Seismic waveform modeling and inversion in acoustic orthorhombic media

Abstract

Three-dimensional seismic waveform inversion (WI) for anisotropic media is highly challenging due to its computational cost, large number of model parameters, and parameter trade-offs. In this thesis, I explore 3D waveform inversion for orthorhombic media in the acoustic approximation. Two mixed-domain seismic wavefield simulators are implemented; one of them is based on low-rank decomposition and the other on the generalized pseudospectral method. Both methods can produce kinematically accurate pure-mode P-wavefields with an acceptable computational cost. The low-rank-decomposition-based method is used to simulate both state and adjoint wavefields due to its higher accuracy and stability. The wave equations from the pseudospectral method are employed to obtain the gradients of the WI objective functionals. To build the initial long-wavelength model for waveform inversion, I use an envelope-based misfit functional, which alleviates the reliance of WI on low-frequency data. The WI gradients are derived for both the conventional data-difference and the envelope-based objective functions. Numerical examples illustrate the performance of the developed wavefield-extrapolation and gradient-computation algorithms for orthorhombic media with realistic complexity. WI is conducted with the help of a limited-memory version of the quasi-Newton optimization algorithm. A test for a modified version of the SEG/EAGE overthrust model validates the proposed approach to waveform inversion in acoustic orthorhombic media.