A Bayesian approach to integrate temporal data into probabilistic risk analysis of monitored NAPL remediation

Upon their release into the subsurface, non-aqueous phase liquids (NAPLs) dissolve slowly in groundwater and/or volatilize in the vadose zone threatening the environment and public health over extended periods of time. The failure of a treatment technology at any given site is often due to the unnoticed presence of dissolved NAPL trapped in low permeability areas and/or the remaining presence of substantial amounts of pure phase NAPL after remediation efforts. The design of remediation strategies and the determination of remediation endpoints are traditionally carried out without quantifying risks associated with the failure of such efforts. We conduct a probabilistic risk analysis (PRA) to estimate the likelihood of failure of an on-site NAPL treatment technology. The PRA integrates all aspects of the problem (causes, pathways, and receptors) without resorting to extensive modeling. It accounts for a combination of multiple mechanisms of failure of a monitoring system, such as bypassing, insufficient sampling frequency and malfunctioning of the observation wells. We use a Bayesian framework to update the likelihood of failure of the treatment technology with observed measurements of concentrations at nearby monitoring wells.

► We examine the risk that contaminants reach receptors without being detected by observations.
► The failure of a well comes from bypassing, sampling frequency and malfunctioning.
► Risk must account for the different sequences of multiple failures associated with the monitoring system.
► An integrated Bayesian PRA approach can update the probability of system failure with new observations.
► Risk can increase with the addition of new information.