Baseflow recession analysis (looking at how streamflow decreases after a rain event) has long been of interest as a potential aid in the prediction of streamflow, especially in ungauged catchments, with the goal being to establish reliable relationships between climatic and physical catchment characteristics and the shape of recession curves. Two opposing methods of analysis, focusing on individual recession events and ensemble statistics respectively, have gained traction within the hydrologic community. A recent development involves using a long short-term memory (LSTM) machine learning model that combines these two approaches to better characterize the patterns of baseflow recession in individual catchments, notably identifying an attractor, or master recession curve, that individual recession events will converge to over time. This work extends the application of the LSTM model from a single test basin to a selection of catchments representing the range of hydroclimatic signatures present throughout the contiguous United States. This allows for preliminary investigation of the total variability in observed baseflow recession patterns across the region and lays the foundation for future work on the degree of uncertainty in the model’s analysis of recession characteristics.
