Mathematics of Planet Earth

Advances in HPC resources have pushed the resolution of Earth System Models to length scales unattainable until now. However, obtaining physically accurate results in an efficient manner at these scales requires more than just increasing the number of grid cells. The formulation of the models themselves needs to be revisited, including hydrostatic assumptions, convection parametrization, meso-scale eddies in the ocean, and machine learning methods to improve/emulate parametrization schemes. Solving these new models requires new numerical approaches to address scalability issues and challenges from resolving new model features. Furthermore, many of the recent and planned HPC systems use a combination of CPU and GPU cores. To obtain the necessary throughput on these hybrid architectures, new computational strategies are needed to enhance conventional models optimized for CPU-only systems. This session will present progress on the new modeling approaches, numerical techniques, and computational strategies that will enable higher resolution for Earth System Models.