On the evaluation of geophysical models

With powerful computing facilities now generally available, it is common to attempt to assess the quality of a geophysical model, and not merely to fit it. If techniques used to make the assessment are themselves flawed, perhaps by oversimplification or unrecognized correlation, then further confusion may arise. Here we give a straightforward framework for thinking about reasonably simple modelling, with examples from the transportation of atmospheric pollutants, and exploration seismology. Further, we comment on two pitfalls for the unwary with regard to implementing ideas discussed in the literature for the assessment of models, namely (i) the error that can result from inadequate specification of the model when using the bootstrap, and (ii) the shortcoming of using high correlation between measurements and predictions as a guide to model adequacy.