Along strike variations in short term slow slip events in the southwest Japan subduction zone

JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 115,, doi:10.1029/2008jb006059, 2010 Along strike variations in short term slow slip events in the southwest Japan subduction zone Shutaro Sekine, 1,2 Hitoshi Hirose, 1 and Kazushige Obara 1,3 Received 1 September 2008; revised 1 May 2009; accepted 21 September 2009; published 23 September 2010. [1] Short term slow slip events accompanied by nonvolcanic deep low frequency tremors and deep very low frequency earthquakes in southwest Japan were investigated systematically by means of ground tilting studies. The change in tilt usually lasts for several days. By using a genetic algorithm and a least squares method, we inverted the data for tilt steps that were caused by slow slip events and were detected by at least four stations situated near the source of the tremor. Fault parameters were estimated for 54 slow slip episodes that occurred mainly in the western Shikoku, northeastern Kii, and Tokai regions from 2001 to 2008. In eastern Shikoku, two slow slip episodes were detected quantitatively for the first time. The fault geometries of all the slow slip events were located within the belt like distribution of tremors in the transition zone between the locked and aseismic slip zones at the plate interface of the subducting Philippine Sea Plate. The spatial extent of the fault geometry corresponds roughly to the distribution of clusters of nonvolcanic tremors and very low frequency earthquakes. The moment magnitudes ranged from 5.4 to 6.2, and the slip was 1 cm for each slow slip event. The rate of moment release by the detected slow slip events was 40 60% of the moment accumulation expected from the relative plate motion, and it showed regional differences. They may reflect the along strike variations in plate convergence and/or the characteristic size of the slow slip fault plane on the plate interface. Citation: Sekine, S., H. Hirose, and K. Obara (2010), Along strike variations in short term slow slip events in the southwest Japan subduction zone, J. Geophys. Res., 115,, doi:10.1029/2008jb006059. 1. Introduction [2] In southwest Japan, the Philippine Sea Plate (PHS) subducts beneath the overlying continental plate, and interplate megathrust earthquakes have occurred repeatedly along the Nankai Trough every 100 150 years or so [Ando, 1975]. During interseismic periods both plates are mutually locked and stress accumulates at the seismogenic zone. Between the deep steady slip zone and the locked zone, there is a transitional zone [Hyndman et al., 1997] where we expect to find intermediate slip features. Recently, some types of slow earthquakes, characterized by various time constants, have been identified in the transitional zone on the plate interface. These discoveries were the result of the establishment by the National Research Institute for Earth Science and Disaster Prevention (NIED) of the highsensitivity seismograph network (Hi net) and the broadband seismograph network (F net) [Okada et al., 2004; Obara 1 National Research Institute for Earth Science and Disaster Prevention, Tsukuba, Japan. 2 Now at Association for the Development of Earthquake Prediction, Tokyo, Japan. 3 Now at Earthquake Research Institute, University of Tokyo, Tokyo, Japan. Copyright 2010 by the American Geophysical Union. 0148 0227/10/2008JB006059 et al., 2005]. Obara [2002] discovered nonvolcanic tremors at the downdip side of the megathrust seismogenic zone along the plate interface. These tremors are characterized by a long lasting wave train with a very weak amplitude and a predominant frequency of 1.5 5 Hz.Ito et al. [2007] detected very low frequency (VLF) earthquakes with a predominant frequency of 0.05 Hz that were coincident with the episodic peak activity of tremor. The hypocentral location and focal mechanism of VLF earthquakes are consistent with the geometry of the subducting plate interface and plate motion (Figure 1). These seismic slow events sometimes accompany the geodetic signature, that is to say the slow slip event (SSE). Obara et al. [2004] reported the occurrence of SSEs in the tremor source region of western Shikoku during peaks of tremors. The SSE is a shear slip movement on the plate interface of the reverse fault type that lasts for several days. In southwest Japan, the presence of a densely distributed global positioning system (GPS) observation network permitted the detection of another type of SSE with a duration ranging from 6 months to 5 years [Hirose et al., 1999; Ozawa et al., 2002]. To distinguish between these two types of SSE, we use the term shortterm SSE for events that last for several days and are accompanied by tremors, and the term long term SSE for events that last for years. Short term SSEs and the accompanying tremor activity usually occur episodically at regular intervals in each region. Such a tremor and slip phenomenon 1of13

Figure 1. Distribution of nonvolcanic tremors and very low frequency earthquakes in southwest Japan. Red dots are epicenters of nonvolcanic tremors estimated by Obara [2010]. Black and white beach balls indicate the location and focal solutions of deep very low frequency earthquakes estimated by Ito et al. [2007, 2009]. Crosses show Hi net stations. Dashed lines are contours of the oceanic Moho boundary of the subducting PHS [Shiomi et al., 2008]. The solid line is the axis of the Nankai Trough. is characterized by migration. In many cases, the tremor activity migrates with a speed of approximately 10 km/d, and sometimes we detect migrating short term SSEs and VLF events that are coincident with tremor activity [Ito et al., 2007; Obara, 2010]. If we consider the magnitude of each member of the slow earthquake family, the short term SSE may be a primary phenomenon that arises from a stickslip event on the plate interface, and other seismic events are triggered by the occurrence of a short term SSE. Therefore, knowledge of the activity and source parameters for shortterm SSEs is essential for clarifying the nature of slip phenomena in the transition zone. So far, short term SSEs in southwest Japan have been detected within three regions: western Shikoku [Obara et al., 2004; Hirose and Obara, 