Updating uncertainty in site characterization and chemical transport using Bayesian Monte Carlo methods

Evaluating risk at a contaminated site requires sequential and iterative analysis of individual models that describe the chemical fate and transport process. However, uncertainties at each stage of the analysis can yield significant uncertainty in end predictions and may be too large to be adequately characterized using current uncertainty analysis methods. These uncertainties, which include characterization of the soil stratigraphy, groundwater flow conditions, locations of sources, and chemical dispersion, decay, and retardation, can result in significant uncertainties in the end estimates and ultimately affect the decisions about projected risks from current or remediated sites.; This work presents a Bayesian Monte Carlo (BMC) uncertainty analysis technique to characterize and reduce these uncertainties. The technique is able to simultaneously: (1) allow for the inclusion of expert judgment at the early stages of site assessment, (2) sequentially evaluate model uncertainty during the evaluation process, (3) update prior uncertainty estimates using field measurements, and (4) quantify conceptual model uncertainty.; The technique was applied to a localized and a regional chlorinated hydrocarbon plume at a Superfund site. Uncertainty in source characterization, in chemical transport parameters, and in hydrogeologic parameters describing groundwater flow was evaluated and updated. Head data updating was able to differentiate between reasonable and unreasonable hydraulic conductivity (HC) fields but could not differentiate between alternative conceptual models of the site geological structure. Chemical data updating reduced some parameter uncertainty and identified parameter correlations and combinations that yield similar chemical transport behavior. Alternative conceptual models describing source release locations and chemicals released were evaluated, with results that are consistent with field data. The chemical data updating was also able to differentiate between the alternative conceptual models of the site geological structure. Alternative chemical likelihood estimation methods for the chemical updating revealed the importance of appropriately accounting for the temporal and spatial chemical data correlation. With the updated uncertainty, remediation alternatives were considered for 30-year model projections.