A Bayesian belief network (BBN) for combining evidence from multiple CO 2 leak detection technologies

A Bayesian belief network (BBN) methodology is developed for integrating CO 2 leak detection inferences from multiple monitoring technologies at a geologic sequestration site. The methodology is demonstrated using two monitoring methods: near‐surface soil CO 2 flux measurement and near‐surface perfluoromethylcyclohexane (PMCH) tracer monitoring, from the Zero Emission Research and Technology (ZERT) release test in 2007. Statistical models are fitted to natural background soil CO 2 flux and background PMCH tracer concentrations to determine critical levels for leak inference. Leakage‐induced increments of soil CO 2 flux and PMCH tracer concentrations are computed through TOUGH2 simulations for different leakage rates and subsurface permeabilities. The background characterizations and the simulation results are subsequently used to determine the conditional probabilities of leak detection in the BBN model. The BBN model is illustrated for use in evaluating the performance of alternative monitoring networks in a network design phase, and for combining inferences from multiple observations in the operational phase of a site. The detection capabilities of combined networks with different monitoring densities for soil CO 2 flux and PMCH tracer concentration are compared. Given the test condition, the greater sensitivity of the PMCH tracer allows it to detect smaller leaks, while detection by the soil CO 2 flux monitors implies that a larger leak is most likely present. © 2012 Society of Chemical Industry and John Wiley & Sons, Ltd