Array Back-Projection Imaging of the 2007 Niigataken Chuetsu-oki Earthquake Striking the World s Largest Nuclear Power Plant

Transcription

1 Bulletin of the Seismological Society of America, Vol. 99, No. 1, pp , February 2009, doi: / Array Back-Projection Imaging of the 2007 Niigataken Chuetsu-oki Earthquake Striking the World s Largest Nuclear Power Plant by Ryou Honda and Shin Aoi Abstract The 2007 Niigataken Chuetsu-oki earthquake occurred near the Kashiwazaki Kariwa nuclear power plant in Japan, the largest in the world. The strong motions were recorded by seven seismometers installed at the foundation slab (base-mat) of the plant and exceeded the design level of the ground motion for the plant. The strong motion observed by the seismographs in and around the plant show high coherency with three significant pulses. In order to understand the cause of these pulses, the rupture process of the earthquake was estimated using these seismograms. The seismograph network was taken into account as a dense array and semblanceenhanced waveform stacking was performed. By projecting the power of the stacked waveforms onto the fault plane, the asperities that generated significant pulses were successfully separated. The first and third pulses were generated at the hypocenter and the southwest edge of the rupture zone, respectively. The rupture propagated toward the southwest and terminated offshore from the power plant. The overall pattern of the imaged asperities coincides well with the slip distribution determined by conventional waveform inversions. Introduction The Niigataken Chuetsu-oki earthquake occurred in Japan on 16 July 2007, and the moment magnitude of the earthquake was estimated as M w (e.g., National Research Institute of Earth Science and Disaster Prevention [NIED], 2007; Matsumoto et al., 2008). The source region of the earthquake was located in the Japan Sea, slightly offshore from the Niigata prefecture in central Japan. The focal region was in the Niigata Kobe Tectonic Zone (NKTZ) (Sagiya et al., 2000), in and around which several large earthquakes have recently occurred (e.g., the 2007 Noto Hanto earthquake and the 2004 mid-niigata prefecture earthquake), as shown in Figure 1. A Japan Meteorological Agency (JMA) seismic intensity of 6 (nearly equal to IX on the modified Mercalli intensity [MMI] scale) was observed around Kashiwazaki City, which was the area that was most severely damaged by the earthquake. The mainshock yielded three strong pulses that commonly appeared in accelerograms recorded in and around the focal region. These pulses are thought to be the cause of the damage suffered at the Kashiwazaki Kariwa nuclear power plant, which was approximately 16 km from the epicenter. The maximum peak ground acceleration (PGA) observed at the power plant was 680 cm=sec 2 at the foundation slab (base-mat; V s is about 500 m=sec) of the reactor, which was in excess of the design level of the power plant, 273 cm=sec 2. At the power plant, acceleration waveforms were recorded at 33 stations with intervals of several hundred meters, 27 stations in the nuclear reactor or turbine buildings, and six stations at the surface or in boreholes. In this study, we attempted to retrieve the sources of the three remarkable pulses, which are indicated by the short time window (time windows 1, 2, and 3) in Figure 2. The pulses appeared in all of the records at the nuclear power plant. It is a rare case that near-field waveforms with high coherency are recorded by a dense seismograph network. The use of such records can allow us to obtain information of the earthquake source with array techniques such as semblance analysis or stacking (e.g., Rost and Thomas, 2002). Overall locations of the asperities were obtained by conventional waveform inversions (Aoi et al., 2008) and the empirical Green s function method (Irikura et al., 2007). However, detailed rupture propagation could not be imaged by such previous studies. The focal region has thick sedimentary layers (Shibutani et al., 2005) that cause waveform distortions or non-source-origin waves such as reflection or surface waves. Therefore, it is difficult to calculate accurate Green s functions for this earthquake; the inaccuracy of the Green s functions leads to an estimation error of the rupture process. Therefore, one of the purposes of this study is to image the rupture propagation using near-field array data and to clarify the origin of the strong pulses that caused remarkable ground motion in the focal region. In order to retrieve the rupture process, a back-projection method based on slant stack and semblance analysis was applied. 141

