Abstract

This chapter discusses the issues associated with the invalidation of computer simulation models, taking environmental scienceEnvironmental science as an example. We argue that invalidationInvalidation is concerned with labelling a model as not fit-for-purposeFit-for-purpose for a particular application, drawing an analogy with the Popperian idea of falsificationFalsification of hypotheses and theories. Model invalidation is a good thing in that it implies that some improvements are required, either to the data, to the auxiliary relations or to the model structures being used. It is argued that as soon as epistemic uncertainties in observational dataObservational data and boundary conditionsBoundary condition are acknowledged, invalidationInvalidation loses some objectivity. Some principles for model evaluation are suggested, and a number of potential techniques for model comparison and rejection are considered, including Bayesian likelihoodsLikelihood, implausibility and the GLUE limits of acceptability approaches. Some problems remain in applying these techniques, particularly in assessing the role of input uncertainties on fitness-for-purpose, but the approach allows for a more thoughtful and reflective consideration of model invalidation as a positive way of making progress in science.