Probing the extended image volume for seismic velocity inversion

In seismic velocity inversion one aims to reconstruct a kinematically correct subsurface velocity model that can be used as input for further processing and inversion of the data. An important tool in velocity inversion is the prestack image volume. This image volume can be defined as a cross-correlation of the source and receivers wavefields for non-zero space and time lags. If the background velocity is kinematically acceptable, this image volume will have its main contributions at zero lag, even for complex models. Thus, it is an ideal tool for wave-equation migration velocity analysis in the presence of strong lateral heterogeneity. In particular, it allows us to pose migration velocity analysis as a PDE-constrained optimization problem, where the goal is to minimize the energy in the image volume at non-zero lag subject to fitting the data approximately. However, it is computationally infeasible to explicitly form the whole image volume. In this paper, we discuss several ways to reduce the computational costs involved in computing the image volume and evaluating the focusing criterion. We reduce the costs for calculating the data by randomized source synthesis. We also present an efficient way to subsample the image volume. Finally, we propose an alternative optimization criterion and suggest a multiscale inversion strategy for wave-equation MVA.