2 142 R. Honda and S. Aoi (a) ' 38 00' 2007 Chuetu-oki Earthquake (Mw 6.7) ' 38 (b) Study area km KKZ5R KKZ6R2 KKZ7R2 KKZ5G ' 20 km Dip Strike L65059 L65058 NIG018 KKZ4R2 KKZ3R2 KKZ2R2 KKZ1R2 KKZ1G1 KSHSG1 Figure 1. (a) Station distribution for the array analysis. Black triangles indicate stations used in the analysis. The bold solid rectangle in the center of the map indicates the assumed fault plane. The black star indicates the epicenter of the 2007 Niigataken Chuetsu-oki earthquake (M w 6.7). Gray and white stars indicate the epicenters of the 2007 Noto Hanto earthquake (M w 6.6) and the 2004 mid-niigata prefecture earthquake (M w 6.7), respectively. The white circle indicates the K-NET station (NIG018). (b) The station distribution at the Kashiwazaki Kariwa nuclear power plant. KKZ1G1 and KKZ5G1 are installed at the surface. The other stations are installed at the foundation slabs (basemat) of the seven reactors or in a borehole. Preparation of Analysis Strong ground motions observed by a dense observer array are useful for conducting array analyses based on slant stack and semblance analyses. Resolution of the source location is strongly affected by the array size. Strong groundmotion records observed at 10 stations in the nuclear power plant and two additional stations installed by the local government (Niigata prefecture) were used to obtain accurate constraints for the source location of coherent waves. Figure 1 shows the station distribution. The stations of the local government (L65058 and L65059) and two stations in the power plant (KKZ1G1 and KKZ5G1) are installed on the ground surface, and the other stations in the power plant are installed on the foundation slabs (base-mat) of the reactors or in a borehole (KSHSG1). Because the precise locations of stations in the nuclear power plant are not given, we obtained the locations from a 1=30; 000 road map of Niigata prefecture, and Table 1 shows the station locations that were used in this study. Because the timing of the seismograms among the stations was not synchronized and no information was obtained regarding absolute timing, we picked the onset of the P-wave arrival and aligned the waveforms by regarding the onset as the origin of time axis. Finally, the time for each trace was shifted according to the theoretical travel time of the P wave, as shown in Figure 2, where the origin of the time axis is the origin time. The waveforms were band-pass filtered in the frequency range of 1.0 to 20.0 Hz. To calculate theoretical travel time, the velocity structure model proposed by Ukawa et al. (1984) for depths deeper than 400 m was used. For the shallower part, three velocity structure models, shown in Figure 3, were introduced to consider station correction. Models 1 and 2 correspond to the velocity structure for the southwestern half of the power plant and that for the northeastern half, using the borehole data from KKZ1G1 and KKZ5G1, respectively. For the other stations (i.e., KSHSG1, L65058, and L65059), model 3 was used. The focal mechanism of the mainshock was a reverse fault type with the P axis in the west-northwest eastsoutheast direction (Matsumoto et al., 2008), which is consistent with the maximum tectonic compressive stress in and around the mid-niigata area. However, the fault plane of the mainshock could not be distinguished from the aftershock distribution. An approximate southwest alignment of the aftershock distribution, determined by NIED routine analysis, could be discriminated; however, it was difficult to reveal the fine structure of the mainshock. Yukutake et al. (2008) estimated precise aftershock distributions using the differential arrival time obtained by both manual picking and waveform cross-correlation analysis (Waldhauser and Ellsworth, 2000). They concluded that the dominant rupture had propagated in