2005], northeastern Kii and Tokai [Hirose and Obara, 2006; Obara and Sekine, 2009]. Similar coupling phenomena between tremors and SSEs have been reported in the Cascadia subduction zone [Rogers and Dragert, 2003]. In the Cascadia margin, Dragert et al. [2001] discovered the occurrence of SSEs from geodetic data obtained from a GPS network. The SSEs in Cascadia recur at intervals of 13 16 months [Miller et al., 2002] and coincide with tremor activity. The source of the SSEs usually migrates with the tremor activity and continues for several weeks. The SSEs in Cascadia are classified as short term SSEs because of their duration and their coupling with tremor activity. Recently, such tremors accompanied by SSEs have been detected in Mexico [Payero et al., 2008] and other subduction zones. The characteristic recurrence interval and duration of the SSE episodes are different in each subduction zone. These slow earthquakes may reflect the subduction process at the downdip side of the megathrust seismogenic zone, so the studies at various subduction zones are important in elucidating the nature of the subduction process. [3] Here, we estimate and compile source parameters for short term SSEs accompanied by tremor activity that were detected from ground tilting data in southwest Japan from 2001 to 2008. We then discuss the spatial characteristics of short term SSEs along the strike of the subducting PHS, and we discuss the moment release in the transition zone on the plate interface. 2. Data and Methods 2.1. Data Acquisition [4] The NIED Hi net stations were constructed in the Japanese islands at a spacing of 20 30 km to improve the capability for detection of microearthquakes. At each station, a three component velocity seismometer is installed at the base of a borehole that is normally 100 m deep. Each Hi net station is equipped with a two component horizontal high sensitivity accelerometer (tiltmeter) that is attached to the velocity seismometer in the same capsule. The NIED Hi net tiltmeter has a wide frequency response range from 2of13

5 Hz to the DC component. Data from the tiltmeter are transmitted at a sampling frequency of 20 Hz, and resampled to 1 h interval data by an averaging process with corrections for the installation azimuth of the sensor direction [Shiomi et al., 2003]. The instrumental offset and spike noise caused by strong seismic waves are manually corrected. The tidal components and the response to atmospheric pressure changes are removed by using the BAYTAP G program [Tamura et al., 1991]. For this correction, we use atmospheric pressure data taken from observations made at the office of the Japan Meteorological Agency (JMA) that is nearest to each NIED Hi net station. Finally, a linear trend corresponding to a long term drift of each sensor is removed for each period used in the analysis. 2.2. Inversion Procedure [5] From the observed tilting data processed as described above, we extracted the tilting changes caused by short term SSEs associated with peaks of tremors in southwest Japan from January 2001 to June 2008, and we estimated the fault geometry and slip parameters for each SSE. In terms of the stability of the inversion process, we estimate the parameters for SSEs in those cases in which we detected coherent changes in tilt at a minimum of four stations when these changes were coincident with tremor activity. The duration of each SSE episode was partitioned on the basis of the tilt change pattern with reference to tremor activity. Then, the difference in tilt between the value at the beginning and that at the end of the period of the SSE episode was measured (Figures 2, 3, and the auxiliary material). 1 Here, we assumed the existence of a single rectangular fault plane with uniform slip for each SSE episode. In the case of distinct migrating episodes with a long duration and a long spatial extent, we divided these into two or three periods and we estimated the SSE source parameters for each rectangular fault plane corresponding to each subperiod; this was done because it is inappropriate to assume a single rectangular fault model for such distinct migrating events. The fault geometry and slip parameters were estimated by applying a weighted least squares method to the linear parameters and using a genetic algorithm to optimize the nonlinear parameters [Obara et al., 2004; Hirose and Obara, 2005]. Seven parameters of the fault geometry (longitude, latitude, depth, length, width, strike, and dip) were estimated by using the genetic algorithm and two parameters (strike and dip slip) were estimated by the weighted least squares method. This inversion procedure uses elastic dislocation modeling based on a homogeneous halfspace medium [Okada, 1992]. The rigidity was assumed to be 40 GPa. [6] We used a bootstrap method to test the reliability of the estimated SSE fault models. We created 200 sets of bootstrap data by adding random noises to synthetic tilting data derived from an optimized result by the inversion process, and we estimated a set of SSE fault parameters for each set of bootstrap data by using the genetic algorithm. The standard deviation of the estimated parameters for the 200 sets of bootstrap data is represented as the error in the inversion process. 1 Auxiliary materials are available in the HTML. doi:10.1029/ 2008JB006059. 2.3. Parameter Setting [7] First, we checked the stability of the inversion process by changing the inversion constraints and SSE duration. Figure 2 shows an example of the observed tilt data, with linear trends removed, and an estimated fault model for the SSE episode of February 2008 in eastern Shikoku. The duration of the SSE was identified to be 5 days from the pattern of tilt changes at stations taken in conjunction with the tremor activity. The difference in tilt data averaged within each 1 day before and after the period of the SSE was measured from the detrended tilt trace. In this case, the observed tilting steps ranged from 0.04 to 0.1 mrad, as shown in Figure 2a. The synthetic tilting data are coincided with the observed tilting data as shown in Figures 2a and 2b. The estimated values and bootstrap errors are as follows; latitude = 33.873 ± 0.089 N, longitude = 134.343 ± 0.233 E, depth = 42 ± 11 km, length = 46 ± 8 km, width = 