Comments of statistical issue in numerical modeling for underground nuclear test monitoring.

摘要

The Symposium concluded with prepared summaries by four experts in the involved disciplines. These experts made no mention of statistics and/or the statistical content of issues. The first author contributed an extemporaneous statement at the Symposium because there are important issues associated with conducting and evaluating numerical modeling that are familiar to statisticians and often treated successfully by them. This note expands upon these extemporaneous remarks. Statistical ideas may be helpful in resolving some numerical modeling issues. Specifically, we comment first on the role of statistical design/analysis in the quantification process to answer the question ''what do we know about the numerical modeling of underground nuclear tests.'' and second on the peculiar nature of uncertainty analysis for situations involving numerical modeling. The simulations described in the workshop, though associated with topic areas, were basically sets of examples. Each simulation was tuned towards agreeing with either empirical evidence or an expert's opinion of what empirical evidence would be. While the discussions were reasonable, whether the embellishments were correct or a forced fitting of reality is unclear and illustrates that ''simulation is easy.'' We also suggest that these examples of simulation are typical and the questions concerning the legitimacy and the role of knowing the reality are fair, in general, with respect to simulation. The answers will help us understand why ''prediction is difficult.''