the P wave velocities of the island arc upper crust and the lower crust as 6.1 and 6.7 km/s, respectively. [36] The upper layer connects continuously with unit B (V p = 5.4 5.9 km/s) beneath the forearc region. The vertical velocity gradient beneath the land area (0.02 s 1 ) is smaller than that of the forearc region (0.05 s 1 ). In particular, the lateral heterogeneity around the coastline is large between ERI2 and OBS 1. The lower layer beneath Shikoku correlates with unit C on the forearc. [37] The resolution values of the interface nodes at the top of unit C in this area are greater than 0.5 at south of ERI3. The resolution values at the top of unit B are also greater than 0.5 at the southern side from ERI3. The resolution value for the unit C beneath land is less than 0.5 because no refracted rays from unit C were recorded between ERI1 and MSK1. [38] The boundary between a land and a marine character detected between OBS 1 and OBS 2 corresponds to the location of the splay fault proposed by Sagiya and Thatcher [1999] and Kato [1983]. Although this study does not detect the splay fault directly, its existence is consistent with the abrupt change in P wave velocities observed between OBS 1 and the land-based seismic stations. It could be that this lateral heterogeneity of the island upper crust relates to the splay fault. 5.1.4. Subducting Oceanic Crust [39] We constructed a P wave velocity model of the nonrupture zone of the 1946 Nankaido earthquake as

ESE 2-10 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 9. Comparisons the observed and calculated travel times. P3, PmP, Pn, Psed, Puc and Plc are the refraction arrivals from the oceanic layer 3, the reflection arrivals from the oceanic Moho, the refraction arrivals from the oceanic Moho, the island arc upper crust, and the island arc lower crust, respectively. (a) OBS 15, (b) OBS 13, (c) OBS 12, (d) OBS 11, (e) OBS 10, (f ) OBS 9, (g) OBS 8, (h) OBS 7, (i) OBS 6, ( j) OBS 5, (k) OBS 4, (l) OBS 3, (m) OBS 2, (n) OBS 1, (o) ERI1, (p) ERI2, (q) ERI3, (r) MSK1. shown in Figure 8. However, good resolution for the top velocity nodes of oceanic layers 2 and 3 is limited in the southern region near OBS 10 and OBS 6, respectively (Figure 11b). Picked reflection arrivals from the top of the subducting oceanic crust beneath the forearc slope are also limited in the area between OBS 7 and OBS 5 (Figures 10i and 11a). The reflection arrivals from the oceanic Moho and the refraction arrivals from the oceanic upper mantle can be traced over much of the section (Figures 10c and 10d). We tried to estimate the subduction angle of the oceanic crust in the northern area from OBS 6 by applying travel time inversion for the reflection and refraction arrivals from the oceanic crust and upper mantle (Figure 12). This test is to confirm change of subduction angle of the oceanic crust landward. We assumed that the velocity of oceanic layer 2 and the thickness of oceanic layers 2 and 3 were the same before and after subduction. We calculated c 2 values against the subduction angle assuming minimum and maximum P wave velocities of oceanic layer 3 of 6.9 and 7.2 km/s, respectively, taking into account the effect of subduction. Consequently, it is suggested that the oceanic crust smoothly subducts with

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-11 Figure 10. Ray diagrams. (a) Refraction arrivals from the oceanic layer 2. (b) Refraction arrivals from the oceanic layer 3. (c) Reflection arrivals of the oceanic Moho. (d) Refraction arrivals from the oceanic upper mantle. (e) Refraction arrivals from the sedimentary wedge. (f ) Refraction arrivals from the island arc upper crust and the 5.4 5.6 km/s layer. (g) Reflection arrivals from the top of the island arc lower crust. (h) Refraction arrivals from the island arc lower crust. (i) Reflection arrivals from the top of the oceanic crust. the angle of 6.5 7.0 at 20 120 km from the trough axis. 5.2. Poisson s Ratio Profile of the Forearc Region [40] We investigated the Poisson s ratios of the units A and B using the PPS converted arrivals to estimate the heterogeneity of the plate boundary structure and infer its composition. The results of applying 2-D ray tracing for the converted PPS arrivals are shown in Figure 14. As described in section 3.2, the converted PPS arrivals were used to estimate the Poisson s ratio because the PSS arrivals could not be picked. Therefore it is