Seismic Velocity Structure in the Crust and Upper Mantle beneath Northern Japan

Transcription

1 J. Phys. Earth, 42, , 1994 Seismic Velocity Structure in the Crust and Upper Mantle beneath Northern Japan Hiroki Miyamachi,l,* Minoru Kasahara,2 Sadaomi Suzuki,2,** Kazuo Tanaka,3 and Akira Hasegawa 4 1 Faculty of Science, Kagoshima University, Kagoshima 890, Japan Faculty of Science, Hokkaido University, Sapporo 060, Japan 2 Faculty of Science, Hirosaki University, Hirosaki 036, Japan 3 4 Faculty of Science, Tohoku University, Sendai 980, Japan We have applied an inverse method to P and S wave arrival time data observed at 52 seismic stations for 349 local earthquakes in order to estimate a three-dimensional (3-D) velocity structure beneath northern Japan. The method is characterized by a simultaneous determination of the two-dimensional depth distributions of velocity boundaries, 3-D velocity distribution and station corrections as well as hypocenter parameters of earthquakes. The Moho discontinuity under eastern Hokkaido is inclined from a depth of 32 km beneath the Hidaka Mountains to 20 km beneath the Konsen Plateau, while the thick crust with a thickness of km distributes widely in western Hokkaido and the Tohoku region. The dip direction of the Moho in eastern Hokkaido is approximately perpendicular to the range of the Hidaka Mountains. The upper boundary of the subducting Pacific plate distributes in the depth range from 50 to 170 km in the surveyed area and the dip angle of the plate boundary in eastern Hokkaido (the Kurile arc) is slightly larger than that in the Tohoku region (the northeastern Japan arc). Estimated P (S) wave velocities are km/s ( km/s) in the crust, km/s ( km/s) in the upper mantle, and km/s ( km/s) in the plate, respectively. The velocities in the crust beneath northern Hokkaido are lower than those beneath eastern Hokkaido and the Tohoku region. The upper mantle has velocities which gradually increase with depth, and the subducted plate contains a high velocity zone with a P wave velocity faster than 8.3 km/s. VP/VS ratios derived from P and S wave velocities generally range from 1.75 to 1.80 in the crust and upper mantle, while the ratios in the high velocity zone within the plate are less than These results suggest that the plate is composed of two layers: the first layer is a low-velocity and high-vp/vs zone with a thickness less than 20 km, and the second layer is a thick high-velocity and low VP/VS zone. The estimated plate thickness is about 90 km in the Kurile arc and about 110 km in the northeastern Japan arc. The relocated hypocenter distribution beneath Hokkaido clearly shows the double seismic plane in the subducting plate. The upper seismic plane is located in the first Received April 27, 1994; Accepted October 5, 1994 * To whom correspondence should be addressed. ** Present address: Faculty of Science, Kyushu University, Fukuoka 812, Japan. 269

2 270 H. Miyamachi et al. layer with a low velocity and the lower seismic plane is in the second layer with a high velocity. The seismic activity in the upper seismic plane is high in the depth range from 60 to 90 km, while that in the lower seismic plane is high at depths deeper than 100 km. The double seismic plane joins at depths from 90 to 130 km. 1. Introduction Three-dimensional (3-D) seismic wave velocity structures in Hokkaido have been investigated by various researchers using seismic arrival time data. Takanami (1982) and Miyamachi and Moriya (1984) applied the modified three-dimensional inverse method of Aki and Lee (1976) to the Hidaka Mountains region and discovered an inclined low velocity zone. They interpreted this zone as the descending crust beneath the mountains formed as a result of the collision of the Kurile and northeastern Japan arcs. They also found that seismicity is very active in the low velocity zone. Miyamachi and Moriya (1987) revealed a complex P wave velocity distribution down to a depth of 20 km in and around the aftershock region of the 1982 Urakawa-Oki Earthquake using the inverse method of Thurber (1983). Nakanishi (1985) applied a tomographic inverse method based on the Algebraic Reconstruction Technique of Herman (1980) to the Hokkaido-Tohoku region and obtained a P wave high velocity zone down to a depth of 120 km. However, these previous studies could not explicitly reveal the depth distributions of the Conrad, Moho and plate boundaries beneath the Hokkaido region. Horiuchi et al. (1982 a, b) revealed two-dimensional (2-D) depth distributions of the Conrad and Moho discontinuities beneath the central part of the Tohoku region using local seismic data. Sato (1981) showed a configuration of the subducting plate boundary together with a 3-D P and S wave velocity distribution by analyzing deep local earthquake data. Hasemi et al. (1984) and Obara et al. (1986) investigated a 3-D fine velocity structure beneath the whole Tohoku region. They showed that the descending Pacific plate is characterized by a high velocity and that the crust and upper mantle have low velocity zones beneath active volcanoes. Obara et al. (1986) also estimated an S wave velocity structure and obtained a spatial distribution of the VP/VS ratio. They found that there is an obvious relation between the VP/VS distribution and seismic activity in the double-planed deep seismic zone (Hasegawa et al., 1978): the lower plane seismic activity is not homogeneous in space and it is very low in the regions with VP/VS ratio larger than Sato et al. (1989) estimated depth distributions of the upper and lower planes of the subducting Pacific plate beneath the northern part of the Tohoku region from their 3-D P wave velocity structure. Zhao et al. (1992 a) and Zhao and Hasegawa (1993) developed a new "pseudo bending method" for calculation of the travel time and they derived a 3-D velocity structure beneath the Tohoku region and the whole Japan by using 2-D depth distributions of the Conrad and Moho discontinuities estimated by Zhao et al. (1990, 1992 b). Explosive seismic observations have been carried out in Hokkaido and the Tohoku region (Yoshii and Asano, 1972; Okada et al., 1973; Research Group for Explosion Seismology, 1986; Iwasaki et al., 1993). The observations provided P wave velocities in the upper and lower crusts as well as those in the uppermost mantle. In addition, J. Phys. Earth

