Google DeepMind's WeatherNext 3 Dominates Operational WeatherBench Leaderboard
The latest innovation in weather forecasting comes from Google DeepMind's WeatherNext 3 model, which has topped the Operational WeatherBench leaderboard. This achievement is a significant milestone in meteorology, outperforming traditional physics-based models that have been used for decades.
Weathernext 3 uses live geostationary satellite mosaics to update global forecasts every hour at up to 5 km resolution for surface variables like temperature and moisture. This is a significant improvement over the traditional models, which refresh every six hours and resolve detail at roughly 25 km grids.
The model's hourly updates and high-resolution forecasts have led to improved probabilistic rain-prediction accuracy of 50-60% compared to its predecessor, WeatherNext 2. Additionally, it has introduced new clean-energy variables such as 100-meter wind speeds, solar irradiance, and cloud cover targeted at grid operators and renewable facilities.
The stakes are high when it comes to cyclones, with the model's extra day of warning potentially saving lives. Google equates this improvement to a decade of meteorological progress, making it a significant leap forward in weather forecasting.