Asperity map along the subduction zone in northeastern Japan inferred from regional seismic data

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 109,, doi:10.1029/2003jb002683, 2004 Asperity map along the subduction zone in northeastern Japan inferred from regional seismic data Yoshiko Yamanaka and Masayuki Kikuchi 1 Earthquake Research Institute, University of Tokyo, Tokyo, Japan Received 10 July 2003; revised 30 January 2004; accepted 9 March 2004; published 17 July 2004. [1] In an attempt to examine the characteristic behavior of asperities, we studied the source processes of large interplate earthquakes offshore of the Tohoku district, northeastern Japan, over the past 70 years. In this area, earthquakes of M7 class have a recurrence interval of about 30 years. Seismic observation using a strong-motion seismometer has been carried out by the Japan Meteorological Agency since the beginning of the 1900s. We collected these seismograms in order to make a waveform inversion. On the basis of the derived heterogeneous fault slip, we identified large slip areas (asperities) for eight earthquakes which occurred after 1930, and we constructed an asperity map. The typical size of individual asperities in northeastern Japan is M7 class, and an M8 class earthquake can be caused when several asperities are synchronized. We propose that the patterns of asperity distribution beneath offshore Tohoku fall into three different categories. In the northern part (40 41.3 N) the seismic coupling in the asperity is almost 100%, and the size is large. In the central part (39 40 N), little seismic moment has been released by large earthquakes, and the asperity size is small. In the southern part (37.8 39 N) the seismic coupling is medium. The weak seismic coupling may be related to submarine topographical features and to the sediment and water along the subducting plate. Our results also suggest a general tendency for the asperities to be located away from the hypocenters (initial break), with aftershocks occurring in the area surrounding the asperity. INDEX TERMS: 7215 Seismology: Earthquake parameters; 7209 Seismology: Earthquake dynamics and mechanics; 7230 Seismology: Seismicity and seismotectonics; 8150 Tectonophysics: Plate boundary general (3040); KEYWORDS: asperity, source process, historical seismograms Citation: Yamanaka, Y., and M. Kikuchi (2004), Asperity map along the subduction zone in northeastern Japan inferred from regional seismic data, J. Geophys. Res., 109,, doi:10.1029/2003jb002683. 1. Introduction [2] The asperity model was proposed in the 1980s based on analysis of the detailed source rupture process [Lay and Kanamori, 1980; Beck and Ruff, 1987; Kikuchi and Fukao, 1985]. Lay and Kanamori [1980] named the area where a large seismic moment is released in a great earthquake an asperity. Lay and Kanamori [1981] examined the asperity size of the circum-pacific subduction zones and found that the size of an asperity is characteristic of its region and can be largely divided into four types, the Chile, Aleutian, Kurile, and Mariana types. Ruff and Kanamori [1983] suggested that the asperity size is the principal control on the eventual earthquake size. Ruff and Kanamori [1980] also proposed that earthquake size is correlated with the age of the subducting plate and the convergence velocity. [3] In northeastern Japan, M7 class earthquakes have repeatedly occurred with an interval of about 30 years. The rupture area of each event was often estimated from 1 Deceased October 2003. Copyright 2004 by the American Geophysical Union. 0148-0227/04/2003JB002683$09.00 the aftershock distribution [Umino et al., 1990] and/or the tsunami source area [Hatori, 1974, 1978, 1996]. Some of these earthquakes repeatedly occurred in the same area over several decades. For example, the 1994 Sanriku-oki earthquake (M w = 7.7) occurred within the source area of the 1968 Tokachi-oki earthquake (M w = 8.2). The source process of the 1994 Sanriku-oki earthquake was investigated by a variety of researchers [Nakayama and Takeo, 1997; Tanioka et al., 1996]. Nakayama and Takeo [1997] analyzed the rupture process of the 1994 Sanriku-oki earthquake using the strong-motion data records from a regional network and compared their result with the moment release distribution of the 1968 Tokachi-oki earthquake obtained by Kikuchi and Fukao [1985]. Nakayama and Takeo [1997] pointed out that the location of a part of the large slip area agrees for the two earthquakes. However, Kikuchi and Fukao [1985] used teleseismic seismograms only, so that the resolution of the spatial distribution of asperity for the Tokachi-oki event was too poor to judge the relationship between the two earthquakes. An analyses based on tsunami records record and GPS displacement data suggests that different asperities were ruptured during these two events [Tanioka et al., 1996]. Nagai et al. [2001] investigated the source processes and aftershock distribu- 1of14

