Model Calibration and Data Assimilation for the National Water Model: A Case Study over California

Abstract Flooding causes significant loss of life and property in the United States. Improved flood forecasts have the potential to reduce these losses. The National Oceanic and Atmospheric Administration (NOAA) and collaborators have developed the National Water Model (NWM), a nationwide hydrological prediction system intended for various applications, including flood forecasting. The NWM core is Noah LSM with multiparameterization options (Noah-MP), a spatially distributed hydrological model that predicts river discharge given observed (or forecasted) precipitation, surface air temperature, and other surface meteorological variables. However, it has had only limited evaluation for flood forecasting in the western United States to date. Here, we report on retrospective flood forecasting experiments using the Noah-MP model, the application of which we enhanced through model calibration, bias correction of precipitation, and adjustment of initial soil moisture through data assimilation of discharge observations. Our study focuses on selected flood events between 2004 and 2023 across three Forecast-Informed Reservoir Operations (FIRO) watersheds in California, the Russian, Yuba–Feather, and Santa Ana River basins. Our results indicate that in the two humid and semihumid basins, reforecasts with Noah-MP produced flood forecast results with accuracy comparable to archived forecasts produced by the National Weather Service California–Nevada River Forecast Center (CNRFC). Incorporating precipitation bias correction and adjustments to initial soil moisture informed by forecasted peaks, Noah-MP reforecast results were slightly more accurate than CNRFC forecasts in all three river basins, particularly for lead times exceeding 2 days.