Google DeepMind's WeatherNext 3 Cuts Precipitation Forecast Errors by Up to 50%

Key Info

Google DeepMind's WeatherNext 3 model, developed with Google Research, improves global precipitation forecasting by learning directly from real-time observations. It delivers up to a 50% reduction in forecast error, with the largest gains in regions that have historically been harder to predict.

Highlights

• Achieves up to 50% lower prediction error for global precipitation, especially in historically unreliable forecast regions.
• Learns directly from real-world, real-time observations to produce more localized forecasts.
• A significant step beyond earlier methods that produced blurry estimates or missed severe storm boundaries.