Details of the talk:
- Date: December 14
- Time: 11:00 a.m. - 12:00 p.m.
- Location: Ground-floor lecture hall, Tübingen AI Center (Maria-von-Linden-Str. 6, 72076 Tübingen)
Talk title: Challenges and Opportunities of Machine Learning for Weather and Climate
Abstract: Weather and climate are of central importance for human societies. Traditionally, physical laws and numerical methods based on their discretization have been used to model and simulate the Earth system. However, a large record of Earth observations and simulation data provides a tremendous opportunity for machine learning-based techniques. For weather forecasting, machine learning models such as PanguWeather and GraphCast have already demonstrated performance comparable or better than the best conventional approaches. The talk will provide an overview of recent developments, including the AIFS developed at the European Centre for Medium-Range Weather Forecasts, and AtmoRep, which shows the potential of large scale representation learning for Earth system modeling. Open research frontiers will be discussed in detail; e.g. probabilistic (generative) models; the development of models that are trained on a wide range of heterogeneous data sources for the entire Earth system; training on observational data, e.g from satellites; and machine learning-based approaches that might be able to help address climate change.
Bio: Christian Lessig is a professor at Otto-von-Guericke Universität Magdeburg where he is affiliated with the Institute for Simulation and Graphics. His research centers around the development of effective simulation techniques for computer graphics, in particular for light and fluids. More recently he has taken an interest in the application of deep learning to weather and climate problems, developing one of the first large-scale foundation models for atmospheric data. He is also part of the recently started initiative by the European Center for Medium Range Weather Forecasting (ECMWF) to develop and deploy its own AI-based model.