109 - Strategies for Evaluating the Performance of NHDPlus-based Continental-scale Hydrologic Models in Predicting Flood and Drought Events

Hydrological models are essential tools for predicting the timing and magnitude of flood and drought, with model results and accuracy impacting the actions to mitigate adverse flood and drought-induced societal and economic impacts. While hydrological model development evaluates model skill based on the comparison between observed and simulated streamflow timeseries, this most widely used method does not emphasize hydrological extremes that emergency responders and planners need to inform critical decisions. We developed the Research-Oriented Streamflow Evaluation Tool (ROSET), a Python-based model-agnostic model evaluation tool, to facilitate a standardized evaluation framework and reduce the barriers to NHDPlus-based model evaluation. Within ROSET, we develop and integrate the Streamflow Extreme Event Dataset (SEED) to identify flood and drought events with characterized return intervals for over 5000 USGS monitoring stations collocated with NHDPlus stream reaches. SEED utilizes the Bulletin 17C recommended Log Pearson Type III (LP3) distribution for determining historical events of varying return intervals (2 – 100 years) and corresponding historical event(s) to evaluate hydrological model skill. While there is no standardized method for extreme droughts, we modify the LP3 method to determine the return intervals of drought events and locate the events within the historical record to evaluate model performance. To demonstrate the utility of ROSET and SEED, we use the National Water Model (NWM) v2.1 for several locations in the United States with various regionally dominant hydrology.