3 Seismic Velocity Structure beneath Northern Japan 271 the depth distributions of the Conrad and Moho discontinuities along the seismic profiles were also derived. There are also several studies for structural analyses using converted waves (Okada, 1971; Nakanishi et al., 1981; Matsuzawa et al., 1986, 1988). For example, Matsuzawa et al. (1986) analyzed travel times of PS waves which are P-to-S converted waves at the plate boundary beneath the Tohoku region. They proposed a two-layered plate model composed of a thin low-velocity upper layer and a thick high-velocity lower layer. They also pointed out that most earthquakes in the upper seismic plane of the double seismic zone occur in the thin low-velocity layer. The velocity structure beneath the Tohoku region has been well investigated by various studies, as mentioned above. However, due to the scanty information about the velocity structure beneath Hokkaido, a problem of a structural relation between the Kurile arc and the northeastern Japan arc still remains to be solved. Moreover, the previous velocity structure analyses are either to estimate the depth distributions of the velocity boundaries under the assigned velocity distribution (Horiuchi et al., 1982 a; Zhao et al., 1990, 1992 b) or to solve the velocity distribution under the fixed boundaries (Takanami, 1982; Miyamachi and Moriya, 1984; Hasemi et al., 1984; Obara et al., 1986; Sato et al., 1989; Zhao et al., 1992 a; Zhao and Hasegawa, 1993). Recently, Miyamachi (1994) developed a method which simultaneously determines the velocity distribution and the boundary depth distribution using the local earthquake data. In this method, the P and S wave slowness distributions and the 2-D depth distributions of velocity boundaries are described using a power series of spatial coordinates. The first aim of this study was to determine a 3-D P and S wave velocity structure beneath northern Japan by applying the inverse method of Miyamachi (1994) to P and S wave arrival time data obtained from local earthquakes. The second was to investigate the relation between the velocity structure and seismic activity. 2. Method for Inverse Analysis We assume that the medium of the study area is a layered structure with some velocity boundaries. Then, according to Miyamachi (1994), the 2-D depth distribution of the boundary H(Įi,_??_i) may be expressed as a function of a power series of latitude and longitude at a given point (0Įi,_??_i): (1) where ľ and _??_ are the normalized latitude and longitude, bk are coefficients, NB is the number of terms of power series, and ock and Ĉk take values of 0, 1, 2, 3, c, espectively. A slowness distribution in each layer is also assumed to be modeled by a standard velocity depending only on depth and by a slowness perturbation described by the power series of latitude, longitude, and depth. The standard velocity distribution V(r) in each layer is expressed by the one-dimensional (1-D) velocity model, (2) Vol. 42, No. 4, 1994

4 272 H. Miyamachi et al. where R is the radius of the earth, H=H(ƒÆi,_??_i) is the depth of the boundary evaluated by Eq. (1), r is the distance measured from the center of the earth, ƒè is an arbitrary coefficient and V0 is the velocity at the uppermost depth of the layer. The slowness perturbation, ƒ s, at a given location with latitude ƒæi,, longitude _??_i and depth hi, in an assigned layer is given by (3) where ck are coefficients, the coefficients ƒ k, ƒàk, and ƒák take values 0, 1, 2, 3, c, Nv is the number of terms in the power series, and ƒæ, _??_, and h are the normalized latitude, longitude, and depth, respectively. From Eqs. (2) and (3), the slowness s(ƒæi, _??_i, hi) at a given point assigned by latitude, longitude and depth in each layer is expressed by (4) where ri=r-hi. In this model description and the given initial estimates of the velocity model and hypocenter parameters, we may write the observation equation for ith earthquake (i= 1, 2, c, NQ) and jth station (j= 1, 2, c, Ns) as (5) where ƒñij is the travel time residual, ÝTij/ Ýxik, ÝTij/ ÝckL, and ÝTij/ ÝbkL, are the partial derivatives of the travel times with respect to the hypocenter, velocity and boundary parameters; ƒ xik, ƒ ckl, and ƒ bkl are corrections for the parameters; and ƒ tj is a correction for the jth station. We construct the observation Eq. (5) toward the whole data set, and iteratively solve the whole system by using the damped least squares method (Aki and Lee, 1976). The details are referred to Miyamachi (1994). As described in Eq. (5), this inverse method is characterized by a simultaneous determination of depth distributions of the boundaries, P and S wave velocity distributions, station corrections for P and S waves and hypocenters, in contrast to the previous inverse methods such as Aki and Lee (1976) and Horiuchi et al. (1982 a). 3. Data 3.1 Arrival time data Figure 1 shows a regional map of the study area and locations of 52 seismic stations used in this study, of which 23 stations belong to Hokkaido University, 3 stations to Hirosaki University, 5 stations to Tohoku University and 21 stations to the Japan Meteorological Agency (JMA). The locations of -alp stations are listed in Appendix 1. From the earthquake catalogue during the period from 1986 to 1989 compiled by the Research Center for Earthquake Prediction (RCEP), Hokkaido University, we selected 329 local earthquakes based on the following criteria: (1) earthquakes are widely distributed in the whole study region, (2) focal depths of earthquakes are less than J. Phys. Earth