Table 1. List of Earthquakes Analyzed in This Study a Date Time, LT Longitude, E Latitude, N Depth, km M J M t M s M CMT 9 March 1931 1249:00.0 143.233 40.191 20.0 7.6 7.2 7.8 3 Nov. 1936 0546:00.0 142.065 38.260 61.0 7.5 7.0 7.2 27 July 1937 0456:00.0 142.050 38.283 40.0 7.1 21 March 1960 0207:26.1 143.433 39.833 0.0 7.2 7.5 7.7 16 May 1968 0948:53.0 143.583 40.733 0.0 7.9 8.2 8.1 12 June 1968 2241:42.8 143.133 39.417 0.0 7.2 7.4 7.3 12 June 1978 1714:25.4 142.167 38.150 40.0 7.4 7.4 7.5 7.6 19 Jan. 1981 0317:23.9 142.967 38.600 0.0 7.0 7.0 7.0 7.0 2 Nov. 1989 0325:33.5 143.057 39.855 0.0 7.1 7.5 7.4 7.4 28 Dec. 1994 2119:20.9 143.748 40.427 0.0 7.5 7.7 7.5 7.7 a Each hypocenter was determined by the Japan Meteorological Agency (JMA). M J, JMA magnitude; M t, tsunami magnitude determined by Abe [1981, 1999]; M s, surface wave magnitude determined by the U.S. Geological Survey; M CMT, M w determined by Harvard University. tions of these earthquakes by using the same method and similar data set (the strong-motion records at a regional network and teleseismic body waves at global networks), and concluded that the 1968 earthquake consists of more than two asperities and the 1994 Sanriku-oki earthquake was caused by one of the 1968 asperities. These examples show that in order to discuss the location of the asperity, it is important to use the same technique and similar regional data for all of the earthquakes under concern. [4] In this paper we examine the heterogeneous seismic slips for large-thrust earthquakes in northeastern Japan and assemble a map of major asperities. The existence of an asperity controls not only the earthquake rupture pattern but also the seismicity or slip rate pattern on the plate boundary. Determining the locations of asperities is very useful for assessing the hazard of future large earthquakes. 2. Data Preparation [5] According to the Japan Meteorological Agency (JMA) bulletin and the Utsu historical catalog, 19 large subduction zone earthquakes with magnitudes of 7.0 or greater occurred after 1900 in the northeastern Japan region (38.0 41.0 N, 141.0 144.0 E). We collected strong-motion seismograms for these events from the JMA. Seismic observation by strong-motion seismometers has been carried out by JMA since the beginning of the 1900s. Table 1 lists these earthquakes. The 1968 Tokachi-oki and the 1994 Sanriku-oki earthquakes have already been analyzed by a technique similar to ours [Nagai et al., 2001]. The hypocenters of the earthquakes analyzed in this study are shown by stars in Figure 1. We used hypocenter information revised by the JMA (N. Hamada, personal communication, 2003). The 1968 Tokachi-oki earthquake (M w = 8.2) was followed by a large aftershock with M = 7.2. These events are referred to as 1968M and 1968A, respectively. [6] The waveforms before the 1970s are written on smoked paper (Figure 2b), and later in ink. Recently the JMA created a microfilm archive of historical seismograms. We used these data to retrieve the waveform records, but sometimes we had to use the original seismograms because of the poor resolution of the microfilm data. The analog records are scanned to be converted into raster data, digitized on a personal computer screen, and saved as vector data. [7] Since the data were drawn in pen by an arm about 30 cm in length (L), a waveform with large amplitude reveals a circular arc. We corrected this arc effect by the following method [Kikuchi et al., 1999]. Figure 2a schematically illustrates seismogram drawn with a finite-length pen lever. a denotes the biased inclination, q the actual oscillation, and V the chart speed of recording paper. Correction was made for the seismogram so that the coordinate system (x, y) is transformed into (q, t). The relation between these two coordinate systems is written as x ¼ Vt L cosðq þ aþþl cos a y ¼ L sinðq þ aþþl sin a; Figure 1. Epicenters of large earthquakes with M J 7.0. Solid stars show the earthquakes analyzed in this study. 2of14

after the transformation of the coordinate systems, while A(t) is the amplitude of displacement wave at time t. Finally, we applied a band-pass filter between 0.5 and 0.02 Hz. [8] Various types of seismographs had been used before the JMA 50-, 51-, and 52-type strong-motion seismometers were installed. In order to calculate synthetic waveforms, we have to know the instrumental constants of these seismographs and the paper feeder speed. This information is usually found on seismograms. For some stations, however, we could not find information on the instrumental constants. In such cases, we substituted the standard value given in the literature [Hamamatsu, 1966]. Sometimes no time mark was recorded on the seismograms. In such cases, we referred to the paper feeder speed given in the above literature or estimated the value based on S-P time. The instrumental parameters are listed in Table 2. In order to reduce the effect of any unknown underground structures, we only used the seismograms taken at shorter epicentral distances. 3. Inversion Method [9] We used the waveform inversion method developed by Kikuchi et al. [2003]. The hypocenter determined by the JMA was adopted as the initial break of the rupture. We assumed a fault plane based on the focal mechanism derived from the P wave initial motion or waveform inversion, if such a solution was found in literature. Then, we took N M grid points equally spaced along the strike and the dip, respectively. The spacing was varied from 5 to 20 km depending on the size of the earthquake. We first assumed a relatively large fault so that less constraint were imposed on the extent of rupture. After the determination of rupture extent, we assume a smaller fault and smaller grid size to represent the slip distribution appropriately with a relatively small number of model parameters. The synthetic waveform at the jth station was given in the following form: y j ðþ¼ t X n X X X D nmlk G nmkj ðt t nm ðl 1ÞtÞ; m l k Figure 2. (a) Schematic of a seismogram drawn with a finite length of pen lever, L. The variable a denotes the biased inclination, q denotes the actual oscillation, and V denotes the chart speed of the recording paper. (b) Original Japan Meteorological Agency (JMA) strong motion seismogram of the 1968M Tokachi-oki earthquake written on smoked paper at Morioka (MRK). (c) Digitized waveform after correcting for the arc effect. where the origin of systems (x, y) is assumed to correspond to that of the system (2, q). The temporal change of displacement is then described as At ðþ¼lqðþ t where G nmkj is the Green s function due to the source element with the fault slips toward the k direction (k =1,2) at the (n, m) grid point, t nm is the start time of the slip, t is the interval of the time step, and D nmlk is the unknown model parameter at the lth time step. We assumed a maximum rupture velocity, V, and allowed each grid point (n, m) to have a fault motion during the time interval between t nm = R nm /V to t nm + l * t, where R nm is the distance from the initial break to the (n, m) grid point. Thus rupture is not necessarily restricted to a circular form. Green s functions were calculated using the reflectivity method [Koketsu, 1985] assuming to the velocity structure obtained by Fujie [1999]. [10] D nmlk was determined by the least squares method with two constraints. Taking the components of the slip vector in two directions, (l 0 +45 ) and (l 0 45 ), where l 0 was the assumed slip angle, we imposed the positivity constraint as well as the smoothness constraint of the slip distribution following the method of Yoshida [1989]. The degree of smoothness is defined by a digital Laplacian: r 2 D n;m ¼ 4D n;m D n 1;m D nþ1;m D n;m 1 D n;mþ1 ; where we omitted the subscript lk of D nmlk. The objective function that should be minimized was given as follows: D ¼ X j Z 2dt w j x j ðþ y t j ðþ t X þ b 2 n X r 2 D m n;m 2¼ min; 3of14

