Five ways to avoid storing source wavefield snapshots in 2D elastic prestack reverse time migration

Abstract

Five alternative algorithms were evaluated to circumvent the excessive storage requirement imposed by saving source wavefield snapshots used for the crosscorrelation image condition in 2D prestack elastic reverse time migration. We compared the algorithms on the basis of their ability, either to accurately reconstruct (not save) the source wavefield or to use an alternate image condition so that neither saving nor reconstruction of full wavefields was involved. The comparisons were facilitated by using the same (velocity-stress) extrapolator in all the algorithms, and running them all on the same hardware. We assumed that there was enough memory in a node to do an extrapolation, and that all input data were stored on disk rather than residing in randomaccess memory. This should provide a fair and balanced comparison. Reconstruction of the source wavefield from boundary and/or initial values reduced the required storage to a very small fraction of that needed to store source wavefield snapshots for conventional crosscorrelation, at the cost of adding an additional source extrapolation. Reverse time checkpointing avoided recursive forward recomputation. Two nonreconstructive imaging conditions do not require full snapshot storage or an additional extrapolation. Timebinning the imaging criteria removed the need for image time searching or sorting. Numerical examples using elastic data from the Marmousi2 model showed that the quality of the elastic prestack PP and PS images produced by the cost-optimized alternative algorithms were (virtually) identical to the higher cost images produced by traditional crosscorrelation.

Full–waveform inversion using the excitation representation of the source wavefield

“…It is intrinsically handled in conventional FWI by utilizing the full source wavefield with all its complexity. However, an intermediate solution of storing more than one maximum of the source wavefield also can ameliorate the excitation method in the presence of multipath (Jin et al, 2015;Nguyen and McMechan, 2015).…”

3D elastic full-waveform inversion for OBC data using the P-wave excitation amplitude

“…Therefore, it is hard to segregate the excitation time of P-wave in elastic media. As suggested by Nguyen and McMechan (2015) in the context of reverse time migration, we compute excitation time of the divergence of wavefields ( S ∇ ), which includes only Pwave motions. Next, we store the original displacement wavefield at this excitation time.…”

Full waveform inversion using envelope-based global correlation norm

“…To perform the inversion on such a large 3D area, we solve the first-order acoustic wave equations using the staggered grid finite-difference method with domain decomposition (Bohlen, 2002). Among the many ways to avoid storing the huge forward (source) wavefields (Nguyen and McMechan, 2015) in 3D FWI, we choose the time- frequency hybrid domain approach (Sirgue et al, 2008), in which the forward and backward wavefields are transformed to the frequency domain on the fly so that we can calculate the gradient direction in the frequency domain within the desired frequency band. To scale the gradient direction, the diagonal of pseudo-Hessian matrix is used (Shin et al, 2001).…”