Google DeepMind AI hurricane forecast has track error of just 19 miles, beating traditional models

Google DeepMind’s AI weather model posted an impressive result on Hurricane Isaias: 72 hours before landfall, its track error was just 19 miles, far better than the 36 miles of the European AI model and the 119 miles of the US GFS traditional numerical model. On track forecasting, AI has now fully overtaken traditional models. Intensity forecasting remains a weak spot, however. The model did not get the Category 2 strength of 105 mph at landfall right; even though early intensity readings were correct, intensity is the next hurdle AI weather models must clear. As background, the US National Hurricane Center has been officially using AI models, including DeepMind’s, in operations since June 2025, and AI forecasting is moving from experiment to operational use. > Source: [read the original](https://www.palmbeachpost.com/story/weather/hurricane/2026/10/14/google-deepmind-ai-hurricane-forecast-for-isaias-excelled-with-track-less-so-for-intensity/92229231007/). This text is a summary for reference only.