Efficient reflection waveform inversion using a locally normalized zero-lag correlative objective function

摘要

Full-waveform inversion is characterized by cycle-skipping when the starting background model differs significantly from the true model and low-frequency data are unavailable. To mitigate this problem, reflection waveform inversion is applied to provide a background velocity model for full-waveform inversion. This technique attempts to extract background velocity updates along the reflection wavepath by matching the reflection waveforms. However, two issues arise during the implementation of reflection waveform inversion: amplitude and efficiency. The amplitude is always underestimated due to the complex subsurface parameter (i.e. the source signature, density, attenuation etc.). This makes it unreasonable to match the reflection amplitude involved in waveforms, especially in the filed data cases. In addition, generating the background velocity gradient requires the simulation of the reflection wavefield. However, simulating the reflection wavefield is time-consuming. To address the former, we introduced a locally normalized objective function, while for the latter, we used an efficient strategy by avoiding the explicit generation of the reflection wavefield. Results show that applying the proposed method to both synthetic and field data can provide a good background velocity model for full-waveform inversion with high efficiency.