Linking indices of biotic integrity to environmental and land use variables: multimetric clustering and predictive models

Abstract

Advanced computerized methods and models of retrieving knowledge from large multiparameter data bases were used to analyze data on fish and macroinvertebrate composition (metrics), habitat, land use and water quality. The research focused on the north central and northeastern United States and involved thousands of sites monitored by the state agencies. The techniques and methodologies included supervised and unsupervised Artificial Neural Networks (ANN) modeling, Principal Component Analysis, Canonical Component Analysis (both linear and nonlinear), Multiple Regression Analyses, and analyses of variance by ANOVA. The research resulted in defining a concept of clusters of sites based on their biotic (fish) community composition, identified cluster dominating factors, and developed meaningful models for predicting fish composition based on environmental and in-stream habitat stresses.