Development and evaluation of predictive models for measuring the biological integrity of streams

The ratio of the number of observed taxa to that expected to occur in the a bsence of human-caused stress (O/E) is an intuitive and ecologically meanin gful measure of biological integrity. We examined how O/E ratios derived fr om stream invertebrate data varied among 234 unimpaired reference sites and 254 test sites potentially impaired by past logging. Data were collected f rom streams in three montane ecoregions in California. Two sets of River in vertebrate Prediction and Classification System (RIVPACS) predictive models were built: one set of models was based on near-species taxonomic resoluti on; the other was based on family identifications. Two models were built Fo r each level of taxonomic resolution: one calculated O and E based on all t axa with probabilities of capture (P-c) > 0; the other calculated O and E b ased on only those taxa with P-c greater than or equal to 0.5. Evaluations of the performance of each model were based on three criteria: (1) how well models predicted the taxa found at unimpaired sites, (2) the degree to whi ch O/E values differed among unimpaired reference sites and potentially imp aired test sites, and (3) the degree to which test site O/E values were cor related with independent measures of watershed alteration. Predictions of s pecies models were more accurate than those of family models, and predictio ns of the P-c greater than or equal to 0.5 species model were more robust t han predictions of the P-c greater than or equal to 0 model. O/E values der ived from both species models were related to land use variables, but only assessments based on the P-c greater than or equal to 0.5 model were insens itive to naturally occurring differences among streams, ecoregions, and yea rs.