39HYDRO Variability of Rainfall-Driven Stormflow Events in the National Water Model Multi-Decadal Continental-Scale Retrospective Simulation

The National Water Model (NWM) version 3.0 retrospective simulation spans 40 years over the contiguous United States, the Hawaiian Islands, the main island of Puerto Rico, and a portion of Alaska. The simulation is driven using hydrometeorological forcings from the Office of Water Prediction’s Analysis of Record for Calibration. Outputs include gridded values of soil moisture, evapotranspiration, snow accumulation and melt, glacial processes, reservoir releases, 2-D surface runoff, and 1-D channel routing on a vector stream network. Most evaluations of this simulation focus on continuous metrics, such the Nash-Sutcliffe Model Efficiency, or categorical metrics like Probability of Detection conditioned on a statistical discharge threshold. We present an alternative evaluation of the NWM Retrospective based on hydrograph features, also called “hydrological signatures.”

We inspected aspects of the simulation related to stormflow generation processes including storm total volume, peak runoff, runoff efficiency, and flashiness. Model evaluations of this nature often involve a laborious process of manual event selection or require assumptions about baseflow recession parameters. We applied a nonparametric method of stormflow “event detection” that is computationally efficient and suitable for use on a long simulation of streamflow at thousands of locations. This evaluation explored the efficacy of using a single modeling approach to simulating stormflow generation processes at continental scale.