Ultra-High Resolution Numerical Weather Prediction with a Large Domain Using the K Computer

In Japan, torrential heavy rain induced debris flow in Izu Ohshima on October 2013 and in Hiroshima on August 2014. To mitigate these kind of heavy rain related disaster, more accurate weather prediction is necessary. This study aims to examine whether an ultra-high resolution numerical weather prediction (NWP) model with a large domain is able to produce more accurate forecast with K super computer, which is the fastest computer in Japan and the fifth fastest computer in the world on August 2016. The five factors of NWP that are (1) grid spacing (5, 2 km and 500, 250 m), (2) turbulence closure model (Mellor-Yamada-Nakanishi-Niino [MYNN] level 2.5 and 3, and Deardorff [DD]), (3) model domain (LARGE: 1100 * 1600 km, SMALL: 200 * 200 m), (4) the lateral boundary conditions, and (5) terrain data were investigated in the above two heavy rain events.

In the Izu Oshima case, the turbulence closure model greatly influenced to the position of the front and associated rain band. In the experiments with DD, the front was simulated at the similar position to observation, while in the experiments with MYNN the front was shifted to the northwest of the island. The models with DD tended to simulate the more intense downdraft in the lee of the front than the models with MYNN. The back-building formations were seen in the experiments with DD, but the finer grid spacing models showed more distinct vertical flows along the front and associated cold pools. These differences in the vertical motions and the cold pool formations affected the position of the rain band. In Hiroshima case, the grid spacings influences to the position of the rain band.

Sensitivity experiments on the domain size and lateral boundary conditions showed the importance to have large domain and include of cloud microphysical quantities in the lateral boundary condition in Izu Ohshima case. However, the domain size and the lateral boundary conditions were little impact on the Hiroshima case.

In both cases, the 250-m grid model with the finest terrain representation showed the best performance in all experiments. These results demonstrate that the very high resolution NWP model with the large domain has the ability to better predict the meso scale rain band and associated precipitation.