important to identify the interface that converts P waves to S waves. Figure 14a shows the vertical velocity contrast profile. Picked PPS arrivals converted at the interfaces that have large velocity contrasts. Therefore the estimated Poisson s ratio is the average value of the layers bounded by interfaces with large velocity contrasts and limited area to just beneath each OBS due to narrow area the S waves travel. [41] The Poisson s ratios of unit A changes south to north from 0.435 to 0.46. The ratios of unit B were estimated to be between 0.36 and 0.41. The Poisson s ratios of main part of the layer were estimated as 0.27 or 0.32 0.34. The ratio of the northern part of unit B is 0.27 and that of the southern part is 0.32 0.34. [42] Errors in picking S wave travel times are 0.2 0.4 s. When it is difficult to distinguish PPS phases and multiple phases, PPS phases were not picked. Since the rough error estimates of Poisson s ratio is about plus or minus 0.05 to 0.2 taking into account each picking error of the PPS phases, the difference of the Poisson s ratio of the southern part of unit B (0.27 or 0.32 0.34) can be resolved. [43] When a structure such as the splay fault beneath an OBS has horizontally strong heterogeneity, Poisson s ratio is strongly affected. Because the P wave velocities of the uppermost sedimentary layer and unit A are strongly heterogeneous along the profile, the differences of these layers Poisson s ratios may also reflect the strong heterogeneity. The heterogeneity in P wave velocities of unit B, however, is very small compared with the above two layers and does not explain the difference of the Poisson s ratio of between 0.27 and 0.32 0.34. 5.3. Comparing This Velocity Model With That of the 1946 Nankaido Earthquake Coseismic Rupture Zone [44] The two velocity models located on either side of the western end of the 1946 Nankaido earthquake coseismic

ESE 2-12 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 11. Resolution values calculated by travel time inversion. (a) Resolution values of the interface node. (b) Resolution values of the P wave velocity node. rupture zone, including the subducting oceanic crust, have following common characteristics (Figure 15). (1) The thickness of the oceanic crust is 5 km. (2) The subduction angle is 7. (3) A lens-shaped unit composed sediments (unit A) corresponding to the sedimentary wedge of Kodaira et al. [2000] is recognized. (4) Unit B with velocity of 5.4 5.9 km/s corresponding to island arc upper crust of Kodaira et al. [2000] thickens landward. (5) The P wave velocity of unit B increases near the coastline. (6) Faster velocity unit below unit B (unit C) corresponding to island arc lower crust of Kodaira et al. [2000] is more homogeneous. [45] Main differences between the two velocity models appear in the thickness of unit A and the seaward part of unit B. The thickness of unit A of the 1946 Nankaido earthquake coseismic rupture zone reaches a maximum of 9 km (5 km in this study). However, the total volume of unit A is almost the same in both models because a length of unit A along a seismic line in this study is longer than that of the coseismic rupture zone. The southern part of unit B in this study, whose top interface is flat, is broader than that of the coseismic rupture zone. Consequently, the length of the contact zone between the oceanic crust and unit A in this study is half that of the coseismic rupture zone proposed by Kodaira et al. [2000], and the contact zone between subducting oceanic crust and unit B in this study is longer than that of the coseismic rupture zone. The lengths of the contact zone of this study area and the coseismic rupture zone are 45 km and 30 km, respectively. 5.4. Geological Interpretations [46] According to Taira et al. [1989] the entire area of southwestern Japan is mostly composed of ancient accretionary prisms from Paleozoic to Tertiary in age and intruded by Mesozoic to Holocene igneous rocks. Four tectonic belts run the length of Shikoku Island is metamorphic belt originated the accretionary prisms. These tectonic belts are called the Shimanto, the Chichibu (CB), the Sanbagawa (SB), and the Ryoke metamorphic belts (RB), respectively (Figure 16). The tectonic lines between the Shimanto belt and the CB, and the SB and the RB are called the Butsuzo tectonic line (BTL) and the median tectonic line (MTL), respectively. [47] The Shimanto belt can be divided two parts. One is the southern Shimanto belt (SSB) that consists of Tertiary accreted sediment. The other is the northern Shimanto belt (NSB) that consists of Cretaceous accreted sediment. The boundary between the northern and the southern parts is called the Aki tectonic line (ATL). The southern part of the

