We develop a localized low-rank approximation algorithm with (space) domain decomposition that significantly improves the scalability of P/S decomposition of anisotropic wavefields. The model and simulated wavefield are decomposed into rectangular blocks, which don’t have to be geologically constrained; low-rank approximations and P/S decomposition are performed separately and simultaneously in each block. An overlap-add method is used to reduce artifacts at block boundaries caused by Fourier transforms at wavefield truncations, and limited communication is required between blocks. Tests with synthetic data show good P/S decomposition results.
Presentation Date: Tuesday, September 26, 2017
Start Time: 1:50 PM
Location: 360D
Presentation Type: ORAL
