Validation of Exposure Models: The Role of Conceptual Verification, Sensitivity Analysis, and Alternative Hypotheses

Abstract

Available methods for evaluating the veracity and reliability of exposure models extend far beyond simple comparisons of observed and predicted values. To be valid, model tests must, first, be tailored to evaluate a model within its intended range of operation. Appropriate tests must then be developed to ensure that the model performs in accord with its designers'; intentions and that significant phenomena have not been neglected or oversimplified. Finally, comparisons between model predictions and realworld data sets can be used to compare the performance of competing models, but such studies usually cannot serve as general proofs of the reliability or validity of an individual model. These points can be elucidated by means of a discussion of some of the “validation” studies executed during the development of the exposure analysis modeling system (EXAMS). The EXAMS was designed to evaluate the probable long-term behavior of newly synthesized organic chemicals in aquatic systems. This design constraint strongly influenced the structure of the computer program and constrained the choice of appropriate tests of the program's capabilities. Conceptual verification, that is, comparison of program outputs with defined special cases, was used to evaluate and modify the program's description of chemical exchanges across the benthic boundary layer. Sensitivity analysis was used to assess the potential error induced by using reaction quantum yields as summary descriptors of the effects on photochemical reactivity of sorption to suspended sediments. “Validation,” in the sense of direct comparisons with observational data, was used to test the relative success of several versions of a two-resistance volatilization model. Complex exposure models can never be fully validated in the dictionary sense of establishing the conclusive truth. Instead, we must rely upon a history of successful applications to impart increasing confidence in the reliability and applicability of the tools of environmental chemistry and computer modeling to environmental problems.