Assessing waveform predictions of recent three-dimensional velocity models of Tibet

Abstract:

High-resolution tomographic models are essential for understanding the physical and compositional properties in the lithosphere and obtaining accurate earthquake source locations and moment tensors. Yet, there are significant disagreements in recent three-dimensional velocity models of the crust and uppermost mantle in Tibet. Question also remains as to whether models constructed from one type of seismic waves (body or surface waves) can be used to predict travel times and waveforms of another. In this study, six global or regional models are selected for Tibet, most of which became publically available in the past five years. A three-dimensional finite-difference method in the spherical coordinates is applied to simulate full-wave propagation of regional P_n (with periods longer than 1 second) and Rayleigh waves (20-75 s period) for ground-truth events located at regional distances. The models are evaluated based on the phase delays and cross-correlation coefficients between synthetic and observed waveforms.

A model generated from full-wave ambient noise tomography by Shen and Zhang (2012) consistently produces the best predictions for Rayleigh waves throughout the dataset and the P_n waves for the paths from the Tarim Basin to central Tibet. LITHO1.0, inverted from surface wave dispersions, shows a relatively stable but intermediate performance in predicting P_n and Rayleigh waves. None of the models provide the best matches to both waves throughout the region. Furthermore, the models constructed from surface waves are not well suited to predict P_n, and vice versa. We attribute this mainly to lack of accurate constraints on radial anisotropy and V_p/V_s ratios in the upper mantle, and Moho topography. We conclude that simultaneous prediction for P, S, and surface waves requires an integrated velocity model constructed with multiple seismic waveforms and consideration of other important properties, such as anisotropy and attenuation.