Uncertainty and sensitivity analysis of a postexperiment simulation of nonexplosive melt-water interaction

Abstract

Results from applications of computer models are subject to uncertainty. This is basically due to model and parameter uncertainties. Model uncertainty arises in areas like phenomenology, model formulation, and numerical algorithms and is often partly due to intentional simplifications. Though uncertainty in model results can be substantial, few applications of computer models have been subjected to an uncertainty and sensitivity analysis. Reasons given are lack of funds, computer resources, and appropriate tools. An uncertainty analysis can provide a quantitative statement about the combined influence of potentially important uncertainties on the results. Often more important, it provides quantitative sensitivity statements that rank the uncertainties with respect to their contribution to model output uncertainty. This helps to guide research, model development, and measurement efforts to reduce model output uncertainty. An analysis approach is demonstrated that is based on well-established concepts and tools from probability calculus and statistics and is suited for uncertainty analysis of applications of computer models in general. It involves as little subjectivity as possible, once the potentially important uncertainties are identified and the current state of knowledge of each is quantitatively expressed by subjective probability distributions. This analysis input clearly does and always will contain subjective elements to a varying degree. This fact should be kept in mind when conclusions are drawn from the analytical results. The practical feasibility of the approach and the resulting uncertainty and sensitivity statements are illustrated by the analysis of a computer model application for a postexperiment simulation of the melt-water interaction processes.