DeepMind's AI predicts hurricanes faster with 80% accuracy

Google’s DeepMind and Google Research have developed WeatherNext, an AI model that predicts hurricane intensity and trajectory with unprecedented accuracy. In a case study during Hurricane Melissa in October 2025, the model forecasted a Category 5 hurricane hitting Jamaica five days before landfall with 80% confidence, outperforming traditional models. Traditionally, improving forecast lead time by a day would require a decade of research, but WeatherNext achieves this by leveraging lower-resolution atmospheric data to capture critical signals missed by conventional methods.

The model addresses a long-standing challenge in AI weather prediction: the rarity of extreme events like hurricanes limits training data. Researchers trained WeatherNext on both general weather patterns and cyclone-specific data, enabling it to predict storm intensity—a feat most AI models previously struggled with. While global models excel at tracking storm direction, they often fail at intensity predictions, which are crucial for emergency response. WeatherNext’s ability to forecast rapid intensification, such as predicting a Category 5 storm from a Category 1 stage, marks a breakthrough in real-time hurricane forecasting.