Table 2. List of Station and Instrumental Parameters Station Code Distance, deg Azimuth, deg Seismograph a Components (T0, V) b 9 March 1931 Sapporo SAP 227.9 335.4 I NS(3.1,2),EW(2.2,2) Yamagata YAM 375.6 210.0 I NS(5.0,2),EW(5.0,2) Akita AKI 271.7 O NE(3.5,2),EW(3.5,2) 3 Nov. 1936 Tohoku Univ. THK 108.4 276.9 I NS(10.0,2),EW(10.0,2) Utsunomiya UTS 247.5 315.4 CMO NS(6.0,2),EW(6.0,2),UD(4.9,5) 27 July 1937 Tohoku Univ. THK 105.1 268.3 I NS(10.0,2),EW(10.0,2) Utsunomiya UTS 272.7 CMO NS(6.0,2),EW(6.0,2),UD(4.9,5) 21 March 1960 Miyako MIY 126.6 261.1 51 NS(5.9,1),EW(5.9,1),UD(5.4,1) Hachinohe MAC 179.2 296.2 51 NS(6.0,1),EW(6.0,1),UD(4.8,1) Morioka MRK 193.8 266.3 52 NS(6.0,1),EW(5.9,1),UD(5.0,1) Ishinomaki ISN 240.9 230.3 50 NS(6.0,1),EW(6.0,1),UD(5.0,1)) 12 June 1968 (1968A) Miyako MIY 103.0 284.6 51 NS(5.9,1),EW(5.9,1),UD(5.4,1) Ofunato OFU 128.0 252.5 52B NS(6.0,1),EW(6.0,1),UD(5.9,1) Hachinohe HAC 184.1 312.5 51 NS(6.0,1),EW(6.0,1),UD(4.8,1) Ishinomaki ISN 192.9 235.7 50 NS(6.0,1),EW(6.0,1),UD(5.0,1) 12 June 1978 Ishinomaki ISN 81.3 292.4 50 NS(6.0,1),EW(6.0,1),UD(5.0,1) Ofunato OFU 110.3 52B NS(6.0,1),EW(6.0,1),UD(5.9,1) Sendai SEN 111.4 276.7 50 NS(5.8,1),EW(5.9,1),UD(5.0,1) Fukushima FKS 159.1 51 NS(5.8,1),EW(6.0,1),UD(4.9,1) Yamagata YAM 159.1 274.7 52 NS(5.9,1),EW(5.9,1),UD(4.9,1) Miyako MIY 167.1 354.2 51 NS(5.9,1),EW(5.9,1),UD(5.4,1) Onahama ONA 177.0 51 NS(6.0,1),EW(6.0,1),UD(4.9,1) Morioka MRK 192.4 333.6 52 NS(6.0,1),EW(5.9,1),UD(5.0,1) 19 Jan. 1981 Miyako MIY 144.6 323.8 51 NS(5.9,1),EW(5.9,1),UD(5.4,1) Ishinomaki ISN 146.0 262.9 50 NS(6.0,1),EW(6.0,1),UD(5.0,1) Sendai SEN 183.9 258.8 50 NS(5.8,1),EW(5.9,1),UD(5.0,1) Morioka MRK 197.2 308.7 52 NS(6.0,1),EW(5.9,1),UD(5.0,1) 2 Nov. 1989 Miyako MIY 95.9 256.3 51 NS(5.9,1),EW(5.9,1),UD(5.4,1) Ofunato OFU 144.8 232.9 52B NS(6.0,1),EW(6.0,1),UD(5.9,1) Hachinohe HAC 149.8 300.3 51 NS(5.9,1),EW(6.0,1),UD(5.1,1) Morioka MRK 162.3 52 NS(6.0,1),EW(5.9,1),UD(5.0,1) Ishinomaki ISN 219.4 50 NS(6.0,1),EW(6.0,1),UD(5.0,1) a I, Imamura strong motion; O, Omori strong motion; CMO, Central Meteorological Observation strong motion. b T0, natural period of strong motion seismogram; V, static magnification. where w j (>0) is the relative weight factor for the jth record, x j (t) is the jth observed data, and b is the degree of smoothness. It is shown that a large value of b makes the solution smoother [Kikuchi et al., 2003]. We assumed b = 0.2 in this study. 4. An Example: The 1978 Miyagi-oki Earthquake [11] As an example, we here describe the 1978 Miyagi-oki earthquake in some detail. On 12 June 1978, an earthquake with magnitude 7.4 (determined by JMA) occurred off the coast of Miyagi, northeastern Japan (Figure 3). The hypocentral parameters given by JMA are origin time = 1714:25.3 LT, epicenter = (38.150 N, 142.167 E), depth = 40.0 km. This event caused extensive damage to the Miyagi prefecture. Considering the distribution of azimuthal coverage, we selected eight strong motion records, Ishinomaki (ISN), Ofunato (OFU), Sendai (SEN), Fukushima (FUK), Yamagata (YAM), Miyako (MIY), Onahama (ONA), and Morioka (MRK). All the digitized records, which were corrected for arm effect and band passed between 0.5 Hz and 0.02 Hz, are shown in Figure 3. Seismograms at four stations with a hypocentral distance of less than 160 km are clipped at the arrival of the S wave. [12] In the inversion, we assumed the mechanism used by Seno et al. [1980]. They used the first motion data and the surface-wave analysis and obtained: strike = N200 E, dip = 20. We found that it is most appropriate to assume 7 9 grid points with a spacing of 10 km on the fixed planar fault. Adopting the epicenter by JMA, the depth of the initial break was assumed to be 37 km based on the depth of the subducting Pacific plate, although the hypocentral depth determined by JMA was 40 km. [13] The source time function and the distribution of the moment release are shown in Figure 4. A comparison between the observed and synthetic waveforms is also 4of14

Figure 3. Digitized waveforms of the low-gain strongmotion seismogram from the 1978 Miyagi-oki earthquake. The star indicates the epicenter determined by JMA. shown in Figure 4. The waveform match was satisfactorily good at almost all the stations. At Fukushima (FKS), Sendai (SEN) and Onahama (ONA), the amplitudes of the synthetic seismograms, especially the horizontal components, could not be well reconstructed. This may be ascribed to the heterogeneity of the underground structure. [14] The main rupture propagated toward the northwest. The total source time was 30 s. The seismic moment was 2.3 10 20 Nm, and the moment magnitude M w was 7.5. Figure 5 indicates the aftershock distribution during the first month following the main shock as determined by JMA. The slip distribution obtained in this study is shown by the slip contour. The maximum dislocation was 2.3 m. We defined the region where the seismic slip is greater than half the value of the largest slip as an asperity. The aftershock distribution is largely divided into two groups, A and B, in Figure 5. We can see that most of the aftershocks occurred in the area surrounding the asperities. As discussed below, this feature has been seen in many other earthquakes. [15] In 1936 an earthquake of magnitude 7.5 occurred close to the site of the 1978 earthquake. The derived asperity (gray contour in Figure 5; for more details, see Appendix A) is located on the south side of the 1978 asperity. As shown in Figure 8, the aftershocks of the 1936 earthquake spread out to the south beyond the after- Figure 4. Final solution of the 1978 Miyagi-oki earthquake: (a) focal mechanism; (b) moment-rate function; (c) slip distribution on the fault plane; and (d) comparison of the observed (upper) and synthetic (lower) waveforms. The number above the station code is the peak-to-peak amplitude in mm ofthe observed displacement waveform. 5of14

