weathernext

How does a weather model get shared with the world? Google DeepMind's WeatherNext 2 ships code, weights, and daily data feeds in this repository.

The repository contains the code for WeatherNext 2, the global, medium range atmospheric and cyclone forecasting model developed by Google DeepMind and Google Research. GitHub lists Python as the primary language and the Apache-2.0 license. The README provides pretrained model weights and calls the code research software with no guarantee of API stability.

For people who want the outputs rather than running the model, daily data feeds are provided across multiple platforms, including Earth Engine, BigQuery, and Vertex AI, WeatherLab with cyclone tracks, and OpenMeteo with an API and interactive builder. The full training data, the README notes, comes from ECMWF and is best accessed as Zarr through WeatherBench2. The README also covers installation, a quick start notebook, and where to find model weights alongside the feed platforms.

This repository is the primary home for the WeatherNext family. Alongside WeatherNext 2 it hosts legacy and specialized models: WeatherNext Graph, the deterministic graph neural network approach published as GraphCast, and WeatherNext Gen, the diffusion based ensemble approach published as GenCast, so the repo covers both deterministic and ensemble approaches to medium range weather forecasting.

The code is under the Apache License 2.0 and other materials under CC BY 4.0. The README calls the models an experimental research project, says they were not produced with or endorsed by any government meteorological agency, and notes they in no way replace official alerts and warnings. The acknowledgements credit ECMWF data and products, modified by Google, along with NOAA's IBTrACS data.

• Source 1 · Project repository
• Source 2 · README