3 Array Back-Projection Imaging of the 2007 Niigataken Chuetsu-oki Earthquake 143 Depth (m) Model 1 Model 2 Model 3 S P Velocity (m/s) Figure 2. The three velocity structures used for the analysis. Models 1 and 2 were used for stations at the southwestern and northeastern sides of the power plant, respectively. Model 3 was used for the others (KSHSG1 and the local government stations). the main fault plane, which dips toward the southeast with partly different dip angles. We employed the geometry of a southeast-dipping fault inferred from well-resolved aftershock distributions. The strike and dip angles obtained for the focal mechanism from F-net data (NIED, 2007) were N49 E and 42, respectively. The fault geometry proposed by Aoi et al. (2008) was used to estimate the rupture process. The hypocenter was N, E, and 8 km deep. The assumed fault plane was 30 km long and 24 km wide, which covered the aftershock distribution, and was divided into 180 subfaults of 2 2 km. Method of Analysis Slant stack processing (e.g., Yilmaz, 1987) was applied to the observed strong ground-motion data after band-pass Table 1 Station Locations Based on WGS84 Station Longitude ( ) Latitude ( ) Depth (m) KKZ1R2* KKZ2R2* KKZ3R2* KKZ4R2* KKZ5R2* KKZ6R2* KKZ7R2* KSHSG1* KKZ1G1* KKZ5G1* L L Station locations in the reactor buildings (indicated by an asterisk) were provided by TEPCO; however, precise geographic positions of the stations were not provided. These locations were obtained from a 1=30; 000 road map of the Niigata prefecture. filtering using the frequency range of 1.0 to 20.0 Hz. Travel times for each subfault and station pair were calculated using the ray theory, and the waveforms were stacked according to the travel-time difference after rotating into radial and transverse directions. A slant-stacked waveform for a particular subfault can be regarded as a time series of energy release at the subfault. The stacking procedure is shown as follows: For the jth source location, the seismograms U are summed to make a stack A j as a function of time t: A j t R ij X N i 1 U t t s j dt ij ; (1) where R ij is the propagation distance from the jth source to the ith station and t s j is the S-wave travel time from the jth source to the reference station, KKZ7R2. dt ij is the traveltime difference between the reference site and the ith site for the jth source. In order to enhance the stacking effect, a semblance-enhanced stack (Matsumoto et al., 1999) was introduced. In this process, the semblance value calculated in a time window is multiplied by the stack A j. The semblance value S is calculated as follows: S j t P Mk 1 P N i 1 U t k dt ij 2 N P M k 1 P N i 1 U 2 t k dt ij ; (2) where M is the number of samples in the time window and N is the number of sites. The length of the time window is 1 sec. The range of t k is from t t s j 0:5 (k 1) tot ts j 0:5 (k M). After calculation of the stacked waveforms A j t at a certain time t, which is the center of the time window, the semblance-enhanced stack on a subfault E j t can be obtained by S j t A j t. The semblance-enhanced stack was independently calculated for two horizontal components (radial and transverse) from the array with respect to each

4 144 R. Honda and S. Aoi Whole time window T.W. 1 T.W. 2 T.W. 3 EW cm/s NS KKZ7R cm/s L KKZ5R KKZ6R KKZ5G KKZ4R KSHSG KKZ3R KKZ2R KKZ1R KKZ1G L sec Figure 3. Observed waveforms that were band-pass filtered between 1 and 20 Hz. The whole time window was used to estimate the total energy release during the earthquake. Short time windows 1, 2, and 3, which separate the strong pulses, were used to obtain the corresponding asperity shown in Figure 4. All of the waveforms were shifted by the theoretical arrival time of the P wave and were normalized by the maximum value in each component. The origin of the time axis is the origin time. Triangles indicate the theoretical arrival time of the S wave from points (a), (b), and (c) indicated in Figure 4. The numbers at the lower right of each trace indicate the maximum value in cm=sec 2. subfault. By repeating this procedure for all subfaults and times, a time series of released energy was obtained for the two horizontal components of each subfault. The estimated time series was obtained with consideration for geometrical spreading, but amplitude variation related to radiation patterns or the directivity effect was ignored. Finally, the integration of je j t j was taken with respect to time, and the average of the integrated value for two components was regarded as the total released energy of the subfault. The stacking procedure can constructively sum the radiated energy from a given source and cancel out the remaining energy present in the seismograms. Therefore, the time series of energy release at each subfault can be automatically determined. This provides us with one of the major advantages of array analysis, that is, rupture velocity or rupture duration time need not be assumed, and synthetic seismograms are not required to retrieve the rupture process (e.g., Ishii et al., 2005; Honda et al., 2008). However, because of the small number of samples (i.e., number of stations), the noise in the time