Iterative modeling migration and inversion (IMMI): Combining full waveform inversion with standard inversion methodology

Abstract

Summary Full Waveform Inversion (FWI) employs full waveform information to estimate subsurface properties through an iterative process by minimizing the difference between the observed data and synthetic data. This inversion method has been widely studied in recent years but still cannot be practiced effectively in industry. The gradient in FWI is similar to an ungained Reverse Time Migration (RTM) image with the cross-correlation imaging condition. The estimation of the velocity update from the migrated section for FWI is analogous to the common process of impedance inversion. We show that the poorly scaled gradient can be improved by preconditioning with an approximate Hessian matrix (source illumination) forms the gradient or image based on deconvolution imaging condition which can be used to estimate the approximate reflectivity section directly. Thus, the deconvolution based gradient can be employed to estimate the impedance perturbation using the standard inversion methodology, denoted as SM. By examining the key concepts in FWI and SM, a hybrid inversion strategy is proposed by combining FWI with SM, namely, Iterative Modeling Migration and Inversion (IMMI). The IMMI method keeps the step of creating reflectivity image tying to wells control in SM and incorporates the concepts of imaging the data residuals and iteration from FWI. Furthermore, to reduce the computation cost for constructing the gradient and source illumination, the phase-encoding technique is introduced. In this paper, we practice the proposed strategies on a modified Marmousi model and the inversion results show that the IMMI method can reconstruct the velocity model efficiently and stably.