A Study on the Prediction of Long-Period Ground Motions from Intraslab Earthquakes Yadab Prasad DHAKAL Candidate for the Doctor of Engineering Supervisor: Prof. Tsutomu SASATANI Division of Architectural and Structural Design Introduction Recently, long-period structures, such as high-rise buildings, huge oil storage tanks, and base-isolated structures, which are susceptible to long-period seismic waves, increase the number in urban areas located on deep sedimentary basins. It is very important for planning of urban seismic disaster mitigation to understand long-period ground motion from future large earthquakes. It is well known that disastrous earthquakes are plate boundary earthquakes such as the 23 Tokachi-oki earthquake (Mw.3) at the subduction zone. However, after severe damage due to the 1993 Kushiro-oki intraslab earthquake (Mw 7.6; depth~ km), we have recognized that large intraslab earthquakes are also disastrous ones at the subduction zone. The 1993 Kushiro-oki earthquake caused damage of many facilities in different sectors such as the water supply, roads and railways, commerce and industry, schools and residential buildings etc. An example of damage of a hospital building from this event is illustrated in Fig. 1. The human casualty was relatively small from this event; only two persons died from the earthquake. However, about people were injured and 116 of them were severely wounded (AIJ, 1995).. Fig. 1 Damage of the Obihiro-dai-ichi hospital from the 1993 Kushiro-oki earthquake. Photo courtesy: AIJ(25). More recently the southern Sumatra earthquake that occurred on September 3, 29 in the subduction zone environment with Mw 7.5 at a focal depth of 1 km, similar to the 1993 Kushiro-oki event, killed more than persons and injured more than 2 people (e.g., EERI, 29). This earthquake also caused damage of many residential buildings and other facilities. Similarly few damaging intermediate depth earthquakes have occurred in the different parts of the world in different tectonic environments (e.g., Frohlich, 26). These observations imply that the future large intermediatedepth events may pose a serious threat to the recently growing number of long-period structures in urban areas located on the deep sedimentary basins because the deep sedimentary basins amplify the long-period ground motions and can generate the basin surface waves with long duration. There are empirical and theoretical methods for the strong motion prediction. It has been found that the empirically predicted long-period response spectra from intraslab earthquakes are several times smaller than the observed ones at the basin sites. This is due to difficulty of effectively including the basin site effects in the empirical attenuation relationship. The long-period ground motions are greatly controlled by the deep velocity structure of the basin. This indicates that we have to rely on the theoretical method based on a threedimensional (3-D) velocity structure model of the sedimentary basin for the prediction of long-period ground motion. The construction of the 3-D basin velocity structures is ongoing in many regions of Japan. However, it is difficult to resolve completely the 3-D velocity structure with a single geophysical exploration method. Therefore, combinations of several kinds of exploration datasets are required for sufficiently resolving the 3-D velocity structure (Koketsu et al., 29). Finally, the resultant velocity model should be checked and verified based on the observed features of earthquake ground motions because the velocity models are constructed primarily for the prediction of long-period ground motions from future earthquakes in Japan. Obtaining a reliable 3-D velocity structure has the key to make the successful prediction by the theoretical method. On the one hand, we have to validate the existing 3-D velocity model based on the observed features of long-period ground motions. On the other hand, it is important to develop methodologies for the improvement of the velocity structure. This study aims to resolve these issues related to the prediction of longperiod ground motions from large intraslab earthquakes. The target area in this study is the Tokachi basin located on eastern part of Hokkaido, Japan; the Tokachi basin extends about km in the north-south direction and about 5 km in the east-west direction (see Fig. 5). Empirical prediction of pseudo-velocity response spectra
