[A-AS05] Weather, Climate, and Environmental Science Studies using High-Performance Computing

[AAS05-07] Efforts in the Field of Weather and Climate towards Next-Generation Supercomputing Infrastructures around 2030

★Invited Papers

*Chihiro Kodama^1,2, Sachiho A. Adachi², Masuo Nakano¹, Rui Tanaka³, Aihisa KAMIJO^1,4, John Harrington-Tsunogai¹, Tatsuya Seiki¹, Seiya Nishizawa², Shin-ichiro Shima^6,2, Shigenori Otsuka², Arata Amemiya², Yousuke Sato⁷, Takuya Kawabata⁵ (1.Japan Agency for Marine-Earth Science and Technology, 2.RIKEN Center for Computational Science, 3.Department of Earth and Planetary Science, Graduate School of Science, The University of Tokyo, 4.Atmosphere and Ocean Research Institute, The University of Tokyo, 5.Meteorilogical Research Institute, 6.Graduate School of Information Science, University of Hyogo, 7.Faculty of Science, Hokkaido University)

Keywords:next-generation supercomputing infrastructures, weather and climate

We will introduce our efforts in the field of weather and climate in the Feasibility Studies on next-generation supercomputing infrastructures (system research) (commonly known as Fugaku-NEXT FS), for which RIKEN had been selected as one of the representative institutions since 2022 summer. Major efforts include creating benchmark sets for the evaluation of the architectures, compiling the expected breakthrough on the next-generation supercomputer around 2030 and necessary spec and issues to achieve the breakthrough, examing modification of model components for GPU, and examing the latest trends in AI weather models and domain-specific languages. As a foundation for these efforts, we had formed a new community for researchers from major modeling groups in Japan to exchange information and discuss related issues.
As a candidate for the next-generation supercomputing infrastructures, an architecture equipped with GPUs in addition to CPUs has been presented, in anticipation of the era of "AI for Science." These GPUs will have high performance in low-precision martix operation, that will accelerate AI training and inference processes. However, in order for conventional physics-based models to use full performance of the architecthre, partial replacement with AI surrogate models will also be an option in addition to making them GPU compatible. From broader perspective, it is expected that research combining simulation and AI will be in the spotlight. We need to discuss on what can be done and what should be done in the next-generation supercomputing infrastructures while keeping an eye on the long-term advancement of the weather and climate field.