Numerical simulations of seismic wave propagation in anisotropic and heterogeneous Earth models---the Japan subduction zone

Abstract

This thesis involved studies of the seismic wave propagation in fully anisotropic and heterogeneous Earth models, and the seismic velocity structures in the Japan subduction zone derived from the observations, using both forward modeling and tomographic methods. All the simulations of seismic wave propagation in this thesis have been carried out using both 2-D finite-difference method (FDM) and 3-D spectral-element method (SEM). We extend the SEM code to incorporate three-dimensionally, fully anisotropic earth models. For weakly anisotropic media, we benchmark the numerical simulations against asymptotic, ray theoretical predictions for both surface waves and body waves. The numerical simulations and asymptotic predictions are in good agreement for anisotropy at the 5% level. Weakly anisotropic body-wave propagation involves all 21 independent elastic parameters. The code is capable of simulating the shear-wave splitting in terms of waveforms. Strong waveform distortions are observed when the seismic waves propagate through a simple and weakly anisotropic Earth model. Our results also further prove the anisotropic effects on body-wave traveltime are completely directional dependent. From the high-density waveform data provided by Japanese data center (NIED), we observed a strong secondary SH arrival recorded in NE Japan. In order to explain this secondary arrival, a thin but strong low velocity layer (LVL) on top of the slab has to be introduced. Our 2-D model suggests the LVL extends down to a depth of 300 km with an S-wave velocity reduction of 14% if a thickness of 20 km is assumed. Further 3-D SEM simulations confirm that this model explains the strong secondary arrival. We interpret this deeply extended LVL as a zone composed of hydrated mafic and/or ultramafic rocks, and more likely ultramafic rocks (serpentinized peridotite) at depths greater than 150 km. The water released from the dehydration reactions in this hydrous zone could cause the abundant arc volcanism, the intermediate-depth intra-slab seismicity (70-30 km), and possible silent slip events. In order to obtain better 3-D seismic velocity models of the Japan subduction zone and the neighboring region beneath Eastern China, we use adjoint tomography to iteratively minimize the misfit between the synthetics and data from Hi-net, F-net, and Global Seismographic Network (GSN) stations. In this study, we are able to maximize the information obtained from three-component seismic records for tomographic inversion by using an automated windowing code. It selects not only the body-wave but also the surface-wave windows. With a dataset of 206 events in the Japan subduction zone, the frequency-dependent traveltime measurements are made in 44,709 windows for the period range of 24-120 s and 119,376 windows for the period range of 6-30 s. The combined adjoint sources are thus constructed based on these traveltime misfit measurements for all the receivers. Given the adjoint sources, we use the adjoint spectral-element method to calculate banana-doughnut kernels for the selected records. The weighted sums of the banana-doughnut kernels for all event-station pairs, with weights determined by the traveltime measurements, can be used to construct misfit kernels, which are the gradients required in a non-linear conjugate gradient algorithm to further refine the existing 3-D model.