Assessing Risks Through the Determination of Rare Event Probabilities

We consider the problem of evaluating the probability of occurrence of rare, but potentially catastrophic, events. The lack of historical data renders conventional statistical approaches inappropriate. The problem is compounded by complex multivariate dependencies that may exist across potential event sites. In order to evaluate the likelihood of one or more such catastrophic events occurring, we provide an information theoretic model for merging a decision maker's opinion with expert judgment. Also provided is a methodology for the reconciling of conflicting expert judgments. This merging approach is invariant to the decision maker's viewpoint in the limiting case of exceptionally rare events. These methods are applied to likelihood assessment of liquid natural gas tanker spills and seismic induced light water nuclear reactor meltdowns.