ABSTRACT
Over the last few decades a controversy has arisen between some environmental scientists about the way in which models of environmental systems should be formulated. On the one hand, there are those who feel that mathematical and computer-based models should reflect the perceived complexity of environmental systems and should have a structure which resembles closely the physical, biological and chemical structure of the real world as it is understood by them, given the contemporary state of scientific knowledge about the system. In such simulation models, the equations, usually in the form of ordinary or partial differential equations, are obtained in various ways, e.g. from dynamic conservation equations (mass, energy, momentum, etc.), which have a direct physical interpretation. On the other hand, scientists with a statistical turn of mind tend to warn of the dangers inherent in such a 'mechanistic' or 'simulation modelling' approach and favour 'data-based' procedures, where the model structure is inferred, and the model parameters are estimated, by reference to experimental data using more objective, statistically based methods. Following from such considerations, the physically-based (or mechanistic) models are normally fairly complex in structure, deterministic in form, and are characterized by many parameters. In contrast, the data-based models are often simple in structure, inherently stochastic in form, and are characterized by only as many parameters as can be justified by the information content of the 11available data (i.e. they are 'parametrically efficient' or 'parsimonious'; see Box and Jenkins, 1970).
