To address the challenges posed by site-specific variability, we first used K-Means clustering to group sites based on average monthly high-flow events, geographic proximity, and elevation. To adequately distinguish stations with different hydrological patterns while allowing for regional generalization, we created fifteen cluster groups. The resulting groups provide a useful and objective framework for aggregate analysis of sites with similar hydrological regimes, and effectively separate climate and hydrologic settings. For example, stations in the snowmelt-dominated New England are distinguished by earlier flood timing from stations in the Rockies. Stations in the high-elevation spine of Appalachia are separated from proximal sites at lower elevations by high-flow peak magnitudes.
For each cluster group and each individual station, we used the Cox-Lewis statistic to test for significant changes in the occurrence rate of high- and low-flow events (10-, 5-, and 3-year recurrence interval), examining annual and seasonal changes. Consistent with observations of changing precipitation regimes, we found changes in seasonal high- and low-flow events to be spatially and temporally variable. In the eastern US and Canada, and across the Upper Midwest, summer and fall high-flow events have become more common, while low-flow events less common. In the western US and Canada, where recent drought has been well documented, we find that low-flow events have also increased in frequency. In addition, drought in the Southeast United States has led to more frequent low-flow events, particularly in the spring and summer. In some regions, seasonal analysis indicates significant changes in high- and low-flow events that are not detectable using annual analysis; however, for several cluster groups no significant changes have occurred in the past 60 years at either the annual or seasonal scale. Our cluster approach provides meaningful categorization of sites with similar typical flood regimes, allowing for objective evaluation of regional trends in extreme events. In addition, it provides a guide for mechanistic studies, which will guide projections for future flood and drought events in the US and Canada.