Accurate source-wavefield reconstruction for large-scale 3D reverse-time migration

3D reverse-time migration using the cross-correlation imaging condition poses a challenge for the computer-memory requirement and computational complexity, because both the forwardpropagated source wavefield and the backward-propagated receiver wavefield should be accessible at the same time step. We develop a new wavefield reconstruction method to balance the computer-memory requirement and computational complexity of reverse-time migration. During the forward simulation of the source wavefield, we store the wavefield at only one or two layers of spatial grid points at/near the model boundaries, and reconstruct the back-propagated source wavefield in the imaging domain using a high-order wave-equation extrapolation scheme. One of conventional reverse-time migration methods uses boundary wavefields stored at multiple layers of spatial grid points. For a finite-difference scheme with the eighth or sixteenth order of accuracy in space, our new method uses only 37.5% of the computer memory required by this conventional method to store the boundary wavefields, and has the almost same accuracy as the latter. We validate our method using synthetic seismic reflection data. Our method produces 2D and 3D migration images of complex subsurface structures as accurate as those yielded using a reverse-time migration method that stores source wavefields at multiple boundary layers to maintain the spatial order of accuracy of the finitedifference scheme near the boundaries.