Effects of Land Cover on Stream Ecosystems: Roles of Empirical Models and Scaling Issues

We built empirical models to estimate the effects of land cover on stream ecosystems in the mid-Atlantic region (USA) and to evaluate the spatial scales over which such models are most effective. Predictive variables included land cover in the watershed, in the streamside corridor, and near the study site, and the number and location of dams and point sources in the watershed. Response variables were annual nitrate flux; species richness of fish, benthic macroinvertebrates, and aquatic plants; and cover of aquatic plants and riparian vegetation. All data were taken from publicly available databases, mostly over the Internet. Land cover was significantly correlated with all ecological response variables. Modeled R2ranged from 0.07 to 0.5, but large data sets often allowed us to estimate with acceptable precision the regression coefficients that express the change in ecological conditions associated with a unit change in land cover. Dam- and point-source variables were ineffective at predicting ecological conditions in streams and rivers, probably because of inadequacies in the data sets. The spatial perspective (whole watershed, streamside corridor, or local) most effective at predicting ecological response variables varied across response variables, apparently in concord with the mechanisms that control each of these variables. We found some evidence that predictive power fell in very small watersheds (less than$1-10\ {\rm km}^{2}$), suggesting that the spatial arrangement of landscape patches may become critical at these small scales. Empirical models can replace, constrain, or be combined with more mechanistic models to understand the effects of land-cover change on stream ecosystems.