Large simulation models: calibration, uniqueness and goodness of fit

Large simulation models of environmental systems which are based on biological and physical mechanisms are useful because of their ability to integrate diverse types of information relevant to the problem under analysis. Inherent in such models is a high degree of both structural and parametric complexity. A number of modeling and simulation studies have demonstrated that there are many parameter sets which produce good fits to calibration data. This lack of uniqueness requires a different perspective on parameter estimation which can be usefully addressed employing computer-intensive methods of multivariate statistical analysis. Regardless of how parameters are estimated, there appears to be a strong case to be made that the ‘best’ parameter estimate is an extended and complex region in a high dimensional space.