EAGLE

May 2025: Initial operating capability for the Global-EAGLE-solo and Global-EAGLE-ensemble demonstrations. 

December 2025: NCEP has implemented the first version of Machine Learning (ML) based Artificial Intelligence Global Forecast Model (AIGFS). AIGFS v1.0 leverages Google DeepMind GraphCast model, and fine tuned with GDAS inputs and ERA5/HRES/GDAS training targets (Tabas et al., 2025). Compared with the operational GFS, AIGFS v1.0 demonstrates significantly improved synoptic patterns and hurricane track forecasts, while achieving remarkable computational efficiency. In AIGFS v1.0, two updates were introduced to improve the hurricane intensity forecast: (1) a spectral harmonic loss function to overcome the double penalty issue caused by the grid point mean square error loss function, (2) revised pressure level weights and variable scaling to refine the vertical structure of the model state. With these enhancements, the hurricane intensity forecasts are significantly improved and now comparable to the GFS forecast.