Google WeatherNext 3 delivers hourly 5-km forecasts from live satellite data

WeatherNext 3, released by Google DeepMind and Google Research, generates hourly forecasts at up to 5-kilometer resolution, compared to the 25-kilometer grid with 6-hour increments of its predecessor WeatherNext 2. The model trains on live geostationary satellite mosaics and sparse weather station observations rather than on outputs from traditional numerical weather prediction models, cutting the six-hour data lag of prior AI weather approaches. Precipitation forecasting accuracy improved by up to 60% on the CRPS metric against NASA's IMERG satellite data, 30% against MRMS, and 10% against rain gauges at early lead times. For planning a day or more ahead, Google states the model delivers up to 50% more accurate precipitation forecasts, with the greatest improvements in regions where forecasts have historically been less reliable. The model includes specialized outputs for renewable energy, including wind speed forecasts at 100-meter height and high-resolution solar radiation data, and supports probabilistic as well as deterministic forecasts. WeatherNext 3 is being integrated into Google Search, Gemini, Google Maps, the Maps Platform Weather API, and Google Earth Engine, with data also available via BigQuery and Cloud Storage.