Harness Google's breakthrough machine learning models for high-resolution global weather prediction, from real-time operational forecast datasets to custom on-demand cloud inference.
Quick start with forecast data Explore models
WeatherNext 3 is Google's flagship operational model, delivering 15-day global probabilistic forecasts initialized hourly across 64 ensemble members, combining live satellite streams with high-fidelity 5 km resolution.

  • Hourly satellite ingestion

    Ingests live geostationary satellite mosaics directly to refresh forecasts every hour for fast-moving storms.

  • Multi-resolution predictions (5–25 km)

    Single-pass architecture producing 5 km station-calibrated surface variables, 10 km surface grids, and 25 km atmospheric levels.

  • Ground-truth station training

    Trained directly against surface weather station measurements to produce forecasts that reflect real-world ground observations.
View model specifications

Build with WeatherNext

Whether you want to query operational forecast datasets, generate custom predictions on Google Cloud, or inspect open source research models, we have you covered.
Access real-time and historical operational forecast data across Google Cloud Storage (Zarr), Earth Engine, and BigQuery.
Quick start guide
Generate tailored forecasts on demand using dedicated accelerators on Google Cloud. Control ensemble size and forecast horizons.
Get started on Google Cloud
Explore research models, pretrained weights, and interactive demo notebooks on GitHub under open source licenses.
View open source models
Get hands-on with our real-time operational forecasts. Visualize live data, track cyclone forecasts, and see weather where you are to experience WeatherNext in action.
Launch Weather Lab
Explore our latest model, access global forecast datasets, and integrate WeatherNext into your applications.
Get started