Effective boundary RNN-based 3D acoustic FWI

Tianze Zhang, Daniel O. Trad, Kristopher A. Innanen

The development of 3D Full Waveform Inversion (FWI) enable geophysicists to obtain high-resolution subsurface models by fully leveraging seismic wavefields. In this study, we aim to develop a 3D acoustic FWI tool that utilizes the automatic differentiation method and the effective boundary method to compute gradients for the P-wave velocity model. To accelerate computational speed, we incorporate the Message Passing Interface (MPI) and GPU processing. We validate our inversion tool using the SEG 3D Overthrust model, and a time-lapes VSP test, demonstrating its accuracy. The inversion results align well with expectations, and the computational time required is affordable, confirming that the algorithm is suitable for field data inversion.