1:45 PM - 3:15 PM
[AAS08-P03] Exploring the frontier of meteorology with AI and HPC technologies
★Invited Papers
Keywords:AI technology, High Performance Computing, Meteorology, Tropical Cyclone
The Typhoon Science and Technology Research Center (TRC), Institute for Advanced Science and Technology, Yokohama National University, established Fujitsu - Yokohama National University Typhoon Science and Technology Research Center Collaborative Research Laboratory (FYRL) in November 2022. Our goal in FYRL is to contribute to the realization of a paradigm shift in society that changes typhoons from a threat to a blessing. Fujitsu is a world leader in the computing technology, for example, the developer of the world's top-class supercomputer "Fugaku" with RIKEN. TRC has leading researchers for typhoon in Japan and is conducting a wide range of studies related to typhoons. We aim to create new analysis methods by jointly using AI and HPC technology. Furthermore, we hope to mentor next-generation players who will play an important role in a wide range of fields around the world through this research and development.
As the starting point, we are focusing on the analysis of typhoons and typhoon precursors using the explainable-AI technology developed by Fujitsu, and the optimization, refinement, and efficiency improvement of numerical weather predicting models using cutting-edge HPC technology. Matsuoka et al. (2018) is a good example of applying AI technology to typhoon analysis. Dr. Matsuoka, one of researchers in FYRL, constructed an AI model that can detect the precursors of typhoons using numerical model simulation results as training data. We are challenging to improve the AI model so that it can detect the typhoon precursors at the early development stage by assessing training data and adding explanatory variables. The explainable-AI can give a set of characteristics for the detected typhoon precursors through machine learning, thereby enabling us to analyze the manifestation of the developing typhoon precursors. These findings will be necessary understandings to discuss the difference between typhoon formation and non-formation.
For HPC studies, the global cloud system resolving model, NICAM (Tomita and Satoh 2004; Satoh et al. 2008; Satoh et al. 2014) and the regional cloud system resolving model, CReSS (Tsuboki and Sakakibara 2002; Tsuboki 2007) are our targeting applications; both models are outstanding numerical models that have produced many results in numerical simulations for tropical atmosphere and typhoons. We aim to find the better approaches to the various computer architectures from the conventional massive parallel machine to the next-generation GPU parallel machine, so that make it possible to seamlessly migrate from the traditional architecture to the future architecture without disrupting the current coding environment. The climate model E3SM developed by the US Department of Energy has been re-written by a combination of C++ and Kokkos freezing the traditional Fortran code to gain the best performance on the GPU machine. Such a large turning point should be a matter of great concern for the model development group. That's why we want Fujitsu, a world leader in computing technology, and application developers to work together to establish a path toward the future computing.
As the starting point, we are focusing on the analysis of typhoons and typhoon precursors using the explainable-AI technology developed by Fujitsu, and the optimization, refinement, and efficiency improvement of numerical weather predicting models using cutting-edge HPC technology. Matsuoka et al. (2018) is a good example of applying AI technology to typhoon analysis. Dr. Matsuoka, one of researchers in FYRL, constructed an AI model that can detect the precursors of typhoons using numerical model simulation results as training data. We are challenging to improve the AI model so that it can detect the typhoon precursors at the early development stage by assessing training data and adding explanatory variables. The explainable-AI can give a set of characteristics for the detected typhoon precursors through machine learning, thereby enabling us to analyze the manifestation of the developing typhoon precursors. These findings will be necessary understandings to discuss the difference between typhoon formation and non-formation.
For HPC studies, the global cloud system resolving model, NICAM (Tomita and Satoh 2004; Satoh et al. 2008; Satoh et al. 2014) and the regional cloud system resolving model, CReSS (Tsuboki and Sakakibara 2002; Tsuboki 2007) are our targeting applications; both models are outstanding numerical models that have produced many results in numerical simulations for tropical atmosphere and typhoons. We aim to find the better approaches to the various computer architectures from the conventional massive parallel machine to the next-generation GPU parallel machine, so that make it possible to seamlessly migrate from the traditional architecture to the future architecture without disrupting the current coding environment. The climate model E3SM developed by the US Department of Energy has been re-written by a combination of C++ and Kokkos freezing the traditional Fortran code to gain the best performance on the GPU machine. Such a large turning point should be a matter of great concern for the model development group. That's why we want Fujitsu, a world leader in computing technology, and application developers to work together to establish a path toward the future computing.