33 ± 6 km, strike = 249 ± 19, dip angle = 26 ± 9, slip = 3.1 ± 0.4 cm, and rake = 106 ± 6. The estimated slip vector agrees well with the direction of the relative plate motion [Miyazaki and Heki, 2001]. Figure 3 shows another example of a fault model for the SSE of March 2007 in northeastern Kii. In this case, the observed tilt steps ranged from 0.01 to 0.05 mrad. The duration of the SSE was identified to be 3 days. The fault model of the SSE is shown by the rectangle in Figure 3b. In this case, we obtained a slip direction of the normal fault type. This may reflect the poor quality of data obtained for observed tilt changes caused by a low signal to noise ratio. In this analysis, we tried to extract SSE episodes that showed more than four measurable tilt steps. Sometimes, the observed tilt step was too weak to allow the application of the inversion process without any restriction. To obtain a more stable solution, we tried to carry out the inversion process with a slip direction that was fixed according to the relative plate motion of N55 W [Miyazaki and Heki, 2001]. Figure 3c show the results of a reestimation of the SSE fault model with the fixed slip direction. The estimated values and bootstrap errors are as follows: latitude = 34.628 ± 0.070 N, longitude = 136.524 ± 0.302 E, depth = 25 ± 6 km, length = 28 ± 11 km, width = 43 ± 8 km, strike = 208 ± 8, dip angle = 25 ± 6, and slip = 0.8 ± 0.1 cm. In this case the results were well estimated compared with the case in which all the parameters were free. We therefore estimate all SSE fault models with a fixed slip direction in this paper. Figure 2c is the estimated fault model for the SSE episode of February 2008 in eastern Shikoku with a fixed slip direction, which is the same as that in Figure 3c. In this case, the estimated fault parameters were similar to those obtained by the all free inversion process (Figure 2b). The estimated moment magnitude is 6.0. The estimated values and bootstrap errors are as follows: latitude = 34.844 ± 0.085 N, longitude = 134.289 ± 0.196 E, depth = 34 ± 7 km, length = 47 ± 8 km, width = 24 ± 5 km, strike = 258 ± 13, dip angle = 29 ± 7, and slip = 2.9 ± 0.2 cm. [8] To check the effect of the duration of the tilting change, we shortened the time duration to 1 day for this SSE episode, according to the maximum activity of tremors, as shown in Figure 2d. The observed tilt steps ranged from 0.02 to 0.08 mrad, and the estimated SSE fault model is shown in Figure 2e; the moment magnitude is 5.7. This value is less than that estimated for the full duration case. 3of13

Figure 2. (a) Time series of observed ground tilting changes caused by the SSE episode of February 2008 in eastern Shikoku. Red and blue lines are tilt changes in the north south and east west component, respectively. Vertical dotted lines show the duration of the occurrence of the SSE. Solid horizontal lines plotted before and after the SSE are the averaged tilt data for 1 day, for use in the inversion process. The red and blue dashed lines are synthetic tilt steps calculated from the estimated fault geometry and slip parameters. Also shown, from top to bottom, are the daily frequency distributions of detected tremors, the atmospheric pressure, and the precipitation as measured at the Tokushima local office of the JMA. (b) The observed and synthetic tilt vectors and the estimated SSE fault geometry for an analysis duration of 5 days obtained by the inversion process with all parameters free. The blue and white arrows correspond to the observed and the synthetic tilt vectors, respectively. The ellipsoid attached to the head of blue arrow is the error in the observed tilt vector. The synthetic tilt vector is calculated on the basis of the estimated fault geometry and slip parameters. The red rectangle is the estimated fault geometry. The bold line on one side of the rectangle indicates the shallower side of the fault plane. The red arrow indicates the estimated slip vector. Orange dots show the locations of tremors that occurred during the SSE episode. Brown star shows the epicenter of VLF. Green square shows the location of the local office of the JMA. (c) Same as Figure 2b except that the slip direction was fixed in the inversion process. (d) Same as Figure 2a except that the duration for the analysis is 1 day. (e) Same as Figure 2c except that the duration for the analysis is 1 day. 4of13

Figure 3. (a) Time series of observed ground tilting change caused by the SSE episode of March 2007 in northeastern Kii. The red and blue lines are the north south and east west components of the tilt changes, respectively. Solid lines plotted before and after the SSE are the averaged tilt data for 1 day, for use in the inversion process. The red and blue dashed lines are synthetic tilt steps calculated from the estimated fault geometry and slip parameters by the inversion process with the slip direction fixed. Also shown, from top to bottom, are the daily frequency distributions of detected tremors, the atmospheric pressure, and the precipitation as measured at Tsu local office of the JMA. (b) The observed and synthetic tilt vector and the estimated SSE fault geometry for an analysis duration of 5 days. The blue and white arrows are the observed and the synthetic tilt vectors, respectively. The ellipsoid attached to the head of the blue arrow is the error in the observed tilt vector. The synthetic tilt vector is calculated on the basis of the estimated fault geometry and slip parameters. The red rectangle is the estimated fault geometry. The bold line on one side of the rectangle indicates the shallower side of the fault plane. The red arrow indicates the estimated slip vector. Orange dots show the locations of tremors that occurred during the SSE episode. Brown star shows the epicenter of VLF. Green square shows the location of the local office of the JMA. (c) Same as Figure 3b except that the slip direction was fixed in the inversion process. This shows that because the duration of the tilting change directly affects the estimation of the size of the SSE fault model, we must be careful to select a period that includes both the beginning and the end of the tilt change. 