A comparison of uncertainty analysis methods using a groundwater flow model

This report evaluates three uncertainty analysis methods that are proposed for use in performances assessment activities within the OCRWM and Nuclear Regulatory Commission (NRC) communities. The three methods are Monte Carlo simulation with unconstrained sampling, Monte Carlo simulation with Latin Hypercube sampling, and first-order analysis. Monte Carlo simulation with unconstrained sampling is a generally accepted uncertainty analysis method, but it has the disadvantage of being costly and time consuming. Latin Hypercube sampling was proposed to make Monte Carlo simulation more efficient. However, although it was originally formulated for independent variables, which is a major drawback in performance assessment modeling, Latin Hypercube can be used to generate correlated samples. The first-order method is efficient to implement because it is based on the first-order Taylor series expansion; however, there is concern that it does not adequately describe the variability for complex models. These three uncertainty analysis methods were evaluated using a calibrated groundwater flow model of a unconfined aquifer in southern Arizona. The two simulation methods produced similar results, although the Latin Hypercube method tends to produce samples whose estimates of statistical parameters are closer to the desired parameters. The mean travel times for the first-order method does not agree with those of the simulations. In additions, the first-order method produces estimates of variance in travel times that are more variable than those produced by the simulation methods, resulting in nonconservative tolerance intervals. 13 refs., 33 figs