161 The Effect of Conservatism on Identifying Influential Parameters (PSAM-0381)

Sensitivity analysis is an important component of any probabilistic risk assessment that provides the foundation for a risk-informed, performance-based approach for protecting public health or making an engineering decision. Results from sensitivity analysis are typically used to derive the risk significance of various aspects of the system being modeled (i.e., parameters, conceptual models, and assumptions). An implicit assumption for conducting sensitivity analysis is that the model and the associated parameters representing the system is realistic (i.e., neither overly pessimistic nor optimistic). However, making conservative assumptions on the values of the parameters (model conservatism) is unavoidable when modeling large and complex systems, such as a high-level radioactive waste disposal system, when the systems have a significant level of uncertainty. This paper presents a systematic investigation of the effects of model using conservative values on the identification and ranking of influential parameters when using sensitivity analysis. The three simple, nonlinear stochastic example problems in this paper clearly illustrate how the ranking of influential parameters changes with the level of assumed conservatism. Such changes could lead to erroneous conclusions that other parameters in the system model are more influential than the ones that are assumed to be conservative.