Time-lapse Full Waveform Inversion and Uncertainty Analysis with Different Survey Geometries

Abstract

We propose a framework for time-lapse full waveform inversion which inverts the baseline and time-lapse datasets jointly, with model regularization. Methods that invert datasets independently, and obtain time-lapse changes by model subtractions, suffer from artifacts caused by illumination differences and convergence issues. Inversion with differenced data (e.g. double-difference waveform inversion) requires highly repeatable acquisitions. In our framework, non-repeatability (as in survey geometry differences) is properly handled, without special data processing. We invert for the background model by exploiting information from both datasets, which mitigates problems from illumination differences. Time-lapse changes are differentiated from the background structures with the help of regularization. The regularization parameter, which serves as a map of the confidence in the time-lapse changes, is derived from the convergence curves of an alternating baseline/monitor inversion process. We test the framework with a synthetic example, where our method markedly outperforms the conventional time-lapse inversion.