Pseudo-velocity response (cm/s) Pseudo-velocity response (cm/s) Empirical prediction equations have been developed and updated for subduction zone events when strong ground motion data have become more available with the operation of dense strong-motion seismograph networks in Japan (e.g., Kanno et al., 26). The prediction of strong ground motion parameters such as the peak ground acceleration and acceleration response spectra are important for developing seismic hazard maps from future earthquakes. In general, the strong ground motion at a site is estimated as a function of earthquake magnitude, distance and site condition. The site condition beneath the strong-motion observation station is variable from site to site. There is a general practice to fit the observed data by grouping them based on soil type or predominant period of the soil below the recording stations (e.g., Zhao et al., 26). Kanno et al. (26) developed a new attenuation relation using the largest number of strong ground motion data available in Japan. They applied a site correction term in their attenuation relations based on the average S-wave velocity in the upper 3 m (AVS3) of the soil. However, the shallow S-wave velocity information does not improve the prediction of long-period ground motions as shown in Fig. 2. In this figure, observed and predicted pseudo-velocity response spectra from the 1999/5/13 Mw 6.1 event are shown for a deep sedimentary site (HKD95, Obihiro). 1 HKD95, R=119 km, 1999/5/13, Mw 6.1 Pre. 1 S.E.1 1 2 3 5 Natural period (s) Fig. 2 Comparison of observed and predicted pseudovelocity response spectra at HKD95 from the 1999/5/13 Mw 6.1 event using the attenuation relation of Kanno et al. (26). The black- and dark grey-line with circles show the observed and predicted values with site effect. The light grey-lines show one standard error of estimation without site effect. We can see in Fig. 2 that the predicted spectral values are huge underestimates of the observed ones in the natural periods longer than about 1 s. Zhao et al. (26) proposed an attenuation relation using site classification based on predominant period. The observed and predicted values using the attenuation relation of Zhao et al. are shown in Fig. 3. In this case also, we can see that the predicted spectral values are several times underestimates of the observed ones in the longer periods. This indicates that the prediction of long-period ground motions using empirical prediction equations is not so reliable at deep basin sites. This large difference between the observed and predicted spectral values at long-periods is the motivation for the theoretical prediction of them in this study. 1 HKD95, R=119 km, 1999/5/13, Mw 6.1 Pre. 1 S.E.1 1 2 3 5 Natural period (s) Fig. 3. Comparison of observed and predicted pseudovelocity response spectra at HKD95 from the 1999/5/13 Mw 6.1 event using the attenuation relation of Zhao et al. (26). The black- and dark grey-line with circles show the observed and predicted values. The light grey-lines show one standard error of estimation. In the above figures, the basin effect or site effect at long-periods is illustrated. The unknown site condition or the inappropriate assignment of the site condition parameter in the empirical prediction equation leads to the large variability between the observed and predicted values using the empirical prediction equation. The assumption of homogeneous path effect throughout the path length between the earthquake source and observation site further adds to the variability in the regions with heterogeneous attenuation structure. There exists a heterogeneous attenuation structure in northern Japan. Therefore, a new attenuation relationship is found by us considering the heterogeneous attenuation structure in northern Japan. For details of the new attenuation relationship, it is recommended to see the paper by Dhakal et al. (29). Here only the results are compared at the HKD95 site for the 1999 event. In Fig., we can see the similar results from all the empirical prediction equations; the predicted spectral values are underestimates of the observed ones in the natural periods longer than about 1 s by a large margin. We can see that our prediction is close to the prediction of Kanno et al. (26) at long-periods and close to the prediction of Zhao et al. (26) at short periods. In Fig., it is also evident that the prediction at short periods is reasonable and improved one with the prediction equation constructed in this study.
