Comparison of models predicting invertebrate assemblages for biomonitoring in the Fraser River catchment, British Columbia

A multivariate, predictive model based on the reference-condition approach is described for the Fraser River catchment in British Columbia. Benthic invertebrate assemblages and environmental descriptors were measured at more than 200 sites from 1994 to 1996, including 219 reference sites. Reference sites were classified into groups representing similar invertebrate assemblages. Five such classifications were produced, using three taxonomic levels (family, genus, and species) and species and family multimetrics. For each of the classifications, discriminant function analysis was used, with environmental descriptors, to develop a predictive model for the reference sites. These models predicted from 43.8% (species) to 61.6% (family) of the reference sites to the correct benthic group. Each model was used to assess deviation from reference condition for 21 test sites exposed to either agriculture, logging, or mining. The models were examined with regard to their sensitivity, robustness, usability, temporal variability, predictive performance, and model certainty. The family and family-metric models were ranked best, followed by the species-metric, genus, and species models. The family-level model is recommended for assessment purposes; its overall performance was slightly superior to the family-metric model and it avoids an extra step in calculation of the metrics. However, the species-level model is recommended for conservation or biodiversity issues.