Updating the US hydrologic classification: an approach to clustering and stratifying ecohydrologic data

Hydrologic classifications unveil the structure of relationships among groups of streams with differing streamflows and provide a foundation for drawing inferences about the principles that govern those relationships. Hydrologic classes provide a template to generalize hydrologic responses to disturbance and stratify research and management needs applicable to ecohydrology. We used a mixed-modelling approach to create hydrologic classifications for the continental USA using three streamflow datasets, a reference dataset compiled under more strict traditional standards and two additional datasets compiled under more relaxed assumptions. A variety of models were applied to each dataset, and Bayes criteria were used to identify optimal models and numbers of clusters. Using only reference-quality gauges, we classified 1715 stream gauges into 12 classes across the USA. By including more streamflow gauges (n=2402 and 2618) of lesser reference quality in subsequent classifications, we observed minimal increases in dimensionality (i.e. multivariate space) at the expense of increasing uncertainty and outliers. Part of the utility of classification systems rests in their ability to classify new objects and stratify data by common properties. We constructed separate random forest models to predict hydrologic class membership on the basis of hydrologic indices or landscape variables. In addition, we provide an approach to assessing potential outliers due to hydrologic alteration based on class assignment. Departures from class membership due to disturbance take into account multiple hydrologic indices simultaneously; thus, classes can be used to determine if disturbed streams are functioning within the natural range of hydrologic variability. Published 2013. This article is a U.S. Government work and is in the public domain in the USA.