Pseudo-velocity response (cm/s) HKD95, R=119 km, 1999/5/13, Mw 6.1 Dhakal et al. (Model-2) Kanno et al. Zhao et al. 1.1 1 2 3 5 Natural period (s) Fig. Comparison of observed and predicted pseudovelocity response spectra at HKD95 from the 1999/5/13 Mw 6.1 event using the empirical prediction equation of Dhakal et al. (29), Kanno et al. (26), and Zhao et al. (26). The black and grey lines (many) with circles show the observed and predicted values. Fig. 5 Index map showing the Tokachi basin and adjoining areas. The black filled stars denote the epicenters of the events used in this study and the beach balls show their focal mechanisms. The white rectangles denote the boundaries of the model regions for the 3-D finite difference simulation. The scale bar shows the depth of seismic basement with S-wave velocity of 3.2 km/s. The white circles denote the strong motion stations used in this study (see Fig. 6). Validation of the Tokachi basin NIED velocity model Here we examine the.5 grade 3-D velocity structure model of the Tokachi basin, constructed by NIED, by 3-D simulation of long-period ground motions for periods longer than 2 s from three nearby earthquakes, which had focal depths of 6 ~ 11 km and Mw of 6.1 ~ 6.7 (Fig. 5). One of the advantages of using these intermediate depth earthquakes for validating the 3-D velocity structure model is that the direct S-wave and the basin-induced surface waves can be easily recognized. The NIED velocity structure model comprises five sedimentary layers with S-wave velocities of m/s, 7 m/s, 1 m/s, 17 m/s, and 22 m/s above the seismic basement with S-wave velocity of 32 m/s. The velocity structure is provided for 1 km mesh size. The velocities of the simplified crustal and upper mantle layers are based on the 1-D velocity structure model of Iwasaki et al. (1991). We use the technique of Graves (1996) to implement the anelastic attenuation in the finite difference method. We apply staggered-grid finite difference method (Graves, 1996), with fourthand second-order accuracy in space and time, respectively, for the 3-D simulation. We consider a horizontal free-surface and apply the zero-stress formulation of Graves (1996) for the free-surface. The absorbing boundary condition of Clayton and Engquist (1977) and non-reflecting boundary condition of Cerjan et al. (195) are applied as boundary conditions for the other sides. Fig. 6 Location of strong motion observation sites used in this study and example of vertical sections of the NIED velocity structure along the selected profiles. Strong motion sites from K-NET, WISE and JMA are denoted by different symbols; the cross symbols denote the microtremor array measurement sites. The white rectangle indicates the area, where we compare the NIED velocity structure model with revised velocity structure. In the cross sections, the velocities are 3.2, 2.2, 1.7, 1.1,.7, and. km/s from the bottom to top. It is not possible to show all the waveforms for the 1999 event in this abstract. The -component records have larger amplitude compared to the other component
Velocity (cm/s) Displacement (cm) Displacement (cm) records from this event due to the radiation pattern of S- wave. Therefore, In Fig. 7, we show an example of the observed and synthetic waveforms for only the component at the basin sites, HKD9, HKD95 and HKD96. In Fig. 7, two points are important to consider. The first is not reasonable NIED velocity model just beneath the HKD96 site due to the large difference between the observed and synthetic S-wave amplitudes. The second is the strength of excitation of the later phases. The later phases are more dominant in the synthetic waveforms at the HKD9 and HKD96 sites than the HKD95 site. 2 15 5 HKD9 HKD95 HKD96 Filter [.5-.5 Hz] Syn. 2 3 5 6 7 9 Fig. 7 An example of observed (black) and synthetic (grey) waveforms at basin sites from the 1999, Mw 6.1 event. Displacement records from two JMA stations (D7, Hiroo and D59, Obihiro) are available in the Tokachi basin for the 191 and 197 events. Due to the small space, we show the waveforms only for the 191 event. The stations, D7 and D59, are located on stiff and deep sedimentary soil sites, respectively. We show the comparison of the observed and synthetic waveforms at the two sites in Fig.. We can see that the excitation of later phases is very strong at the D59 site, compared with those at the stiff soil site, D7. The synthetic S waveforms have relatively good agreement with the observed ones at both sites. This indicates that the NIED velocity model just beneath the sites is reasonable. However, a strong later phase, about 3s after the S-wave arrival, on the -component of the synthetic waveform at the D59 site, cannot be seen on the observed record. We find from examination of a synthetic record section along the E-W profile including the D59 site that this later phase is surface wave induced at the western edge of the Tokachi basin. This indicates that the western basin edge structure is unreasonable. Tuning the deep velocity structure model We propose an improvement in tuning the deep velocity structure by forward modeling of observed long-period S-waveform. We have found that direct long-period S- wave at a basin site from 3-D simulation for a deep basin structure is essentially the same as that from 1-D simulation for a flat layer structure just beneath the site. Therefore, we can check the velocity structure just beneath the site by comparing the synthetic S-waveform from 1-D simulation with the observed one. In this comparison, it is important to select an appropriate band-width of the bandpass filter based on the S-wave amplification factors. This is a simple principle of the improvement in tuning the velocity structure just beneath the basin site by 1-D simulation. For details, see the paper by Dhakal et al. (29). (a) 6 2 191 D7 1.26 1..37.17.13.2 S Syn. 2 6 (b) 6 2 191 D59 2.2 3.3 1. 