3. Results [9] In total, we detected 54 SSE episodes in southwest Japan from January 2001 to June 2008, as shown in Figure 4. All the source parameters for the regions of western Shikoku, eastern Shikoku, northeastern Kii, and Tokai are listed in Tables 1 to 4, respectively. Half the detected SSEs were located in western Shikoku. The source parameters of some SSEs have already been reported in previous papers [Obara et al., 2004; Hirose and Obara, 2005, 2006; Obara and Sekine, 2009], but we reestimated the source parameters of these episodes, together with those of other episodes, using the same procedure. Some episodes that showed distinct migrations were subdivided into two or three separate SSE episodes, and these are indicated by the suffixes A, B, or C. The suffixes D and E for episodes in Tokai (Table 4) indicate that. these episodes were parts of an event that showed a pronounced migration from Kii to Tokai [Obara and Sekine, 2009]. Observed changes in tilt and estimated fault geometries for all SSE episodes are shown in the auxiliary material. Figure 5 shows the spatiotemporal distribution of the detected SSEs. Each SSE is plotted as a rectangle with horizontal and vertical sides that correspond to the length of the SSE fault along the strike and slip, respectively. SSEs in western Shikoku have been detected since 2001, whereas the detection of SSEs in Tokai and northeastern Kii began slightly later. Below, we describe the characteristics of SSEs in each region. 3.1. Western Shikoku Region [10] In western Shikoku, 25 SSEs were recognized from crustal deformation detected by tiltmeters, and their source parameters were estimated. As shown in Figure 5, SSEs occur at approximately 6 monthly intervals in the years 2001 and 2002. During the latter half of 2003 and the beginning of 2004, SSEs occurred at intervals of approximately 3 months. Since the end of 2004, the SSEs have recurred at intervals of approximately 6 months. The change in the recurrence interval during 2003 2004 may have been related to the occurrence of a long term SSE in the Bungo Channel [Hirose and Obara, 2005]. The long term SSE occurred in the southwestern part of the source region of the short term SSEs and it could have triggered the more frequent occurrence of short term SSEs with a 3 month interval. The duration of SSEs ranged from 3 to 9 days in each episode, including migrating events. The average length and width of 5of13

Figure 4. Distribution of SSE fault geometries estimated from NIED Hi net tilt data for southwest Japan. The rectangles show a map view of the fault planes, and the bold line on one side of the rectangle is the shallower side of the fault plane for each SSE. The red dots are tremors estimated by Obara [2010]. The dashed lines are contours of the oceanic Moho boundary in the PHS derived by Shiomi et al. [2008]. The crosses are Hi net stations. The blue dashed line with a distance scale is a reference baseline along the strike of the subducting PHS used in Figures 5 and 7. SSEs were 62 km and 34 km, respectively. The averaged magnitude of the SSEs was 5.9, which is larger than that in the other regions. The slip ranged from 0.6 to 2.6 cm. If we compare the SSE fault geometry with the tremor and VLF activities, the location of the SSE fault geometry can be seen to coincide with areas that have a high population of VLFs and tremors. 3.2. Central and Eastern Shikoku Regions [11] In eastern Shikoku, crustal deformation is rarely detected, unlike the case for western Shikoku. Obara and Hirose [2006] have pointed out that ground tilting was sometimes observed, but only at a few stations that were close to the tremor source area during the peak of the tremor. Since 2006, however, two SSE episodes have been recognized; these occurred in November 2006 and February 2008. The tilt changes for these SSEs were detected at more than four stations, so we could estimate their source parameters. These SSE fault planes overlapped the location of the source of VLFs and tremors. [12] For central Shikoku, two SSE fault planes in western Shikoku extend into this region. Otherwise, no clear tilt change that was coincident with tremor activity was recorded in this region. On rare occasions, we can detect weak changes in tilt at a few stations only, so it is not yet possible to estimate the SSE source parameters, although a concentration of VLF and tremor activity has been observed. 3.3. Kii Region [13] In the Kii region, tremors are distributed with a length of 150 km along the strike of the subducting PHS; however, 14 SSEs that have been detected were all concentrated solely in the northeastern part of the region (Figure 4 and Table 3). The recurrence interval of SSEs was approximately 6 months as shown in Figure 5. The average length and widths of the SSE were 49 km and 42 km, respectively. The average moment magnitude of the SSEs in this region was 5.8, which is smaller than that in western Shikoku. The slip is ranging from 0.3 to 1.8 cm. The fault plane located at the southernmost part of the region is for the SSE of 200601A (Table 3); this was the first rupture event of January 2006 and it showed a pronounced migration in a northeasterly direction, ending in the Tokai region [Obara and Sekine, 6of13

Table 1. Estimated Fault Geometries and Slip Parameters of the Short Term SSE in Western Shikoku Episode Start Day End Day Duration (days) Latitude Longitude Depth Strike Dip Length Width Slip (cm) Mo (10 27 Nm) Mw Rake Number of Stations 200101 2001/01/04 2001/01/12 8 33.528±0.133N 133.004±0.124E 19±9 223±10 22±9 71±17 44±10 0.77±0.15 9.6±3.7 5.9 97 8 200108A 2001/08/16 2001/08/18 2 33.517±0.131N 132.869±0.096E 26±5 226±9 1±4 53±9 47±7 1.02±0.10 10.1±2.6 5.9 101 7 200108B 2001/08/18 2001/08/20 2 33.729±0.061N 133.279±0.069E 38±6 235±7 16±5 80±11 36±8 0.70±0.13 8.0±2.6 5.9 110 7 200202A 2002/02/10 2002/02/15 5 33.489±0.056N 133.170±0.055E 28±4 250±8 22±4 64±9 33±7 1.15±0.086 9.6±2.6 5.9 123 7 200202B 2002/02/15 2002/02/18 3 33.424±0.209N 132.907±0.241E 34±11 242±11 25±6 62±15 31±12 1.48±0.12 11.4±5.4 6.0 115 7 200208A 2002/08/06 2002/08/09 3 33.438±0.059N 132.835±0.132E 29±9 221±8 23±5 48±7 36±9 0.918±0.085 6.4±1.9 5.8 95 8 200208B 2002/08/09 2002/08/13 4 33.599±0.089N 133.126±0.228E 28±6 237±9 25±5 62±10 18±8 1.17±0.13 5.2±2.4 5.7 110 8 200304 2003/04/17 2003/04/21 4 33.166±0.069N 132.471±0.070E 28±5 