Elastic wave mode decomposition in anisotropic media with convolutional neural network

Abstract

Seismic anisotropy is widely distributed in the subsurface of Earth. In an anisotropic medium, P and S waves are intrinsically coupled due to the nature of elastic wave propagation. For the elastic reverse time migration (ERTM), it is essential to isolate the P and S wave modes before applying the imaging condition in order to avoid the image crosstalks. However, the polarization direction of P or S waves in anisotropic media, which is crucial for the mode separation, is no longer parallel or perpendicular to the propagation direction and will be spatially varying with the change of model parameters. Therefore, the wave mode separation or vector decomposition of elastic wavefields becomes infeasible because of the expensive computational cost to calculate the polarization.