series of this analysis is not sufficiently suppressed. As a result, energy was leaked outside of the appropriate time range and the total energy release was overestimated. To obtain a correct time series for the energy release, a limitation for the length of the time window in which energy could be released was introduced to each subfault. To introduce the limitation, the upper limit of the rupture velocity was set to 2800 m=sec, and the duration time was set to less than 4 sec. The amplitude of the stacked wave outside of the time window is replaced by 0, and the gaps at both ends of the time window are smoothed with a cosine taper. Because the assumed upper limit of the rupture velocity is 76% 85% of the S-wave velocity at the fault depth in the velocity model, these assumptions for rupture velocity are reasonable. Results The rupture process and released energy distribution were estimated from dense strong-motion array data. The sources of three significant pulses apparent in the waveforms were clearly imaged. Figure 4a d shows the released energy distributions corresponding to time windows 1, 2, 3, and the whole time window shown in Figure 2, respectively. The first and third pulses were generated around the hypocenter and the southwest part of the fault plane, respectively. Although the source of the second pulse can be imaged as a peak in Figure 4b, the peak is not clearly seen in Figure 4d, which shows the total released energy distribution. As shown in Figure 2, the first and the third pulses are broad and large, and they appear coherently in all records. On the other hand, the second pulse appears to be less coherent, because the second pulse consists of higher frequencies than the first and third pulses. This leads to poor stacking power for the second pulse. Because a 1 sec time window was applied to calculate the semblance value, the weight of the second pulse, includ-

5 Array Back-Projection Imaging of the 2007 Niigataken Chuetsu-oki Earthquake 145 ing the high-frequency component, is lowered. The gray circles in Figure 4 indicate aftershocks that were located with hypodd using both manually picked and cross-correlation data from Yukutake et al. (2008). These aftershocks occurred at the outer regions of a large energy release. The overall image of energy release agrees with that of the waveform inversion analysis reported by Aoi et al. (2008). Remarkable slip regions appear close to the hypocenter and northwest of the hypocenter. The differences between the present results and those of Aoi et al. (2008) may indicate a complex source process, where the source of high-frequency radiation and low-frequency radiation is different (e.g., Madariaga, 1971; Spudich et al., 1984), because the present results obtained from acceleration waveforms possibly include the effect of high-frequency radiation. However, significant pulses apparent in the records consist of the lower frequency phases, and the present results are characterized by the lower frequency phases, the same as those reported by Aoi et al., (2008), because a 1 sec time window was adopted to calculate the semblance value. Waveform inversion analysis using observations in a wide area must be affected by complex velocity structures, which lead to poor Green s functions. In addition, the waveform inversion analysis requires parameter constraints, which are not required for the present analysis. Our interpretation of the difference between both results is that they arise from the parameter configurations of each analysis rather than the frequency range used in the analysis. Hereafter, we describe the regions of large energy release as asperities. The rupture propagation was also successfully imaged by plotting the released energy every second, as shown in Figure 5. The ruptured area expanded to the southwest of the fault plane and terminated offshore from the power plant. The rupture started at the hypocenter and unilaterally propagated toward the southwest. The rupture time is shown as a function of the epicentral distance in Figure 5b. When the accumulated energy reached 10% of the maximum on each subfault, it was taken as the rupture time. The rupture propagation velocity estimated by the least-squares method was approximately 2450 m=sec. Discussion and Conclusions Figure 4. (a) Total released energy distribution for time window 1, as shown in Figure 2. The total released energy was normalized by the maximum value. Gray circles indicate aftershocks determined by the double difference method (Yukutake et al., 2008). The star represents