Improving Image-domain Least-squares Reverse Time Migration Close to High Velocity Contrasts

Reverse time migration (RTM) and least-squares reverse time migration (LSRTM) are derived under a single scattering assumption, implying the use of smooth background migration velocity models. However, to obtain the best kinematic image, it is frequently the case that high-contrast velocity discontinuities are built into the migration velocity model. For example, salt bodies are typically interpreted at high resolution and included without smoothing to provide the best subsalt RTM image. Low-frequency artefacts arising from including such high-contrast discontinuities in the migration velocity model are typically handled with a Laplacian filtering workflow. Assuming that we want LSRTM to retain this accurate kinematic imaging whilst attempting to improve upon the resolution and illumination compensation of RTM, we must run LSRTM with the same migration velocity model and mitigate the effects induced by such discontinuities. Herein, we focus on an image-domain implementation of LSRTM using point-spread functions (PSFs). We propose two complementary approaches to reduce the imprint of the velocity discontinuities in the PSFs used for inversion, and the use of a mask (defining geological regions) to avoid interpolating PSFs across boundaries. We show that the combination of these approaches provides more reliable results close to high velocity contrasts.