Approach to the preliminary analysis of environmental systems

In the preliminary analysis of environmental problems, mathematical modelling studies can sometimes aid in hypothesis development and in the integration of preliminary data. Circumstances usually require models used in this way to be simulation models closely based on traditional scientific descriptions of component processes. As a result, such models contain many ill-defined parameters, a fact which severely limits the reliance that can be placed on the outcome of any single simulation. In an attempt to overcome this difficulty, it has been proposed that parameters be assigned statistical distributions which reflect the degree of parametric uncertainty and that these distributions be used in Monte Carlo simulation analyses. The authors propose a variation on this theme in which they first stipulate the systems' problem-defining behavior and define a classification algorithm to be applied to the model's output. This algorithm results in each simulation run being classified as a behavior, B, or not a behavior, anti B. The parameters leading to the result are stored according to the behavioural outcome. Subsequently, all parameter vectors are subjected to analysis to determine the degree to which the a prior; distributions separate under the behavioral mapping. This separation, or lack thereof, forms the basis for amore » generalized sensitivity analysis in which parameters and their related processes important to the simulation of the behaviour are singled out. The procedure has been applied to a eutrophication problem in the Peel-Harvey Inlet of Western Australia with encouraging results.« less