Coupling Approaches for Next-Generation Architectures (CANGA)

Computer models for simulating the Earth system combine component models of the atmosphere, ocean, land surface, sea-ice and land-ice to project how future changes in climate will impact our energy choices. CANGA scientists are developing (1) a new approach for assembling Earth System Models (ESMs) to better utilize new high-performance computing architectures, (2) new methods for transferring data between models to improve the accuracy and fidelity of the fully coupled system, and (3) improved analyses and techniques for integrating multiple components forward in time in a stable and robust manner.  We will be exploring task-parallel programming models in which the E3SM model is formulated as a collection of computational tasks and data dependencies that are then distributed and scheduled based on a dependency graph and a mapping to computational elements. Task-parallel models will enable us to exploit additional parallelism, improve resilience and manage both model software and computational complexity. We will be developing new remapping algorithms for transferring data beween meshes that include support for irregular and time-dependent meshes as well as preserving desired properties of vector and scalar fields. Finally, coupled climate models are currently integrated forward in time based on knowledge of relevant timescales and engineering decisions to improve computational performance. To date, little attention has been paid to the stability and accuracy implications of those decisions. We will perform mathematical analyses of the full coupled system and develop new algorithms for more stable and robust time integration.

Project Term: 
2017 to 2022
Project Type: 
Laboratory Funded Research

Publications:

None Available

Research Highlights:

None Available