248±8 15±5 44±10 36±8 1.43±0.19 9.0±3.1 5.9 123 8 200308A 2003/08/27 2003/08/30 3 33.222±0.043N 132.628±0.059E 22±5 250±4 45±6 80±14 20±6 1.28±0.30 8.3±3.6 5.9 130 5 200308B 2003/08/30 2003/09/04 5 33.373±0.034N 132.837±0.025E 32±3 245±3 10±4 28±12 39±5 2.64±0.18 11.6±5.2 6.0 103 5 200311A 2003/11/07 2003/11/13 6 33.324±0.053N 132.923±0.025E 45±2 243±2 38±4 66±12 44±6 2.41±0.20 28.1±6.7 6.2 104 5 200311B 2003/11/19 2003/11/25 6 33.682±0.120N 133.346±0.108E 30±7 247±9 12±7 66±17 28±8 1.429±0.097 10.5±4.2 5.9 121 10 200402 2004/02/10 2004/02/13 3 33.172±0.070N 132.566±0.162E 42±3 235±4 28±5 28±13 47±7 2.10±0.34 11.1±5.6 6.0 118 5 200404 2004/04/19 2004/04/26 7 33.395±0.055N 132.923±0.062E 28±4 248±7 25±4 75±11 44±8 1.02±0.10 13.5±3.4 6.0 121 9 200412 2004/12/27 2005/01/01 5 33.403±0.098N 132.869±0.054E 22±7 233±12 29±7 62±14 47±11 0.798±0.031 9.3±3.0 5.9 106 8 200505A 2005/05/12 2005/05/14 2 33.255±0.041N 132.687±0.033E 28±4 237±6 33±3 66±10 42±6 1.09±0.15 12.2±3.1 6.0 108 7 200505B 2005/05/14 2005/05/18 4 33.609±0.253N 132.931±0.082E 26±6 237±10 19±7 55±16 39±8 0.633±0.072 5.5±2.0 5.8 111 7 200510 2005/10/23 2005/10/27 4 33.393±0.072N 132.850±0.070E 24±7 234±7 28±6 62±11 21±9 0.97±0.12 5.1±2.5 5.7 107 6 200604 2006/04/15 2006/04/20 5 33.450±0.153N 132.858±0.118E 33±6 231±6 16±7 62±12 37±9 1.27±0.15 11.8±4.0 6.0 105 5 200609 2006/09/08 2006/09/17 9 33.383±0.225N 132.920±0.101E 40±8 250±9 20±9 71±14 39±10 2.64±0.28 29.3±10 6.2 123 6 200703 2007/03/13 2007/03/17 4 33.544±0.166N 132.947±0.252E 25±7 240±15 20±10 55±14 18±8 1.177±0.032 4.7±2.5 5.7 113 8 200709A 2007/08/27 2007/09/03 7 33.629±0.089N 133.051±0.080E 24±4 217±10 20±5 71±12 23±6 0.85±0.43 5.5±3.3 5.8 92 5 200709B 2007/09/09 2007/09/12 3 33.495±0.088N 132.862±0.042E 26±6 226±7 27±7 78±11 26±9 0.644±0.050 5.2±2.0 5.7 100 9 200712 2007/12/19 2007/12/25 6 33.108±0.037N 132.774±0.057E 33±4 260±8 32±4 73±9 41±6 1.667±0.061 19.9±4.0 6.1 130 5 200803 2008/03/14 2008/03/18 4 33.467±0.093N 132.910±0.135E 28±8 248±11 14±11 62±15 25±9 1.34±0.11 8.2±3.7 5.9 123 9 7of13

Table 2. Estimated Fault Geometries and Slip Parameters of the Short Term SSE in Eastern Shikoku Episode Start Day End Day Duration (days) Latitude Longitude Depth Strike Dip Length Width Slip (cm) Mo (10 27 Nm) Mw Rake Number of Stations 200611 2006/11/07 2006/11/11 4 34.002±0.045N 134.294±0.094E 36±7 231±7 10±7 78±11 41±9 0.78±0.11 9.9±2.9 5.9 105 5 200802 2008/02/13 2008/2/18 5 33.844±0.085N 134.290±0.196E 35±7 258±13 29±7 47±8 25±5 2.91±0.24 13.5±3.6 6.0 129 5 Table 3. Estimated Fault Geometries and Slip Parameters of the Short Term SSE in Northeastern Kii Episode Start Day End Day Duration (days) Latitude Longitude Depth Strike Dip Length Width Slip (cm) Mo (10 27 Nm) Mw Rake Number of Stations 200210 2002/10/13 2002/10/19 6 34.654±0.064N 136.947±0.060E 21±7 237±11 16±6 53±13 54±8 1.113±0.081 12.6±3.8 6.0 112 6 200401 2004/01/09 2004/01/15 6 34.527±0.079N 136.785±0.124E 29±6 242±12 17±5 37±9 54±8 1.85±0.14 14.7±4.3 6.0 116 7 200406 2004/06/08 2004/06/11 3 34.543±0.063N 136.644±0.046E 11±4 217±9 10±4 64±14 39±6 0.389±0.044 3.9±1.1 5.7 92 6 200411 2004/11/29 2004/12/02 3 34.588±0.108N 136.747±0.090E 24±6 224±11 12±7 37±17 45±9 1.093±0.023 7.3±3.6 5.8 99 5 200507 2005/07/09 2005/07/13 4 34.582±0.069N 136.693±0.202E 25±5 228±16 13±9 45±16 38±11 1.228±0.093 8.3±3.8 5.9 103 7 200601A 2006/01/07 2006/01/09 2 34.333±0.068N 136.314±0.065E 11±9 205±11 12±10 48±19 23±11 0.310±0.081 1.37±0.95 5.4 80 5 200601B 2006/01/10 2006/01/13 3 34.626±0.103N 136.828±0.202E 27±7 215±11 22±8 51±14 49±9 0.928±0.088 9.2±3.2 5.9 98 9 200601C 2006/01/13 2006/01/16 3 35.079±0.061N 137.122±0.146E 22±7 211±8 25±6 71±14 37±8 0.30±0.11 3.2±1.5 5.6 87 5 200606 2006/05/29 2006/06/03 5 34.551±0.135N 136.761±0.125E 25±7 226±9 17±12 21±11 45±12 1.43±0.12 5.6±3.2 5.8 108 5 200611 2006/11/04 2006/11/09 5 34.780±0.180N 136.989±0.097E 26±6 221±8 3±6 78±13 33±8 0.921±0.070 9.3±2.8 5.9 65 4 200703 2007/03/23 2007/03/26 3 34.627±0.070N 136.524±0.302E 20±6 208±8 25±6 28±12 44±8 0.756±0.089 3.7±1.7 5.6 84 5 200710 2007/10/16 2007/10/20 4 34.809±0.153N 136.980±0.099E 12±6 221±9 23±6 48±11 45±8 0.319±0.082 2.8±1.1 5.6 42 4 200803 2008/03/05 2008/03/08 3 34.629±0.200N 136.975±0.182E 21±5 237±13 39±7 51±14 37±9 0.996±0.038 7.6±2.7 5.9 108 5 200806 2008/06/15 2008/06/19 4 34.373±0.167N 136.447±0.099E 25±4 221±8 17±7 53±14 34±10 0.886±0.078 6.4±2.6 5.8 96 10 8of13

Table 4. Estimated Fault Geometries and Slip Parameters of the Short Term SSE in Tokai Number of Stations Rake Mo (10 27 Nm) Mw Slip (cm) Width Length Dip Strike Depth Longitude Latitude Duration (days) Episode Start Day End Day 200305 2003/05/29 2003/06/03 5 34.830±0.029N 137.718±0.100E 31±5 260±7 3±6 80±13 50±9 0.749±0.072 12.1±3.1 6.0 135 7 200401 2003/12/31 2004/01/04 4 34.909±0.069N 137.791±0.098E 22±7 270±8 32±8 78±16 45±9 0.483±0.079 6.8±2.2 5.8 140 7 200402 2004/02/12 2004/02/18 6 35.025±0.029N 137.490±0.066E 16±5 229±6 7±4 39±10 28±7 0.439±0.058 1.92±0.76 5.5 104 7 200412 2004/12/17 2004/12/21 4 34.964±0.188N 137.324±0.170E 35±2 246±14 36±7 30±9 29±10 1.76±0.19 6.3±3.0 5.8 116 6 200507 2005/07/20 2005/07/24 4 34.929±0.042N 137.718±0.047E 29±2 258±9 30±4 41±8 41±6 0.801±0.080 5.3±1.5 5.7 129 6 200601D 2006/01/16 2006/01/19 3 34.944±0.086N 137.223±0.115E 25±7 249±12 8±7 30±10 29±8 0.62±0.10 2.2±1.0 5.5 124 9 200601E 2006/01/19 2006/01/23 4 34.884±0.070N 137.418±0.037E 28±4 267±4 1±5 13±7 27±6 1.06±0.16 1.50±0.94 5.4 142 8 200608 2006/08/27 2006/09/02 6 35.257±0.117N 138.065±0.149E 26±9 231±8 12±7 28±13 49±10 0.88±0.13 4.8±2.5 5.7 105 5 200702 2007/02/04 2007/02/09 5 35.081±0.182N 137.387±0.253E 33±5 219±7 23±5 37±10 25±7 1.11±0.10 4.0±1.6 5.7 65 7 200709 2007/09/25 2007/10/02 7 35.055±0.075N 137.698±0.066E 27±4 258±5 18±4 52±9 20±6 0.679±0.018 2.78±0.99 5.6 132 6 200710 2007/10/05 2007/10/10 5 34.911±0.044N 137.297±0.072E 28±3 252±4 27±3 39±8 26±7 0.876±0.054 3.6±1.2 5.6 124 8 200801 2007/12/31 2008/01/07 7 35.261±0.107N 138.124±0.101E 31±5 229±9 10±4 51±10 33±10 1.201±0.081 7.9±2.9 5.9 84 9 200805 2008/05/14 2008/05/18 4 34.974±0.041N 137.243±0.044E 30±6 240±10 12±6 33±8 25±7 1.181±0.084 3.9±1.5 5.7 115 6 2009]. However, very few SSEs extended to the southern part of the region in this manner, and the SSE fault geometries were mainly located near the coastline of Ise Bay. As shown in Figure 1, the VLF earthquakes appear to be concentrated at the northeastern edge of the tremor distribution around the coastline of Ise Bay. The distribution of SSE fault planes is very similar to that of VLF earthquakes. [14] In the southern part of the Kii region, the tremor sources are very crowded, whereas fewer VLF earthquakes are observed compared with the northeastern part of the region. No short term SSE were detected by NIED Hi net tiltmeters but, recently, the installation of strainmeters in this area has permitted the detection of some SSEs associated with tremor activity [Fukuda and Sagiya, 2007]. 