Comparison of Sampling Methods with Local Conditioning to Seismic Data

We focus on Bayesian inversion, exemplified by that of predicting reservoir variables conditional on seismic amplitude data. Our interest is to study local approaches for obtaining approximate samples from the posterior distribution. The posterior is defined via prior samples and a likelihood model for the geophysical observations. This likelihood model involves a non-linear expectation term, commonly termed the forward model. In this paper we conduct a comparison of a local rejection sampler and a local ensemble transform Kalman filter for this challenge. Seismic AVO data from the top-reservoir formation of the Alvheim field in the North Sea are used in the study