EVALUATING FOREST GROWTH-MODELS

Effective model evaluation is not a single, simple procedure, but comp rises several interrelated steps that cannot be separated from each ot her or from the purpose and process of model construction. We draw att ention to several statistical and graphical procedures that may assist in model calibration and evaluation with special emphasis on those us eful in forest growth modelling. We propose a five-step framework to e xamine logic and bio-logic, statistical properties, characteristics of errors, residuals, and sensitivity analyses. Empirical evaluations ma y be made with data used in fitting the model, and with additional dat a not previously used. We emphasize that the validity of conclusions d rawn from all these assessments depends on the validity of assumptions underlying both the model and the evaluation. These principles should be kept in mind throughout model construction and evaluation. (C) 199 7 Elsevier Science B.V.