PNNL

Abstract

This report to the Nuclear Regulatory Commission (NRC) describes the development and application of a methodology to systematically and quantitatively assess predictive uncertainty in groundwater flow and transport modeling that considers the combined impact of hydrogeologic uncertainties associated with the conceptual-mathematical basis of a model, model parameters, and the scenario to which the model is applied. The methodology is based on a n extension of a Maximum Likelihood implementation of Bayesian Model Averaging. Model uncertainty is represented by postulating a discrete set of alternative conceptual models for a site with associated prior model probabilities that reflect a belief about the relative plausibility of each model based on its apparent consistency with available knowledge and data. Posterior model probabilities are computed and parameter uncertainty is estimated by calibrating each model to observed system behavior; prior parameter estimates are optionally included. Scenario uncertainty is represented as a discrete set of alternative future conditions affecting boundary conditions, source/sink terms, or other aspects of the models, with associated prior scenario probabilities. A joint assessment of uncertainty results from combining model predictions computed under each scenario using as weight the posterior model and prior scenario probabilities. The uncertainty methodology was applied to modeling of groundwater flow and uranium transport at the Hanford Site 300 Area. Eight alternative models representing uncertainty in the hydrogeologic and geochemical properties as well as the temporal variability were considered. Two scenarios represent alternative future behavior of the Columbia River adjacent to the site were considered. The scenario alternatives were implemented in the models through the boundary conditions. Results demonstrate the feasibility of applying a comprehensive uncertainty assessment to large-scale, detailed groundwater flow and transport modeling and illustrate the benefits of the methodology I providing better estimates of predictive uncertiay8, quantitative results for use in assessing risk, and an improved understanding of the system behavior and the limitations of the models.