Factors Affecting Risk Predictions: Can Some Risk Metrics Be More Accurately Predicted Than Others?

Abstract

A numerical experiment was performed in order to examine the ability of multiple Monte Carlo realizations of a numerical model to reproduce the risk from a hypothetically known waste disposal situation. In the analysis, the risk was summarized by several risk metrics that could be chosen by a regulatory agency to set a risk standard. In the numerical experiment, the parameters in the numerical model are systematically varied to adjust bias (conservative or nonconservative) and to increase uncertainty relative to the hypothetically known future. The influence of parameter bias and uncertainty on the accuracy of each risk metric in predicting the nominal risk was evaluated and presented graphically. These analyses concluded that the peak-of-the-mean metric provides the least stable and least accurate risk predictions, whereas the cumulative release metric and mean of the peaks are more stable and accurate. The peak-of-the-mean and peak-of-the-median metrics exhibit risk dilution (i.e., a decrease in the predicted risk with increased uncertainty) and tend to underpredict risk. Additionally, these results illustrated how risk predictions that are made using what may be considered “conservative” assumptions can be moved in a direction that may or may not be expected or intended. Simulation relative to a hypothetical future (i.e., the nominal case) provides insight into the numerical behavior and potential accuracy of our risk assessment tools and potential issues with setting regulatory standards.