microsoft/aurora

Implementation of the Aurora model for Earth system forecasting

What it solves

Aurora is designed to predict atmospheric variables, such as temperature, across the Earth system. It addresses the challenge of creating a forecasting model that can be generalized across different atmospheric tasks without requiring massive amounts of task-specific training data for every new application.

How it works

Aurora is a foundation model for the Earth system. It is first trained on a vast amount of general atmospheric data, allowing it to learn general patterns. It can then be adapted (fine-tuned) to specialized forecasting tasks using relatively small amounts of data. The project provides four specialized versions of the model for:

  • Medium-resolution weather prediction
  • High-resolution weather prediction
  • Air pollution prediction
  • Ocean wave prediction

Who it’s for

This tool is intended for researchers and developers working in global environmental modelling, weather prediction, and air pollution forecasting.

Highlights

  • Foundation Model Architecture: Uses a general-to-specific training approach to reduce the need for task-specific data.
  • Multi-Task Capability: Capable of predicting temperature, air pollution, and ocean waves.
  • High Resolution: Supports both medium and high-resolution weather predictions.
  • Broad Data Integration: Trained on diverse datasets including ERA5, CMIP6, and CAMS reanalysis.

Related

  • Project
    ant-intl/Falcon-TST

    A family of large-scale time series foundation models, including univariate and multivariate versions, for efficient and accurate forecasting.

  • Project
    NVIDIA/earth2studio

    A Python-based AI inference pipeline toolkit for weather and climate science that provides a unified API to integrate various AI models and data sources.

  • Project
    AIStream-Peelout/flow-forecast

    An open-source deep learning framework for time series forecasting, classification, and anomaly detection featuring a wide array of state-of-the-art models.

  • Project
    google-research/weatherbench2

    A benchmark framework for evaluating and comparing data-driven and traditional numerical global weather forecasting models using scalable evaluation code and cloud-optimized datasets.

  • Project
    openclimatefix/graph_weather

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