WeatherAI

Our WeatherAI team pioneers the use of advanced deep learning methods for Earth System modelling, with a particular focus on weather forecasting.

Over the past few years, machine learning (ML)-driven weather forecasting models have transformed atmospheric modelling. These models now outperform classical physics-based numerical weather prediction systems in various aspects, pushing the limits of predictability and showing promise in forecasting rare, yet high-impact extreme weather events in a changing climate. By leveraging vast amounts of environmental monitoring data, our research extends beyond purely ML-driven forecasting models. In this area, the WeatherAI team excels in developing Foundation Models—comprehensive AI systems that learn a rich representation of the Earth System and can be flexibly applied to multiple applications.

Beyond weather forecasting, we also focus on air quality prediction, aiming to mitigate the adverse effects of air pollution on both society and nature. Through close collaborations with national and international partners, we are committed to advancing the next generation of Earth system modelling using cutting-edge deep learning approaches.

Building on the success of the AtmoRep initiative and our contributions to the EuroHPC project MAELSTROM and the Destination Earth use case on air quality, we are a proud partner of the EU Horizon project WeatherGenerator and lead the Federal Ministry of Education and Research (BMBF)-funded RAINA project.

Team Goals

  • Develop WeatherGenerator, an advanced foundation model for the Earth system
  • Seamlessly integrate observational, geospatial, and numerical simulation data
  • Enhance high-resolution weather forecasting, with a particular emphasis on extreme events
  • Explore innovative methods for air quality prediction
  • Optimise deep learning scalability on high-performance computing (HPC) systems

Projects

Members

Last Modified: 15.04.2025