Downscaling FWI images for reducing structural uncertainty

SUMMARY Despite ongoing improvements in the accuracy and efficiency of seismic interpretation, resolving small-scale geological structures remains challenging due to inherent limitations. Indeed, seismic imaging techniques are based on frequency band-limited data, and therefore they provide smooth representations of the real complex geology. This means that, within the resolution limit of seismic data, there is room for alternative valid interpretations at the small-scale. We suggest the use of a downscaling (or inverse homogenization) approach to make quantitative inferences and reduce uncertainty on small-scale geological structures. In particular, from a smooth velocity model obtained through the full waveform inversion (FWI) technique, the downscaling inversion aims to recover all the finer scale models compatible with it. In this context, the non-periodic homogenization serves as forward operator to construct smooth long-wavelength equivalent models. This approach can effectively be applied to specific areas of interest where seismic imaging is deemed reliable but the resolution is insufficient to support decision making. In this study we present an application in which downscaling is used to properly detect a fault and estimate small-scale P-wave velocity variations. In particular, the approach uses geostatistical simulation to generate velocity model realizations in the pre-faulting state and a kinematic modelling approach to generate fault displacement. The inverse problem is solved with a Bayesian stochastic framework in which fault-related and stratigraphic uncertainties are jointly incorporated into the inversion process. The methodology, validated on a synthetic example, is shown to be a reliable approach to reduce uncertainty on both fine-layering velocity and fault structure. Despite the additional challenges to be considered for real-case studies, we believe this target-oriented approach represents an efficient and cost-effective strategy to locally reduce FWI interpretation uncertainty.