Adaptive variable-grid least-squares reverse-time migration

Abstract

Variable-grid methods have the potential to save computing costs and memory requirements in forward modeling and least-squares reverse-time migration (LSRTM). However, due to the inherent difficulty of automatic grid discretization, conventional variable-grid methods have not been widely used in industrial production. We propose a variable-grid LSRTM (VG-LSRTM) method based on an adaptive sampling strategy to improve computing efficiency and reduce memory requirements. Based on the mapping relation of two coordinate systems, we derive variable-grid acoustic wave equation and its corresponding Born forward modeling equation. On this basis, we develop a complete VG-LSRTM framework. Numerical experiments on a layered model validate the feasibility of the proposed VG-LSRTM algorithm. LSRTM tests on a modified Marmousi model demonstrate that our method can save computational costs and memory requirements with little accuracy loss.