5 Seismic Velocity Structure beneath Northern Japan 273 (a) Regional map (b) Station distribution Fig. 1. (a) A regional map in Hokkaido and the northern part of the Tohoku district. The hatched area is the Japan Trench which is more than 7,000 m deep. (b) A map showing the station locations with the code names used in this study. Solid circles, solid triangles, solid squares, and open circles are the stations operated by Hokkaido University, Hirosaki University, Tohoku University, and the Japan Meteorological Agency (JMA), respectively. 250 km, and (3) earthquakes should be observed by more than 10 stations for P waves and more than 5 for S waves. Because we use the catalogue by RCEP to select the earthquakes, most of the earthquakes are distributed beneath Hokkaido. Therefore, in addition to the 329 earthquakes, we selected 20 earthquakes located in the Tohoku region from the Japan University Network Earthquake Catalogue (from 1986 to 1988). The epicenters of the selected earthquakes are shown in Fig. 2. P and S wave arrival time data for the university's seismic stations are read visually on an electric display using the digital waveform data. S wave data at only 18 stations among the university's stations are available for this study (see Fig. 2). On the other hand, for the JMA's stations, we collect P and S arrival time data with a phase rank "IP," "P," "IS," or "S" reported in JMA Bulletins. We also used arrival time data listed in the Japan University Network Earthquake Catalogue for the earthquakes selected from the catalogue. Finally, we collect about 8,400 P wave arrival time data and about 4,500 S wave arrival time data. We assign a relative weight to each arrival time, as 1.0, 0.8, 0.5 for P waves and 0.5, 0.2, 0.1 for S waves, according to the sharpness of the onset. 3.2 Initial model Before applying the inverse method described in Sec. 2, we must give some initial values to the model parameters. On the basis of the results in the previous studies, we Vol. 42, No. 4, 1994

6 274 H. Miyamachi et al. Fig. 2. A map showing the initial epicentral distribution of the earthquakes used in the inversion. Open circles, open squares, and open triangles indicate the epicenters of the earthquakes which are located in the crust, the upper mantle and the subducting Pacific plate, respectively. Solid circles and solid triangles are stations with both P and S wave data and only P wave data, respectively. Thick lines, A-A', B-B', C-C', D-D', E-E', and F-F' are the locations of the vertical sections. Earthquakes located in the hatched area including each line are plotted in the vertical section. constructed the initial 3-D P and S wave velocity structures. In the construction, we assume that the study area is divided into three layers by the Moho discontinuity and the plate boundary. We do not consider the Conrad discontinuity in the crust because of a small number of earthquakes occurring in the crust. We also consider only the upper plate boundary in the modeling. Explosion seismic studies have revealed that P wave velocities in the upper and lower crust and the uppermost mantle are almost km/s, km/s, and.7.5 km/s, respectively (Yoshii and Asano, 1972; Okada et al., 1973; Research Group for Explosion Seismology, 1986; Iwasaki et al., 1993). The velocity boundaries, the Conrad and Moho discontinuities, are also estimated to be located around 10 and 30 km deep in Hokkaido, and 18 and 30 km deep in the Tohoku region. We also refer to results obtained by Horiuchi et al. (1982 a, b), Sato (1981), Hasegawa et al. (1983), Zhao et al. (1990, 1992b), Takanami (1982), Miyamachi and Moriya (1984) and Nakanishi (1985). In the following, we show the initial values of the parameters for the inversion. We simply establish the standard P wave velocity (VP) distribution beneath the study area in the form of Eq. (2) to be J. Phys. Earth

7 Seismic Velocity Structure beneath Northern Japan 275 for the crust, for the upper mantle, and for the subducting plate, where R is the radius of the earth, HM(ƒÆ,ƒÓ) and HP(ƒÆ,ƒÓ) are the depths of the Moho discontinuity and the subducting plate boundary at a location (ƒæ,ƒó), respectively. S wave velocity (VS) is assigned so that the VP/VS ratio is 1.73 for all of the layers. he initial values for all the coefficients of the power series describing the T slowness perturbations for P and S waves are assigned to be zero. Considering the number of data and the available stations, we assume that P wave slowness perturbations in the crust, the upper mantle and the plate are described by power series of four orders (35 coefficients), four orders (35 coefficients) and three orders (20 coefficients), respectively. On the other hand, the power series of three orders (20 coefficients) are given for S wave slowness perturbations in all of the layers. Though the 2-D depth distribution of the Moho discontinuity, HM(ƒÆ,ƒÓ), is described by a power series of eight orders with 45 coefficients, the initial distribution is assumed to be a constant depth, namely, the coefficient b0 for the Moho is set to be 30 km deep and the other 44 coefficients are all zero. The initial 2-D depth distribution of the plate boundary, HP(ƒÆ,ƒÓ), is estimated from the results of Zhao et al. (1992 b) and the deep hypocenter distribution by Hasegawa et al. (1983). Assuming a power series with seven orders, we determine 36 coefficients bk of Eq. (1) using the classical least squares method. Under the initial velocity and boundary distributions mentioned above, we carried out a hypocenter determination using the classical least squares method to obtain the initial hypocenter parameters of 349 earthquakes used in the inversion. Figure 2 shows the initial epicentral distribution obtained. We simply adopt, as initial P and S wave station corrections, a mean value of the travel time residuals at each station derived from the hypocenter determination. We fix the station correction for P waves at MSN to be 0.0 s during the inversion. The number of unknown model parameters which should be estimated by the inversion is 321 : 150 for P and S wave slowness perturbations, 81 for the Moho and plate boundaries, 90 for the P and S wave station corrections. Moreover, there are the hypocenter parameters of the earthquakes used in the inversion (349 ~4=1,396). Therefore, the total of 1,717 parameters should be determined by 8,400 P wave data nd 4,500 S wave data. A damping parameter used in the damped least squares method a is fixed to be 0.10 during the inversion. Vol. 42, No. 4, 1994