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-13 Figure 12. Result of the subduction angle analysis. (a) Schematic location of the oceanic crust. Two rectangles and an arrow indicate images of the subducting oceanic crust and this angle analysis. In this analysis a position at 140 km offsets from the southern end of this line was fixed. (b) Subduction anglenormalized c 2 profile for the refraction and reflection arrivals into the oceanic crust. Black line and broken line are the results of the travel time inversion using 6.9 and 7.2 km/s as the average P wave velocity of the oceanic layer 3, respectively. NSB consists mainly of upper Cretaceous accreted sediment that includes many melange units. Comparing our velocity measurements with the geological mapping shows that the NSB corresponds to the more homogeneous, landward part of unit B whose P wave velocity is 6.2 6.4 km/s. On the other hand, the SSB appears to correspond to the transition zone between the homogeneous part of unit B and southern part of this layer whose velocity is 5.4 5.9 km/s. The landward part of unit B with P wave velocity of 6.2 6.4 km/s in this study consists of accreted Cretaceous metasediments and intruded igneous rocks. In this study, our velocity model does not have the resolution to distinguish accreted sediments from granitic rocks and reflection phases are not observed in unit B. [48] Unit A beneath the forearc region probably consists of sediments accreted during Miocene to Pliocene time [e.g., Kodaira et al., 2000]. The southern part of this layer from OBS 8 to OBS 11 with relatively slow P wave velocities probably corresponds to most recently accreted sediments. The northern part between OBS 4 and OBS 8, whose P wave velocity is faster than that of southern part, corresponds to a region of imbricate thrusts including some splay faults (Figure 2). 5.5. High Poisson s Ratio of Southern Part of 5.4 5.6 km/s Layer [49] To obtain more detailed characteristics of the plate boundary velocity structure, a profile of the Poisson s ratio was estimated. The horizontal inhomogeneous Poisson s ratios of rocks in the accretionary prism probably are influenced by the heterogeneity of the structure caused by the many faults within the sedimentary wedge shown Figure 2.

ESE 2-14 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 13. Poisson s ratio profile beneath the forearc region of the island arc. (a) The velocity contrast profile calculated from the structure. (b) The Poisson s ratio profile. Numerals are Poisson s ratio. Thick lines show the interface that we can identify the conversion from P wave to S wave. Alphabet characters show the estimated error of the Poisson s ratio: a, ±0.005; b, ±0.01; c, ±0.02. We discuss the high Poisson s ratio of the southern part of unit B comparing with that of northern part. [50] The Poisson s ratios of the southern and the northern parts of unit B are 0.32 0.34 and 0.27, respectively (Figure 13). This difference is resolvable even when considering the picking error and the effect on the error of the Poisson s ratio of the shallow part (Figures 13 and 14). [51] Johnston and Christensen [1992] measured Poisson s ratios of dry sedimentary rock samples at 100 MPa pressure. According to their study the Poisson s ratios of sandstone (porosity 1.2 6.4%), argillaceous sandstone (porosity 1.3 1.5%) and shale (porosity 1.0 1.2%) are 0.02 0.22, 0.19 0.29, and 0.31 0.32, respectively. The Poisson s ratio increases as the clay content increases. In 10 50 MPa pressure, Han [1986] and Blangy [1992] measured the Poisson s ratios of sandstone samples. According to their studies the Poisson s ratio of sandstone with 4 30% porosity is 0.13 0.28, and that of poorly consolidated sandstone with 22 36% porosity is 0.31 0.34. The Poisson s ratio also increases as the porosity increases. Clearly, a high Poisson s ratio is related to high porosity and/or high clay content. [52] Eberhart-Phillips et al. [1989] estimated the relationship between a P wave velocity and an S wave velocity, porosity, clay content, and pressure using sandy samples at<60 MPa. They determined the