An optimal transport distance for full-waveform inversion: Application to the 2014 Chevron benchmark data set

International audienceMitigating cycle skipping in full waveform inversion is a long term issue. In this study, a modification of the misfit function based on an optimal transport distance is proposed. A specific numerical strategy based on the Kantorovich-Rubinstein norm and a proximal splitting algorithm is designed to compute this misfit function and its gradient efficiently. An example of application on the Marmousi model shows that the optimal transport misfit function is more robust to cycle skipping than the conventional L2 misfit function. When applied to the Chevron 2014 benchmark data-set, a reliable velocity estimation is computed, using only a multi-scale strategy in frequency, while the L2 misfit function is trapped into a local minimum following the same workflow