8 276 H. Miyamachi et al. 4. Results of Inversion We have about 8,400 P wave arrival time data and 4,500 S wave arrival time data derived from 349 local earthquakes. However, the effective number of the arrival time data was reduced to about 7,600 for P waves and 4,200 for S waves during the inversion, because our algorithm for searching the ray paths between earthquakes and stations could not determine about 800 P wave ray paths and 300 S wave ray paths. The cause for this reduction will be discussed in Sec. 5. Figure 3 shows spatial distributions of intersection points of P and S wave ray paths at the Moho and plate boundaries. The numbers of the intersection points both for P and S wave ray paths at each boundary are about 10,400 and 7,800, respectively. From the distributions, we can intuitively recognize areas where the solutions are well resolved by the inversion. We expect that the Moho discontinuity can be estimated beneath a wide region in the seismic network and the plate boundary can be obtained beneath a region between the eastern part of the Hidaka Mountains and the Shimokita Peninsula. In Figs. 4 and 5, we also indicate the graphical distributions of the ray paths for P and S waves in the vertical sections along lines shown in Fig. 2. We can recognize that ray paths are passing through deep areas beneath the upper plate boundary. Accordingly, we can anticipate that the deep velocity distribution in the plate is obtained by the inversion as well as those in the crust and the mantle wedge. As shown in Fig. 6, the standard deviation of the travel time residuals conquers the criterion for convergence in the inversion after 6 iterations. In the initial model it is 0.53 s and has decreased to 0.38 s in the last one (the variance reduction is 52%). In (a) Moho Discontinuity (b) Plate Boundary Fig. 3. A distribution of intersection points between the P and S wave ray paths and the boundaries: (a) the Moho discontinuity (11,432 points) and (b) the upper plate boundary (7,833 points). Open circles are station locations. J. Phys. Earth

9 Seismic Velocity Structure beneath Northern Japan 277(a)(b)(c)(d)(e) Fig. 4. A graphical distribution of P wave ray paths in the vertical sections along the lines shown in Fig. 2: (a) A-A', (b) B -B', (c) C-C', (d) D-D', and (e) E-E'. Thick lines are the Moho and the upper plate boundary. the figure, the standard deviation at the first iteration becomes larger than that at the 0th iteration (the initial model). The cause for this undesirable change, as discussed in Miyamachi (1994), may be a discontinuous change of the coefficients of the power series during the inversion. In the following sections, we describe the result of each parameter obtained by the inversion. 4.1 Boundary depth distribution Figure 7 shows the depth distributions of the Moho and upper plate boundaries. The resolution for each coefficient of the power series is very poor (less than 0.1), but the standard error is small. The estimated coefficients are listed in Appendix 2. According to the results obtained by Miyamachi (1994), even if the resolution is poor, the boundary depth distribution calculated by the estimated coefficients is reliable in the well-resolved region beneath the seismic network. The standard errors in depth for the Vol. 42, No. 4, 1994

10 278 H. Miyamachi et al.(a)(b)(c)(d)(e) Fig.5. A graphical distribution of S wave ray paths in the vertical sections along the lines shown in Fig. 2: (a) A-A', (b) B-B', (c) C-C', (d) D-D', and (e) E-E'. Thick lines are the Moho and the upper plate boundary. Fig. 6. Change in the standard deviations of travel time residuals versus iterations in the inversion. J. Phys. Earth

11 Seismic Velocity Structure beneath Northern Japan 279 (a) Moho Discontinuity (b) Plate Boundary Fig. 7. A depth distribution of the boundary in km estimated by the inversion: (a) the Moho discontinuity and (b) the upper plate boundary. Solid circles are station locations. Arrows in the figures indicate the dip direction of the boundaries. Moho and plate boundaries calculated by the obtained coefficients range from }2 to 3 km and from }3 to }4 km in the seismic network, respectively. } The depth distribution of the Moho discontinuity is revealed as follows: (1) the Moho is 36 km deep in an area between the Shimokita Peninsula and off Urakawa, (2) the Moho is clearly inclined from a depth of 32 km beneath the Hidaka Mountains to 20 km beneath the Konsen Plateau, and the dip direction seems to be perpendicular to the NNW strike direction of the Hidaka Mountains, and (3) the Moho abruptly becomes shallow in an area of the Oshima Peninsula and the Tsugaru Peninsula. The crust with a thickness of more than 34 km is widely distributed beneath the Shakotan Peninsula, the Ishikari Plain, off Urakawa, and the Shimokita Peninsula. The plate boundary distributes from 50 to 180 km depth beneath the seismic network. We found that the configuration of the contours in the figure is basically parallel to the Japan Trench. The dip direction of the boundary is gradually changed: NNW at central and eastern Hokkaido, NW at western Hokkaido and W at southern Hokkaido and the northern part of the Tohoku region. The dip angle in central and eastern Hokkaido is larger than that in southern Hokkaido and the northern part of the Tohoku region. 4.2 P and S wave velocity distributions Coefficients for P and S wave slowness perturbations in the crust, the mantle wedge and the plate obtained by the inversion have resolutions ranging from 0.1 to 0.9, and Vol. 42, No. 4, 1994

12 280 H. Miyamachi et al. their standard errors are somewhat large. The values of the coefficients are listed in Appendix 3, We examined the standard velocities defined in Sec. 3. Because the first term of the power series for the slowness perturbation is not a function of depth, latitude, and longitude, this term can be considered to be a correction for the standard velocity. Accordingly, from the assigned standard velocity and the first term in the slowness perturbation, we can calculate the new standard velocity value for P (S) wave at the top of each layer, 6.0 (3.4) km/s at the surface, 7.6 (4.3) km/s at the Moho discontinuity and 8.1 (4.6) km/s at the plate boundary. On the basis of the assumption that an effect (a) 5 km depth (b) 20 km depth (c) 40 km depth Fig. 8. Contour maps showing the P wave (solid lines) and S wave (dashed lines) velocity distributions in km/s with the contour interval of 0.1 km/s: (a) 5 km depth, (b) 20 km depth, and (c) 40 km depth. Solid circles are station locations. J. Phys. Earth

