Imaging with prestack reverse-time migration (RTM) is typically approached via a zero-lag crosscorrelation between source and receiver wavefields, which imposes unnecessarily stringent requirements for computational resources and disk storage. The imaging principle for reflectivity is analyzed and we demonstrate that a single maximal energy arrival event is often sufficient for migration imaging. Methods to alleviate the cost of crosscorrelation imaging are proposed and categorized into reconstructive and non-reconstructive schemes.;Source wavefield reconstruction treats the source extrapolation as a method of providing the auxiliary conditions for an initial-boundary value problem. A first-pass (forward-time) extrapolation for the source wavefield identifies the boundary and/or initial values necessary to uniquely reconstruct it using a second (reverse-time) backward propagation. Mixed value, or hybrid, reconstruction is proposed as the most accurate alternative to storing the source wavefield time history. Reconstructing the source wavefield reduces storage costs by up to two orders of magnitude without an appreciable loss of image quality. Boundary value and initial value reconstruction methods are extended from acoustic to elastic RTM.;Non-reconstructive approaches deviate from the conventional imaging paradigm, as only the most salient information required for imaging is kept. A maximal energy arrival event (termed the `excitation amplitude') imaging condition is explored as the direct analog for the theoretical reflection coefficient for acoustic isotropic media, and extended for elastic RTM. Sparse crosscorrelation is proposed as an equivalent method to standard crosscorrelation where the migrated image is now represented with a minimized data set. Time-binning is dynamic sorting algorithm with linear time complexity proposed for use with both excitation amplitude and sparse crosscorrelation approches to further expedite imaging. These parsimonious imaging methods reduce data storage by up to four orders of magnitude, which also effectively minimizes computational I/O bottlenecks.
展开▼