A GPU implementation of 2nd acoustic least-squares reverse time migration

Abstract

Least-squares reverse time migration (LSRTM) is a seismic imaging method that can provide higher-resolution images of the subsurface structures compared to other methods. However, LSRTM is computationally expensive. To reduce the computational time of LSRTM, GPU can be utilized. This leads to the objective of this work which is to develop a GPU implementation of LSRTM. In this work, the two-dimensional first-order acoustic wave equations were solved using the second-order finite difference method on a staggered grid, and a perfectly matched layer was used as an absorbing boundary condition. The adjoint-state method was used to compute the gradient of the objective function concerning model parameters. A linear conjugate gradient method was used to minimize the objective function. Both forward- and backward-propagation of wavefields using the finite-difference method were performed on a single GPU using the NVIDIA CUDA library. For verification purposes, the GPU program of LSRTM was applied to a synthetic data set generated from the Marmousi model. Numerical results showed that LSRTM could provide an image with higher quality compared to a conventional RTM image. The GPU-version of LSRTM has acquired a speedup factor of 12-to-13 times compared to the serial CPU-version of LSRTM.