(Bloomberg) -- The newest AI weather model from Google DeepMind not only forecasts how hard the wind blows at turbine height and how much sunlight reaches a solar farm, but refreshes those projections every hour from satellite images.
Most Read from Bloomberg
Uber to Cut 10% of Jobs, or 3,300 Roles, to Slash Bureaucracy - World's Unusually High Dollar Exposure Risks Fueling Selloff - Apple Maps Renames Lake Ontario 'Lake America' After Trump Order - China-US Spat at G20 Largely Came Down to Dispute Over One Word - Massive Tax Bill Clouds Sale of Singapore Land by Malaysian King's Son
The WeatherNext 3 model, released Thursday, generates detailed forecasts comparable with other major models. And it generates them more quickly and frequently because it's no longer limited by the six-hourly government data cycles they typically rely on.
"It gets much more accurate by not waiting for the next analysis date and using the most recent information," said Ilan Price, a senior research scientist at DeepMind.
The new model shows improved temperature and precipitation predictions compared with previous versions and major European models, and adds new solar and cloud cover estimates, according to DeepMind. The model's forecasts are yet to be independently verified.
That combination may prove especially useful for forecasting weather conditions that move global energy markets, said Dan Travers, co-founder of Open Climate Fix, a London-based nonprofit that helps energy traders and utilities forecast renewable generation. Google says the group has been testing WeatherNext 3.
Wind and solar are reshaping power grids and weather is increasingly driving energy supply, not just demand, while trading has shifted from month-ahead contracts toward 30- and 15-minute markets throughout the day, he said.
"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," Travers said.
Major AI weather models developed by governments, commercial forecasters and tech companies typically depend on large datasets that combine historical observations with forecasting models to reconstruct a consistent picture of past atmospheric conditions.
The most commonly used dataset, produced by the European Centre for Medium-Range Weather Forecasts, is assembled every six hours and takes about five hours to complete. Leading AI models, including previous versions of DeepMind's, could only produce a new forecast when each new version of that snapshot arrived.