Efficient 3D Bayesian Full Waveform Inversion and Analysis of Prior Hypotheses

Abstract

Spatially 3-dimensional seismic full waveform inversion (3D FWI) is a highly nonlinear and computationally demanding inverse problem that constructs 3D subsurface seismic velocity structures using seismic waveform data. To characterise non-uniqueness in the solutions we demonstrate Bayesian 3D FWI using an efficient method called physically structured variational inference applied to 3D acoustic Bayesian FWI. The results provide reasonable posterior uncertainty estimates, at a computational cost that is only an order of magnitude greater than that of standard, deterministic FWI. Furthermore, we deploy variational prior replacement to calculate Bayesian solutions corresponding to different classes of prior information at low additional cost, and analyse those prior hypotheses by constructing Bayesian L-curves. This reveals the sensitivity of the inversion process to different prior assumptions. Thus we show that fully probabilistic 3D FWI can be performed at a cost that may be practical in small FWI problems, and can be used to test different prior hypotheses.