Springe direkt zu Inhalt

MSDI2: Learning approaches for multiscale atmospheric dynamics

Mar 02, 2021 | 11:00 AM - 04:30 PM

Chairs: Nikki Vercauteren, Stephan Pfahl, Annette Müller, Tom Dörffel, Lisa Schielicke

Atmospheric models are classically based on the equations of motion, for which the unresolved scales need to be parameterised. In a complementary approach, statistical and machine learning methods are used to leverage observations and high-resolution model data by providing data-driven models. This mini symposium aims at the application of machine learning approaches for a better understanding of atmospheric dynamics across all scales. Different approaches, techniques and challenges applying machine learning to atmospheric phenomena across all scales will be discussed.

Find all abstracts for all talks here or linked below:


Philipp Hess (FU Berlin)

Inferring precipitation from atmospheric general circulation model variables with deep learning


Gabriele Messori (Stockholm University - University of Uppsala, Sweden)

Different applications of neural networks for weather forecasting


Peter Dueben (ECMWF, UK)

Machine learning for numerical weather predictions

12:30 Open Discussion


Lunch Break


Davide Faranda (CNRS/LSCE, Paris-Saclay)

Can machine learning predict the behavior of non-stationary chaotic dynamical systems? Some preliminary results


Stephan Rasp (ClimateAi)

The optimization dichotomy: The long way towards improving climate models with machine learning


Janni Yuval (MIT)

Physics-guided machine-learning parameterizations of subgrid processes for climate modeling


Open Discussion