The testing and evaluation of models
JL Black · 1995
Testing and evaluation of models is an extremely important but difficult exercise. The term testing is generally taken to mean that the model is mathematically, numerically and logically correct, that is, it is free from ‘bugs’. There are several procedures that can be adopted to reduce the chances of making errors and to increase the opportunity for locating those that have occurred. These include the logical definition of variables, a modular construction of the model to enable small components to be tested independently of the remainder, precise coding rules, examination of equation predictions outside the model structure, a check of equation parameter dimensions and a check for correct balances of state variables. Model evaluation is concerned with establishing the appropriateness and accuracy of predictions over a wide range of simulated conditions. The fact that a model predicts accurately under one set of circumstances does not mean that it is valid. However, the wider the circumstances under which the model predictions are accurate, the more confidence is developed in the appropriateness of the concepts and parameters upon which it is based and the more useful will be its predictions. There are several phases to the validation process, including examining the general behaviour of the model, identifying the variables and equation parameters to which the model outputs are highly sensitive, and making direct comparisons with experimental results.