Robust ADMM-based wavefield reconstruction inversion with phase retrieval

Description

Wavefield Reconstruction Inversion (WRI) extends the linear regime of Full Waveform Inversion (FWI) by relaxing the wave equation with a feedback term to the data during wavefield simulation before updating the subsurface model by minimizing the source residuals this relaxation generates. Wavefield reconstruction and parameter estimation are efficiently nested with the alternating-direction method of multipliers (ADMM), leading to the so-called iteratively-refined WRI (IR-WRI). Capitalizing on the bilinearity of the wave equation, the alternating-direction strategy recasts the parameter estimation as a linear subproblem, which can be easily regularized with bound constraints and nonsmooth regularizations in the ADMM framework. Although the promise of regularized IR-WRI to image large-contrast media has been shown, the robustness of the parameter estimation subproblem can be further improved by phase retrieval during the early stages of IR-WRI, that is when the phase of the reconstructed wavefields is inaccurate in the deep part of the velocity model. Then, the velocity model inferred from phase retrieval at low frequencies is used as initial model for the subsequent steps of classical IR-WRI. This new workflow successfully reconstructs the BP salt model from a sparse fixed-spread acquisition using a 3 Hz starting frequency and a homogeneous initial velocity model.

Abstract

International audience

Additional details