On the evaluation of geophysical models