How can I improve the accuracy of weather forecasts for extreme and high impact? One of the feasible methods is the first-step simulation based on the weather system and atmospheric physicochemical conditions (i.e., numerical simulation of the atmosphere). However, simulating the global atmospheric system is not an easy task.
But that hasn’t stopped scientists from working on it. On April 1, the reporter learned from the University of Science and Technology of China that the research group of An Hong, a professor at the School of Computer Science and Technology, and the research group of Zhao Chun, a professor at the School of Earth and Space Sciences, jointly realized the atmospheric physics-chemistry fully coupled numerical simulation of the global 3-kilometer space resolution for up to 7 days on China’s new generation of Shenwei supercomputer, and this process took less than 9 hours. Recently, the study was published online in Science Bulletin under the title “Establishing a non-hydrostatic global atmospheric modeling system at 3-km horizontal resolution with aerosol feedbacks on the Sunway supercomputer of China”Science Bulletin) on.
Schematic diagram of the efficiency of massive parallel simulations of 3 km spatial resolution of the global atmosphere Source: Thesis
In less than 9 hours, atmospheric physics-chemistry fully coupled numerical simulations of global 3 km spatial resolution for up to the next 7 days are undoubtedly boons for detailed and accurate forecasts of the future. This achievement also begins with the challenge of simulating the global atmospheric system.
Global high-resolution numerical simulations of the evolution of atmospheric composition are still difficult challenges to break through. There are two reasons for this: first, the atmospheric numerical model needs to have a finer spatio-temporal resolution – the current state-of-the-art Earth system numerical simulation device has a maximum global model horizontal resolution of 10 to 25 kilometers, which is difficult to improve the accuracy; second, the atmospheric numerical model also needs to include many complex physical and chemical processes – the previous weather forecast only focused on elements such as moisture, and the coupling of sulfur dioxide, PM2.5 and other elements in the air will increase the complexity of the simulation by geometric levels. Therefore, to build such an atmospheric numerical model, it will bring about an exponential increase in the amount of simulated computation and the need for extremely large-scale data reading and writing.
Conceptual illustration of atmospheric physics-chemistry fully coupled numerical patterns at high spatial resolution Source: Thesis
Breakthroughs in supercomputing system performance and a range of supercomputing application technologies provide a turnaround for this challenge. According to the paper, the research work is based on a new generation of domestic Shenwei supercomputing platform, and developed the global high-resolution non-static equilibrium atmospheric numerical model iAMAS, which includes the evolution process of atmospheric composition, and fills the gap in large-scale data reading and writing speed, parallel computing efficiency, scale scalability, runtime effectiveness and other aspects.
An Hong introduced to China Science Daily that according to the characteristics of the new generation of Shenwei supercomputer communication system and file system, the joint research team made full use of the read and write bandwidth provided by the system and released the file system performance by reconstructing the data reading and writing strategy, and solved the reading and writing bottlenecks that seriously affected the scalability of iAMAS simulation; they also comprehensively optimized the parallel computing methods at all levels from the system to the chip, gave full play to the computing characteristics of the new generation of Shenwei heterogeneous many-core processor chip architecture, and realized the process-level, The parallelism of thread-level and data-level greatly improves the parallel computing efficiency of atmospheric physics-chemistry fully coupled numerical simulations.
In the end, when the research team expanded the scale of the numerical simulation test to nearly 40 million processor cores (about 600,000 core groups), the parallel efficiency remained at 76.2%, and under the scenario of frequent large-scale data reading and writing, the speed of simulating the weather process for 20 hours per hour was reached, and the efficient numerical simulation test of 3 km atmospheric physics-chemical coupling in the world was realized, reaching the international leading level.
“That is to say, at present, we can complete the fine forecast of the 7-day weather process in 8.54 hours, and we have achieved good timeliness.” An Hong told China Science Daily: “If there is no new generation of Shenwei supercomputer such a large system, without a series of supercomputing technology innovations and breakthroughs, this simulation may not be completed in a month, it will not play a predictive role.” ”
An Hong told reporters that the team is making further efforts to control the more accurate and detailed medium- and long-term (more than 10 days) weather process and air quality forecast time within a few hours, laying the foundation for long-term high-resolution climate simulations.
Coping with a warming global climate and frequent extreme and high-impact weather is one of the major challenges facing the scientific community in the 21st century. In recent years, the frequent occurrence of extreme and high-impact weather and air pollution events, such as heavy rain, blizzards, high temperatures, droughts, thunderstorms, typhoons, hail, cold waves, haze, sandstorms, etc., not only have a serious impact on people’s production and life, but also cause immeasurable losses of life and property.
An Hong said that this study initially reveals the global high-resolution atmospheric physico-chemical coupling simulation, which can significantly improve the accuracy of extreme and high-impact weather event forecasting, and show the significance of high spatial resolution and coupled chemical feedback effect for numerical weather and air quality forecasting, which has great application prospects.
“In the future, we plan to further optimize this atmospheric model, improve its computational efficiency and stability, and couple it into China’s high-resolution Earth system model, so that long-term high-resolution climate simulation experiments can be carried out to explore the laws of climate change and its impact on human beings, and serve the country’s ‘double carbon’ strategy.” In addition, zhao chun said, the global high-resolution atmospheric physico-chemical coupling simulation is also suitable for studying the evolution of planetary atmospheric composition. The team will also develop a high-resolution atmospheric model for planets based on this research to serve the country’s deep space exploration program and related scientific research.
The research has been co-funded by the National Natural Science Foundation of China, the National Key Research and Development Program, the “Double First-Class Project” Research Fund of the University of Science and Technology of China, and the Strategic Key Research Program of the Chinese Academy of Sciences. The results were completed with the close cooperation of relevant researchers from the National Meteorological Administration, the National Supercomputing Wuxi Center, Peking University and Tsinghua University. Qingdao Marine Science and Technology Pilot National Laboratory, Supercomputing Center of University of Science and Technology of China, National Supercomputing Wuxi Center, National Supercomputing Jinan Center, National Supercomputing Guangzhou Center and other units have provided supercomputing resource support for the project research. (Source: China Science Daily Zhao Guangli)
Related paper information:https://doi.org/10.1016/j.scib.2022.03.009