Recommendations for SZ/TSPA Model Uncertainty Analysis Concerning the Yucca Mountain Project

Abstract

This report considers the problem of how best to evaluate the stability (i.e., sampling reliability) of Monte Carlo outputs obtained for two Yucca Mountain Project (YMP) modeling components, namely, outputs for the Saturated Zone (SZ) model, and for the Total System Performance Assessment (TSPA) model. One approach considered is the one that has been employed to date, namely, the application of Monte Carlo methods. Also considered in this context are potential improvements that might be obtained by the additional use of a Monte Carlo ''quitting rule'', such as that defined by Woo (1991), to select the number of Monte Carlo sample runs to perform. By the Monte Carlo approach, each output-value sample (realization) is calculated as a function of a sample-value vector of stochastic realizations, each of which in turn corresponds to a value of a corresponding distributed input variable. Abstractions from the SZ model and the Biosphere model are both used as input to the TSPA model. Sets of stochastic realizations required for SZ and TSPA abstractions ''expensive'' to generate, so the practical issue addressed by a ''quitting rule'' is how to determine what number of realizations is ''enough'' for the purpose of characterizing sampling error in the Montemore » Carlo estimate obtained for a specified model output of concern. In the TSPA context, the model output of concern is generally considered the time evolution of the arithmetic mean value, an estimator of the expected value, of TSPA-generated annual dose D(t) to the defined receptor within 10,000 y after waste-repository closure. Recommendations below specifically address: (1) whether or not a Monte Carlo approach (such as one employing the Woo quitting rule) is an appropriate basis for undertaking SZ- or TSPA-related uncertainty analysis, and (2) what other method might be more appropriate. While the following discussion and recommendations focus particularly on SZ applications, they also apply to other YMP model components including the TSPA. For this reason, some TSPA-specific issues are also addressed.« less