13 Seismic Velocity Structure beneath Northern Japan 281 of the sub-surface layer is sufficiently removed by the station corrections, the new standard velocity of 6.0 km/s is considered to be a velocity in the basement underlying the sub-surface layer. The calculated P and S wave velocity distributions are shown in Fig. 8 at the specified depths of 5, 20, and 40 km. P (S) wave velocity ranges from 5.9 (3.4) to 6.3 (3.7) km/s at 5 km deep, and 6.6 (3.7) to 7.3 (4.2) km/s at 20 km deep. In the crust, the P wave velocity is clearly low beneath northern Hokkaido, and is slightly high in southern Hokkaido and the Tohoku region. It abruptly becomes high in eastern Hokkaido. A similar tendency is also found for S wave velocity. P (S) wave velocity at 40 km depth beneath the Moho discontinuity takes values from 7.4 (4.2) to 7.8 (4.5) km/s. It seems that the P and S wave velocities increase from the Japan Sea to the Pacific Ocean side in Hokkaido and the Tohoku region. Figures 9 and 10 represent the P and S wave velocity distributions in the upper mantle projected on the vertical sections whose horizontal locations are shown in Fig. (a)(b)(c)(d)(e) Fig. 9. P wave velocity distributions in the upper mantle and the plate in the vertical sections along the lines shown in Fig. 2: (a) A -A', (b) B-B ', (c) C -C', (d) D-D', and (e) E-E'. The contour interval is 0.1 km/s. Thick lines are the Moho and upper plate boundaries. Hatched circles are the hypocenters relocated in each section. Vol. 42, No. 4, 1994

14 282 H. Miyamachi et al.(a)(b)(c)(d)(e) Fig. 10. S wave velocity distributions in the upper mantle and the plate in the vertical sections along the lines shown in Fig. 2: (a) A-A', (b) B-B', (c) C-C', (d) D-D', and (e) E-E'. The contour interval is 0.1 km/s. Thick lines are the Moho and upper plate boundaries. Hatched circles are the hypocenters relocated in each section. 2. In the upper mantle, the P (S) wave velocity ranges from 7.4 (4.2) to 8.0 (4.5) km/s. The velocities gradually increase with depth in the well-resolved area. In the subducting plate, the P (S) velocity value is evaluated to be from 8.1 (4.6) to 8.6 (4.9) km/s. We find that the velocities in the plate are 2-5% higher than those in the overlying mantle wedge. These figures show that a pattern of the velocity distribution in the plate is obviously different from that in the mantle wedge. A high velocity zone with a P velocity from 8.3 to 8.4 km/s is distributed inside the plate, which is clearly seen in the three vertical sections for the P wave velocity distribution (C-C', D-D', and E-E'). Moreover, it seems that a relatively low velocity layer with a P wave velocity of 8.1 km/s is located just beneath the plate boundary and its thickness is less than 20 km. We can also recognize a similar pattern for the S wave velocity distributions. Though the resolution in this deep area is relatively poor, such a layered structure composed of the upper thin low velocity layer and the lower thick high velocity zone in the plate is probably the real image of the subducted plate. J. Phys. Earth

15 Seismic Velocity Structure beneath Northern Japan Station corrections Station, corrections for P and S waves at each station obtained by the inversion are graphically shown in Fig. 11 and listed in Appendix 1. These are relative values toward the fixed station correction for P waves, 0.0 s, at MSN. The resolution for the corrections ranges from 0.2 to 0.9, and the standard errors are less than 0.02 s. It is easy to see that the corrections for both P and S waves are distributed systematically in space. Large delayed corrections for both P and S waves are dominant along the coastal side of the Japan Sea in the Tohoku region and southern Hokkaido. On the other hand, the stations with advanced corrections are located along the coast of the Pacific Ocean in the Tohoku region and near the Shakotan Peninsula. In the area from the Ishikari Plain to Rumoi, the corrections are slightly delayed. The distribution of the corrections in the Hidaka Mountains area is complicated. In eastern Hokkaido, large advanced corrections for both P and S waves are distinct. We investigated a relation between the P and S wave station corrections. Figure 12 indicates that S wave arrivals at the stations with solid symbols are relatively faster than those at the stations with open symbols. We find that stations in eastern Hokkaido (open squares) have advanced P wave corrections and relatively delayed S wave corrections. At the stations located along the Pacific Ocean side of the Tohoku region, (a) P wave station correction (b) S wave station correction Fig. 11. A distribution of station corrections in seconds estimated by the inversion: (a) P wave station corrections and (b) S wave station corrections. A cross symbol in (a) is the location of MSN where the P wave station correction is fixed to be 0.0 s during the inversion. Solid circles mean delay; open circles advance. Vol. 42, No. 4, 1994

16 284 H. Miyamachi et al. Fig.12. Relation between P wave station corrections and S wave station corrections. Symbols are explained in Fig. 13. both P and S wave station corrections are advanced. We can divide the stations into two groups, Group-A and Group-B, denoted by the solid and open symbols. Group-A is composed of stations located in eastern Hokkaido and the western side of the Hidaka Mountains, and Group-B mainly consists of stations located in the eastern side of the Hidaka Mountains, southern and northern Hokkaido and the Tohoku district. Because a gradient of the line in the figure corresponds to the VP/VS ratio, the ratios in the subsurface velocity structure in Group-A and Group-B are 1.91 and 1.84, respectively. 4.4 Hypocenter distribution Earthquakes relocated in the inversion are plotted on the vertical sections in Figs. 9 and 10. Though the number of earthquakes occurring in the plate is not large, the double seismic plane is found, as previously pointed out by Hasegawa et al. (1978), Suzuki and Motoya (1981) and Hasegawa et al. (1983). Earthquakes occurring in the upper seismic plane are located on or just beneath the upper plate boundary and those in the lower seismic plane are located 30 to 40 km apart from the plate boundary. The seismic activity in the upper seismic plane is obviously high at depths shallower than 100 km. On the other hand, the activity in the lower seismic plane seems to be low at shallow depths and high at depths deeper than 100 km. These two seismic planes seem to be located within the upper low velocity layer and the lower high velocity layer in the plate, respectively. We notice that the upper seismic plane has a tendency to separate gradually from the plate boundary at depths more than 90 km, though the resolution for the deep position of the plate boundary is thought to be poor. The double seismic plane gradually becomes vague at depths of more than 130 km. It is interesting to compare the final hypocenters relocated in the inversion with the initial hypocenters determined by the initial velocity model. The relation between the final and initial hypocenters is shown on a horizontal plane in Fig. 13 and on vertical sections in Fig. 14. The mean difference in horizontal distance between the two hypocenters is 5.9 km. (the maximum difference is 20 km). There are some systematic features as follows: (1) the final hypocenters shift westward in the Tohoku region, (2) the shifting direction is NW in the Oshima Peninsula, (3) the direction is widely J. Phys. Earth

