Evaluating hydrologic region assignment techniques for ungaged basins in Alaska, USA

Building continental-scale hydrologic models in data-sparse regions requires an understanding of spatial variation in hydrologic processes. Extending these models to ungaged locations requires techniques to group ungaged locations with gaged ones to make process importance and model parameter transfer decisions to ungaged locations. This analysis (1) tested the utility of fundamental streamflow statistics (FDSS) in defining hydrologic regions across Alaska, USA; (2) evaluated if the hydrologic regions represented different hydrologic processes; and (3) tested the ability of random forest and direct assignment techniques, informed by statistically estimated FDSS (FDSSest) and basin characteristics (BCs), to correctly assign ungaged locations to hydrologic regions. Six hydrologic regions were identified across the domain using FDSS. Differences in mean flow, phase shift of the seasonal cycle, and skewness were the primary characteristics defining each region. Two regions represented arctic and continental climates, generally in the northern portion of the domain; four regions represented the southern, maritime portion of the domain. Random forest modeling with BCs (67% success rate) outperformed FDSSest (58% success rate) suggesting that no statistically estimated streamflow was needed to assign ungaged locations to a region. For regions with many sites, most region assignment techniques performed similarly. Random forest modeling performance declined when BCs and FDSSest were both used to predict region membership, suggesting FDSSest had little information in addition to BCs. This analysis demonstrated that FDSS-based hydrologic regions discern process differences across a data-sparse and hydrologically diverse landscape. Process importance rankings from random forest-derived BCs provided model-independent information for making modeling decisions.