1.9.3 1.6 S BISW Syn. 2 6 Fig. Observed (black) and synthetic (red) waveforms at the D59 and D7 sites from the 191, Mw 6. event. S: S-wave; BISW: basin induced surface waves. Here we try to revise the NIED velocity structure in the western part of the Tokachi basin; the target area is shown in Figure 5 (the rectangular region). The strong motion observation sites are not enough for revision of the NIED velocity structure. Fortunately there are twelve microtremor array measurement sites in the target area as shown in Fig. 6. The phase velicities of Rayleigh waves at the microtremor measurement sites except for the TKCH6 site were estimated by MATSUSHIMA (199). We also conducted the array mesurements of microtremors and applied the spatial autocorrelation method (SPAC-method; OKADA, 23) to estimation of the phase velocities at the TKCH6 site. We use these phase velocity data to estimate the deep velocity structure. In this process, we tentatively assume that the sedimentary layer number and the S-wave velocity of each layer are the same as those of the NIED velocity structure model. The GA (genetic algorithm) inversion method (YAMANAKA and ISHIDA, 1996 ) is applied to estimation of the thickness of each layer. We also revise the deep velocity structure just beneath the HKD96 site using the tuning method mentioned above. We apply a linear interpolation between the three neighbouring 1-D velocity structures to interpolate the velocity structure in the western basin edge. The contour maps of the top of the seismic basement for the NIED and revised velocity structures are shown in Fig. 9a and 9b, respectively. Fig. 9c and 9d show the vertical cross sections of the NIED and revised velocity structures along a profile B-B (a part of the A-A profile in Fig. 6). We can see a change in the thickness and interface geometry of the sedimentary layers between the two models.
Velocity (cm/s) Velocity (cm/s) Displacement (cm) revised model is fairly good for the both 191 and 197 events at the D59 site. Fig. 9 Upper panel: depth of seismic basement with S- wave velocity of 3.2 km/s for the NIED (a) and revised (b) velocity models. The white lines (B-B ) denote profiles of the vertical sections shown in the middle and lower figures. Middle and lower: vertical cross section of the velocity structure along the profile B-B of the NIED (c) and revised (d) velocity structure models; the layering corresponds to S-wave velocity of, 7, 1, 17, 22, and 32 m/s, respectively from the top to bottom. 3-D Simulation using the revised velocity structure model 1 6 2-2 - -6 - - D59 NIED. Rev. 2.2 3.3 1.75 1. 1.9 2.21.3 1.6.26 2 6 Fig. 11 Comparison of the observed and synthetic waveforms at D59 using the NIED and revised velocity structure for the 191 event. Prediction of long-period ground motions for the 1993 Kushiro-oki earthquake The 1993 Kushiro-oki earthquake (Mw 7.6, Focal depth 3 km) is the largest intermediate depth intraslab earthquake that occurred nearby the Tokachi basin (Fig. ). As a preliminary prediction of long-period ground motions for this event, an image map of peak ground velocity (PGV) from the bandpass filtered (.5-.5 Hz) waveforms is shown in the Tokachi basin (Fig. 13). The source model of Morikawa and Sasatani (2) is used with some modifications for the 3-D simulations. Fig. shows the comparison between the observed and synthetic waveforms for the 1999 event at two sites. At HKD95, the amplitude of the basin-induced surface waves for the revised model show somewhat underestimate in comparison to those for the NIED model although the S-waveforms are essentially the same. On the other hand, at the HKD96 site, we obtain the good agreement between the observed and synthetic S-waveforms for the revised model. The basin-induced surface waves are considerably reduced in amplitudes for the revised model. HKD95 16.13 3.2 - - - -16 3.5.93.53.56.56.7.36 NIED Rev. 2 6 HKD96 16 - - - -16 1.59 2.7 1..3 1.7.3.16.3.21 NIED Rev. 2 6 Fig. Comparison of the observed and synthetic waveforms at HKD95 (left) and HKD96 (right) using the NIED and revised velocity structure. The large amplitude basin-induced surface waves on the synthetic waveforms (for the 191 event) for the NIED model are substantially diminished for the revised model (Fig. 11). Consequently, the agreement between the observed and synthetic waveforms for the Fig. Index map showing the epicenter of the 1993 Kushiro-oki earthquake and the strong-motion observation sites that recorded the event. The white filled star denotes the epicenter, triangles the strongmotion observation sites. Conclusion This study has been made with the aim of resolving current issues related with the prediction of long-period ground motions from intraslab earthquakes. Particularly, this study has focused on simulation of long-period ground motions in a deep sedimentary basin from intraslab earthquakes.
