Convergence Analysis of MCMC Methods for Subsurface Flow Problems

Abstract

In subsurface characterization using a history matching algorithm subsurface properties are reconstructed with a set of limited data. Here we focus on the characterization of the permeability field in an aquifer using Markov Chain Monte Carlo (MCMC) algorithms, which are reliable procedures for such reconstruction. The MCMC method is serial in nature due to its Markovian property. Moreover, the calculation of the likelihood information in the MCMC is computationally expensive for subsurface flow problems. Running a long MCMC chain for a very long period makes the method less attractive for the characterization of subsurface. In contrast, several shorter MCMC chains can substantially reduce computation time and can make the framework more suitable to subsurface flows. However, the convergence of such MCMC chains should be carefully studied. In this paper, we consider multi-MCMC chains for a single–phase flow problem and analyze the chains aiming at a reliable characterization.