Elastic full waveform inversion using plane-wave

Abstract

Elastic full waveform inversion (FWI) is used to invert multicomponent data to obtain more accurate subsurface information; however, this method has a much higher computational cost than the acoustic method. In this study, the weighted pseudo-Hessian, which gives a weighting factor based on depth, was used as a preconditioner in the limited-memory Broyden-Fletcher-Goldfarb-Shanno algorithm for stable and fast convergence of elastic FWI. The multiscale approach was also applied to avoid the cycle-skipping problem. Despite fast convergence, wavefield modelling is cumbersome; thus, we performed plane-wave domain elastic FWI to reduce the amount of computation. The plane-wave domain approach requires a proper shot interval and an appropriate number of ray parameters to reduce aliasing and crosstalk noise. If these two conditions are not satisfied, artefacts may appear in the inversion results. The effects of these conditions on the plane-wave domain inversion results were investigated systematically with the Marmousi-2 model. Some artefacts appeared and the quality of inversion was slightly lower when using approximately a quarter of the number of ray parameters required for planewave encoding. However, the inversion result was comparable to that of the case with the satisfying condition; additionally, the inversion was more efficient. If aliasing exists in the observed data due to a large shot interval, the plane-wave domain approach can still be applied successfully, as the effect of aliasing is suppressed through the summation process of the effects by multi-ray parameters. (C) 2019 Elsevier B.V. All rights reserved.