A modeling study of preprocessing considerations for reverse-time migration

Abstract

Much of the thinking behind conventional geophysical processing assumes that we wanted to image energy that propagates down from the surface of the earth, scatters from a reflector or diffractor, and then propagates back up to the recording surface without being reflected by any other feature. Such travel paths conform to the assumptions of one-way wave propagation, and most contemporary migration schemes are designed to image such data. In addition, the moveout behavior of these primary reflection events in the various prestack domains is well understood, and many of our standard data-preprocessing techniques relied on the assumption that this behavior adequately describes the events we wanted to preserve for imaging. As a corollary, events that do not conform to this prescribed behavior are classified as noise, and many of our standard preprocessing techniqueswere designed to remove them. We assessed the kinematics of moveout behavior of events that arise from two-way wave propagation and the effect of certain preprocessing techniques on those events. This was of interest to us because the recent rapid increase in available cost-effective computing power has enabled industrial implementation of migration algorithms—particularly reverse-time migration—that in principle can image events that reflect more than once on their way from source to receiver. We used 2D synthetic data to show that some conventional data-processing steps—particularly those used in suppression of complex reverberations (“multiples”)—remove nonreverberatory primary events from seismic reflection data. Specifically, they remove events that have repeated or turning reflections in the subsurface (such as double-bounce arrivals) but that otherwise are imageable using reverse-time migration.