Validation and Equifinality

Abstract

In this chapter, the concept of equifinality of model representations is discussed, from a background of model applications in the environmental sciences. Equifinality in this context is used to indicate that there may be many different model structures, parameter sets and auxiliary conditions that might appear to give equivalent output predictions or acceptable fits to any observation data available for use in model calibration. This does not imply that the resulting ensemble of models will give similar predictions when used to predict the future under some changed conditions.As newinformation becomes available to allowmodel validation, this can be used to constrain the ensemble ofmodels within a Bayesian updating framework, although epistemic sources of uncertainty can make it difficult to define appropriate likelihood measures. It seems likely that the equifinality concept will persist into the future in the form of ensembles of (stochastic) model runs being used to estimate prediction uncertainties. However, more research is needed into the limitations of model structures, information content of data sets subject to epistemic uncertainties and means of evaluating and validating models in the inexact sciences.