Building Uncertainty and Sensitivity Analysis Into the TRIM Framework. D. H. Bennett and T. E. McKone, Lawrence Berkeley National Laboratory, One Cyclotron Road, 90-3058, Berkeley, CA 94720; and M. G. Dusetzina, U.S. EPA, OAQPS (MD-15), Research Triangle Park, NC 27711
An important aspect of the Total Risk Integrated Model (TRIM) is the integration of sensitivity and uncertainty analysis methods into the model framework. Many of the parameters used in modeling of natural systems are uncertain or variable. It is critical to confront sources and ranges of parameter variance for several reasons. Among them are (1) the need to determine the range of possible outcomes of the model and (2) the need to determine what parameters are the important contributors to the range of outcome values generated by the model. In a contaminant transport model, such as TRIM, uncertain parameters are typically associated with chemical properties and transformation rate factors; variability is more typically associated with climate data and human/animal behaviors and activities. The TRIM framework is designed provide for tiered uncertainty/sensitivity analyses in several ways. All inputs to TRIM are entered in parameter tables where value distributions are the default option and the labels "uncertain" or "variable" can be applied to make initial classifications. The capability to conduct a joint uncertainty and variability analysis will be provided. Some limited assessment of model uncertainty is provided through the option of selecting from alternate transport/transformation algorithms from an algorithm library. Many specific types of sensitivity/uncertainty analyses are possible. We will focus on the use of probabilistic analyses, which use correlation, rank correlation, regression or other means to examine how much of outcome variance is attributable to particular inputs or assumptions. We also give specific attention to regional sensitivity analysis (RSA) methods and the use of mulitvariate sensitivity analyses to assess the performance of the TRIM models. This process facilitates the comparison of model predictions to limited multimedia environmental data.