Local
AI model correctly tracked Isaias toward Florida Panhandle, but forecasters still weigh storm strength
Google DeepMind’s model pointed Hurricane Isaias toward the western Florida Panhandle while conventional models initially sent it toward Louisiana. The National Hurricane Center combines AI and physics-based forecasts, with human forecasters translating changing projections into public guidance.
An artificial intelligence model from Google DeepMind correctly predicted Hurricane Isaias would strike the western Florida Panhandle, while conventional models initially projected the storm would move farther west toward southeastern Louisiana.
Isaias made landfall near Destin in early October as a Category 2 hurricane. The AI system performed better than other models on the storm’s path, but it was less accurate than the National Hurricane Center’s blended forecast at predicting the storm’s strength.
The contrast illustrates both the promise and limits of AI forecasting for Gulf Coast communities facing hurricane risks. The National Hurricane Center began incorporating the Google DeepMind model into its official forecasts last year. It now considers that system alongside traditional physics-based models and other forecasts rather than relying on any single prediction.
Traditional models use current atmospheric measurements, including temperature, humidity, air pressure and wind shear, in equations that simulate how a storm may move and intensify. AI models instead identify patterns in large collections of past storms and atmospheric data.
Michael Lowry, a Miami-based meteorologist and hurricane expert who has advised the Federal Emergency Management Agency and the National Hurricane Center, said the DeepMind system has become an unusually strong performer. But AI systems can be difficult to interpret: forecasters may not know why a model reaches a particular conclusion, or why it gets a forecast wrong.
National Hurricane Center Director Michael Brennan said each storm brings different forecasting challenges, making it important to use a consensus of strong models. Forecasters issue updates every six hours during active hurricanes, and they must make sense of model projections that can shift sharply between updates.
Human judgment remains central to the process, Brennan said. Forecasters use their expertise to turn predictions into guidance about hazards that residents can act on.
The center also gives Google DeepMind regular feedback to improve its model. Lowry said AI forecasting systems can be updated in weeks or months, while traditional models may be revised annually at best. That pace could accelerate improvements, though the models remain works in progress and do not replace forecasters.