An Empirical Reevaluation of Streamflow Recession Analysis at the Continental Scale

Streamflow recession analysis is a widely used hydrologic tool that uses readily available discharge measurements to estimate otherwise unmeasurable watershed-scale properties, predict low flows, and parameterize many lumped hydrologic models. Traditional methods apply the simplifying assumptions of outflow from a Boussinesq aquifer, which predicts the slope of the recession curve relating streamflow to its derivative in log-log space to decrease from early-stage to late-stage recession. However, this prediction has not been validated in actual watersheds. Also, recent studies have shown that slopes of observed recession events are often much greater than traditional methods that predict with data point clouds. We analyze recession behavior of 1,027 streams from across the continental United States for periods of 10 to 118 years, identifying over 155,000 individual recession events. We find that the average slope of observed recession events is greater than that of the point cloud for all streams. Further, recession slopes of observed events decrease with time in only 10% of cases and instead increase with time in 74% of cases. We identify only nine watersheds where observed streamflow behavior often conforms to the predictions of traditional recession analysis, each of which is arid and flat with low permeability. Analysis of our extensive empirical results with a regionalization of catchment hydrologic characteristics indicates that heterogeneity of subsurface flow paths increases the nonlinearity and convexity of observed recession, likely as a function of watershed memory. The practical implications of our analysis are that streamflow is more stable during periods of extended drought than generally predicted.