Figure 5. Map view of the fault slip and aftershock distribution for 1 month following the 1978 Miyagi-oki earthquake. Black and gray stars show the hypocenters of the earthquakes of 1978 and 1936, respectively. The black contour line shows the moment release distribution obtained in this study. We painted the area within the value of half the maximum slip as an asperity. The 1978 aftershock distribution is largely divided into two groups, A and B. The aftershocks mainly occurred in the area surrounding the asperity. The gray contour line indicates the moment release distribution of the 1936 event. The initial breaks of these two events are close together, but the ruptures proceeded in different directions, and the asperities are different. shocks of the 1978 event, although the hypocenter determination was not very good. [16] The 1936 and 1978 earthquakes produced tsunamis. Hatori [1974, 1978] estimated the source area of these earthquakes using the tsunami data. Though the source areas of the tsunami are larger than our asperities, the relative locations are consistent with both results. 5. Asperity Map Along the Subduction Zone in Northeastern Japan [17] Figure 6 shows the hypocenters (stars) and slip distributions (contour lines) for the eight earthquakes analyzed in this study. Each earthquake is distinguished by color. The interval of the contour lines is 0.5 m. For earthquakes which have a maximum slip of less than 0.5 m, only the hypocenters are plotted in Figure 6. The slip distributions of the 1968M Tokachi-oki earthquake and the 1994 Sanriku-oki earthquake obtained by Nagai et al. [2001] are also plotted. Since their data and method of analysis are quite similar to ours, their results may be directly compared with ours. The source parameters obtained for these earthquakes are summarized in Table 3. As mentioned above, we defined the region in which slip was greater than or equal to half of the maximum slip as an asperity. [18] As shown in Figure 6, seismic slip on some asperities occurred more than once. The 1968M Tokachi-oki earthquake consisted of at least three asperities, namely A, B, and C in Figure 6. Asperity B again generated the 1994 Sanriku-oki earthquake [Nagai et al., 2001]. It seems likely that the 1931 earthquake was associated with the same asperity. On the other hand, asperity A is the largest one, and apparently has a longer repetition time. Asperity C generated large earthquakes three times: 1960, 1968M, and 1989, although the size of this asperity is small compared with the other asperities. This evidence suggests that asperities are fixed in space. [19] For hazard assessment of a future large earthquake, it is very important to judge whether or not the 1936 and 1978 earthquakes are associated with the same asperity. The present analysis indicates that the rupture of the 1978 earthquake propagated toward the west, while that of the 1936 earthquake propagated toward the south. Although there were only two observation stations for the 1936 earthquake, it is clear from the P wave polarity at THK that the rupture propagated toward the south. [20] Figure 7 shows the slip distribution in comparison with tsunami source areas for individual events obtained by Hatori [1974, 1978, 1996]. His tsunami source areas in general tend to be wider and show considerable overlap. It can be seen from Figure 7 that the tsunami source area and the asperity are mostly in agreement in the southern region. In the northern region, on the other hand, the tsunami source area seems to be located seaward of the asperity. For an earthquake located near the trench, the dislocation in the shallower part can more effectively contribute to the sea bottom movement. 6. Discussion 6.1. Characteristics of the Rupture Pattern [21] The main feature of the northern part of the Tohoku district is that each asperity is of class 7 7.5 magnitude, but more than one adjacent asperity may be synchronized to generate an earthquake of class 8 magnitude. The 1968M Tokachi-oki earthquake was a multiple asperity event, while the 1994 Sanriku-oki earthquake was a single asperity event. [22] Figure 8 shows the aftershock distribution in the first month for all of the earthquakes that we analyzed. Aftershocks generally occur in the area surrounding the asperity, while asperities tend to be located in areas with fewer aftershocks. Such a relationship between a main shock and aftershocks has already been pointed out by several researchers [Mendoza and Hartzell, 1988; Houston and Engdahl, 1989]. With respect to the recurrence of large earthquakes, the distribution of aftershocks around an asperity seems to be invariable with respect to the location of the asperity. As an example, the location of aftershocks for the 1968M Tokachi-oki and the 1994 Sanriku-oki earthquakes [see Nagai et al., 2001, Figure 13] is compared with the slip distributions of these earthquakes in Figure 8d. As stated in the previous section, asperity B was ruptured when the 1968M and 1994 events occurred. It is clearly observed in Figure 8d that the spatial distribution of aftershocks near asperity B is quite similar for the two events. Similarity in the aftershock distribution is also obtained for the 1960 (Figure 8c) and 1989 (Figure 8h) events; each of the events ruptured asperity C. Thus if the distribution of aftershocks is obtained accurately, the location of the asperity can be estimated from the area with fewer aftershocks. 6of14

Figure 6. Asperity map along the subduction zone in northeastern Japan. Stars show the main shock epicenters. Contour lines show the moment release distribution. The interval of the contour lines is 0.5 m. Each earthquake is distinguished by color. We painted the area within the value of half the maximum slip as an asperity. See color version of this figure at back of this issue. Table 3. List of Source Parameters Obtained in This Study a Date (LT) h 0 M w Strike Dip Slip M o T s D max D a S 9 March 1931 (1249) 20.0 7.3 156 20 35 0.9 30 0.9 0.7 48 3 Nov. 1936 (0546) 25.0 7.5 200 20 81 2.2 16 2.7 2.0 17 27 July 1937 (0456) 30.0 7.2 200 20 81 0.9 25 0.4 0.3 52 21 March 1960 (0207:26.1) 15.0 7.3 184 15 60 1.0 30 1.6 1.1 17 16 May 1968 (0948:53.0) 9.0 8.3 156 20 38 35.0 90 9.3 Asperity A 9.3 6.2 32 Asperity B 6.5 4.3 24 Asperity C 3.0 2.6 8 12 June 1968 (2241:42.8) 20.0 7.0 184 15 64 0.4 20 0.6 0.4 20 12 June 1978 (1714:25.4) 37.0 7.5 200 20 95 2.3 30 2.3 1.6 10 19 Jan. 1981 (0317:23.9) 10.0 7.1 186 15 56 0.6 20 0.6 0.4 27 2 Nov. 1989 (0325:33.5) 10.0 7.0 183 20 64 0.4 25 0.8 0.5 21 28 Dec. 1994 (2119:20.9) 10.0 7.7 184 15 70 4.4 60 4.0 2.4 18 a h 0, hypocenter depth (km); M w, moment magnitude; M o, seismic moment (10 20 Nm); T s, rupture time (s); D max, maximum seismic slip (m); D a, averaged seismic slip (m); S, rupture area (10 2 km 2 ). 7of14

