We discuss the ways in which atmospheric numerical models can be shown to have both predictive and explanatory
value. This argument rests largely on the established predictive value of numerical weather forecasts, and can be
extended to atmospheric models at larger- and smaller scales. We demonstrate that atmospheric modellers have not been
particularly critical about the logical basis of model evaluation, and recommend strategic approaches to remedy this. Our
suggestions involve the creation of a spatio-temporal scale-dependent and context-relative ordinal scale for model evaluation
that must be applied in an a-priori, context-dependent fashion.