Evaluating modelling uncertainty for model selection

Modelling uncertainty is evaluated with respect to model complexity, sensitivity and error. The hypothesis that more complex models simulate reality better, but with more sensitivity and less error, is tested. An Index of Complexity is proposed. Improvements in fit of more complex models are weighed against the increase in model sensitivity. A simple index of utility is then proposed for model selection. The index of utility evaluates model sensitivity (response to changes in input) and model error (closeness of simulation to measurement). Model utility is evaluated for several models in two case studies, a simple system involving sorption of metals on sediments, and a more complex system, involving groundwater transport of a hydrophobic contaminant. Moderately complex models are found to be the more utile of those tested.