17 Seismic Velocity Structure beneath Northern Japan 285 Fig. 13. Relation in epicenter between the initial locations and the final locations (open circles) estimated by the inversion. Other symbols except the solid circles indicate station locations used in Fig. 12. distributed from west to north in the Hidaka Mountains and off Urakawa, but the horizontal distance is short, (4) the direction is from WNW to NNW in the Konsen Plateau, and (5) the direction is NNW in northern Hokkaido. These systematic differences in the shifting direction are thought to be caused mainly by the lateral velocity variation. We must note that the horizontal difference is generally large in the surrounding areas of the seismic network. The focal depths of the relocated hypocenters are, on the average, about 10 km shallower than the initial ones. In particular, the large difference in focal depth is obviously found for earthquakes occurring in the subducted plate. We suppose that two causes may give rise to the difference: one is the horizontal velocity variation and the other is the location of the plate boundary. Because the plate boundary obtained by the inversion is clearly shallower than the initial location, the latter seems to be the main reason for the difference. It is found that the differences in focal depth between the final and initial hypocenters of the earthquakes occurring in the crust are smaller than those in the upper mantle. 4.5 VP/VS distribution By using the P and S wave velocities obtained by the inversion, a distribution of VP/VS ratios is shown on horizontal planes in Fig. 15 and vertical planes in Fig. 14. It is found that the VP/VS ratio ranges from 1.70 to 1.75 at a depth of 5 km in the upper crust, and from 1.70 to 1.80 at 20 km in the lower crust. Areas with a high ratio (more than 1.75) in the upper crust distribute in northern Hokkaido, the Ishikari Plain, the Hidaka Mountains and the eastern coast of the Tohoku region. In the lower crust, the Vol. 42, No. 4, 1994

18 286 H. Miyamachi et al.(a)(b)(c)(d)(e) Fig. 14. A relation between the initial hypocenters and the final hypocenters (open circles) with the VP/VS distributions in the vertical sections along the lines shown in Fig. 2. A contour interval for the VP/VS distribution is Thick lines are the Moho and upper plate boundaries. Japan Sea side in the northern part of Hokkaido and the Ishikari Plain have a high ratio of 1.80, and the high ratio area extends to the northern part of the Kitakami Mountains through off Urakawa. In the upper mantle, the ratio takes values from 1.75 to From the vertical sections along A-A' and B-B', the area with a high ratio of 1.80 is found to be distributed near the plate boundary, but the high ratio area is diminished in the C-C' s ection, and seems to disappear in the D-D ' and E-E' sections. In the subducting plate, the distribution of the ratio seems to be complicated. Areas with a high ratio (more than 1.75) are distributed just along the plate boundary and zones with a relatively low ratio less than 1.75 are widely located inside the plate. Consequently, we have come to assume that these two zones correspond to the upper thin low velocity layer and the lower thick high velocity layer described in Subsec J. Phys. Earth

19 Seismic Velocity Structure beneath Northern Japan287 (a) 5 km depth (b) 20 km depth (c) 40 km depth Fig. 15. A map showing the distributions of the VP/VS ratios at depths of (a) 5 km, (b) 20 km, and (c) 40 km. A contour interval is Solid circles are station locations. 5. Discussion In this section, we compare the velocity structure obtained by the present inversion with previous seismological investigations. We also compare the velocity structure with earthquake occurrence, gravity data and geological structure. As mentioned in Sec. 4, the effective number of arrival time data is reduced during the inversion. Our algorithm for searching ray paths could not determine about 1,100 ray paths. As discussed in Miyamachi (1994), the determination of the ray trajectory between earthquakes and stations is based on the shooting method by Julian and Vol. 42, No. 4, 1994

20 288 H. Miyamachi et al. Gubbins (1977), and Snell's law is strictly applied when the ray paths cross the boundaries. Therefore, it sometimes happens that the ray paths can not be identified by our algorithm due to a discontinuous change of the ray trajectory. In addition, as shown in Fig. 7(a), the Moho discontinuity abruptly becomes shallow beneath the Konsen Plateau, and the Moho depth beneath areas around Nemuro is close to the earth surface. As a result, the effective number of data at two stations, NMR and NEM, is significantly diminished. 5.1 Crustal structure In Subsec. 4,2, we obtained the averaged P wave velocities of 6.0 km/s for the upper crust and 7.6 km/s for the upper.---most mantle. According to the results obtained by explosion seismic observations operated in Hokkaido and the Tohoku region, P wave velocities are 5.9 km/s for the upper crust, 6.6 km/s for the lower crust and 7.5 km/s (inland area) to 8.1 km/s (sea area) for the uppermost mantle (Yoshii and Asano, 1972; Okada et al., 1973, 1979; Research Group for Explosion Seismology, 1986; Iwasaki et al., 1993). The velocities obtained by the present study are well consistent with those by the previous explosion studies. In our result shown in Fig. 7, the Moho discontinuity with a depth more than 34 km is widely distributed in and around the Ishikari Plain, the eastern part of the Oshima Peninsula and the Shimokita Peninsula, and this deep discontinuity seems to extend to the Tohoku region. We also found that a dip direction of the Moho beneath the area from the Hidaka Mountains to the Konsen Plateau is perpendicular to the strike of the Hidaka Mountains range. Zhao et al. (1992 a) investigated the Conrad and Moho discontinuities beneath the whole of Japan by applying an inverse method to arrival time data obtained from local earthquakes. A comparison of our Moho distribution with that by them reveals two distinct differences: one is the deepest location of the Moho and the other is the dip direction of the Moho beneath the area from the Hidaka Mountains to the Konsen Plateau. In Zhao et al. (1992 a), the deepest Moho of 36 km is located on the east side of the Ishikari Plain, the Moho is shallow beneath the northern side of the Shimokita Peninsula, and the dip direction of the Moho seems to be perpendicular to the Japan Trench. Taking account of the number of data, the number of stations and the distribution of the intersection points on the Moho used in the inversion, we suppose that our depth distribution of the Moho is more reliable. Okada et al. (1973) estimated from explosion seismic data that the Moho discontinuity beneath the western area of the Ishikari Plain is about 30 km deep. Our estimation for the Moho discontinuity in this area is about 35 km deep, and could not be accurate because there are a few intersection points of the ray paths as shown in Fig. 3. In the Tohoku region, low velocity zones were detected beneath active volcanoes in the crust and upper mantle (Hasemi et al., 1984; Obara et al., 1986; Sato et al., 1989; Zhao et al., 1992; Zhao and Hasegawa, 1993). However, this was not confirmed in our study because of a different way to model the velocity distribution. Our model representation using power series can express a large-scale velocity distribution, but is weak in expressing local velocity variations. To detect such a relation between the low velocity zones and the volcanoes in Hokkaido, the velocity distribution should be expressed by blocks or grids. J. Phys. Earth

