Google’s DeepMind hurricane model is helping forecasters make faster and more accurate predictions. The AI tool was released in June and quickly proved its value during the 2025 Atlantic hurricane season. Forecasters at the National Hurricane Center now rely on the model to predict storm paths and intensity.
Predicting Hurricane Melissa’s Path
Philippe Papin, a National Hurricane Center meteorologist, used the DeepMind model to forecast Hurricane Melissa. He predicted the storm would become a category 4 hurricane within 24 hours. No forecaster had ever issued such a bold forecast for rapid strengthening. According to The Guardian, Papin explained on social media that roughly 40 to 50 DeepMind ensemble members showed Melissa becoming a category 5 storm.
The model proved correct. Melissa made landfall in Jamaica at category 5 strength. The storm was one of the strongest landfalls ever documented in nearly two centuries of Atlantic basin record-keeping. Papin’s forecast likely gave people extra time to prepare and possibly saved lives and property.
How the Model Works
Google DeepMind spots patterns that traditional physics-based weather models may miss. The AI tool works much more quickly than older models. Michael Lowry, a former National Hurricane Center forecaster, said the computing power is less expensive and time consuming. The model takes just a few minutes to produce results on a desktop computer. Traditional government models require some of the biggest supercomputers in the world and can take hours to run.
Performance and Limitations
Through all 13 Atlantic storms so far this year, Google’s model has been the best at predicting hurricane tracks. It even beat human forecasters on track predictions. James Franklin, a retired forecaster, said the sample is now large enough to show this is not beginner’s luck.
But the model has weaknesses. It sometimes gets high-end intensity forecasts wrong. The tool struggled with Hurricane Erin earlier this year and with Typhoon Kalmaegi, which hit the Philippines on Monday. Franklin said he plans to talk with Google about making the output more helpful. He wants additional data to understand why the model produces its answers.
The US and European governments also have their own AI weather models in the works. These tools have shown improved skill over previous non-AI versions. Startup companies are now trying to solve tougher problems like tornado outbreak warnings and flash flooding predictions.
