An efficient and accurate finite-difference operator using adaptively discretized grids and its application for 3D least-squares reverse-time migration

Abstract

Least-square reverse-time migration (LSRTM) is a powerful tool to image subsurface reflectivity with high resolution. It has the ability to reduce migration artifacts, balance amplitudes, and improve imaging resolution. However, the large amount of computation cost is one of its challenging problems, especially for 3D problems. We propose an efficient and accurate finite-difference modeling operator using an adaptive variable grid strategy. The resampled model’s grid intervals adapt to local velocity and wave frequency, ensuring that dispersion is mitigated to some extent. Furthermore, we apply the modeling operator to 3D LSRTM with graphics processing unit (GPU) implementation in order to mitigate the large calculation costs. The 3D modeling is applied to two synthetic examples to validate its feasibility, accuracy, and efficiency. The imaging results of the 3D SEG/EAGE overthrust model demonstrated that adaptive grid LSRTM (AGLSRTM) is capable of reducing computing time and memory requirement while producing the same imaging accuracy as traditional LSRTM.