Relaxing the Initial Model Constraint for Crustal-Scale Full-Waveform Inversion with Graph Space Optimal Transport Misfit Function

Abstract

Summary We investigate the potential of the recently proposed Graph-Space Optimal Transport (GSOT) misfit function, to mitigate the cycle-skipping problem at the initial stage of the crustal-scale Full-Waveform Inversion (FWI) from the wide-angle stationary-receiver seismic data. Ultra-long offset seismic data contain very rich information about the deep crust, recorded along the various diving, refracted and reflected wavepaths. This diversity of the arrivals representing different propagation regimes - combined with relatively sparse receiver spacing - increases the non-linearity of the inverse problem and makes it difficult to find the initial FWI model which satisfies the cycle-skipping criterion. Here we search for the panacea to this problem through the application of more convex misfit function - namely GSOT. We apply the acoustic time-domain FWI to the synthetic marine seismic dataset generated in the realistic model of the complex subduction zone. We show that FWI with GSOT misfit function, combined with proper data selection, allows for reconstruction of the true velocity model starting from the simple 1D initial model. Despite the significant cycle-skipping problem at the initial stage of inversion, the GSOT misfit function is still able to overcome local minima and guide the inversion toward the correct solution.