| Abstract/Description or Keywords |
Management of fish biodiversity and individual fish species requires the ability to understand and predict expected species distributions. Models predicting species distributions can provide insight into habitat requirements, help identify unique outlier populations, and determine potential biodiversity hotspots. Our goal was to determine whether reliable models of species distributions could be developed for freshwater fish in British Columbia using large-scale macrohabitat data. We surveyed 48 stream sites in a British Columbia drainage with high species diversity (the Blackwater River) and developed statistical models based on macrohabitat variables to describe fish species distributions. Classification rates (i.e., the proportion of sites correctly classified by the model) of our logistic regression models based solely on map-based variables were generally high (73-90%) for most fish species found in the Blackwater River drainage. Including field-measured variables produced significantly better models for most species, but improvements in classification rates were generally marginal. Application of the Blackwater River species models to data from a geographically distant drainage (the Similkameen River) was not successful (i.e., classification rates were poor for all shared species). Classification rates for species in the Similkameen River drainage were improved considerably by using the same macrohabitat variables to generate logistic models that were unique to that watershed. Combining data for the two watersheds also generated significant logistic regressions. Quality of these combined models was significantly improved for most species by incorporating a categorical variable that presumably captured broad differences in habitat conditions between the watersheds; however, classifications were poorer than models specific to individual drainages. Further refinements to quantify variation in habitat conditions among watersheds should permit the development of regional fish distribution models as a layer in geographic information systems. In conjunction with map-based macrohabitat information, distribution models could provide a powerful diagnostic and predictive tool for improving watershed planning and management practices. |
