Particle.news

Google Unveils WeatherNext 3 Hourly Satellite-Driven Global Forecasts

The model uses live geostationary satellite data to produce hourly, high-resolution forecasts that Google says boost day-ahead precipitation accuracy.

Overview

  • Google announced WeatherNext 3 on Thursday as a new AI weather model that ingests live geostationary satellite mosaics to initialize forecasts every hour.
  • Google says the model resolves some surface variables down to about 5 kilometers and delivers up to 50% better day-ahead precipitation accuracy and up to 60% CRPS gains versus IMERG in medium-range tests.
  • WeatherNext 3 is already being rolled into Google Search, Maps, the Gemini app and Google Cloud tools so developers and enterprises can access hourly forecasts and energy-specific outputs like 100-meter wind and solar radiation.
  • Independent benchmarks run by Brightband’s Operational WeatherBench place WeatherNext 3 at the top of live leaderboards, but a startup called WindBorne disputes Google’s claim to be first to use raw observations and meteorological agencies say more independent validation and operational testing are still needed.
  • The release continues a multi-year AI forecasting arc from GraphCast and GenCast to WeatherNext 2 and now WeatherNext 3, showing faster, cheaper machine-learning forecasts that can help renewables, agriculture and underserved regions while still needing careful integration with physics-based systems.