• Risk assessment;
  • Multivariate statistics;
  • Nonmetric clustering;
  • Endpoints;
  • Artificial intelligence


Ecological risk assessment has evolved so that the interaction among the components is now an implicit assumption. Unlike single species-based risk assessments, it is often crucial in environmental or ecological risk assessments to be able to describe a system with many interacting components. In addition, some quantifiable description of how different biological communities respond upon the addition of a toxicant or some other stressor is required to adequately describe risk at the ecosystem level. Three methods have been applied at this level: the mean strain measurement used by K. Kersting, the state-space analysis pioneered by A.R. Johnson, and the nonmetric clustering developed by G. Matthews for ecological data sets and for analysis of standardized aquatic microcosm data. Each method has direct application to the description of an affected ecosystem with-out reliance upon a single specific and perhaps misleading endpoint. Each also can assign distance or probability measures in order to compare the control to treatment groups. Nonmetric clustering (NMC) has the advantage of not attempting to combine different types of scales or metrics during the multivariate analysis and is robust against interference by random variables. Applications of these methodologies into an ecological risk assessment should have the benefit of combining large interactive data sets into distinct measures to be used as a measure of risk and as a test of the prediction of risk. The primary impact of these methods may be in the selection and interpretation of assessment and measurement endpoints. Much recent debate in toxicological studies has focused on appropriate measurement endpoints for tests. Nonmetric clustering and other multivariate techniques should aid in the selection of these endpoints in ways meaningful at the ecosystem level. We suggest that the search for assessment and measurement endpoints be left to the appropriate multivariate computation algorithms in the case of multispecies situations. Application of these methods in the verification and validation process of risk assessment will serve to check the selection of endpoints during modeling exercises and to improve the presentation of assessment criteria.