Frequency Domain Least-Squares Reverse Time Migration Using Low-Rank Green’s Function for High Memory Efficiency

Abstract

Summary Least-squares reverse time migration (LSRTM) in the frequency domain can efficiently produce the high-resolution reflectivity model if the Green’s function could be stored in memory. This is because wavefield extrapolation could be done only once for entire iterations of LSRTM. However, saving the Green’s function requires the huge memory storage. To overcome this memory issue, we propose the efficient frequency domain LSRTM scheme with low-rank Green’s function. By exploiting the singular value decomposition (SVD), Green’s function is decomposed to one rectangular diagonal matrix and two unitary matrices. Therefore, Green’s function can be accurately reproduced with little memory. In the numerical example, it is demonstrated that our LSRTM scheme can generate the identical migration results of the frequency domain conventional LSRTM with reduced memory storage.