3.4. Tokai Region [15] In the Tokai region, 13 SSEs have been recognized by means of NIED Hi net tiltmeters, and these can be Figure 5. Spatiotemporal distribution of the SSE activity in southwest Japan. Each rectangle corresponds to an SSE episode. The horizontal axis indicates the location and spatial extension of the fault plane along the strike of the subducting PHS, and the vertical axis indicates the slip for each SSE. The solid line on the uppermost side of each rectangle indicates the origin time of the occurrence of the SSE. 9of13

Figure 6. Frequency distribution of observed tilt steps. The black, blue, and red lines indicate the distribution of observed tilt vectors combined from the steps in north south and east west components in the Shikoku, Kii, and Tokai regions, respectively. roughly divided into two groups: a northeastern group and a western group (Figure 4 and Table 4). SSE activity in the western group occurred more frequently than that in the northeastern group. The interval between SSEs in the western group was about 6 months to 1 year. The average length and width of the SSEs were 45 km and 33 km, respectively. The average moment magnitude of the SSEs in this region was 5.7. The slip is ranging from 0.4 to 1.7 cm. The estimated moment magnitude is smaller than that in western Shikoku. This is caused by difference in the distribution of observed tilt steps as shown in Figure 6. Many VLF earthquakes also occurred in the western group. This concentration of VLF and SSE fault planes is the same as in the northeastern Kii region. [16] In the Tokai region, the SSEs in this region were detected after 2003. In the southeastern side of the shortterm SSE source regions, a long term SSE occurred from the end of 2000 to 2005 and was detected by GPS observation network [Ozawa et al., 2002]. After the first detection of a short term SSE in the Tokai region in December 2004 by NIED Hi net tiltmeters [Hirose and Obara, 2006], Kobayashi et al. [2006] succeeded in detecting strain changes that were coincident with tremor activity, and they classified the short term SSE activities from 1984 to 2005. As the result, the short term SSEs were found to have occurred before the beginning of the long term SSE. However, during of the long term SSE, weak short term SSEs also occurred frequently. This increase in the frequency of short term SSEs is similar to the case in the Bungo Channel. 3.5. Moment Release Rate of SSE [17] Figure 7 shows the estimated spatial variation in cumulative moment release through the occurrence of shortterm SSEs from 2001 to 2008. Some spatial peaks coincide with the distribution of VLF earthquakes, as shown in Figure 1. To investigate the temporal history of the moment release in the SSE source area, we evaluated the moment release for three rectangular areas that included spatial peaks in the distribution of SSE fault planes in the Shikoku, Kii and Tokai regions. These rectangular areas for evaluation of the moment release were selected by using the average size of the SSE fault model estimated for each region. [18] Figure 8 shows the temporal history of cumulative moment release by SSEs. Here we choose the detected SSEs that included the spatial peaks of moment release for each of the three regions, as shown in Figure 7. The moment accumulation by the subducting PHS in the evaluation rectangle is obtained by multiplying the subducting rate of 6.2 cm/yr [Miyazaki and Heki, 2001] by the area of the averaged SSE fault plane and the rigidity of 40 GPa. By comparing the moment calculated from plate subduction with the cumulative moment release estimated from SSEs for each region, we obtained rates of seismic coupling of 65%, 39%, and 35% for the western Shikoku, northeastern Kii, and western Tokai regions, respectively. 4. Discussion [19] From tilting observation made by the NIED Hi net, short term SSEs accompanied by tremor activity were Figure 7. Accumulated moment release of the SSE along the strike of the subducting Philippine Sea Plate derived from estimated SSEs. Figure 8. Estimated cumulative moment release of SSEs in three regions: western Shikoku, northeastern Kii, and Tokai. The three bold lines are cumulative moments of estimated SSEs for each region. The dashed lines denote the moments accumulated from the plate motion of the subducting PHS against the Amurian Plate with a convergence rate of 6.2 cm/yr [Heki and Miyazaki, 2001]. 10 of 13

Figure 9. Cumulative frequency distribution of SSEs against the moment magnitude. detected mainly in three regions: western Shikoku, northeastern Kii and Tokai. These SSE episodes recurred at intervals of approximately 6 months. The estimated moment magnitude of the SSEs ranged from 5.4 to 6.2, and the estimated slip was approximately 1 cm for each SSE. In northern Cascadia, similar geodetic SSEs recur at intervals of 13 16 months [Miller et al., 2002], the moment magnitude is 6.5 6.8, and the slip is 2 4 cm[dragert et al., 2004; Schwartz and Rokosky, 2007]. The extent of the segment in northern Cascadia is longer than 200 km, so the spatial extent of the segment could be directly related to the magnitude of the SSEs and may control the recurrence interval. Obara [2010] investigated segmentation in the belt like tremor zone along the strike of the subducting PHS on the basis of the spatiotemporal characteristics of tremor activity and obtained a segment length of approximately 100 km in these regions. Therefore, the size of SSEs and crustal deformation caused by SSEs in southwest Japan are smaller than the corresponding values in Cascadia. [20] The rate of moment release by the detected SSEs is