Accelerating Multi-Dimensional Optimal Transport Full Waveform Inversion with Mini-Batch Optimization and Anisotropic Diffusion Filter

Full waveform inversion (FWI) has become an industry standard tool for building high-resolution subsurface models. Conventional FWI using least-squares misfit function is known to suffer from cycle skipping and can be sensitive to noise. Recently, misfit functions based on the optimal transport (OT) have been proposed to replace the least-squares norm and show promising results in addressing the cycle skipping issue. Specifically, the multi-dimensional optimal transport FWI based on the Kantorovich-Rubinstein (KR) norm has demonstrated improved velocity update compared to least-squares FWI in various field data applications. However, solving for the KR norm in the multi-dimensional optimal transport is computationally quite expensive. In this paper, we present a workflow to accelerate multi-dimensional optimal transport FWI with mini-batch optimization and anisotropic diffusion filter.