DeepMind weather model outperforms government forecasters in accuracy benchmarks

← Back to the feed

DeepMind weather model outperforms government forecasters in accuracy benchmarks

Developing story first seen 39 minutes ago

· 39 minutes ago

Google has introduced WeatherNext 3, an artificial intelligence weather forecasting system that diverges from conventional meteorological methods by combining machine learning with real-time satellite imagery and historical weather patterns. The model generates detailed hourly predictions with improved accuracy for precipitation forecasting, representing a significant upgrade over its predecessor and traditional supercomputer-based simulation approaches.

WeatherNext 3 will be integrated into Google's consumer products—Search, Maps, and the Gemini assistant—while also being made available to researchers through Google Cloud. The system has demonstrated its effectiveness by achieving the top ranking on Operational WeatherBench, an independent evaluation framework, where it surpassed competing deep-learning weather models from Microsoft and Nvidia.

  • Google released WeatherNext 3, combining AI, live satellite data, and historical patterns for improved hourly weather forecasts, especially for rain and snow
  • The model integrates across Google Search, Maps, and Gemini, with researcher access via Google Cloud
  • WeatherNext 3 ranked first on Operational WeatherBench, outperforming AI weather systems from Microsoft and Nvidia

Coverage

AI Art Business Celebrity Companies Culture Entertainment Research Science Software Technology

Read the full article at the source →