Measurement uncertainties, e.g., low statistical power due to insufficient observations, difficulties in making physical measurements, inappropriateness of measurements, and natural variability in organic responses to stress;
Conditions of observation, e.g., spatiotemporal variability in climate and ecosystem structure, differences between natural and laboratory conditions, and differences between tested or observed species and species of interest for risk assessment;
Inadequacies of models, e.g., lack of or knowledge concerning underlying mechanisms, failure to consider multiple stresses and responses, extrapolation beyond the range of observations, and instability of parameter estimates.
Implications of Uncertainty for Ecological Risk Assessment
Most of the above uncertainties affect human health risk assessments, as well as ecological risk assessments. The consensus of the group was that knowledge-based uncertainties are often more important than uncertainties in parameter estimates. The usual statistical measures of uncertainty, p values and variance, measure only uncertainty due to random variation within the model; they do not account for uncertainties due to use of an incorrect model.
It was generally felt that the degree of uncertainty in ecological risk assessments increases with the level of biological organization. Models of ecosystem stress have higher uncertainties than models of populations and models of individual organism response. That is due in part to the increase in the number of end points available for modeling. Organism-level studies, such as single-species toxicity tests, usually have simple end points, such as survival and reproductive success. Ecosystem studies have the same end points plus additional ones that account for species interactions and measure community effects. Because of those uncertainties, ecological risk assessments still require substantial reliance on expert judgment and cannot be strictly model-based. Judgment-based approaches, such as the quotient approach to pesticide hazard assessment (described by Dr. Slimak in his plenary presentation) are often preferable to models for regulatory risk assessment.