the epicenter. The white dotted line indicates the source region, from which the possible arrival time is shown in Figure 6, and the arrow labeled (a) indicates the peak. (b) Total released energy distribution for time window 2. The white dotted line and the arrow labeled (b) represent the same as described in (a). (c) Total released energy distribution for time window 3. A white dotted line and an arrow labeled (c) represent the same as described in (a). (d) Total released energy distribution for the whole time window. (e) Slip distributions obtained by Aoi et al. (2008). Here we attempt to determine the source of the three strong pulses observed at the K-NET (Kinoshita, 1998) station, NIG018, which could not be used for this analysis or waveform inversions, due to the strong nonlinear effect on the waveform. The station is in Kashiwazaki City, where the damage by the earthquake was the most severe. Figure 6 shows the accelerogram observed at NIG018. Dotted lines indicate the time window, which allows the arrival of signals from the peaks of each asperity, which are shown by the white dotted lines in Figure 4. Arrival times from sources, indicated by (a), (b), and (c) in Figure 4, are shown by arrows. Although the onset of each pulse is not easily recog-

6 146 R. Honda and S. Aoi a b c a b c NIG018 EW NS sec Figure 6. Observed waveforms at NIG018. The theoretical S-wave arrival time from points (a), (b), and (c) in Figure 3 are shown by arrows. Possible arrival time windows of the S wave from peaks surrounded by dotted lines in Figure 3 are indicated by dotted lines. Figure 5. (a) Snapshots plotting the energy release every second. The major rupture unilaterally propagated from the hypocenter to the southwest. The numbers in the upper left of each snapshot indicate the time after initiation. (b) Rupture time as a function of the hypocentral distance. The rupture velocity estimated by the least-squares method was 2450 m=sec. The solid line indicates the rupture times corresponding to 2450 m=sec. nizable, due to pulses overlapping with each other, the three pulses are separated by the estimated arrival times (a), (b), and (c), and are shown in the time window as a dotted line. Consequently, it can be concluded that the sources of the three strong pulses observed at NIG018 are identical to the three asperities shown in Figure 4. This supports our assumption that if waveforms observed in Kashiwazaki City are used, despite nonlinear effects, the peak may become clear. Because the resolution of the array analysis is dependent on the array size, even though the waveform is distorted by the nonlinear effect, sufficient improvement of the resolution by array extension can be expected if the timing and width of the pulse are not significantly affected. There are some previous studies of the source process using array data (e.g., Ishii et al., 2005; Honda et al., 2008). They successfully imaged the rupture process or location of asperity; however, the size of the earthquakes that they could analyze was constrained down to M 8 because they used a nationwide seismograph network (Hi-net or K-NET) with site separations longer than 20 km. Fletcher et al. (2006) tried to identify the rupture front of the 2004 Parkfield, California, earthquake (M 6) using high-frequency arrivals at a shortbaseline seismic array. They obtained the source and timing of high-frequency phase initiation and estimated the rupture velocity. In this article, we obtained an average rupture velocity of 2450 m=sec and images of the rupture propagation. The rupture time plotted in Figure 5b indicates the time variation of rupture initiation. The rupture velocity near the hypocenter appeared to be faster than that of the distant part. This implies a complex rupture process on the fault plane. Although the rupture velocity near the edge of the fault plane is close to 2800 m=sec, as shown in Figure 4, this does not mean that the rupture accelerates, because the energy release at the edge is very small and the stacked waves are composed of noise. Our results clearly show that analyses using array data provide an effective way for imaging rupture propagation and determining asperity locations as the source of coherent waves, even in the case of an M 7 class earthquake. In addition, even in the case of a thick sedimentary layer, which may mislead results of waveform inversions, or in the case of rupture velocity variations on the fault plane (Honda et al., 2008), the back-projection method can be a useful tool to obtain the rupture process.

