Testing Alternative Models Using Cross Validation and Computationally Frugal Methods

Many methods can be used to test alternative models. Tests considered here are for model discrimination (identifying models most likely to produce accurate predictions) and sensitivity analysis (identifying important observations and parameters). Computationally efficient model discrimination criteria and local sensitivity analysis and demanding cross-validation methods are compared using groundwater models of the Maggia Valley, Italy. Results show opportunities and difficulties of the computationally frugal methods considered.