Dealing with uncertainty in environmental model predictions



There is a growing trend toward the use of complex, integrated models of environmental systems to support policy making in such diverse areas as human health and ecological risk assessments, safety analysis of nuclear waste repositories, and analysis of global climate change. Historically, modeling studies for such problems have been based on a deterministic framework, in which “point” estimates of specific parameters are combined to generate a single estimate of the outcome of interest. These analyses are typically carried out either with conservative—for example, “worst-case”— assumptions, or more realistic—for example, “best-guess”—estimates for the model parameters.

In general, the actual likelihood of the resulting single-point deterministic model predictions cannot be quantified without additional information. However, it can be shown that a systematic use of conservative parameter values yields results that have a very small probability of occurrence due to the problem of “compounded conservatism.” Given that uncertainty is certain, the utility of deterministic model predictions is therefore likely to be diminished without some demonstration of the associated uncertainties.