A Broadscale Fish-Habitat Model Development Process: Genesee Basin, New York
James E. McKenna, Jr., Richard P. McDonald, Chris Castiglione, Sandy S. Morrison, Kurt P. Kowalski, Dora R. Passino-Reader
Abstract.—We describe a methodology for developing species–habitat models using available fish and stream habitat data from New York State, focusing on the Genesee basin. Electrofishing data from the New York Department of Environmental Conservation were standardized and used for model development and testing. Four types of predictive models (multiple linear regression, stepwise multiple linear regression, linear discriminant analysis, and neural network) were developed and compared for 11 fish species. Predictive models used as many as 25 habitat variables and explained 35–91% of observed species abundance variability. Omission rates were generally low, but commission rates varied widely. Neural network models performed best for all species, except for rainbow trout Oncorhynchus mykiss, gizzard shad Dorosoma cepedianum, and brown trout Salmo trutta. Linear discriminant functions generally performed poorly. The species–environment models we constructed performed well and have potential applications to management issues.