Targeted nullspace shuttling in time-lapse FWI

Abstract

Time lapse inversion plays an important role in monitoring applications. Uncertainties in seismic inversion mean that are many time-lapse changes in subsurface properties consistent with any given baseline and monitor survey, including changes that are implausible given our prior knowledge. Several time-lapse inversion methodologies have been formulated, with the goal of minimizing spurious differences while preserving real changes. We propose an approach that explicitly navigates the inversion nullspace to find the data-consistent time-lapse model best satisfying our prior knowledge. In synthetic examples, this approach demonstrates a significant capacity to mitigate the effects of non-reproducible noise and changing acquisition, and to identify when time-lapse differences fall below the confidence threshold described by nullspace shuttling.