Least-squares reverse-time migration guided full-waveform inversion

Full-waveform inversion (FWI) has become a powerful tool for high-resolution velocity building. However, FWI suffers from the local-minima problem, particularly when both the initial velocity model is inaccurate and low-frequency data are absent. To alleviate this problem and improve the convergence rate, we develop a new full-waveform inversion method using least-squares reverse-time migration (LSRTM) to guide interface updates and an efficient, implicit wavefield separation scheme to alternatively update the low-wavenumber and high-wavenumber components of velocity models. During each iteration step, our new method first employs a migration-like kernel to update high-wavenumber velocity perturbations using LSRTM, and then utilizes a tomography-like kernel to recover the low-wavenumber background velocity. To accurately compute these two types of kernels, we employ an efficient, implicit wavefield-separation scheme, rather than using the reflection waveform inversion under the Born-approximation. We validate our new FWI method using synthetic data for the Marmousi model. We demonstrate that that our new method is more robust than conventional FWI, particularly when initial velocity models are poor and low-frequency data are not available.