| description abstract | Nova Scotia is a small maritime province with limited capacity to gauge its abundant water resources, but it possesses a diverse geologic setting and surface water distribution. The common engineering practice of using nearest neighbor catchments as hydrologic surrogates may be unreliable in regions such as this. Catchment classification provides a tool to identify and explain variability in hydrologic regimes and inform data transfer across catchments. Here we develop a catchment classification framework using hydrometric, climatic, and landscape data from Nova Scotia, Canada. An inductive classification approach was first used to identify five generalized hydrologic metaclasses based on streamflow signatures derived from 47 long-term streamflow records. We then attempted to replicate this classification using deductive approaches, and identified key physiographic and meteorological variables that could be useful in classifying ungauged catchments. Due to the limited number of gauged catchments, two supervised deductive classification methodologies were applied for comparison: (1) an automated approach often used in more data-rich scenarios (random forests and classification and regression trees); and (2) a nonautomated approach, which involved manual construction and testing of decision trees. The products of the automated approach (random forests), although more robust, may be challenging to apply, while the manually constructed decision tree, which was guided by a combination of local knowledge and theoretical reasoning, could be a useful tool for practitioners. Climate did not emerge as a particularly strong controlling factor in hydrologic variability in this region, but surface water storage had an important role in flow regime across the province. Results demonstrate that this type of hybrid approach can be effective for understanding hydrologic variability and identifying surrogate watersheds in data-limited regions. | |