Figure 7. Comparison between asperities and tsunami source areas obtained by Hatori [1974, 1978, 1996]. [23] It is also noteworthy that the rupture initiation point is significantly far from the asperity, and yet the correspondence between the initial break and the asperity is not invariant. For example, the initial rupture points of the 1936 and the 1978 Miyagi-oki earthquakes were very similar, but the rupture of the 1936 earthquake propagated southward and that of the 1978 earthquake propagated northwestward. Figure 9 illustrates the relationship between the hypocenter, asperity, seismic slip area, and distribution of aftershocks obtained in this study. 6.2. Seismic Coupling [24] In order to estimate the seismic coupling, we plot the space-time distribution of seismic moment released by each asperity in Figure 10. In Figure 10b asperities are projected against the latitude direction. We calculated the average seismic slip (D a ) in the asperities that are hatched in Figure 10a. The average seismic slip (D a ) and the area (S) for each asperity are listed in Table 3. The width of the rectangle (time span) in Figure 10b is the corresponding plate convergence time, namely D a /V plate, where V plate (= 0.1 m yr 1 ) is the subduction velocity of the Pacific plate. We illustrated the space-time distribution of seismic moment release assuming the slip-predictable model, so that the right end of each rectangle in Figure 10b corresponds to the earthquake occurrence time. If the asperities are fully coupled, the rectangles should fill the horizontal axis. Considering the variation in the seismic coupling of asperities along the subduction zone in northeastern Japan, we defined three different domains in this region. [25] In the northern region (40 41.3 N: R1 in Figure 10a), the seismic coupling was almost 100%. Nagai et al. [2001] calculated the seismic coupling for asperity B in Figure 6, and suggested that the moment accumulated from plate convergence between the 1968M Tokachi-oki earthquake and the 1994 Sanriku-oki earthquake was nearly the same as the seismic moment released in the 1994 Sanriku-oki earthquake. [26] In the central region (39 40 N: R2 in Figure 10a), only a slight seismic moment was released by large earthquakes, and seismic coupling was found to be weak. The maximum earthquake size that occurred in this region during the last century had a magnitude of 7. The maximum slip was significantly smaller than that in the R1 and R3 regions. [27] In the southern region (37.8 39 N: R3 in Figure 10a), the seismic coupling is medium. The seismic moment released by earthquakes over the last few decades was smaller than the moment accumulated by the subducting plate. Ueda et al. [2001] obtained the postseismic deformation from tide gauge and precise leveling data, and found that the afterslip continued for at least 4 years after the 1978 Miyagi-oki earthquake in and around the main fault. The total aseismic moment release reached 2.0 10 20 Nm. The postseismic moment release was as large as the coseismic moment release. [28] The slip rate distribution on the plate boundary in the northern Japan region obtained from the GPS [Nishimura et al., 2000] showed a rough coupling pattern similar to ours; that is, plate coupling was stronger in the R1 and R3 regions and weaker in R2. Igarashi et al. [2003] investigated the occurrence of small repeating earthquakes on the plate boundary off the Sanriku region. They suggested that these earthquakes are caused by the recurrent rupture of the same small-sized asperity located outside large asperities. It was also shown that many repeating earthquakes occurred in region R2, which suggests that the seismic coupling is relatively weak there. [29] Some researchers obtained the averaged seismic coupling in the Japan Trench using the seismic catalog. Peterson and Seno [1984] obtained the ratio of the seismic slip rate to the relative plate velocity, a = 0.24, using the RM2 plate motion model [Minster and Jordan, 1978] and an earthquake catalog compiled by Abe [1981]. Pacheco et al. [1993] also estimated the seismic coupling a = 0.18 using the Nuvel 1 plate motion model [DeMets et al., 1990] and a more complete and homogeneous earthquake catalog [Pacheco and Sykes, 1992]. Their values are much smaller than the average seismic coupling for asperities inferred in our study, because they considered that the strain accumulates uniformly in the whole area by the subducting plate. Our study suggests that asperities are very strongly coupled but occupy only a fraction of plate contact area. 6.3. Origin of Asperities [30] It is interesting to note that the variation in seismic coupling along the Japan Trench is correlated with the bathymetry of the subducting plate. Tanioka et al. [1997] classify the topography of the subducting plate from the bathymetry profile data into three parts. Their classification is very similar to ours. According to the multichannel seismic survey in the northern Japan region reported by 8of14

Figure 8. Map view summary of asperities (contour line) and the first month of aftershocks as determined by JMA. Tsuru et al. [2000], the Pacific plate with clear horst-graben structures is subducted around 39 N. Such a horst and graben structure can transport sediments and water to the deep region of the subducting plate. Ultraslow events with the time constant on 1 2 days frequently occur in this area [Kawasaki et al., 2001]. We think that the subducted sediments and water weaken the coupling between the upper and lower plates. [31] In the Tohoku district, tsunami earthquakes sometimes occurred near the Japan Trench. The weak coupling is a characteristic of the R2 region. Tanioka et al. [1997] also pointed out the submarine topography near the Japan trench off Sanriku was complicated in the area where huge tsunami earthquakes have occurred. Ruff [1989] considered that the horst and graben structure can transport sediments and water that fill the bucket on the subducting plate [Hilde, 1983], and the subducted sediments and water could be a cause of tsunami earthquakes. [32] The seismic reflection and refraction experiments indicate that a strong reflected wave from the plate boundary can be seen between the 1968A asperity and the 1981 asperity [Fujie et al., 2002]. In this region neither great earthquakes nor microearthquakes have occurred. Fujie et al. [2002] suggested the existence of a very thin and lowvelocity layer on the plate boundary. We think that the subducting sediment and water strongly affect the rheology of the plate coupling, and that no earthquake will occur. It is possible that by using the intensity of the reflected wave Figure 9. Schematic of the relationship between the hypocenter, asperity, seismic slip area, and aftershock distribution. 9of14

