*Doctoral Dissertation 2011*

### Study on Seismic Wave Propagation Characteristics in a Sedimentary Basin and Waveform Inversion for Three-dimensional Basin Boundary Shape

Asako Iwaki

When seismic waves propagate into a sedimentary basin, ground motions at periods ~1 s and longer with large amplitude last for a long time in the basin because of the amplification due to low seismic wave velocity sediment and the generation of surface waves due to lateral heterogeneity of the basin velocity structure.Studying the ground motion characteristics and their relation to the three-dimensional (3D) velocity structure of a sedimentary basin is an interesting and important topic in seismology for understanding the propagation of seismic waves (~1 s) within heterogeneous media. In this thesis, we study the long-period (~3--20 s) ground motion characteristics in the Osaka sedimentary basin and the basin velocity structure.

We first evaluate the applicability of a current existent velocity structure model of the Osaka basin based on long-period ground motion simulations. Two types of long-period ground motion characteristics are examined at strong motion stations that are densely distributed within the basin: (1) peak ground velocity (PGV) and pseudo velocity response spectra (pSv) of strong ground motion from a M6.5 earthquake and (2) horizontal-to-vertical spectral ratio (HV) as site-specific event-independent ground motion characteristics. We compare the ground motion characteristics of type (1) from observed records and a finite-difference simulation of ground motion using objective indices. The simulation is able to reproduce PGV within a range of roughly 3/4 to 4/3 of the observed PGV. For type (2), we first demonstrate that HV from coda part of ground motion shows stable values near the peak periods regardless of the depth or the position of the earthquake. The observed HV was compared with the HV computed from ground motion simulations (3DHV). The peak periods of the observed HV ranges from $\sim$3 to 8 s in Osaka, and the differences between the peak period of the observed HV and 3DHV are less than $\pm$1 s at most stations, while they are more than +1.5 s at some stations in the north and northeast part of the basin. We also compare 3DHV with the theoretical ellipticity of the fundamental mode Rayleigh wave computed from 1D stratified velocity structure beneath each site (1DHV). There are some discrepancies in the peak periods of 3DHV and 1DHV at some areas, because 3DHV is affected by the lateral heterogeneity of the structure around the site, especially the complicated topography of the sediment/bedrock interface. It suggests the importance of introducing a method that can take into account 3D wave propagation within a basin in order to further improve the current velocity structure model.

In the latter part of the thesis, we propose a method to estimate the 3D basin boundary shape by waveform inversion in the Osaka sedimentary basin. We estimate the coefficients of B-spline function that describes the basin boundary shape by minimizing the L2-norm of the difference between the observed and synthetic waveforms. The observation equation is linearized and solved iteratively. Jacobian, which corresponds to the sensitivity of ground motion to the model parameters, is computed by finite-difference ground motion simulations for each model parameter. Examination of the spatial distribution of sensitivity using different time windows of seismic waveforms and events with different azimuths suggests that (1) later phases of ground motion are more sensitive to model perturbation than body-wave portion, (2) later phases are affected by a wider area than body-wave portion and (3) ground motion is more sensitive to the model parameters located in the direction of the earthquake source. Based on these results, we propose a two-step multi-event inversion method that involves two types of time windows and two earthquakes with different azimuths. The method is applied to real seismic data observed in the Osaka basin. After 8 iterations, the waveform fitting improves at most of the stations. The bedrock depth of the updated model is shallower than the initial model by up to $\sim$250 m in a region that extends along the bay and around the Uemachi elevation. The bedrock depth of the updated model is consistent with that obtained from deep borehole digging data.