Why trust Greenbot
Google DeepMind launched Weather Lab on Thursday, an interactive website featuring experimental artificial intelligence (AI) hurricane predictions that forecasters say could revolutionize storm warnings.
The new platform showcases Google’s tropical cyclone model alongside traditional physics-based forecasts. Experts at the National Hurricane Center are now testing the AI predictions in real-time to enhance their official warnings for the 2025 Atlantic hurricane season.
Hurricane forecasting carries huge importance. Tropical storms have caused $1.4 trillion in economic losses over 50 years. Earlier warnings could potentially save thousands of lives through better evacuation planning.
Google’s AI model generates 50 possible storm scenarios up to 15 days in advance. It predicts formation, track, intensity, size, and shape simultaneously — solving a fundamental trade-off that has limited traditional forecasting for decades.
Physics-based models typically excel at either tracking storm paths or predicting intensity, but rarely both effectively. Google’s system overcomes this limitation using stochastic neural networks trained on 45 years of cyclone data and millions of global weather observations.
Source: Google
Testing showed remarkable accuracy improvements. The AI model’s 5-day cyclone track predictions averaged 140 km closer to actual storm locations than ENS, the leading European physics-based model. That performance matches traditional 3.5-day forecasts.
“This is comparable to the accuracy of ENS’s 3.5-day predictions — a 1.5-day improvement that has typically taken over a decade to achieve,” Google researchers reported.
For intensity predictions, the AI outperformed the National Oceanic and Atmospheric Administration’s (NOAA) Hurricane Analysis and Forecast System, a leading regional high-resolution model that has struggled with this challenge.
Dr. Kate Musgrave, a research scientist at Colorado State University who evaluated the technology, found “comparable or greater skill than the best operational models for track and intensity.”
Wallace Hogsett, the hurricane center’s science and operations officer, described the partnership as potentially “really powerful” for “helping people protect themselves.”
Weather Lab displays live and historical predictions from multiple AI models alongside European physics-based forecasts. Users can explore over two years of historical data for analysis and comparison.
Google emphasized that Weather Lab remains a research tool. Live predictions are experimental and not official warnings. The company directed users to local meteorological agencies for authoritative forecasts.
The launch coincides with NOAA’s prediction of an “above average” Atlantic hurricane season featuring 13 to 19 named storms.
Ferran Alet, a DeepMind research scientist, said the AI model “looks very, very strong on both tracks and intensities.” He noted initial skepticism from hurricane experts has shifted toward acceptance as the technology demonstrated its capabilities.
The collaboration comes as the Trump administration has reduced National Weather Service staffing. The partnership planning began more than a year ago, predating the current administration.