estimated to be roughly 40 60% in the three active regions along the tremor source belt, and is higher in western Shikoku than in northeastern Kii and Tokai. There are two main reasons that might explain the along strike variations in the rate of moment release by SSEs. The first is the difference in the plate convergence rate in those regions. Heki and Miyazaki [2001] pointed that the rate of convergence between the PHS and the North American Plate is only half that between the PHS and the Amurian Plate. The Tokai region, which has a low rate of moment release by SSEs, is located in the boundary area of the Amurian and North American Plates. Taking into consideration the along strike variation in the plate convergence, the rate of the moment release might be nearly the same as that in western Shikoku. Therefore, an estimation of the relative plate motion is essential for evaluating the contribution of SSEs to the release of strain accumulated by plate subduction. [21] The second possible reason is the difference in the capability to detect SSEs of different sizes. In this analysis, the estimated moment magnitudes of SSEs were larger than 5.4. The cumulative frequency distribution of the SSE against the moment magnitude, as shown in Figure 9, shows that SSEs of magnitudes greater than 5.7 are detected completely, and smaller SSEs are sometimes not detected; this is related to the capability of NIED Hi net tiltmeters to detect SSEs. Weak SSEs are difficult to be detected by means of ground tilting observations because of the presence of noise. If some SSEs are not detected, the rate of the cumulative moment release is underestimated in comparison with its true value. Therefore, it is possible that undetected SSEs occur, particularly, in the Tokai and Kii regions. Obara [2010] proposed a linear relationship between the moments of detected SSEs and the number of tremor solutions in each corresponding tremor episode. This suggests that minor tremor episodes could be caused by weak SSEs, that are not detected geodetically because of their weak signals. Actually, the average moment magnitude of SSEs estimated in this paper is different in each region. In western Shikoku, the average moment magnitude is 6.0, which is larger than the values for the Kii and Tokai regions. This indicates that the source size of SSEs in the Kii and Tokai regions could be smaller than that in western Shikoku. Therefore, the lower rate of moment release by short term SSEs in the Kii and Tokai regions suggests that the characteristic size of the SSE fault is smaller than that in western Shikoku. This could reflect an along strike variation in the seismic coupling property. [22] In this analysis, SSEs in the Kii and Tokai regions have been detected since 2002 and 2003, respectively, whereas the SSEs in western Shikoku have been detected since 2001. This is the result of an improvement in detection capability through an increase in the number of Hi net stations during this period. In these regions, the size of SSEs is smaller than that in western Shikoku, and the detection capability is strongly affected by the distribution of the observation stations. Actually, a few stations in the Kii and Tokai regions did observe tilt steps accompanied by tremor episodes between 2001 and 2003. In addition, the occurrence of the long term SSE [Ozawa et al., 2002] may affect the detection of short term SSEs in the Tokai region. In particular, during the Tokai long term SSE from 2003 to the beginning of 2004, the rate of moment release caused by the long term SSE was accelerated, and tremor episodes occurred frequently [Obara, 2010]. During this high activity of tremors, short term SSEs were frequently identified in this region by means of the strainmeters used by the JMA; these SSEs were reported by Kobayashi et al. [2006], who estimated that moment of each SSE in this period was smaller than that of the SSEs that occurred in other periods at 6 monthly intervals. The network of Hi net stations in the Tokai region has been strengthened since 2003. We were therefore able to detect the frequent occurrence of relatively small SSEs by using the NIED Hi net tiltmeters. This frequent occurrence of small SSEs could reflect presence of small asperities that were ruptured as the result of strain accumulation by the occurrence of the long term SSE. This suggests the existence of small size inhomogeneity on the transition zone in the Tokai region. [23] In this analysis, we succeeded in estimating the source parameters of SSEs in the eastern Shikoku region for the first time. The results confirm that the tilt step with a small amplitude that is coincident with tremor activity in eastern Shikoku was caused by a short term SSE. On the other hand, in the southern part of the Kii region, SSEs 11 of 13

accompanied by tremor activity have not been detected by NIED Hi net tiltmeters, but have been detected by strainmeters [Fukuda and Sagiya, 2007]. This emphasizes that tremor activity is triggered by SSEs. The detection of such small crustal deformations depends on the site location of the detectors in relation to the SSE fault geometry and the noise level. Compared with the detection of SSE, the detection of tremor is easier in these regions. [24] The inversion method used in this paper assumes a homogeneous slip distribution in a rectangular fault geometry for each SSE. However, the slip distribution is expected to be inhomogeneous on the fault plane, because the distribution of tremors and VLFs is inhomogeneous. Hirose and Obara [2010] developed a time evolution analysis to estimate the slip distribution by setting many subfaults on the possible fault geometry. As the result, although the slip area for each SSE is smaller and the maximum slip is larger than those estimated in this paper, the moment is nearly the same. In particular, if the seismic moments that we divided into several independent faults in the case of distinct migrating events are added together, the result is equal to that obtained by Hirose and Obara [2010]. [25] As shown in Figure 4, the located SSE fault geometries are not homogeneously distributed, but are concentrated in some regions. This inhomogeneous distribution of SSE faults is similar to that of VLF earthquakes. For example, on