Diagnostic variables as predictors of ecological risk

The state of an ecosystem may be represented by a multidemensional state vector,x. The goal of ecosystem management is to insure that the ecosystem remains within some setX of acceptable states, such thatx ∈X. Since ecosystem management decisions must be based on limited knowledge, a small number of diagnostic variables must be found which accurately reflect ecosystem state. If the vector of diagnostic variables, ω, is found to be within a specified set Ω, the state vectorx is predicted to be withinX. The selection and use of such diagnostic variables is examined in the context of an aquatic ecosystem simulation model. Techniques used in searching for diagnostic criteria include multiple linear regression, discriminant analysis, and visual inspection of graphical data displays. The adequacy of a diagnostic criterion as a predictor of ecological risk is demonstrated to be a function of the associated rates of type I and type II statistical errors. A simple cost-benefit analysis is undertaken to illustrate one approach for choosing an optimal balance between these error rates.