Surface wave tomography with transfer learning and Moho constraints: Method and application to China mainland

We develop a novel method based on deep learning (DL) to simultaneously obtain shear wave velocity (Vs) structure, sedimentary layer thickness and Moho depth from Rayleigh-wave phase and group velocities. Compared with previous studies, the proposed method constrains the inversion process through forward information to enhance the physical interpretability, and uses transfer learning to significantly improve the inversion results. Furthermore, in order to alleviate the multiple solution problem in surface wave inversion, we constrain it by introducing Moho information. We fully considered the influence of the parameter settings in the inversion process, and selected appropriate parameters during the inversion, such as sedimentary layer, attenuation, density, Vp/Vs ratio. The effectiveness and robustness of the proposed method are demonstrated through synthetic tests. Using recently measured surface wave dispersion data, we obtained the lithospheric Vs structure in East Asia based on the proposed approach. Results from application of our model to real data show consistency with previous research, fits to known geological structure, and results in smaller misfit to observed data. Our findings support our method as an impactful tool for Vs structure inversion.. In the mantle, the tearing of the Indian plate beneath the Tibetan Plateau (TP) can be seen at approximately 85°E and 90°E. The high-velocity block outlines the possible intrusion form of the Indian plate, extending as far as about 35° N position. The lithosphere in the western Sichuan Basin (SB) may have been modified by plateau material from the southeastern edge of the TP, showing low-velocity anomalies at a depth of about 100 km.