both sides of Ise Bay, the VLF earthquakes and SSEs are concentrated near Ise Bay in the Kii and Tokai regions. The coherency between SSE and VLF activities suggests a common process in the occurrence of these events. For example, a VLF earthquake may be a part of an SSE and the integration of a small slip by a VLF earthquake could be represented as an SSE. 5. Conclusion [26] The source parameters for short term slow slip events (SSEs) in southwest Japan from 2001 to 2008 were systematically compiled from NIED Hi net tiltmeter data. The estimated SSE fault geometries were distributed in the tremor source area along the strike of the subducting PHS. There are three active regions of SSEs: western Shikoku, northeastern Kii, and Tokai. Moreover, we detected the fault parameters for two SSE episodes in eastern Shikoku. In the three active regions, SSEs occur repeatedly at intervals of approximately 6 months. The characteristic fault width of an SSE is 40 km in all regions. This suggests that the average width of the transition zone in the dip direction has a similar length scale. The moment magnitudes of detected SSE ranged from 5.4 to 6.2, and their slip length was approximately 1 cm. The frequency distribution of the moment magnitude showed a regional difference and was larger in western Shikoku than in the Kii and Tokai regions. The rate of moment release by the detected slow slip event is 40 60% of the expected moment accumulation from the relative plate motion of PHS against the Amurian Plate. This alongstrike variation in the rate of moment release may reflect regional differences in the convergence rate and/or a variation in the characteristic size of the SSE fault plane. This may reflect differences in frictional properties along the plate interface. [27] Acknowledgments. The authors would like to thank two anonymous reviewers for their valuable comments and kind suggestions. We would like to thank Y. Ito, T. Matsuzawa, and T. Maeda for providing the VLF catalog. We also extend our gratitude to the staff of NIED for their valuable assistance. We used the Generic Mapping Tools [Wessel and Smith, 1998] for drawing maps. This research was carried out as a part of a NIED project entitled Research Project for Crustal Activity Based on Seismic Data. References Ando, M. (1975), Source mechanisms and tectonic significance of historical earthquakes along the Nankai Trough, Japan, Tectonophysics, 27, 119 140, doi:10.1016/0040-1951(75)90102-x. Dragert, H., K. Wang, and T. S. James (2001), A silent slip event on the deeper Cascadia subduction interface, Science, 292, 1525 1528, doi:10.1126/science.1060152. Dragert, H., K. Wang, and G. Rogers (2004), Geodetic and seismic signatures of episodic tremor and slip in the northern Cascadia subduction zone, Earth Planets Space, 56, 1143 1150. Fukuda, M., and T. Sagiya (2007), Precursory slow crustal deformation before short term slow slip event in January 2006, recorded at Shingu borehole station southern Kii Peninsula, Eos Trans. AGU, 88(52), Fall Meet. Suppl., Abstract T21A 0361. Heki, K., and S. Miyazaki (2001), Plate convergence and long term crustal deformation in central Japan, Geophys. Res. Lett., 28, 2313 2316, doi:10.1029/2000gl012537. Hirose, H., and K. Obara (2005), Repeating short and long term slow slip events with deep tremor activity around the Bungo channel region, southwest Japan, Earth Planets Space, 57, 961 972. Hirose, H., and K. Obara (2006), Short term slow slip and correlated tremor episodes in the Tokai region, central Japan, Geophys. Res. Lett., 33, L17311, doi:10.1029/2006gl026579. Hirose, H., and K. Obara (2010), Recurrence behavior of short term slow slip and correlated nonvolcanic tremor episodes in western Shikoku, southwest Japan, J. Geophys. Res., 115, B00A21, doi:10.1029/ 2008JB006050. Hirose, H., K. Hirahara, F. Kimata, N. Fujii, and S. Miyazaki (1999), A slow thrust slip event following the two 1996 Hyuganada earthquakes beneath the Bungo Channel, southwest Japan, Geophys. Res. Lett., 26, 3237 3240, doi:10.1029/1999gl010999. Hyndman, R. D., M. Yamano, and D. A. Oleskevich (1997), The seismogenic zone of subduction thrust faults, Isl. Arc, 6, 244 260, doi:10.1111/ j.1440-1738.1997.tb00175.x. Ito, Y., K. Obara, K. Shiomi, S. Sekine, and H. Hirose (2007), Slow earthquakes coincident with episodic tremors and slow slip events, Science, 315, 503 506, doi:10.1126/science.1134454. Ito, Y., K. Obara, T. Matsuzawa, and T. Maeda (2009), Very low frequency earthquakes related to small asperities on the plate boundary interface at the locked to aseismic transition, J. Geophys. Res., 114, B00A13, doi:10.1029/2008jb006036. Kobayashi, A., T. Yamamoto, K. Nakamura, and K. Kimura (2006), Shortterm slow slip events detected by the strainmeters in Tokai region in the period from 1984 to 2005 (in Japanese with English abstract), J. Seismol. Soc. Jpn., 59, 19 27. Miller, M. M., T. Melbourne, D. J. Johnson, and W. Q. Sumner (2002), Periodic slow earthquakes from the Cascadia subduction zone, Science, 295, 2423, doi:10.1126/science.1071193. Miyazaki, S., and K. Heki (2001), Crustal velocity field of southwest Japan: Subduction and arc arc collision, J. Geophys. Res., 106, 4305 4326, doi:10.1029/2000jb900312. Obara, K. (2002), Nonvolcanic deep tremor associated with subduction in southwest Japan, Science, 296, 1679 1681, doi:10.1126/science.1070378. Obara, K. (2010), Phenomenology of deep slow earthquake family in southwest Japan: Spatiotemporal characteristics and segmentation, J. Geophys. Res., doi:10.1029/2008jb006048, in press. Obara, K., and H. Hirose (2006), Non volcanic deep low frequency tremors accompanying slow slips in the southwest Japan subduction zone, Tectonophysics, 417, 33 51, doi:10.1016/j.tecto.2005.04.013. Obara, K., and S. Sekine (2009), Characteristic activity and migration of episodic tremor and slow slip events in central Japan, Earth Planets Space, 61, 853 862. Obara, K., H. Hirose, F. Yamamizu, and K. Kasahara (2004), Episodic slow slip events accompanied by non volcanic tremors insouthwest Japan subduction zone, Geophys. Res. Lett., 31, L23602, doi:10.1029/ 2004GL020848. Obara, K., K. Kasahara, S. Hori, and Y. Okada (2005), A densely distributed high sensitivity seismograph network in Japan: Hi net by National 12 of 13