ABSTRACT The increasing frequency of hydrological extremes highlights the need for event‐based approaches to evaluate hydrological model performance and support water resource management. Traditional long‐term continuous simulations often overlook model behavior during critical flood and drought periods, limiting their operational value. To address this gap, we developed a coupled SEED–CSES framework for large‐sample, event‐based benchmarking. SEED identifies flood and drought events using the Log‐Pearson Type III (LP3) distribution for multiple return intervals (2, 5, 10, 25, 50, and 100 years), while CSES evaluates model skill. We demonstrate the framework by assessing the extreme‐event prediction performance of the National Water Model (NWM) v3.0 at more than 7000 USGS NWIS stations, including over 600 CAMELS basins. Across the CONUS domain, NWM 3.0 shows higher skill for flood events (median KGE ≈0.20) than for drought events (median KGE ≈−0.78). Wetter eastern, southeastern, and northwestern regions perform better (median KGE ≈0.387), while arid western and southwestern regions show low performance (median KGE ≈−0.447), illustrating how event‐based benchmarking reveals hydrological behaviors masked in long‐term evaluations. The integrated SEED–CSES framework provides a standardized and automated platform for hydrological model assessment, supporting improved flood forecasting, drought monitoring, and climate resilience.
Alam et al. (Sun,) studied this question.