Bayesian Seismic Lithology/Fluid Inversion Constrained by Rock Physics Depth Trends

Abstract

In this study we consider 2D seismic lithology/fluid inversion constrained by rock physics depth trends and a prior lithology/fluid Markov random field. A stochastic relation from porosity and lithology/fluid to seismic observations is established. The inversion is done in a Bayesian framework with an approximate posterior distribution. Block Gibbs samplers are used to estimate the approximate posterior distribution. Two different inversion algorithms are established, one with the support of well observations and one without. Both inversion algorithms are tested on a synthetic reservoir and the algorithm with well observations is also tested on a data set from the North Sea. The classification results with both algorithms are good. Without the support of well observations it is problematic to estimate the level of the porosity trends, however the classification results are approximately translation invariant with respect to porosity trends.