Fig. 7: Iterative inversion for the differential adjoint tomography using low-frequency data.

A Top: topography along the LASSIE array with fault locations. Bottom: total sensitivity for all 36 virtual sources in the training set at the initial evaluation of the kernel. B Total sensitivity for all 36 virtual sources at the 13th evaluation of the kernel, which uses the optimal velocity model that minimize the validation set. C Training and validation misfit (loss) functions versus iteration. The optimal velocity parameter is reached at the 13th iteration, where the validation misfit is minimum.