Session descriptions

Very high-resolution modeling (Chairs: Daniel Klocke, Florian Ziemen)

Current high performance computers already allow to look into the future of weather and climate modeling. Global coupled storm- and ocean eddy-resolving simulations will mature from being just possible to being a practical approach even on climate time scales with the next generation of supercomputers. There is hope that this new class of climate models will provide a new perspective on the climate system and will contribute to advancing our understanding of the climate system.

To this session we invite contributions analyzing storm- and eddy-resolving simulations, especially in the context of the DYAMOND model intercomparison, as well as contributions dealing with solutions to the challenges in the design, initialization and tuning of such setups.       

Performance portability (Chairs: Sophie Valcke, Mario Acosta, Kim Serradell, Reinhard Budich)

With the deployment of pre-Exascale machines using many-core technologies and based on advanced architecture solutions (most likely on accelerators), there is a need for climate applications to run effectively on these new architectures, as well as on standard “cluster” technologies. This new complexity makes portability difficult and forces application teams to spend considerable effort on porting and optimizing their applications for the new era.

In this session we will share experiences, ideas and recent work to identify major challenges and solutions towards performance and portability on the next pre-Exascale platforms. This session includes topics such as developing new algorithms and methods (such as DSL, separating science issues from the platform requirements) or examples of substantial re-coding, in order to take advantage of the new possibilities of the hardware and accelerators for hybrid computation. 

Machine learning for parameterization schemes (Chairs: Jean-Claude Andre, Peter Düben)

The session “Machine learning for parametrization schemes" will concentrate on two main aspects:

  • The use of machine learning techniques to derive improved parameterization schemes from observations, high-resolution simulations, or model configurations with increased complexity.
  • The use of machine learning for the emulation of existing parameterization schemes with particular focus on HPC efficiency.

Submission of short papers dealing with such issues are encouraged. Each of the selected contributions will be given 15 minutes for oral presentation.

Challenges in exascale data processing and visualization (Chairs: Sandro Fiore, Niklas Röber)

Earth system sciences belong to the most data-intensive scientific disciplines with a progressive data production. Current climate models, implemented with a higher systems complexity that is paired with an increased resolution, produce larger and larger data sets that need to be analyzed, displayed and understood. This session focuses on tasks to foster data analysis of very large simulations. We welcome abstracts that share ideas or present recent work demonstrating how to create effective and efficient visualizations for complex and large simulation output. Short paper submissions are encouraged from all areas. Selected submissions will be given 15 minutes to present their work.