Japan Geoscience Union Meeting 2023

Presentation information

[J] Oral

A (Atmospheric and Hydrospheric Sciences ) » A-AS Atmospheric Sciences, Meteorology & Atmospheric Environment

[A-AS08] Weather, Climate, and Environmental Science Studies using High-Performance Computing

Sun. May 21, 2023 3:30 PM - 5:00 PM 304 (International Conference Hall, Makuhari Messe)

convener:Hisashi Yashiro(National Institute for Environmental Studies), Tomoki Miyakawa(Atmosphere and Ocean Research Institute, The University of Tokyo), Chihiro Kodama(Japan Agency for Marine-Earth Science and Technology), Shigenori Otsuka(RIKEN Center for Computational Science), Chairperson:Chihiro Kodama(Japan Agency for Marine-Earth Science and Technology)


4:15 PM - 4:30 PM

[AAS08-09] A few hundred-meter global simulations

*Shuhei Matsugishi1, Tomoki Ohno2, Junshi Ito3, Masaki Satoh1, Yoshiyuki Kajikawa4, Yuta Kawai5, Masuo Nakano6, Hiroshi G. Takahashi7, Daisuke Takasuka1, Hirofumi Tomita5, Hisashi Yashiro8 (1.Atmosphere and Ocean Research Institute, The University of Tokyo, 2.Meteorological Research Institute, Japan Meteorological Agency, 3.Tohoku University, 4.Research Center for Urban Safety and Security, Kobe University, 5.RIKEN Center for Computational Science, 6.JAMSTEC Japan Agency for Marine-Earth Science and Technology, 7.Department of Geography, Tokyo Metropolitan University, 8.National Institute for Environmental Studies)


Keywords:Cloud resolving model, Fugaku, Turbulence scheme

The representation of clouds and convection is essential in weather and climate simulations. Extreme weather events such as torrential rains, so-called Senjo-Kosuitai, and storms represented by intense typhoons are caused by cumulonimbus clouds in which upward motion is concentrated in a narrow region. Therefore, an accurate representation of the upward motion or vertical wind in the numerical model is essential for reproducibility. In recent years, numerical models have become increasingly higher resolution, and global nonhydrostatic models that cover the globe with a kilometer (km)-scale mesh have been developed by various organizations worldwide. These models are highly expected as next-generation models that can explicitly calculate cumulus clouds (Slingo et al. 2022). However, cumulus clouds' representation is insufficient even in the global km-scale model. To understand the limitation of global km-scale models and for further precise simulations of various phenomena, a comparison with global simulations with a sub-km-scale resolution, which more accurately represents the vertical motion of the atmosphere, is needed.
In this study, we used the Supercomputer Fugaku, Japan's latest flagship machine, to simulate the global atmosphere with a horizontal mesh of several hundred meters by a global cloud-resolving model NICAM (Satoh et al. 2014). We compare the differences in mean states of simulations and representation of convection across the globe between several resolutions. The resolution between km to sub-km is called a gray zone (Honnert et al. 2020) in the turbulence parameterization used in atmospheric simulations, which is the boundary between both Reynolds Averaged Navier-Stokes (RANS) models and Large Eddy Simulation (LES) models. Therefore, we conducted experiments using MYNN (Nakanishi and Nino 2004), a RANS scheme, and Smagorinsky scheme, an LES scheme, respectively and compared them. A global simulation with a horizontal resolution of 220 m was possible by taking full advantage of Fugaku's performance. The 220m simulation was performed using 81920 nodes on Fugaku.
We investigate the dependence of the large-scale state on resolution. The zonal mean humidity and precipitation distribution do not change significantly. A decrease in the lower cloud is noticeable for increasing resolution. The distribution of precipitation is noteworthy: very weak rainfall, such as drizzle with less than 1 mm/hr, is reduced. However, when we focus on rainfall distribution over 1 mm/hr, weak rainfall increases with higher resolution, and the rainy area expands. Coarse-grained rainfall distribution is smoother in the sub-km model than in the km model.
In comparing the results of the RANS and LES schemes, the lower clouds cover and the representation of lower clouds also differed significantly: the LES scheme showed more low cloud cover than the RANS scheme. The simulation with the LES scheme also showed more variation in the cloud-top and cloud-bottom heights of the low clouds than in the RANS scheme.