Building a GPU Accelerated Multi-RRM High-resolution Global Climate Model to Characterize Extreme Events over U.S. Pacific Ocean Islands

Developing a climate model in an Exascale-GPU environment is crucial for simulating high-resolution, climatological-scale data through mid-century across the Pacific Ocean, continental United States (US), and the US islands of Hawaii, Guam, Northern Mariannas, and American Samoa. This work employs the GPU-ready model SCREAM (Simple Cloud Resolving E3SM Atmosphere Model) with a regionally refined mesh (RRM) to enhance computational efficiency in extended production runs. Accurately capturing U.S. islands in the Pacific Ocean presents challenges in mesh generation and model set up due to their small sizes and complex topography. Some islands are volcanic with distinct climate features, while others are flat and featureless, making them indistinguishable from their oceanic surroundings. Addressing these unavoidable challenges requires the highest horizontal resolution possible to adequately capture future climate features over the targeted U.S. islands. This work focuses on the development of a multiple-Regional Refined Mesh (M-RRM) configuration in a GPU environment. The new M-RRM capability is tested using different horizontal grid spacing in each mesh to optimize computational efficiency in simulating the highest resolution possible over confined regions (i.e. very small U.S. Islands). The SCREAM-GPU simulations have been built on three DOE Exascale supercomputers, including ALCFs Polaris, OLCF’s Frontier, and NERSC’s Perlmutter. By building SCREAM-GPU with an M-RRM configuration and increased spatial resolution over the Pacific ocean and US islands, the climatological scale data produced with this method will significantly improve our ability to simulate extreme events related to climate changes thorough mid-century in several data poor regions.