21 Seismic Velocity Structure beneath Northern Japan 289 According to the VP/VS distribution in the Tohoku region shown in Fig. 15, the ratio on the Pacific Ocean side is higher than that on the Japan Sea side, and this high VP/VS area extends to Hokkaido. On the other hand, the 3-D velocity structure beneath the Tohoku region obtained by Obara et al. (1986) showed the opposite result, that the VP/VS ratio in the crust is low along the Pacific coast. 5.2 Upper mantle and plate structures The previous studies using the inverse approach (Takanami, 1982; Miyamachi and Moriya, 1984; Hasemi et al., 1984; Nakanishi, 1985; Obara et al., 1986; Sato et al, 1989) have provided many 3-D deep velocity structures beneath Hokkaido and the Tohoku region. Takanami (1982) and Miyamachi and Moriya (1984) revealed a low velocity zone (LVZ) inclined at depths from 20 to 70 km beneath the Hidaka Mountains. They also suggested a possibility that the subducting Pacific plate is bent under the inclined LVZ. We can not explicitly confirm the inclined LVZ and the bent plate boundary from our results because of the different method used for the model representation. However, the maximum depth of the LVZ obtained by Miyamachi and Moriya (1984) coincides with the location of the upper plate boundary of the subducting Pacific plate estimated by the present study. Accordingly, it is possible that the LVZ and the plate interact with each other. Nakanishi (1985) showed the subducting plate with a velocity contrast of 4% higher than the surrounding upper mantle in a depth range from 60 to 100 km. Sato et al. (1989) revealed that the subducting Pacific plate is composed of a high velocity zone with P wave velocity 5% higher than the surrounding upper mantle in the northern part of the Tohoku region. These results of the high velocity zone in the plate are consistent with our model. Hasemi et al. (1984) and Sato et al. (1989) pointed out that the high velocity zones distribute discontinuously in the subducting plate. This discontinuous distribution of the high velocity zones may correspond to the velocity variation in the plate shown in Figs. 9 and 10. Obara et al. (1986) showed that the VP/VS ratio is low in the subducting plate beneath the Tohoku region. They also found that seismicity of the lower plane of the double seismic zone is locally high in areas with low VP/VS ratios and is low in areas with high ratios. As shown in Fig. 14, the ratio in the plate obtained in our study is lower than that in the surrounding upper mantle, which is in agreement with the findings of Obara et al. (1986). We speculate about thickness of the plate. Because our model does not include the lower plate boundary as an unknown parameter, we try to guess the depth distribution from our obtained velocity distribution. As presented in Subsec. 4.2, the P wave velocity just at the upper plate boundary is 8.1 km/s. We assume that the lower plate boundary coincides with a velocity contour of 8.1 km/s. Then, Fig. 16 shows the distributions of the upper (solid line) and lower (dashed line) plate boundaries along F-F' in Fig. 2. It is obvious that the plate is about 110 km thick in southwestern Hokkaido and 90 km thick in northeastern Hokkaido. We can recognize similar thickness distributions in Fig. 9. The plate in Hokkaido is thinner than that in the Tohoku region. In addition, it seems that the plate thickness may decrease in the deep region. This implies that there is only a weak velocity contrast between the subducting plate and the surrounding upper mantle in the deeper region. Moriya (1986) proposed an overlapped structure of Vol. 42, No. 4, 1994

22 290 H. Miyamachi et al. Fig. 16. A schematic model at the collision region along a line F-F' in Fig. 2 showing the relation among the northeastern Japan arc, the Kurile arc, and the subducting Pacific plate, based on the results of Miyamachi and Moriya (1984) and the present study. Large arrows in the figure indicate a collision between the northeastern Japan arc and the Kurile arc. Fig. 17. A schematic model showing the velocity structure inside the subducting plate. The hatched areas are the upper and lower seismic planes in the double seismic plane. the subducting plate from the hypocenter distribution, namely, the plate in the Kurile arc is overlapped by that in the northeastern Japan arc. However, we can not find such a structure in Moriya (1986) from our smoothed depth distribution of the upper plate boundary. A schematic model of the plate beneath Hokkaido is shown in Fig. 17. As mentioned in Subsec. 4.2, the plate may consist of at least two layers. These layers are characterized as follows: (1) the upper layer is a thin low velocity zone with a thickness less than 20 km in which the P (S) wave velocity is less than 8.2 (4.7) km/s with the high VP/VS ratio, and (2) the lower layer is a thick high velocity zone with P (S) wave velocity from 8.3 (4.7) to 8.4 (4.9) km/s and a VP/VS ratio less than Okada (1971) and Nakanishi J. Phys. Earth

