Incorporating Network Connectivity into Stream Classification Frameworks

Stream classification frameworks are important tools for conserving aquatic resources. Yet despite their utility, most classification frameworks have not incorporated network connectivity. We developed and compared three biologically informed stream classification frameworks considering the effects of variables indexing local habitat and/or connectivity on stream fish communities. The first framework classified streams according to local environmental variables largely following the precedent set by previous stream classifications. The second framework classified streams according solely to network connectivity variables, while the third framework considered both local and connectivity variables. Using fish community data from 291 wadeable streams in South Carolina, USA, we used conditional inference tree analyses to identify either seven or eight discrete types of wadeable streams within each framework. Classifications were evaluated on their ability to describe community composition at a subset of sites not used in model training, and canonical correspondence analysis suggested that each framework performed similarly in describing overall community variation, with about 19% of variation explained. After accounting for the effects of biogeography and land use in our analytical approach, each classification explained a substantially higher amount of community variation with 46% of variation explained by our connectivity-informed classification and 42% explained by our locally informed classification. Classifications differed in their ability to describe elements of community structure; a classification incorporating connectivity predicted species richness better than the one that did not. This study ultimately addresses an important knowledge gap in the classification literature while providing broader implications for the conservation of aquatic organisms and their habitats.