Using MCMC Sampling to Calibrate a Computer Model of a Geothermal Field

We take a Bayesian approach to the calibration of an eight-parameter model of a geothermal field, analyzing measured well-test data. The posterior distribution over parameters for each of three scenarios, using different training data subsets, is explored using Markov chain Monte Carlo sampling. A novel parallel rejection algorithm is used to reduce 1 computation time. Caparison across scenarios indicates model error. Comparison of one scenario with a previous least-squares estimate for the same model and data set shows that sample-based statistics give a more robust estimate than gradient-based least-squares, in less compute time.