Google's new AI weather model WeatherNext 3 learns from live satellite data.
It provides hourly forecasts at five-kilometer resolution, five times sharper than before.
This leads to faster, more accurate predictions for storms and rain.
What happened
Google Research and DeepMind released WeatherNext 3, an AI weather model that learns directly from live satellite data instead of traditional physics simulations. It generates a fresh forecast every hour at up to five-kilometer resolution, about five times sharper than its predecessor WeatherNext 2.
Why it matters
Faster updates mean earlier and more accurate predictions for fast-developing storms and precipitation. The model also shows medium-range forecast improvements of up to 60 percent over IMERG and 30 percent over MRMS, and it already powers weather features in Google Search, the Gemini app, and Google Maps.
What to watch
The accuracy of precipitation forecasts for users planning a day or more ahead is expected to improve by up to 50 percent. The biggest gains are likely in regions like Latin America, Africa, and the Asia-Pacific, which have been underserved by traditional regional models due to high compute costs.
Ask the AI about this article →
Summaries like this, in your inbox every morning.
WeatherNext 3 marks a departure from previous AI weather models, which were trained on numerical weather prediction (NWP) data from supercomputer simulations. Those simulations carried a six-hour delay, which could introduce errors for fast-changing variables like rainfall. By processing live geostationary satellite data, the new model aims to capture current conditions more accurately and update its forecasts hourly.
This shift toward higher resolution and more frequent updates directly addresses longstanding challenges in weather prediction. The model's ability to resolve narrow rain bands more sharply and predict wind speeds at 100 meters (roughly turbine height) could prove valuable for renewable energy planning and grid management. Google states that regions like Latin America, Africa, and the Asia-Pacific stand to gain the most, as high compute costs for traditional regional models have left them underserved.
The model's widespread availability across Google products, including Search, Maps, and Gemini, suggests it will have a significant reach. However, Google itself notes that the atmosphere will always be unpredictable and points users to national weather services for official warnings. The true test will be how well its promise of more accurate precipitation forecasts, up to 50 percent better when planning a day or more ahead, holds up in real-world conditions and delivers the biggest gains in those regions where predictions have historically been less reliable.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
