Different prior hypotheses analysed in seismic full waveform imaging problems

What is it about?

Spatially 3-dimensional seismic full waveform inversion (3D FWI) constructs subsurface seismic velocity structures, and is a highly computationally demanding problem. Understanding non-uniqueness in estimates of subsurface structure is important as it directly impacts risks connected to resource estimates and operational decisions. We estimate uncertainties using a new method that is highly efficient compared to previous methods, yet which still produces reasonable results. The computational cost is an order of magnitude greater than that of a single deterministic FWI (to be compared with many orders of magnitude more for previous comparable methods). Furthermore, we introduce a second new method that updates the solution by applying different classes of prior information, such as different spatial constraints or geological concepts, at almost zero marginal cost, and discriminates between them using the data. This opens the possibility that fully probabilistic 3D FWI, which might be the most computationally expensive problem in seismology, can be performed accurately and efficiently and can be used to test different geological hypotheses directly.

Featured Image

Photo by Dan Meyers on Unsplash

Photo by Dan Meyers on Unsplash

Why is it important?

We performed efficient Bayesian 3D FWI using variational inference, and demonstrated that prior information can be updated at low additional cost. A variety of prior hypotheses were therefore analysed, among which optimal ones were selected with almost zero additional computations.