GeoAI workshop: Foundation models for weather and climate

  • * Register (or log in) to the Neural Network to add this session to your agenda or watch the replay

  • Date
    26 March 2025
    Timeframe
    15:00 - 17:00 CET Geneva
    Duration
    120 minutes
    Share this session

    The last few years have seen rapid progress at the intersection of AI and atmospheric physics. While there always have been many uses of techniques from statistics, ML and AI in the domain of weather and climate, we are now at a point where fully data driven AI models arguably rival the performance of complex physical simulations running on HPC systems. Given the successes of foundation models in fields such as language, vision and earth observation, the obvious question and active direction of research is to what extent it is possible to train and use foundation models in this space, and how to efficiently address multiple use cases with a single pretrained model. 

    In this tutorial session we will introduce Prithvi WxC, a 2.3B parameter foundation model for weather and climate. After introducing the model’s main characteristics, we will focus on the problems encountered when fine tuning FMs for weather and climate to new use cases. As a concrete example we will discuss how to use Prithvi WxC to downscale atmospheric data to increase the spatial resolution. 

    Participation 

    Anyone with basic skills in machine learning is invited to register and participate in this workshop. However, support for cloud credits to follow along the labs will be restricted to 200 on first come first serve basis. The first 200 registered users will be asked to provide a GitHub handle and will be able to get AWS resources. Other users will be provided a notebook on Colab for people to follow the training.  

    Workshop assistants: Romeo Kienzler and Daniel Salles.

    Please post in the Neural Network chat any questions you may have during the lab

    Are you sure you want to remove this speaker?