The U.S. National Hurricane Center (NHC) shifted its Oct. 7 forecast for Tropical Storm Isaias east as AI models pointed farther east than many conventional models, affecting decisions about coastal alerts.
At 11 a.m. EDT, the revised track still lay west of the Google DeepMind ensemble mean (GDMI) and the Hurricane Forecast Improvement Program Corrected Consensus Approach (HCCA). The center said stronger and faster projections on the eastern side of its guidance led it to issue a variety of coastal watches for wind and storm surge risks.
The NHC’s 11 p.m. EDT forecast continued to give greater weight to GDMI and HCCA; the center was issuing warnings by then. Isaias made landfall near Destin, Florida, at about 9:30 p.m. EDT Oct. 9 (01:30 UTC Oct. 10), with maximum sustained winds of 105 mph, the NHC said.
The storm shows how AI guidance can enter an operational forecast when models disagree. Historical performance figures offer context, though they do not establish which model was most accurate throughout Isaias.
In comparable cases from the 2025 hurricane season, GDMI had a small track accuracy advantage over NHC forecasts issued 12 to 72 hours ahead, the center’s verification report found. The comparison excluded the season’s first two storms because the model was unavailable for them.
A separate study of cyclones in 2023 and 2024 reported an average five-day track error of 230 kilometers for WeatherNext Cyclones, against 370 kilometers for a conventional ensemble from the European Centre for Medium-Range Weather Forecasts. The researchers said improved guidance can give forecasters and communities more time to prepare. Those figures come from a historical evaluation, not an Isaias forecast score.
Track and intensity pose different challenges. NHC Director Michael Brennan told NPR that the center’s forecast, which blended guidance, performed better than Google DeepMind’s model at predicting Isaias’s intensity. The NHC uses multiple models when preparing its official forecasts.
