A Critique of the Use of Distribution-Based Extrapolation Models in Ecotoxicology

A fundamental unanswered question in ecotoxicology concerns the extent to which ecosystem-level effects of pollutants can be understood or predicted from tests at lower levels of organization (Forbes & Forbes 1993). Much attention during the 1970s and early 1980s has been directed towards developing new and better test methods and identifying ideal test species, indicator organisms and biomarkers. Given the impracticality of testing all species for their sensitivity to toxicants, reliance has usually been put on data gathered for a few selected species. In contrast to toxicological studies where data for a few surrogate species are extrapolated to humans, ecotoxicological testing requires extrapolation from a small number of test species to a vast number of species varying in taxonomy, size, life history, physiology and geographic range (Cairns & Mount 1990). In this paper we examine the features and assumptions of recently developed distribution-based extrapolation models that are currently in use (OECD 1991; Van Leeuwen et al. 1992). All these presume an underlying interspecific distribution of sensitivities to a toxicant. We suggest that the rational use of these models as ecotoxicological tools for environmental regulation requires additional basic knowledge in two areas. The first concerns the relationship between structure and function in ecosystems. The second involves the nature of the statistical distribution of toxicity end-points in natural assemblages of species. Because of deficiencies in basic knowledge in these areas, statistical extrapolation does not presently offer an improvement over much simpler arbitrary assessment factors. We conclude that the added complexity inherent in their use is not outweighed by the benefits obtained.