Figure 10. (a) Asperity map. (b) Space-time distribution of the seismic moment release based on the slip-predictable model. The width of the rectangle indicates the moment accumulation given by time D a /V plate, where D a is the average seismic slip in asperities and V plate = 0.1 m yr 1. See color version of this figure at back of this issue. from the plate interface, the location of the weak coupling may be estimated. 7. Summary [33] We analyzed the source process of eight large earthquakes with magnitudes greater than 7 which occurred after 1930 in northeastern Japan, and constructed an asperity map on the subducting plate along the Japan Trench. We found that some asperities in this region repeatedly generated large earthquakes during the last 70 years. Asperity B in Figure 6 was associated with the earthquakes of 1931, 1968M, and 1994, while asperity C was associated with the earthquakes of 1960, 1968M, and 1989. [34] On the basis of the seismic coupling of the asperities, northeast Japan was divided into three regions. In the R1 region (40 41.3 N) the seismic coupling in the asperity is almost 100%, and the size of the asperity is large. In the R2 region (39 40 N), little seismic moment was released by large earthquakes, and the seismic coupling is weak. In this region, slow earthquakes frequently occur. In the R3 region (37.8 39 N), the seismic coupling is medium. [35] The general characteristics of asperities found by this study are as follows. (1) The location is fixed in space; (2) the asperity tends to be located away from the initial break; and (3) aftershock activity is rather low in the asperity. Appendix A A1. 9 March 1931 [36] At 1249 LT on 9 March 1931, a large earthquake (M J 7.6) occurred off the coast of Sanriku, the northern part of Honshu, Japan. Recently the hypocenter was redetermined by JMA to be at 143.233 E, 40.181 N, depth 0 km. We assume the same mechanism for this earthquake (Figure A1a) as for the 1994 Sanriku-oki earthquake, because this event occurred near the site of the 1994 Sanriku-oki earthquake. The depth of the hypocenter was assumed to be 10 km considering the plate configuration. Figure A1. Result of the 1931 earthquake. See the caption of Figure 4 for an explanation. 10 of 14

Figure A2. Result of the 1936 earthquake. See the caption of Figure 4 for an explanation. Seismograms from only 3 JMA stations, namely Akita (AKI), Yamagata (YAM) and Sapporo (SAP), were available for the analysis. The epicentral distance exceeded 270 km even at the nearest station, Akita. Therefore we sought only a rough estimation of the moment release. The spatial distribution of the moment release is shown in Figure A3. Result of the 1937 earthquake. See the caption of Figure 4 for an explanation. Figure A4. Result of the 1960 earthquake. See the caption of Figure 4 for an explanation. Figure A1b. Arrows in this figure show the slip vector on the fault plane. Figure A1c shows the observed (upper trace) and synthetic (lower trace) waveforms for this event. The total seismic moment M o is 0.8 10 20 Nm (M w = 7.2), and the source duration time is about 30 s. The maximum slip is about 0.7 m. The location of the asperity overlapped with one of the 1968M asperities and the 1994 asperity. [37] The source area of the tsunami was estimated by Iida [1956] and Hatori [1974]. The source area estimated by Iida [1956] using the first arrival times of the tsunami was 50 km long. Reexamination is needed using the new hypocenter information. A2. 3 November 1936 [38] The earthquake (M J = 7.5) occurred at 0546 LT on 3 November 1936 off the coast of Miyagi. Since the epicenter of this earthquake is very close to that of the 1978 event, this earthquake has so far been regarded as an event of the previous generation. A tidal wave of around 20 cm in height was observed on the coast. The tsunami magnitude was 7.0 [Abe, 1981]. [39] Only two records, from Tohoku Univ. and JMA Utsunomiya station, were available for this analysis. We assumed the same fault plane as the 1978 Miyagi-oki earthquake. The result is shown in Figure A2. The total seismic moment M o is 1.7 10 20 Nm (M w = 7.4), and the maximum slip is 2.0 m. The asperity was located about 11 of 14

30 km southeastward of the initial rupture point. The location is quite consistent with the tsunami source area obtained by Hatori [1974] and Miyabe [1937]. A3. 27 July 1937 [40] On 27 July 1937, an earthquake with a magnitude of 7.1 occurred off the coast of Miyagi. The epicenter was very close to the 1936 earthquake epicenter. We found only two seismograms for this event (Tohoku University, Utsunomiya). P wave particle motion at Tohoku University indicated that the asperity of this event was located to the north of the 1936 earthquake. In this earthquake, no tsunami was observed. [41] We carried out the analysis assuming the mechanism of the 1978 Miyagi-oki earthquake obtained by Seno et al. [1980]. As shown in Figure A3, the area of large seismic moment is 30 km (length) 50 km (width), and the total moment M o is 0.6 10 20 Nm (M w = 7.1). The maximum slip is 0.4 m. The rupture propagated to about 30 km northwest of the hypocenter in 20 s. A4. 21 March 1960 [42] The earthquake (M J 7.2) occurred at 1249 on 21 March 1960. The records from several observation stations are available. We analyzed the nearest three records (MIY, HAC, MRK), assuming that the fault plane was the same as that of the 1994 Sanriku-oki earthquake. The result is shown in Figure A4. The total seismic moment M o is Figure A6. Result of the 1981 earthquake. See the caption of Figure 4 for an explanation. Figure A5. Result of the 1968A earthquake. See the caption of Figure 4 for an explanation. 1.1 10 20 Nm (M w = 7.3), and the duration time is about 30 s. The maximum slip is about 1.9 m. The rupture propagated to the deep direction from the hypocenter. The location of the asperity overlapped with asperity C of the 1968M Tokachi-oki earthquake. [43] Aftershocks spread toward the trench side (Figure 8c). The tsunami source area obtained by Hatori was located near the trench containing these aftershocks (Figure 7). In the aftershock region, many similar earthquakes have occurred [Igarashi et al., 2003]. The seismicity in this region seems to become active after a large slip occurs in the deeper region. A5. 12 June 1968 (1968A) [44] This earthquake occurred about 1 month after the Tokachi-oki earthquake. It had a moment magnitude of 8.2 and was located about 150 km northward. The focal mechanism was obtained by Yoshioka and Abe [1976] using P wave first motions at the World Wide Standard Seismograph Network (WWSSN) and JMA stations as well as the surface wave radiation pattern. Their solution is (strike, dip, slip) = (151, 30, 29). First, we used their mechanism in our analysis, but the fit of the waveforms was poor. The strike deviated from the strike of the trench. Considering the Harvard CMT solutions for other earthquakes in this area and the trench axis, we fixed the fault plane on (strike, dip) = (184, 15). Our result is shown in Figure A5. The total seismic moment is 0.4 10 20 Nm, which is consistent with Yoshioka and Abe s [1976] result. Unlike other earthquakes, the rupture propagated toward 12 of 14

