New model · Google DeepMind
WeatherNext 3
Google DeepMind's global weather model now updates forecasts every hour using raw satellite data
· 7 claims checked against 1 source
What it is
WeatherNext 3 is Google DeepMind's most advanced global weather model to date. It is the first global weather model that generates forecasts every hour of the day, rather than on the fixed six or twelve hour cycles used by traditional numerical weather prediction systems.
How it works
Traditional numerical prediction models are physically grounded but costly to run and limited to periodic updates. WeatherNext 3 instead draws directly from raw satellite imagery, allowing it to refresh its forecast continuously and track fast changing conditions such as rain and snow as they form.
What is new
Surface variables such as temperature and humidity are predicted at resolutions of up to 5km, with wind at 10km. Google reports accuracy gains even at locations that were absent from the model's training data. WeatherNext 3 also forecasts industry relevant variables such as solar radiation and cloud cover, intended to help wind and solar farm operators manage output.
Why it matters
The National Hurricane Center used WeatherNext 3 while tracking Hurricane Melissa's landfall in Jamaica. This shift from fixed six hour cycles to continuous, satellite driven updates means finer grained tracking of rapidly evolving storms. The model's forecasts also feed into Google Search, Maps, and Gemini, putting hourly forecasts in front of consumer and enterprise users without the computational overhead of traditional numerical weather prediction.
Availability
WeatherNext 3 is live now inside Google's consumer and enterprise products. No separate release of model weights, code, or training data has been announced.
The evidence
Each statement above was written with the sentence from the source that supports it, and a program (not a model) checked that every quote and every figure appears in the pages it read. Here they are.
- 01
WeatherNext 3 is the first global weather model that generates forecasts every hour of the day
“WeatherNext 3 is the first global weather model that generates forecasts every hour of the day” S1
- 02
WeatherNext 3 draws directly from raw satellite imagery rather than periodic assimilated data
“Unlike previous models, WeatherNext 3 draws directly from raw satellite imagery” S1
- 03
The model predicts surface variables like temperature and humidity at 5km resolution and wind at 10km
“weather station targeted temperature and humidity at 5km resolution, and other surface variables like wind at 10km” S1
- 04
It results in more accurate forecasts even in locations the model has not seen in training
“This results in significantly more accurate forecasts – even in locations the model hasn’t encountered in training” S1
- 05
WeatherNext 3 covers variables like radiation and cloud cover for wind and solar farm operators
“For wind and solar farms, WeatherNext 3 covers variables like radiation and cloud cover, helping operators to manage more efficiently” S1
- 06
Forecasts feed into Google Search, Maps, and Gemini
“It delivers local data for people and enterprises, into Google products like Search, Maps, and Gemini” S1
- 07
WeatherNext helped the National Hurricane Center better predict Hurricane Melissa's landfall in Jamaica
“How WeatherNext helped the National Hurricane Center better predict Hurricane Melissa’s historic landfall in Jamaica” S1
