Evaluating predictive models of species’ distributions: criteria for selecting optimal models