Highlight

Seismic full-waveform inversion~(FWI) is a nonlinear computational imaging technique that aims to obtain detailed estimates of subsurface geophysical properties. There are two categories of methods for solving FWI: conventional physics-driven approaches and recent data-driven approaches. These physics-driven methods primarily rely on heuristic optimization techniques that are computationally intensive and highly sensitive to data coverage. Once developed, however, they can be robust and relatively easy to generalize to new data sets. To alleviate the computational issues of physics-driven methods, data-driven inversion techniques have been recently proposed and developed. The data-driven techniques can be extremely efficient once fully trained, but the validity of the data-driven approaches is limited by the size and range of the training set, i.e., the predictive model might not generalize well. In this work, we developed a new hybrid computational approach to solve FWI that combines physics-driven models with data-driven methodologies. In particular, we develop a data augmentation strategy that can not only improve the representativity of training set, but also incorporate important governing physics into training and therefore improve the inversion accuracy.

The full paper can be found on arxiv:

Physics-informed Data-driven Waveform Inversion through Data Augmentation