mproved resolution in Bayesian lithology / fluid inversion from prestack eismic data and well observations : Part

The focus of our study is lithology/fluid inversion with spatial coupling from prestack seismic amplitude variation with offset AVO data and well observations. The inversion is defined in a Bayesian setting where the complete solution is the posterior model. The prior model for the lithology/fluid LF characteristics is defined as a profile Markov randomfield model with lateral continuity. Each vertical profile is further given as an inhomogeneous Markov-chain model upward through the reservoir. The likelihood model is defined by profile, and it relates the LF characteristics to the seismic data via a set of elastic material parameters and a convolution model. The likelihood model is approximated. The resulting approximate posterior model is explored using an efficient block Gibbs simulation algorithm. The inversion approach is evaluated on a synthetic realistic 2D reservoir. Seismic AVO data and well observations are integrated in a consistent manner to obtain predictions of the LF characteristics with associated uncertainty statements. The predictions appear very reliable despite the approximation of the posterior model, and errors in seismic data are the major contributions to the uncertainty. Resolution of the inversion is improved considerably by using a spatially coupled prior LF model, and LF units of 1–3 ms thick can be identified even with a seismic signal-tonoise ratio of two. The inversion results appear robust toward varying model parameter values in the prior model as a result of the discretization of LF characteristics and seismic data with good spatial coverage.