Imaging the subsurface of Taiwan using ambient noise tomography and full waveform inversion

Abstract

The results of this study show strong evidence for continental subduction in the southern part of the island. Throughout the entire island, lower shear wave speeds indicate that the crust thickens below the Coastal Mountain Range, forming a root up to ~50 km depth and extending to 90 km depth in the southern part of the island. The west half of the island is generally characterized by a thinner crust and the slowest shear wave speeds in the model. Continental subduction is not inferred from the resulting model of the northern part of the island, where the crustal root is ~60km deep; however, some low-velocity mantle structures are present.

Arc-continent collision is a process that over millions of years built most of the existing continents. Continental crust is thought to remain on the surface during these plate interactions, and its mass loss is accounted for by shortening. Remnants and clues about the mechanics of this process are available on the surface of Earth, but to understand the mechanics one must probe the subsurface of modern day arc-continent collisions. Taiwan is such an example: it is the result of a complex, actively deforming tectonic boundary between the Eurasian and Philippine Sea plates. This seismically active location provides an excellent venue for seismically imaging processes related to arc-continent collision, which is what the TAIGER (TAiwan Integrated GEodynamics Research) project was created to accomplish. In this thesis, data collected by the TAIGER deployment, supplemented by observations from the permanent BATS (Broadband Array in Taiwan for Seismology) network, is used to create a 3D elastic wave velocity model of the crust and upper mantle beneath Taiwan. This model addresses an outstanding question about arc-continent interactions using the tectonic structure of Taiwan: do arc-continent collisions involve the consumption of continental crust?

Ambient noise tomography techniques create a 3D Vs model by using continuous ambient noise, which is whitened and cross-correlated between stations to construct empirical Green's functions of Rayleigh waves. The correlations are graded by the signal to noise ratio prior to recovering group and phase velocities of the fundamental mode for periods between 6 and 30 seconds. The results are combined to generate a 3D Vs model from which a Vp model is calculated using a constant Vp/Vs ratio of 1.7. This model, combined with the arrival time model of Kuo-Chen et al. (2012), is used as a starting model for full waveform inversion of teleseismic body and surface waves using the 2.5D technique of Roecker et al. (2010).