DeepMind's newest model predicts wind speeds at turbine level and how much sun actually hits a solar farm, and it refreshes those snapshots every hour using satellite images. The company says temperature and precipitation forecasts have improved versus earlier versions and leading European models, and it now provides updated assessments of solar irradiation and cloud cover. DeepMind also says the outputs have not yet been independently checked.
Most AI weather systems lean on big blended datasets that reconstruct past atmospheric conditions from observations and forecast models. The most widely used one from the European Centre for Medium-Range Weather Forecasts is updated every six hours and takes roughly five hours to build, which meant top AI models, including DeepMind's earlier releases, typically waited for the next finished snapshot before issuing a new run.
WeatherNext 3 cuts that wait. It can update hourly by using already built ECMWF forecasts and folding in incoming satellite imagery that captures a three dimensional read on the atmosphere. DeepMind's researchers say that reduces the typical information age in a forecast from about seven hours to around three to four. "It gets much more accurate by not waiting for the next analysis date and using the most recent information," said Ilan Price, senior research scientist at DeepMind.
When headlines shift, steady investing over time builds real results, download the free Always Be Buying E-Book
According to Google, Open Climate Fix - a nonprofit based in London that assists energy traders and utilities in predicting renewable output - has been trialing WeatherNext 3. Co-founder Dan Travers said the model's mix of skills could be especially handy for anticipating weather that moves global energy markets. He noted wind and solar are reshaping grids, weather now steers supply as well as demand, and trading has shifted toward 30 minute and 15 minute markets throughout the day. As he put it, "A large bank of clouds coming across Norfolk, where there's a lot of solar farms, could take out the equivalent of an entire coal-fired power station in an hour."
DeepMind says WeatherNext 3 produces detailed forecasts on par with other major models and avoids the six-hourly government update bottleneck, but the outputs have not yet been independently verified. Watch how third party testing shakes out and whether energy players start using it in live markets.
