Update: The final program is now available. Online registration is still open: https://indico3-jsc.fz-juelich.de/event/113/ (click 'Register Now', then you can select in person or online; online participants will receive a Zoom link in due course). To allow also participants from non-European regions to participate in the workshop, recordings of most sessions and talks will be made available to registered participants.

The last year has seen tremendous progress on large scale machine learning for numerical weather prediction. In the workshop we will take stock of the developments and discuss possible next steps and open research frontiers. For this, the workshop will bring together researchers from industry, modeling centers and academia and provide a forum for detailed discussions on the subject.

The workshop will be a mixture of talks and discussions to allow for a free exchange of ideas and help propel the field forward. Topics include:

Information

Date and time: Faculty Club, University of Bonn; 4 September 2023, 9:30 - 5 September 2023, 18:00

Organizers: Martin Schultz (Jülich Supercomputing Center), Matthew Chantry (ECWMF), Christian Lessig (ECMWF), Jürgen Gall (University of Bonn),

We gratefully acknowledge support by ECMWF, the TRA Modelling at the University of Bonn, and the ERC Advanced Grant IntelliAQ.

Program

Monday, 4 September:

Time Content
9:30 - 9:45 Introduction by the organizers
9:45 - 11:00 AI-based forecasting I
Global weather forecasting with GraphCast (Ferran Alet / Alvaro Sanchez, Google DeepMind)
PanguWeather (Lingxi Xie, Huawei Research)
11:00 - 11:30 Coffee break
11:30 - 12:15 Discussion: Evaluation (led by Stephan Rasp, Google Research, with an introduction to WeatherBench 2)
12:15 - 12:45 Short talks I
GraphCast for a limited area numerical weather prediction model (Joel Oskarsson)
Multi-Scale Deep Learning Framework for Projecting Weather Extreme (Antoine Blanchard)
12:45 - 14:00 Lunch
14:00 - 15:15 AI-based forecasting II
FengWu (Lei Bai, Shanghai Artificial Intelligence Laboratory)
The Moonshot Project: towards multi-source AI forecasting (Tom Dunston, UK Met Office)
15:15 - 15:45 Coffee break
15:45 - 16:30 Discussion: Results on evaluation of current models (Zied Ben Bouallegue, ECMWF)
16:30 - 17:15 Discussion: ERA6 and future data sets (Matthew Chantry, ECMWF)
19:00 - 22:00 Workshop dinner (Tuscolo; location, menu)

Tuesday, 5 September:

Time Content
9:30 - 10:45 AI-based forecasting III
Neural General Circulation Models (Stephan Hoyer, Google Research)
Fourier Neural Operators (Boris Bonev, NVIDIA)
10:45 - 11:15 Coffee break
11:15 - 12:00 Discussion: sub-seasonal, seasonal forecasting and climate projections
12:00 - 12:45 Short talks II
Creating skillful and reliable probabilistic forecasts using machine learning (Mariana Clare)
Generative DL for high-resolution regional forecasting: a proof-of-concept (Laure Raynaud)
Uncertainty quantification for data-driven weather models (Sebastian Lerch)
12:45 - 14:00 Lunch
14:00 - 14:35 AtmoRep: Large Scale Representation Learning of Atmospheric Dynamics (Christian Lessig, ECMWF)
14:35 - 15:15 Discussion: Data for and learning from observations (Jörg Schultz, EUMETSAT)
15:15 - 15:45 Coffee break
15:45 - 16:30 Discussion: Towards operationalization (Baudouin Raoult, ECMWF)
16:30 - 17:00 Closing remarks

Registration and Abstract submission

Registration for online participitation is still possible: https://indico3-jsc.fz-juelich.de/event/113/ (Click 'Register Now', then you can select in person or online; online participants will receive a Zoom link in due course).

Abstract submission has been closed.

Accomodation

Bonn has a variety of hotels at various price points. One with a good money/value ratio and very close to the workshop location is Motel One Beethoven. We recommend to book early since Bonn has many business travelers and hotels tend to become booked out and/or expensive.