relationship V p = 5.77 6.94f 1.73C 0.5 + 0.446(Pe e 16.7Pe ) and V s = 3.70 4.94f 1.57C 0.5 + 0.361(Pe e 16.7Pe ). V p, V s, f, C, and Pe are P wave velocity (km/s), S wave velocity (km/s), porosity, clay content, and effective pressure (kbar), respectively. [53] Assuming an average P wave velocity of 5.5 km/s for the southern part of the 5.4 5.6 km/s layer, and a Poisson s ratio of 0.33, the average S wave velocity is calculated to be 2.77 km/s. Adapting these formulas to the results of V p and V s measured in this study, the porosity and the clay content are estimated as unrealistic negative values and greater than 100% values, respectively. Either the P wave velocity is too fast or the S wave velocity is too slow for obtaining realistic porosity and clay content values. It is likely that these realistic values derive from the faulty assumption that the materials consist only of sandy rocks or that the pressure range of the laboratory measurements was too low to apply to our field observations. [54] According to Christensen [1996], high pressure only slightly increases the Poisson s ratio when is no change in composition. Consequently, we believe the cause of this high Poisson s ratio is not likely to be high porosity but high clay content. [55] Other parameters such as SiO 2 content affect Poisson s ratio. Christensen [1996] investigated trends in Poisson s ratio with SiO 2 content and found that rocks with 55

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-15 Figure 14. Comparisons the observed PPS travel time arrivals and synthetic ones. Abbreviations show interpretations of picked phases. For example, a PPSuc-sed means the PPS arrivals refracted from the island arc upper crust and converted at the top of the sedimentary wedge just beneath the OBS. A lc means the island arc lower crust. (a) OBS 3, (b) OBS 4, (c) OBS 5, (d) OBS 6, (e) OBS 7, (f ) OBS 8. 75% SiO 2 show a linear increase in Poisson s ratio with silica content. However, rocks with silica content between 40% and 55% show considerable scatter in Poisson s ratio. He concluded that the strong relationship between the Poisson s ratio and the degree of metamorphosis or temperature could not be seen. [56] According to Taira et al. [1989], the forearc region of southwestern Japan consists of sedimentary or metasedimentary rocks. Assuming that the southern part of unit B also consists of sedimentary materials or sedimentary metamorphic rocks, one of the likely candidates for materials of the southern part of unit B is a metamorphic argillaceous rock with high clay content and the relatively low porosity of slate or shale. P wave velocities of slate or shale are relatively fast and their silica contents are 55%. The Poisson s ratios of slate and shale are estimated as 0.30 and 0.31 0.32, respectively [Christensen, 1996; Johnston and Christensen, 1992]. Moreover, P wave velocities of these metamorphic rocks are faster than those of rocks with no metamorphism [Christensen, 1996]. It is suggested that the 5.4 5.6 km/s layer originates from the underplating of clayrich materials just beneath the decollement at a depth of 10 km and metamorphism of slate or shale. [57] It may be possible that a part of oceanic basalt from oceanic layer 2 is also underplated and included in the 5.4 5.6 km/s layer. One of the characteristics of the 5.4 5.6 km/ s layer with high Poisson s ratio is relatively high P wave velocity. Rocks with relatively high Poisson s ratio (more than 0.29) and high density (more than 2800 kg/m 3 )are likely to be the basalt and gabbro [Christensen, 1996]. Christensen et al. [1973] estimated the Poisson s ratio of oceanic basalt from the Pacific Ocean floor (density 2.87 Mg/m 3 ) between 60 and 200 MPa and showed that these Poisson s ratios are between 0.29 and 0.30 and the P wave velocities are between 6.1 and 6.2 km/s. If the oceanic basalt whose Poisson s ratio is 0.29 0.30 extends into this

