Detection of Extreme Streamflow Reoccurrence Patterns over the Southeast United States

Hydrologic extremes such as floods or droughts are associated with diverse drivers across multiple time scales. Short-term conditions, such as heavy rainfall or rapid snowmelt, link to instantaneous meteorological traits. While these often tie to convective or frontal processes, accurate simulation is possible through physical models that convert weather into hydrologic processes. In addition, long-term processes significantly impact extreme hydrologic conditions, and can be divided into seasonal shifts tied to regional precipitation variability and long-term changes due to large-scale atmospheric circulation or groundwater fluctuations. Simulating long-term processes with physical models is challenging as they stem from a mix of meteorological and climatic forcing mechanisms; however, statistical models that capture persistent data patterns and translate them into numerical descriptions of variability prove more effective. To that end, the objective of this study is to identify recurring patterns in the occurrence and magnitude of high and low streamflow events within rivers located in the southeastern United States using an improved Harmonic Oscillator Seasonal Trend (HOST) model, which outlines seasonal and long-term drivers of hydrologic extremes using a statistical decomposition approach. Within the HOST model framework, time series components are extracted from hydrologic streamflow records using the Seasonal and Trend decomposition using Loess (STL) approach, and a set of wave functions is fitted to effectively capture the underlying patterns within the dataset. A final model is built based on the waveform synthesis of the best-fit functions, to comprehensively explain the variations in combined seasonal and long-term extreme flow behaviors.

The data used in this work include daily streamflow values from nearly 62,000 river segments obtained from the National Water Model (NWM) retrospective data (v.2.1), as well as over 300 USGS streamflow gauges over the Southeast US (defined by the USGS Region 3—Atlantic-Gulf region). The results of the work present the spatial distribution of wave parameters and assess the seasonal and long-term reoccurrence of hydrological extremes, where the length of the period of the fitted functions provides information on the expected variability of the occurrence of extremes both spatially and temporally. A comparison of the accuracy of statistical models built using the NWM and USGS data provides information on how accurately the NWM dataset can simulate seasonal and multiannual streamflow behavior, defining the viability of the retrospective data for research of extreme hydrologic events within a distributed model framework. Furthermore, the methodology can be applied as a tool for stakeholders to improve water management strategies, while the results can be used in a long-term change detection framework as a bias correction method for physical models.