Targeted nullspace shuttling in time-lapse FWI

Scott Keating, Kristopher A. Innanen

Time lapse inversion plays an important role in monitoring applications. Conventional approaches rely primarily on differencing strategies in either data or model space. The results of the model difference based approaches can be strongly influenced by the choice of starting model for baseline and monitor inversions. Different choices can result in different levels of mitigation of non-reproducible survey effects (for instance noise). We propose an approach that substitutes the importance of the initial model choice with explicit navigation of the inversion nullspace. In this strategy, targeted nullspace shuttling is used to find the baseline and monitor models that minimize the difference between models while preserving a desired level of data-fitting. In synthetic examples, this approach demonstrates a significant capacity to mitigate the effects or non-reproducible noise and changing acquisition, and to identify when time-lapse differences fall below the confidence threshold described by nullspace shuttling.