Linking changes in macroinvertebrate community composition to sources of water quality impairment in Minnesota streams

In the United States, both multimetric indices and multivariate modeling techniques are widely used in conjunction with biological data to monitor and assess the health of streams. However, methods to identify specific drivers of observed changes in stream biota are still largely unavailable. Here we describe the combined use of multivariate models and weighted-average inference models to identify potential sources of water quality impairment in Minnesota streams. Using data collected by the Minnesota Pollution Control Agency from 748 stream sites statewide, we developed weighted average inference models capable of predicting stream temperature and sediment conditions based on the abundance of macroinvertebrate taxa. Subsequently, a RIVPACS-type multivariate model was created to generate lists of macroinvertebrate taxa expected at stream sites in the absence of human disturbance. Expected taxa lists were combined with inference models to generate expectations about stream temperature and sediment conditions in the absence of disturbance. Finally, these expectations were compared to estimates of actual temperature and sediment conditions derived from field observations. A difference between observed and expected estimates of temperature or sediment conditions indicates a shift in macroinvertebrate communities that is linked to those conditions.