Semi?Supervised Surface Wave Tomography With Wasserstein Cycle?Consistent GAN: Method and Application to Southern California Plate Boundary Region

Machine learning algorithm has been applied to shear wave velocity (Vs) inversion in surface tomography, where a set of starting 1-D Vs profiles and their corresponding synthetic dispersion curves are used network training. Previous studies showed that the performance such trained is dependent on diversity training data set, which limits its application previously poorly understood regions. Here, we present an improved semi-supervised algorithm-based takes both model-generated observed process. The termed Wasserstein cycle-consistent generative adversarial networks (Wasserstein Cycle-GAN [Wcycle-GAN]). Different from conventional supervised approaches, GAN architecture enables inclusion unlabeled (the dispersion) process can complement set. cycle-consistency metric significantly improve stability proposed algorithm. We benchmark Wcycle-GAN method using 4,076 pairs fundamental mode Rayleigh phase group derived periods 3 16 s Southern California. final 3-D model given by best shows large-scale features consistent with geology. resulting reasonable misfits provides sharper images structures near faults top 15 km compared those machine methods.