the trench side, and the aftershocks occurred toward the land side. Although this earthquake occurred near the site of the 1960 earthquake, the seismicity on the trench side was not activated. A6. 19 January 1981 [45] The earthquake (M J = 7.0) occurred at 0317 on 19 January 1981. We analyzed the nearest four records (MIY, ISN, SEN, and FKS) by assuming the Harvard CMT solution: (strike, dip) = (186, 15). Figure A6 shows our result. The seismic moment is 5.9 10 19 Nm (M w = 7.1), which is larger than that of the Harvard CMT solution (M o = 3.7 10 19 Nm). The rupture propagated southward. The maximum slip is about 0.9 m. The amplitude of the NS component recorded at Sendai cannot be well reconstructed. This is probably due to the local effect of the structure. As seen in Figure 8g, aftershocks occurred in the area surrounding the asperity. A7. 2 November 1989 [46] The earthquake (M J = 7.1) occurred at 0325 on 2 November 1989. This event occurred near the site of the 1960 earthquake. We analyzed the nearest five records (MIY, OFU, HAC, MRK, and ISN). The assumed fault plane is (strike, dip) = (183, 20). The result is shown in Figure A7. The total seismic moment M o is 4.5 10 19 Nm (M w = 7.0), which is smaller than that of the Harvard CMT solution (M w = 7.4) and the tsunami magnitude (M t = 7.5). We collected the DWWSSN data from IRIS-DMC, and analyzed the teleseismic long-period body wave. The results Figure A8. Result of the 1989 earthquake obtained from the long-period teleseismic data. See the caption of Figure 4 for an explanation. The number above the station code is the peak-to-peak amplitude in mm of the observed displacement waveform, and the number below is the source-to-station azimuth. are shown in Figure A8. The obtained seismic moment is 1.3 10 20 Nm, 3 times as large as the result obtained from the regional strong motion data. This difference may be due to the spectral content of the source time function, that is, the smoothness of the rupture time history. The same tendency was seen in the earthquakes of 1960 and 1968A, which occurred near this event. The maximum slip is about 0.9 m. The rupture propagated to the downdip direction from the hypocenter over 25 s. The asperity of this earthquake is the same as that of the 1960 event, although the 1960 earthquake started at a different point on the trench. The waveforms of these two events (for example, MIY station) are very similar. The aftershock distributions are also very similar between the two events. [47] Acknowledgments. We thank K. Yoshikawa, Y. Ishigaki, and many staff members of JMA for providing us with strong motion seismograms. We also thank Satish Singh, Heidi Houston, and Teruo Yamashita for many useful comments. This research was partially supported by Grantin-Aid for Scientific Research (C) 12640404, The Ministry of Education, Science, Sports and Culture. This paper is dedicated to coauthor Masayuki Kikuchi, who passed away in October 2003. Figure A7. Result of the 1989 earthquake. See the caption of Figure 4 for an explanation. References Abe, K. (1981), Magnitudes of large shallow earthquakes from 1904 to 1980, Phys. Earth Planet. Inter., 27, 52 92. 13 of 14

