| Abstrakt: |
This report explains the steps and specific methods used to predict fluvial fish occurrences in their native ranges for the conterminous United States. In this study, boosted regression tree models predict distributions of 271 ecologically important fluvial fish species using relations between fish presence/absence and 22 natural and anthropogenic landscape variables. Models developed for the freshwater portions of the ranges for species represented 28 families. Cyprinidae was the family with the most species (87 of 271) modeled for this study, followed by Percidae (34) and Ictaluridae (17). Model predictive performance was evaluated using four metrics: area under the receiver operating characteristic curve, sensitivity, specificity, and True Skill Statistic, which are all from tenfold cross-validation results. The relative importance of the predictor variables in the boosted regression tree models was calculated and ranked for each species. The three strongest natural predictors of fish distributions were network catchment area, the mean annual air temperature of the local catchment, and the maximum elevation of the local catchment, while the three strongest anthropogenic predictors were downstream main stem dam density, distance to downstream main stem dam, and the percentage of pasture/hay land use area within network catchment boundaries. Study results showed 61 fish species were sensitive to climate variables, and 40 fish species were sensitive to anthropogenic stressors. The models developed in this study can be used to derive critical information regarding habitat protection priorities, anthropogenic threats, and potential effects of climate change on habitat suitability, aiding in efforts to conserve fluvial fishes now and into the future. [ABSTRACT FROM AUTHOR] |
