Google’s DeepMind has unveiled WeatherNext 2, the latest iteration of its AI-powered weather forecasting model, which the company says delivers more efficient, more accurate, and higher-resolution global predictions. The upgraded model can now produce reliable forecasts up to two weeks in advance, covering temperature, pressure, and wind.
Researchers say WeatherNext 2 also significantly improves the ability to track tropical storms. Forecasts for hurricane paths can now remain accurate up to three days before impact — a full day longer than what the previous model could achieve. The system additionally introduces hourly forecasting. Overall, Google reports that WeatherNext 2 runs eight times faster than its predecessor, a boost that could help industries such as energy trading make more informed decisions, according to Bloomberg.
“This gives you a more granular forecast,” DeepMind AI researcher Akib Uddin explained. “Many industries are very interested in these one-hour steps. It helps them make more precise decisions and strengthens their resilience to weather.”
The model’s improvements stem from a new design approach described in a recently published research paper. Earlier methods relied on machine learning architectures originally developed for image and video generation, requiring multiple processing cycles to refine predictions. WeatherNext 2 achieves its results in a single step, reducing computational demands and reliance on expensive AI hardware.
While AI has its limitations, weather prediction is an area where the technology excels, with newer models increasingly outperforming traditional forecasting systems — even those running on supercomputers.
Still, WeatherNext 2 is not without shortcomings. Google acknowledges that the model may struggle with predicting rare or extreme rain and snow events due to gaps in available training data. “It’s a limitation of our forecast, but one we’re actively working to improve,” DeepMind research scientist Ferran Alet told Bloomberg.
AI-based weather forecasting is quickly becoming a competitive space, with companies such as NVIDIA, Microsoft, AccuWeather and Huawei all developing their own systems.
