We present results on the geophysical inversion of geobodies in a Bayesian framework. Although the methodology applies to all near-homogeneous geobodies, we focus on salt diapirs as they are known to create trapping mechanisms for hydrocarbon reservoirs. Despite a great interest shown by the exploration geophysics community in recent years, the inversion of salt bodies remains a challenge because of severely limited penetration of seismic waves through them, which obscures the view underneath. Hence, with such scarce information in the seismic data, it becomes imperative to quantify uncertainty in inverted salt geometries. Inverse uncertainty quantification (a.k.a Bayesian inversion) has a multiyear history in geophysics research. Notwithstanding, the state of the art uncertainty quantification (UQ) methods are simply not practical yet, mainly because a) governing forward models are computationally taxing, and b) the inversion spaces are very high-dimensional owing to the cell-based representation of earth. The crux of this work relies on a lowdimensional representation of salt bodies, which in a Bayesian setting can be viewed as a geologic prior. Equipped with such a prior, we revisit some of the well-known Bayesian methods and reassess their feasibility from a practical standpoint. Specifically, we compare the computational footprint (in terms of number of forward and adjoint solves) of different methods, highlight promising ones, and comment on the future research directions towards practical UQ in geophysical inverse problems.

This content is only available via PDF.
You can access this article if you purchase or spend a download.