23 Seismic Velocity Structure beneath Northern Japan 291 et al. (1981) pointed out that the existence of a low velocity layer intervening between the descending oceanic plate and the upper mantle is required to explain the large amplitude of the ScSp phase. Matsuzawa et al. (1986) also proposed from a travel time analysis of PS-converted wave that the subducting plate beneath the Tohoku region is composed of a thin low velocity upper layer whose thickness is less than 10 km and a thick high velocity lower layer. Though the thickness of the upper layer differs from that in our study, our plate model agrees with the result of Matsuzawa et al. (1986). 5.3 Velocity structure and hypocenter distribution We examined the relationship between the hypocenter distribution and the 3-D velocity structure. As shown in Fig. 9, the double seismic plane in the subducting plate is found in a depth range less than 90 km. In general, it seems that the seismic activity in the upper seismic plane is high in the shallow region and the activity in the lower plane becomes high in the deep region. It is clear that the upper seismic plane is included in the thin low velocity upper layer and the lower plane is located within the thick high velocity lower layer in the plate. In a depth range from 90 to 130 km, earthquakes have a tendency to occur in an area far from the upper plate boundary. We can not clearly discriminate the double seismic plane at depths deeper than 140 km. In other (a) P wave station corrections (b) S wave station corrections Fig. 18. Bouguer anomaly distributions of short wavelength with P and S wave station corrections. A contour interval is 10 mgal and thick solid lines are the 0-mgal contours. The thin solid lines (hatched areas) and the thin dotted lines indicate the positive and negative anomalies, respectively. Open and solid circles for the station corrections are explained in Fig. 11 (after Hagiwara (1967), slightly modified). Vol. 42, No. 4, 1994

24 292 H. Miyamachi et al. words, the double seismic plane seems to merge at depths from 90 to 140 km. Hasegawa et al. (1983) estimated the upper and lower seismic planes of the double seismic plane beneath the Hokkaido-Tohoku region from a spatial distribution of earthquakes precisely determined in a flat-layer velocity model. If their estimated upper seismic plane approximately coincides with the upper plate boundary in the shallow region, their upper plate boundary is systematically deeper than that of this study. This systematic difference in the depth distribution of the plate ranges from 5 to 20 km and is caused by a velocity model used in their hypocenter determination, which is different from our model. In an area deeper than 90 km depth, the difference conspicuously becomes large. 5.4 Relation with gravity data and geology Figure 18 shows the Bouguer anomaly distribution of short wavelength obtained by Hagiwara (1967) together with P and S wave station corrections obtained by our inversion. In general, the advanced and delayed station corrections for P waves correspond to the positive and negative Bouguer anomalies, respectively. A similar tendency is also found for S wave station corrections. The station corrections also closely correlate to surface geology in Hagiwara (1967), as shown in Fig. 19. Stations in the area of pre-neogene outcrops except the Hidaka Mountains have large advanced station corrections for P and S waves, and those in (a) P wave station corrections(b) S wave station corrections Fig. 19. A geological map with P and S wave station corrections (after Hagiwara (1967), slightly modified). Open and solid circles for the station corrections are explained in Fig. 11. J. Phys. Earth

25 Seismic Velocity Structure beneath Northern Japan 293 Neogene and Quaternary formations have delayed corrections. Stations in Mesozoic and Cenozoic volcanics have small advanced or delayed corrections. In the Hidaka Mountains composed of pre-neogene outcrops, the distribution of the advanced and delayed station corrections are complex. This suggests that the geology in the Hidaka Mountains is complicated. In the present modeling, the station correction is introduced as one of the unknown model parameters which explains the travel time due to the subsurface velocity structure just beneath the station. Therefore, it is reasonable that there is basically no correlation among the station corrections. However, as mentioned above, we found a distinct correlation among the distributions of the station corrections, the Bouguer anomalies and geology. This may imply that the station corrections obtained by this study include the travel time anomalies caused by two factors, namely the subsurface structure and the regional structure. 5.5 Tectonic view Our results clearly reveal that the crust is thick in western Hokkaido and the Tohoku region and thin in eastern Hokkaido. The boundary of this difference in the crustal thickness seems to be located along a line passing through the western foot of the Hidaka Mountains, the eastern side of the Ishikari Plain and the eastern side of the Ishikari Bay. On the basis of the location of this boundary, we interpret that western Hokkaido and eastern Hokkaido belong to the northeastern Japan arc and the Kurile arc, respectively. We estimate that the collision between the two arcs possibly takes place along the boundary. Miyamachi and Moriya (1984) found an inclined low velocity zone in a depth range from 20 to 70 km beneath the Hidaka Mountains. They interpreted that the low velocity zone is a crust of the northeastern Japan arc subducted into the upper mantle. As shown in Fig. 16, we suppose that the thin crust of the Kurile arc collides with the thick crust of the northeastern Japan arc, and the former is thrust up by the latter. As a result, the thick crust will be subducted into the upper mantle under the thin crust. The deepest point (70 km deep) of the low velocity zone estimated by Miyamachi and Moriya (1984) approximately corresponds to the location of the upper plate boundary at the collision region beneath the Hidaka Mountains: the head of the subducted crust will collide with the upper boundary of the subducting Pacific plate at a depth of about 65 km. In these complicated situations at the collision region, the subducted crust, must be under the influence of at least two forces: one force results from the collision between the northeastern Japan arc and the Kurile arc, and the other from the collision between the descending Pacific plate and the two arcs. Accordingly, a complicated stress field in the subducted crust is expected. In fact, high seismic activity and various types of focal mechanisms of earthquakes are observed in the area corresponding to the subducted crust (Suzuki et al., 1983; Umino et al., 1984). Chapman and Solomon (1976) proposed that the North American, Eurasian, and Pacific plates collide with each other and form a so-called triple junction beneath the present study area. On the other hand, Nakamura (1983) pointed out a possible nascent trench along the eastern Japan Sea as the convergent boundary between the Eurasian and North American plates. Our results indicate that the structural boundary between the Kurile and northeastern Japan arcs exists along the west coast of the Hidaka Vol. 42, No. 4, 1994