Quantifying the Effects of National Water Model Freshwater Flux Predictions on Estuarine Hydrodynamic Forecasts

Abstract

Accurate streamflow forecasts are critical for modeling and managing estuarine water quality, as freshwater fluxes significantly influence coastal dynamics. The National Water Model (NWM) provides high-resolution streamflow predictions, which are valuable for hydrodynamic modeling in poorly gauged coastal regions. However, inaccuracies in NWM forecasts can limit our ability to predict estuarine and nearshore water quality effectively. First, this study evaluates the accuracy of NWM predictions for 14 coastal reaches in southwest Florida's Charlotte Harbor and Caloosahatchee River estuaries from 2018 to 2024, where hydrologic management has impacted water quality. NWM forecasts showed varying bias and variance, with Nash-Sutcliffe efficiencies (NSE) ranging from −2.26 to 0.77. Next, hydrodynamic simulations for the flow-managed Caloosahatchee River Estuary (CRE) were performed using both NWM forecasts and observed streamflows, revealing that errors in NWM predictions during high-flow events caused significant deviations in the position of ecologically relevant isohalines, lasting weeks. Finally, to address these issues, a Long Short-Term Memory (LSTM) network was developed to bias-correct NWM forecasts, improving NSE from 0.41 to 0.53. However, the LSTM's inability to “learn” managed discharge schedules highlights the need for advanced data assimilation and simulation techniques in flow-managed coastal systems.