Bayesian Networks for Decisions under Uncertainty in Basin Modeling

In this work, we propose a structured workflow for calibrating model results to observed data, ranking prospects, determining the value or utility of obtaining new information, and updating models with its retrieval. Bayesian Networks (BN) help understand uncertainty as well as complexity by incorporating probability theory and graph theory into Basin and Petroleum Systems Modeling. The graphical formulation allows for modularity, and risk components like source, reservoir, and trap can be evaluated separately as well as recombined for accumulation level uncertainty analysis. Conditioning over a subset of variables allows Bayesian Networks to work well even when the parameter space is sparsely sampled. This is an advantage over typical response surface models because coupled simulations are computationally expensive, and number of runs needed rises exponentially with number of uncertain parameters.