How resolution matters - the future of AI climate modelling

→ Europe/Berlin
InnoDom - Cologne

InnoDom - Cologne

Weyertal 109, 50931 Cologne-Lindenthal
Description

We would like to invite you to participate in our upcoming CESOC/ELLIS workshop, “How Resolution Matters: The Future of AI Climate Modelling”, supported by the HClimRep project, the CESOC ML working group, and the ELLIS Unit NRW.

What resolution do AI climate models really need in training? This workshop will bring together experts in atmospheric physics, numerical modelling, and AI to discuss the role of resolution in future climate AI modelling and explore strategies for developing more efficient and reliable AI-based models.

Florentine Weber/Agnes Lohs
    • 08:30 → 09:00
      Registration
    • 09:00 → 09:10
      Welcome Message
      Conveners: Agnes Lohs (JSC-ESDE), Martin Schultz (JSC)
    • 09:10 → 10:30
      Perspective talks

      4 talks of 15 mins + 5 mins discussion

      Convener: Florentine Weber
      • 09:10
        The atmospheric boundary layer and AI-based models - challenges and opportunities 20m

        Climate models, whether based on physical equations or AI-based, have a finite resolution, yet some of the small-scale processes are climatically important and relevant for society, such as turbulence or cloud formation. These processes, and more generally processes occuring within the atmospheric boundary layer, connect the land surface with all its complexity to the larger-scale flow and its dynamics. Representing these processes or their impact on the atmospheric state with data-driven approaches is tempting, but limitations in data availability lead to risks of overfitting and loss of generalization capabilities. We will discuss approaches that leverage the strengths of traditional process-based understanding and of artificial intelligence to model processes occuring within the atmospheric boundary layer.

        Speaker: Prof. Nikki Vercauteren (Universität zu Köln)
      • 09:30
        From Bedrock to Top of Atmosphere: Resolution, Heterogeneity, and AI in Terrestrial Systems Modelling 20m

        Water and energy moves in both directions between the bedrock and the top of the atmosphere. This exchange is inherently nonlinear and spans many scales in space and time. Heterogeneity in physical parameters and land cover is everywhere, and so is human intervention, from irrigation and groundwater pumping to land-use change. Physics-based models remain too coarse to resolve the relevant processes and parameters accurately. Observations are sparse and often taken at very different spatial and temporal scales from the models. This presentation looks at how model resolution shapes simulations of terrestrial systems. It also explores the challenges and potential of AI for improving terrestrial models, both by combining them with observations and by supporting the analysis of their output.

        Speaker: Prof. Stefan Kollet (Forschungszentrum Jülich)
      • 09:50
        Harnessing the Capabilities of WeatherGenerator to Explore High-Resolution Climate Projections 20m

        In this talk, I will present the challenges we encountered when applying WeatherGenerator to climate modeling, along with practical strategies we developed to address them. I will then discuss ongoing downscaling activities within the scope of the WeatherGenerator project. Finally, I will examine the challenges of generating high-resolution, climate-scale projections, which is important for capturing fine-scale phenomena such as turbulence and mesoscale ocean eddies.

        Speaker: Dr Ankit Patnala (FZJ/JSC)
      • 10:10
        Potential for AI to augment high resolution climate projections 20m

        At the UK Met Office we have two AI flagship projects spearheading our efforts to explore the potential of AI to provide improved climate information. In this talk I will briefly outline progress on these. I will then discuss the implications of this for km-scale climate modelling, and our plans for this work at the Met Office. I hope to emphasize the continuing value of km-scale modelling and process understanding, and how these different modelling developments are complementary. This includes the critical role high resolution modelling will have for AI model training, alongside the opportunities that AI will open-up in terms of enabling advances in km-scale modelling and application.

        Speaker: Prof. Lizzie Kendon (Met Office UK)
    • 10:30 → 10:50
      Coffee Break
    • 10:50 → 11:20
      Short Lightning Talks

      up to 10, 3 mins each

      Convener: Agnes Lohs (JSC-ESDE)
    • 11:20 → 12:00
      Perspective Talks

      2 talks of 15 mins + 5 mins discussion

      Convener: Agnes Lohs (JSC-ESDE)
      • 11:20
        Probing the Physical Limits of Extreme Events with Differentiable Weather Models 20m

        The importance of resolution is realized most critically through extreme events, which often occur on short space and time scales. I will talk about discovering new extreme events, and the extent to which they depend on resolution.

        Speaker: Prof. Greg Hakim (University of Washington)
      • 11:40
        Resolution matters - an NWP perspective 20m

        I will briefly review the role of resolution increases in numerical weather prediction and the improvements seen in uncertainty quantification and extreme event representation on weather & climate time scales. Considering the interactions of scales of motion in the atmosphere, I will reflect on different approaches for correcting vs downscaling vs resolving the history of a future forecast or climate state, which may inspire future directions.

        Speaker: Dr Nils Wedi (ECMWF)
    • 12:00 → 12:30
      Plenary discussion - First replies to perspectives and lightning talks
      Convener: Martin Schultz (JSC)
    • 12:30 → 13:30
      Lunch
    • 13:30 → 13:45
      Planning of Breakout Sessions
      Convener: Martin Schultz (JSC)
    • 13:45 → 14:45
      Breakout Sessions

      TBD

    • 14:45 → 15:15
      Coffee Break

      Rapporteurs prepare discussion summaries

    • 15:15 → 16:00
      Plenary report of breakouts and discussion
      Convener: Martin Schultz (JSC)
    • 16:00 → 16:30
      Final plenary discussion and wrap-up
      Convener: Martin Schultz (JSC)