It is found that the predicted long-period response spectra from intraslab earthquakes are several times smaller than the observed ones at the central basin site. From the waveform comparisons, it is concluded that the NIED Tokachi basin velocity model is fairly good, but it requires some modification at the basin edges. A method for tuning the deep velocity structure by 1-D simulation of long-period S-wave has been proposed. Revision of the NIED model is carried out for the western basin edge structure. However this is incomplete. As a preliminary prediction of long-period ground motions (2 ~ 2 s), the synthetic PGV values exceed 2 cm/s in wide areas of the Tokachi basin for the 1993 event; the maximum value reaches near 75 cm/s. Fig. 13 Long-period ground motion hazard map showing PGVs in the Tokachi basin for the 1993 Kushiro-oki earthquake. The triangles denote the K- NET and other strong-motion observation sites for reference. The PGVs are derived at frequencies.5-.5 Hz. The vertical scale bar shows the PGVs in cm/s. The contours indicate the PGVs. Acknowledgements NIED, JMA, K-NET, and WISE are acknowledged for data. References AIJ (Architectural Institute of Japan) (1995). Report on the damage investigation of the 1993 off Kushiro earthquake and the south-west off Hokkaido earthquake, pp 97. Cerjan, C., Kosloff, D., Kosloff, R., and Reshef, M. (195), A nonreflecting boundary condition for discrete acoustic and elastic wave equations, Geophysics 5, 75-7. Clayton, R. W. and Engquist, B. (1977), Absorbing boundary conditions for acoustic and elastic wave equations, Bull. Seism. Soc. Am. 67, 1529-15. Dhakal, Y. P., N. Takai, T. Sasatani, 2. Empirical Analysis of Path Effects on Prediction Equations of Pseudo-Velocity Response Spectra in Northern Japan. Jour. Earthq. Engg. & Struc. Dyn., 39:3 61. Dhakal, Y., Sasatani, T., and Takai, N. (29), Tuning the deep velocity structure model by 1-D simulation of long-period S-waves, Proc. of the 9th SEGJ Intl. Symp. Imaging and Interpretation -, Sapporo, Japan, Paper ID 1 (CD-ROM). Frohlich, C. (26). Deep Earthquakes, Cambridge University Press, pp 5. Graves, R.W. (1996), Simulating seismic wave propagation in 3D elastic media using staggeredgrid finite differences, Bull. Seismol. Soc. Am. 6, 91-16. Iwasaki, T., Hirata, N., Kanazawa, T., Urabe, T., Motoya, Y., and H. Shimamura (1991), Earthquake distribution in the subduction zone off eastern Hokkaido, Japan, deduced from oceanbottom seismographic and land observations, Geophys. J. Int., 5, 693 711. Kanno T, Narita A, Morikawa N, Fujiwara H, Fukushima Y. (26), A new attenuation relation for strong ground motion in Japan based on recorded data, Bull. Seismol. Soc. Am., 96(3), 79-97. Koketsu, K., Miyake, H., Afnimar, and Tanaka, Y. (29), A proposal for a standard procedure of modeling 3-D velocity structures and its application to the Tokyo Metropolitan area, Japan, Tectonophysics, 72, 29-3. Matsushima, T., 199, Studies on determination of deep geological structures using long-period microtremors, Ph. D. thesis, Hokkaido University, 1-133 (in Japanese with English abstract). Morikawa, N. and Sasatani, T. (2), Source models of two large intraslab earthquakes from broadband strong ground motions, Bull. Seismol. Soc. Am., 9, 3-17. Okada, H. (23), The Microtremor Survey Method, Geophysical Monograph Series, Society of Exploration Geophysicists, 132. Yamanaka, H. and Ishida, H. (1996), Application of genetic algorithms to an inversion of surface-wave dispersion data, Bull. Seism. Soc. Am., 6, 36-. Zhao JX, Zhang J, Asano A, Ohno Y, Oouchi T, Takahashi T, Ogawa H, Irikura K, Thio HK, Somerville PG, Fukushima Y, Fukushima Y. (26), Attenuation relations of strong ground motion in Japan using site classification based on predominant period, Bull. Seismol. Soc. Am., 96(3), 9-913.