M B Beck* and J Chen**
*Warnell School of Forest Resources
University of Georgia
Athens, Georgia 30602-2152, USA**Department of Environmental Science and Engineering
TsingHua University
Beijing, PR China
Abstract
Validation (or invalidation) of a model has never not been a problem. It seems to provoke as much interest today- which is considerable - as it did a quarter of a century or more ago. In general, what constitutes validation has conventionally been composed of a set of procedures of evaluation, which can be separated broadly into quantitative methods of matching history, including the analysis and assessment of uncertainty, and the more qualitative techniques of peer review. Matching history, while extremely important, has its limitations, however. There are circumstances - and perhaps they are precisely the circumstances under which the construction of a model is most necessary - for which there is no history to be matched. For example, in seeking to predict the fate and effect of a newly synthesized chemical, there are no empirical data available on the past history of that substance as it moves around the environment. The paper explores what, other than peer review, might be done to assure the quality of a model to be used to fulfil such a task. As a motivating case study one of the EPA’s Multi-Media models is assessed with respect to its capacity to screen the magnitude of potential contamination of groundwater by leachates from facilities for storing hazardous materials. The approach to assuring the quality of this model has been motivated by a particular methodological framework, usually referred to as a Regionalized Sensitivity Analysis. Through the application of this method to the Multi-Media model it has been possible to construct a measure, a distribution of values for a Kolmogorov-Smirnov statistic, having the potential for discriminating between a model that is well suited to its task and one that is not. These are preliminary, prototypical results, however, and are presented as such.
In Mathematical and Statistical Methods for Sensitivity Analysis (A Saltelli, K Chan,