Multi-parameter acoustic full-waveform inversion: a comparison of different parameterizations and optimization methods

Wenyong Pan, Kris Innanen, Yu Geng


Full-waveform inversion methods allow to provide high-resolution estimates of subsurface elastic properties, which are very important for reservoir characterization. However, multi-parameter FWI suffers from parameter crosstalk artifacts arising from the inherent ambiguities (or coupling effects) among different physical parameters, which significantly increase the non-linearity of the inverse problems. For multi-parameter acoustic FWI, density is difficult to be inverted, which maybe caused by the strong parameter trade-off from velocity. Different parameterizations have different resolving abilities. An appropriate parameterization in multi-parameter FWI can help avoid parameter crosstalk and reconstruct the model parameters efficiently. Scattering patterns due to different physical parameters have been employed to study the resolving abilities of different parameterizations. In this research, we will illustrate the scattering patterns for different parameterizations in acoustic media. It has been proved that the second-order derivative (namely Hessian operator) is capable to suppress the parameter crosstalk artifacts. The Hessian-free optimization methods are generally employed for avoiding construct the multi-parameter Hessian explicitly. In this research, we compare the inversion efficiencies of different parameterizations with different optimization methods for multi-parameter acoustic FWI.

View full article as PDF (2.17 Mb)