Abe, K. (1999), Quantification of historical tsunamis by the Mt scale, J. Seismol. Soc. Jpn., 52, 369 377. Beck, S. L., and L. J. Ruff (1987), Rupture process of the great 1963 Kurile Islands earthquake sequence: Asperity interaction and multiple event rupture, J. Geophys. Res., 92, 14,123 14,138. DeMets, C., R. G. Gordon, D. F. Argus, and S. Stein (1990), Current plate motions, Geophys. J. Int., 101, 425 478. Fujie, G. (1999), A new travel time inversion using refracted and reflected arrivals and its application to the estimation of the plate boundary structure off Sanriku, Japan (in Japanese), Ph.D. thesis, Univ. of Tokyo, Japan. Fujie, G., J. Kasahara, R. Hino, T. Sato, M. Shinohara, and K. Suyehiro (2002), A significant relation between seismic activities and reflection intensities in the Japan Trench region, Geophys. Res. Lett., 29(7), 1100, doi:10.1029/2001gl013764. Hamamatsu, O. (1966), Historical table of seismographs for routine observations in J. M. A. Network (in Japanese), J. Seismol. Soc. Jpn., 19, 286 305. Hatori, T. (1974), Tsunami sources on the Pacific side in northeast Japan (in Japanese), J. Seismol. Soc. Jpn., 27, 321 337. Hatori, T. (1978), The tsunami generated off Miyagi prefecture in 1978 and tsunami activity in the region, Bull. Earthquake Res. Inst., 53, 1177 1189. Hatori, T. (1996), The 1994 Sanriku-Oki tsunami and distribution of the radiating tsunami energy in the Sanriku region, J. Seismol. Soc. Jpn., 49, 19 26. Hilde, T. W. C. (1983), Sediment subduction versus accretion around the Pacific, Tectonophysics, 99, 381 397. Houston, H., and E. R. Engdahl (1989), A comparison of the spatiotemporal distribution of moment release for the 1986 Andreanof Islands earthquake with relocated seismicity, Geophys. Res. Lett., 16, 1421 1424. Igarashi, T., T. Matsuzawa, and A. Hasegawa (2003), Repeating earthquakes and interplate aseismic slip in the northeastern Japan subduction zone, J. Geophys. Res., 108(B5), 2249, doi:10.1029/2002jb001920. Iida, K. (1956), Earthquakes accompanied by tsunamis occurring under the sea off the islands of Japan, J. Earth Sci., 4, 1 43. Kawasaki, I., Y. Asai, and Y. Tamura (2001), Space-time distribution of interpolate moment release including slow earthquakes and the seismogeodetic coupling in the Sanriku-oki region along the Japan trench, Tectonophysics, 330, 267 283. Kikuchi, M., and Y. Fukao (1985), Iterative deconvolution of complex body waves from great earthquakes The Tokachi-oki earthquake of 1968, Phys. Earth Planet. Inter., 37, 235 248. Kikuchi,M.,M.Nakamura,M.Yamada,M.Fushimi,Y.Tatsumi,and K. Yoshikawa (1999), Source parameters of the 1948 Fukui earthquake inferred from low-gain strong-motion records (in Japanese), J. Seismol. Soc. Jpn., 52, 121 128. Kikuchi, M., M. Nakamura, and K. Yoshikawa (2003), Source rupture processes of the 1944 Tonankai earthquake and the 1945 Mikawa earthquake derived from low-gain seismograms, Earth Planets Space, 55, 159 172. Koketsu, K. (1985), The extended reflectivity method for synthetic nearfield seismograms, J. Phys. Earth, 33, 121 131. Lay, T., and H. Kanamori (1980), Earthquake doublets in the Solomon islands, Phys. Earth Planet. Inter., 21, 283 304. Lay, T., and H. Kanamori (1981), An asperity model of large earthquake sequences, in Earthquake Prediction: An International Review, Maurice Ewing Ser., vol. 4, edited by D. W. Simpson and P. G. Richards, pp. 579 592, AGU, Washington, D. C. Mendoza, C., and S. H. Hartzell (1988), Aftershock patterns and main shock faulting, Bull. Seismol. Soc. Am., 78, 1438 1449. Minster, J., and T. Jordan (1978), Present-day plate motion, J. Geophys. Res., 83, 5331 5354. Miyabe, N. (1937), Tsunami associated with the Sanriku earthquake that occurred on November 3, 1936, Bull. Earthquake Res. Inst., 15, 837 844. Nagai, R., M. Kikuchi, and Y. Yamanaka (2001), Comparative study on the source process of recurrent large earthquakes in Sanriku-Oki region: The 1968 Tokachi-Oki earthquake and the 1994 Sanriku-oki earthquake (in Japanese with English abstract), J. Seismol. Soc. Jpn., 52, 267 289. Nakayama, W., and M. Takeo (1997), Slip history of the 1994 Sanriku- Haruka-Oki, Japan, earthquake deduced from strong motion data, Bull. Seismol. Soc. Am., 87, 918 931. Nishimura, T., S. Miura, K. Tachibana, K. Hashimoto, T. Sato, S. Hori, E. Murakami, T. Kono, K. Nida, M. Mishina, T. Hirasawa, and S. Miyazaki (2000), Distribution of seismic coupling on the subducting plate boundary in northeastern Japan inferred from GPS observations, Tectonophysics, 323, 217 238. Pacheco, J. F., and L. R. Sykes (1992), Seismic moment catalog for large shallow earthquakes from 1900 to 1989, Bull. Seismol. Soc. Am., 82, 1306 1349. Pacheco, J. F., L. R. Sykes, and C. H. Scholz (1993), Nature of seismic coupling along simple plate boundaries of the subduction type, J. Geophys. Res., 98, 14,133 14,159. Peterson, E. T., and T. Seno (1984), Factors affecting seismic moment release rates in subduction zones, J. Geophys. Res., 89, 10,233 10,248. Ruff, L. (1989), Do trench sediments affect great earthquake occurrence in subduction zones?, Pure Appl. Geophys., 129, 263 282. Ruff, L., and H. Kanamori (1980), Seismicity and the subduction process, Phys. Earth Planet. Inter., 23, 240 252. Ruff, L., and H. Kanamori (1983), The rupture process and asperity distribution of three great earthquakes from long-period diffracted P-waves, Phys. Earth Planet. Inter., 31, 202 230. Seno, T., K. Shimazaki, P. Somerville, K. Sudo, and T. Eguchi (1980), Rupture process of the Miyagi-oki, Japan, earthquake of June 12, 1978, Phys. Earth Planet. Inter., 23, 39 61. Tanioka, Y., L. Ruff, and K. Satake (1996), The Sanriku-oki, Japan, earthquake of December 28, 1994 (M W 7.7): Rupture of a different asperity from a previous earthquake, Geophys. Res. Lett., 23, 1465 1468. Tanioka, Y., L. Ruff, and K. Satake (1997), What controls the lateral variation of large earthquake occurrence along the Japan Trench?, Island Arc, 6, 261 266. Tsuru, T., J. Park, N. Takahashi, S. Kodaira, Y. Kido, Y. Kaneda, and Y. Kono (2000), Tectonic features of the Japan trench convergent margin off Sanriku, northeastern Japan, revealed by multichannel seismic reflection data, J. Geophys. Res., 105, 16,403 16,413. Ueda, H., M. Ohtake, and H. Sato (2001), Afterslip of the plate interface following the 1978 Miyagi-Oki, Japan, earthquake, as revealed from geodetic measurement data, Tectonophysics, 338, 45 57. Umino, N., A. Hasegawa, and A. Takagi (1990), The relationship between seismicity patterns and fracture zones beneath northeastern Japan, Tohoku, Geophys. J., 33, 149 162. Yoshida, S. (1989), Waveform inversion using ABIC for the rupture process of the 1983 Hindu Kush earthquake, Phys. Earth Planet. Inter., 56, 389 405. Yoshioka, N., and K. Abe (1976), Focal mechanism of the Iwate-oki earthquake of June 12, 1968, J. Phys. Earth, 24, 251 262. Y. Yamanaka, Earthquake Research Institute, University of Tokyo, 1-1-1 Yayoi, Bunkyo-ku, Tokyo 113-0032, Japan. (sanchu@eri.u-tokyo.ac.jp) 14 of 14

Figure 6. Asperity map along the subduction zone in northeastern Japan. Stars show the main shock epicenters. Contour lines show the moment release distribution. The interval of the contour lines is 0.5 m. Each earthquake is distinguished by color. We painted the area within the value of half the maximum slip as an asperity. 7of14

Figure 10. (a) Asperity map. (b) Space-time distribution of the seismic moment release based on the slip-predictable model. The width of the rectangle indicates the moment accumulation given by time D a /V plate, where D a is the average seismic slip in asperities and V plate = 0.1 m yr 1. 10 of 14