Date of this Version
Modeling species distributions is in most instances, we believe, better if perceived as an exercise in modeling spatial patterns in habitat conditions. This perspective forces the modeler to think about factors and processes that influence local habitat and also to account for as many of these factors as possible in the modeling process. Local habitat conditions in riverine ecosystems (for example, pH, temperature, turbidity, permanence of flow, depths, velocities, substrate, cover, primary production, etc.) are influenced by a wide array of factors and processes operating at multiple spatial and temporal scales (Matthews 1998; Fausch et al. 2002). However, of primary importance is the interplay of watershed and local conditions (Hynes 1975; Richards et al. 1996; Rabeni and Sowa 2002). For instance, local substrate conditions are influenced by water and sediment delivery which are largely determined by watershed conditions and also local geomorphic conditions (for example, channel gradient) that affect sediment transport (Jacobson and Pugh 1999).
Until recently it has been essentially impossible to quantify watershed conditions for thousands of streams segments across large geographic areas (for example, entire states). For this and other reasons, species distribution models developed for the Missouri Aquatic GAP Project were based on only a handful of local habitat variables (Sowa et al. 2007). This pilot project illustrated the importance and utility of these local variables for modeling the distribution of riverine biota, however, the resulting models had relatively low accuracy. We recently completed a project, involving development of statewide predicted distributions for fishes of Nebraska, in which we were able to quantify both watershed and local conditions for essentially all stream segments in the state and use them in the modeling process. Results from this project, which is the focus of this article, provide a specific example of how using both watershed and local variables for modeling the distribution of riverine biota can significantly improve model accuracy.