Effects of biotic assemblage, classification, and assessment method on bioassessment performance

Biomonitoring requires thorough evaluation of methods used to detect impairment. Using a data set of 202 reference sites and 66 test sites from the Fraser River, British Columbia, Canada, we analyzed the effects of assemblage (benthic macroinvertebrates and periphyton) and reference site classification (ecoregion, stream order, null models, and biotic groups) on two bioassessment methods (BEAST (BEnthic Assessment of SedimenT) and RIVPACS (River InVertebrate Prediction And Classification Scheme)). Although largely undisturbed, the Fraser River is affected in some areas by logging, mining, agriculture, pulp mill effluent, and urban land use. Overall performance was evaluated using the harmonic mean of precision, accuracy, and two measures of sensitivity. Invertebrates and periphyton were equally accurate and precise, but invertebrates were more sensitive. Biotic groups were the least accurate and precise classification, but also the most sensitive and had the greatest overall performance. BEAST was slightly less accurate and precise than RIVPACS, equally sensitive to simulated disturbance, and more sensitive to real disturbance. Assessments with higher sensitivity frequently had lower accuracy, indicating a possible trade-off among these aspects of performance.