Variance Decomposition in the Presence of Epistemic and Aleatory Uncertainty

Abstract

AbstractVariance-based global sensitivity analysis is a powerful approach for understanding the importance of model input variables or groups of variables in driving model output variation. However, input variance is often attributable to both aleatory (irreducible) and epistemic (reducible) uncertainties. This paper presents an approach whereby variance decomposition is used in conjunction with probabilistic analysis. Epistemic uncertainty associated with a model’s probabilistic response is decomposed based on probability distribution uncertainty, deterministic model uncertainty, and other epistemic uncertainty sources. The proposed methodology allows for the identification of the epistemic uncertainty sources having the largest contributions to the uncertainty in the model’s response. As demonstrated in the numerical example, the proposed methodology may be used to support resource allocation decisions in modeling and simulation activities.KeywordsEpistemic UncertaintyVariance DecompositionEvidence TheoryAleatory UncertaintySouthwest Research InstituteThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.