ESE 2-16 TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE Figure 15. Comparison the P wave velocity structure deduced by this study and that of 1946 Nankaido earthquake rupture zone [after Kodaira et al., 2000] and their interpretations. Contour interval is 0.1 km/s. (a) 1946 Nankaido earthquake coseismic rupture area. (b) This study. layer, clay-rich metamorphic rocks whose Poisson s ratio is 0.30 0.32 are also needed. While the oceanic basalt fragment with a relatively fast P wave may exist in the southern part of unit B, it is small in volume. [58] To satisfy the high Poisson s ratio of the southern part of the 5.4 5.6 km/s layer, the existence of clay-rich metamorphic rocks is required. If accreted oceanic basalt is in the high Poisson s ratio layer, clay-rich metamorphic rocks seem to be needed. It is suggested that the main body of this part probably consists of clay-rich metamorphic rocks. It is inferred that the underplating and the metamorphism of the clay-rich sediments just below the decollement formed the southern part of the 5.4 5.6 km/s layer. 5.6. Relationship Between the Velocity Structure and the Coseismic Rupture Area [59] Kodaira et al. [2000] concluded that the upper limit of the 1946 Nankaido earthquake coseismic rupture zone was located within the contact zone between the oceanic crust and unit A, and they compared this interpretation with the upper limit proposed by Ando [1975]. Park et al. [2000] proposed that there are two fluid-filled splay faults from the contact zone between the oceanic crust and unit A, and that these splay faults are essentially the upper limit of the coseismic rupture zone. [60] As described in section 5.5, the differences of the velocity structure between the coseismic rupture area of the 1946 Nankaido earthquake and the nonrupture area can be seen in a thickness of unit A and the southern tip of unit B (Figure 14). The southern tip probably consists of metamorphic argillaceous rocks with high clay content, which comprises the underplated clay-rich materials just beneath the decollement. [61] The contact zone between the oceanic crust and unit B which we interpret to have a high clay content is longer than that of the 1946 Nankaido earthquake coseismic area and its depth is shallower than that of the coseismic area. It is suggested that the differences of the length and the depth of the contact zone result in the coupling between the subducting oceanic crust and the crust of the forearc region. If the area of the high-clay content layer of the nonrupture area is greater than that of the coseismic area, it may mean that the coupling of the nonrupture area is stronger than that of the coseismic area. The differences of the coupling between two crusts may explain why the 1946 Nankaido earthquake stopped. The slow slip of the 1946 Nankaido earthquake that propagated from east to west probably faced the strong coupling around the area Z or the western part of the area A and stopped there. [62] We suggest that the area Z does not have creep phenomena. The relationship between creep, ultraslow earthquakes and slow earthquakes has been discussed in relationship to the rupture process of great earthquakes [e.g., Nakayama and Takeo, 1997]. A weak seismic coupling could be due to the presence of unconsolidated or partly consolidated sediments between crusts of the arc side of the

TAKAHASHI ET AL.: STRUCTURE OF WEST NANKAI SEISMOGENIC ZONE ESE 2-17 Figure 16. Comparison of geologic information and the velocity structure. (a) Geologic map [after Omori et al., 1997]. Abbreviations are same as that shown in Figure 1. Solid parts represent melange units. Phrases in parentheses show the age of each layer. The northern Shimanto belt (NSB) consists of two layers whose ages are upper and lower Cretaceous, respectively. Rectangle is the 1946 Nankaido earthquake coseismic rupture zone as same in Figure 1. (b) A geological interpretation superimposed on the velocity structure. Numerals represent the P wave velocity (km/s). oceanic [e.g., Hartog and Schwartz, 1996]. At the Japan trench seismogenic zone area, it is known that the seismic coupling is <20% [e.g., Pacheco et al., 1993] and a lowvelocity zone exists between the subducting oceanic crust and the forearc crust of the northeastern Japan arc [e.g., Takahashi et al., 2000; Miura et al., 2000]. On the other hand, Ito et al. [1999] showed that an average back-slip rate of the western Nankai trough area is 4 cm/yr and that the seismic interplate coupling is strong. The low-velocity layer as observed in the Japan Trench seismogenic zone does not probably exist between two crusts in the Z area. If such a low-velocity layer is present, we should be observed reflected phases from the entire interface between the oceanic layer 2 and the southern part of unit B. The picked reflected phases from the top of the oceanic crust are likely distributed depending on only its reflectivity (Figure 12i).