AI for Climate Science seminar series: ​​​​​​​AI for (sub)seasonal forecasting - boosting predictability and understanding teleconnections

Join us for the next session of the IIASA-wide seminar series, AI for Climate Science!

We are pleased to announce the next talk in the series, featuring Dim Coumou from Vrije Unversiteit Amsterdam.

Dim Coumou is Professor of Climate Extremes & Societal Risk at VU Amsterdam and co-founder of Beyond Weather. He is an expert on extreme weather events like heat waves, heavy rainfall, storms, hurricanes and droughts. With his team he pioneered the use of artificial intelligence (AI) methods to gain new insights into the underlying physics and to improve predictability of extremes. Coumou holds a PhD in Natural Sciences from ETH-Zurich (2008) and has worked for top-ranking climate institutes including the Potsdam Institute for Climate Impact Research (PIK) and Institut Pierre-Simon Laplace (Sorbonne University, Paris). He authored over 100 publications, is a highly cited scientist (top 1%) and his work has been extensively covered by international media. Currently he leads project XAIDA which unites 16 European research institutes to develop and apply novel AI methods for extreme weather research. Dim is Chief Science officer at the start-up company Beyond Weather - which uses AI to create targeted seasonal-to-subseasonal forecasts for different societal sectors.

For online participation, a registration is necessary.

Please register here.

Please note that the seminar is going to be recorded.

Title:
AI for (sub)seasonal forecasting - boosting predictability and understanding teleconnections

Abstract:
The increase in compute power and the amount of climate data available to us, enables us to study the climate system in exciting new ways. Massive datasets and new AI methods provide enormous opportunities for climate scientists but the challenge will be to create actual new knowledge from this deluge of data. How can we harvest new insights from all of this big data?

In my presentation I will discuss recently developed data science methods and how those can help in improving predictability and at the same time give us new insights into the climate systems. I will focus on Explainable Artificial Intelligence (XAI) and Causal Discovery methods and their application in season to sub-seasonal (S2S) forecasting. I will present a range of studies, mostly from the XAIDA and EXPECT Horizon projects, that use AI methods, from rather basic approaches to advanced deep learning approaches like transfer learning of Foundation Atmosphere models.

These set of studies suggest that S2S predictability of extremes – particularly heat waves and droughts – have longer lead predictability than classically thought. Moreover, numerical weather prediction models seem to miss aspects of some of the associated teleconnection patterns. XAI methods can give us new physical insights on the sources of predictability and on teleconnection pathways. These studies illustrate that these advanced machine learning methods can not only improve predictability but also provide us new insights into climate teleconnections. This paves the way for accurate AI-based forecast systems, which are more needed than ever with extreme weather increasing around the globe.