Multioffset Inversion by Differential Semblance Optimization

A still-unresolved problem in reflection seismology is the estimation of wave velocities directly from waveform data, bypassing expensive manual steps such as event-picking in before-stack data. This paper suggests a waveform inversion algorithm, differential semblance optimization for shot-gather data, designed to yield accurate velocity estimates even in the absence of accurate a priori information. The algorithm described here is a variant of least-squares inversion, which has proven incapable of estimating large-scale velocity features. We explain the obstacles to success of least-squares velocity inversion, and offer both theoretical and numerical evidence that our modification should overcome these obstacles. An appendix describes our approach to accurate calculation of the gradient of the modified cost functional