7 Array Back-Projection Imaging of the 2007 Niigataken Chuetsu-oki Earthquake 147 Data and Resources Seismograms in the nuclear power plant were collected by the Tokyo Electric Power Company (TEPCO). Data can be obtained by submitting an order to the Association for Earthquake Disaster Prevention ( jp.com/). Seismograms observed at L65058 and L65059 cannot be released to the public. All figures in this article were made using the Generic Mapping Tools ( gmt; Wessel and Smith, 1998). Acknowledgments We would like to thank TEPCO and the Niigata prefectural government for providing the strong ground-motion data. We are very grateful for valuable comments by anonymous reviewers. References Aoi, S., H. Sekiguchi, N. Morikawa, and K. Kunugi (2008). Source process of the 2007 Niigata-ken Chuetsu-oki earthquake derived from nearfault strong motion data, Earth Planets Space (in press). Fletcher, J., P. Spudich, and L. Baker (2006). Rupture propagation of the 2004 Parkfield, California, earthquake from observation at the UPSAR, Bull. Seismol. Soc. Am. 96, S129 S142. Honda, R., S. Aoi, H. Sekigushi, and H. Fujiwara (2008). Imaging an asperity of the 2003 Tokachi-oki earthquake using a dense strong-motion seismograph network, Geophys. J. Int. 172, no. 3, Irikura, K., T. Kagawa, K. Miyakoshi, and S. Kurahashi (2007). Ruputre process and strong ground motions of the 2007 Niigataken Cyuetsu- Oki earthquake directivity pulses striking the Kashiwazaki Kariwa nuclear power plant, in The 2007 Southern California Earthquake Center Annual Meeting, Palm Springs, California. Ishii, M., P. M. Shearer, H. Houston, and J. E. Vidale (2005). Extent, duration and speed of the 2004 Sumatra Andaman earthquake imaged by the Hi-Net array, Nature 435, , doi /nature Kinoshita, S. (1998). Kyoshi Net (K-NET), Seism. Res. Lett. 69, Madariaga, R. (1971). High-frequency radiation from crack (stress drop) models of earthquake faulting, Geophys. J. R. Astr. Soc. 51, Matsumoto, T., H. Matsubayashi, and T. Kazakami (2008). Spatial distribution for moment tensor solutions of the Niigataken Chuetsu-oki, earthquake in 2007 and aftershocks, Earth Planets Space (in press). Matsumoto, S., K. Obara, K. Yoshimoto, T. Saito, A. Hasegawa, and A. Ito (1999). Imaging of inhomogeneous structure of the crust beneath Ou backbone range, northeastern Japan, based on small aperture seismic array observations, Zisin 52, (in Japanese with English abstract). National Research Institute of Earth Science and Disaster Prevention (2007). F-net Broadband Seismograph Network, Earthquake mechanisms information, /update1/index.html (last accessed October 2008). Rost, S., and C. Thomas (2002). Array seismology: methods and applications, Rev. Geophys. 40, no. 3, 1008, doi /2000RG Sagiya, T., S. Miyazaki, and T. Tada (2000). Continuous GPS array and present-day crustal deformation of Japan, Pure Appl. Geophys. 157, Shibutani, T., Y. Ito, S. Matsumoto, H. Katao, T. Matsushima, S. Ohmi, F. Takeuchi, K. Uehira, K. Nishigami, B. Enescu, I. Hirose, Y. Kano, Y. Kohno, M. Korenaga, Y. Mamada, M. Miyazawa, K. Tatsumi, T. Ueno, H. Wada, and Y. Yukutake (2005). Aftershock distribution of the 2004 mid Niigata prefecture earthquake derived from a combined analysis of temporary online observation and permanent observations, Earth Planets Space 57, Spudich, P., and L. N. Frazer (1984). Use of ray theory to calculate high-frequency radiation from earthquake sources having spatially variable rupture velocity and stress drop, Bull. Seismol. Soc. Am. 74, Ukawa, M., M. Ishida, S. Matsumura, and K. Kasahara (1984). Hypocenter determination method of the Kanto Tokai observational network for microearthquakes, Rep. Natl. Res. Cent. Disaster Prev. 53, 1 88 (in Japanese with English abstract). Waldhauser, F., and W. L. Ellsworth (2000). A double-difference earthquake location algorithm: method and application to the Northern Hayward fault, Bull. Seismol. Soc. Am. 90, Wessel, P., and W. H. F. Smith (1998). New, improved version of the Generic Mapping Tools released, Eos Trans. AGU 79, 579. Yilmaz, Ö. (1987). Seismic Data Processing, The Society of Exploration Geophycists, Tulsa, Oklahoma, 526 pp. Yukutake, Y., T. Takeda, and K. Obara (2008). Well-resolved hypocenter distribution using the double-difference relocation method in the region of the 2007 Chuetsu-oki earthquake, Earth Planets Space (in press). Hot Springs Research Institute of Kanagawa Prefecture Iriuda 586, Odawara Kanagawa , Japan ryou@onken.odawara.kanagawa.jp (R.H.) National Research Institute for Earth Science and Disaster Prevention Tennodai 3-1 Tsukuba, Ibaraki , Japan (S.A.) Manuscript received 13 May 2008