An MCMC algorithm for sampling from the posterior conditioned to production history

An MCMC algorithm for sampling from the posterior conditioned to production history


by Håkon Tjelmeland (Norwegian University of Science and Technology, Norway), Statistics No 2/2000, Department of Mathematical Sciences, Norwegian University of Science and Technology, to be presented at Geostats 2000 - South Africa.
In evaluation of petroleum reservoirs, observed production history contains valuable information. A natural approach to integrate this type of data with other sources of observations is to adopt a Bayesian model. To sample from the resulting posterior distribution, MCMC algorithms have to be applied. Even if MCMC provides a very flexible framework in which to construct simulation algorithms, so far only standard MCMC algorithms have been employed for sampling from this type of posterior. As a result, a large number of iterations has been necessary to obtain convergence. In this paper, we first discuss the characteristics of the type of distribution in question and thereafter define an MCMC algorithm specifically designed to cope with these features.