The Impact of Seismic Inversion Methods on Facies Prediction

Abstract

Summary We present a simultaneous Bayesian inversion framework for prediction and uncertainty quantification of lithology/fluid classes, petrophysical properties and elastic attributes from seismic amplitude-versus-offset observations. Prediction of these reservoir variables are of interest to generate initial models of the reservoir to model fluid flow. We consider a Gauss-linear likelihood. We compare the result based on two distinct prior models for the lithology/fluid classes. The first prior model is based on a first order neighborhood Markov random field in three dimensions, where both horizontal and vertical spatial coupling is assumed. The second prior model is based on a set of independent vertical first order Markov chains, where only vertical spatial coupling is assumed. We assess the posterior densities by a Markov chain Monte Carlo algorithm. The two models are demonstrated and compared on a gas discovery in the Norwegian Sea. A reduction in prediction error at a blind well location is obtained based on the Markov random field prior.