Modeling stream fish biogeography to assess water quality in a species-rich region

Natural resource managers commonly monitor fish to assess the water quality condition of streams. I have evaluated a biological assessment metric, based on observed and expected fish species richness, for application in Georgia. A multivariate species distribution model was built using a Random Forest machine-learning algorithm to predict expected fish taxa based on a streams environmental characteristics, such as elevation, slope, and flow. The ratio of observed to expected taxa richness was then used to estimate whether a stream was close to, or far from, a least-disturbed condition. The model was useful in the species-rich northern half of the state above the Fall Line, but inadequate in the less speciose southern half. This research analyzed the biogeography of Georgias fishes, demonstrated a tradeoff when including rare species in bioassessment, exhibited the sensitivity of fluvial taxa to human disturbance, and provided an additional method for assessing stream conditions in the state.