openclimatefix/metnet

PyTorch Implementation of Google Research's MetNet and MetNet-2

What it solves

This project provides PyTorch implementations of the MetNet family of models, designed for short-term weather and precipitation forecasting. It enables users to predict weather patterns by leveraging satellite imagery and other atmospheric observations.

How it works

The repository implements several versions of the MetNet architecture:

  • MetNet: The original model for short-term forecasting.
  • MetNet-2: An extension that uses a larger context image to predict weather up to 12 hours ahead.
  • MetNet Global: A version designed for global precipitation forecasting.

These models process input data—such as GOES-16 satellite and MRMS precipitation data—to generate forecasts. The project also provides pretrained weights for MetNet and MetNet-2 trained on cloud mask and satellite imagery data via the HuggingFace Hub.

Who it’s for

This is for researchers and developers working on meteorological AI, weather forecasting, and atmospheric science who want to use PyTorch-based implementations of Google Research's MetNet models.

Highlights

  • Multiple Architectures: Includes implementations of MetNet, MetNet-2, and MetNet Global.
  • Pretrained Weights: Available for MetNet and MetNet-2 via HuggingFace.
  • Data Integration: Integration with HuggingFace Datasets for GOES-16 and MRMS data.
  • PyTorch Native: Built entirely in PyTorch for ease of training and inference.

Related

  • Project
    openclimatefix/graph_weather

    A PyTorch library implementing various graph neural networks for weather forecasting and data assimilation, including models like GenCast and Aurora.

  • Project
    sktime/pytorch-forecasting

    A PyTorch-based library for time series forecasting that provides a high-level API and state-of-the-art deep learning architectures like Temporal Fusion Transformers and N-BEATS.

  • Project
    torchgeo/torchgeo

    TorchGeo is a PyTorch library that makes working with satellite and aerial imagery easy. It offers ready‑made geospatial datasets, smart samplers for patch extraction, multi‑spectral pre‑trained models, and tight integration with PyTorch Lightning, letting researchers train and evaluate remote‑sensing models with just a few lines of code.

  • Dispatch
    WeatherNext 2 release notes / what's new

    Google DeepMind has released WeatherNext 2, an AI weather forecasting model that generates forecasts 8x faster than previous versions with resolution up to 1-hour and improved accuracy across 99.9% of variables.

  • Project
    hysts/pytorch_mpiigaze

    A PyTorch implementation of gaze estimation methods from the MPIIGaze and MPIIFaceGaze papers for training and evaluation of appearance-based gaze tracking.