Predictability of In-Stream Physical Habitat for Wisconsin and Northern Michigan Wadeable Streams Using GIS-Derived Landscape Data

Abstract Quantifying spatial patterns of physical and biological features is essential for managing aquatic systems. To meet broad-scale habitat assessment and monitoring needs, we evaluated the feasibility of predicting 25 instream physical habitat measures for wadeable stream reaches in Wisconsin and northern Michigan using geographic information system (GIS) derived stream network and landscape data. Using general additive modeling and boosting variable selection, predictions of reasonable accuracy were obtained for 10 widely used in-stream habitat measures, including bankfull depth and width, conductivity, substrate size, sand substrate, thalweg water depth, wetted width, water depth, and widthto- depth ratio. Biased predictions were obtained for habitat measures such as bank erosion, large woody debris, fish cover, canopy shading, and substrate embeddedness. Model predictions for many commonlyused habitat variables were judged acceptable based on several criteria, including correspondence between prediction errors and observed interannual and inter-site variability in habitat measures and agreement in correlation analyses of fish assemblage metric data with both predicted and observed values. Prediction of physical habitat variables from widely available GIS datasets represents a potentially powerful and cost-effective approach for broad-scale (e.g., multi-state, national) assessment and monitoring of in-stream conditions, for which direct measurement is largely impractical because of resource limitations.