Stepsize sharing in time-lapse full-waveform inversion
Xin Fu, Kristopher A. Innanen
Full waveform inversion (FWI) methods can produce high-resolution images of the physical properties of the subsurface. It has become a powerful tool for time-lapse or 4D seismic inversion, with applications in the monitoring of reservoir changes with injection and production, and potentially long term storage of carbon. Current time-lapse FWI strategies include the parallel strategy (PRS), the sequential strategy (SQS), the double-difference strategy (DDS), the common-model strategy (CMS), and the central-difference strategy (CDS). PRS time-lapse inversion is affected by convergence differences between the baseline and monitoring inversions, as well as non-repeatable noise and non-repeatable acquisition geometries between surveys. The other strategies above are largely efforts to fix these sensitivities of PRS. In this paper, we introduce and examine two strategies, which we refer to as stepsize-sharing PRS (SSPRS) and stepsize-sharing CMS (SSCMS). As the name suggests, they are characterized by a sharing of update stepsizes between baseline and monitoring stages of the time-lapse FWI. Synthetic data tests indicate that stepsize-sharing reduces artifacts caused by the PRS convergence variability. In particular, the stepsize-sharing common-model strategy (SSCMS) appears to be adept at reducing artifacts caused by all of convergence differences, non-repeated noise, non-repeatable source locations, and biased starting models. This breadth of robustness does not appear in any of the other approaches tested. Especially given that SSCMS through its sharing incurs half of the computational cost of CMS and CDS, we regard the workflow as being worth further study.