Parallelization of a next generation large/local-scale flood model and its application to now-casting

在本项目的框架下，开展了洪水模型并行化和实时洪水预报系统的基础研究。首先，建立了Yodogawa河分布降雨-径流/洪水-淹没（DRR/FI）模型。Yodogawa DRR/FI模型的分辨率为250m （427 × 565=241255个节点）。在研究开始的1小时模拟中，即使使用最快的商用CPU Intel Xeon， CPU时间也在5分30秒左右。然而，通过使用Intel Xeon的14核的Open MP线程并行化，该CPU时间减少到1分30秒。之后，通过与SPIRE （Strategic Programs for Innovative Research）的合作，将Yodogawa DRR/FI模型转移到K超级计算机上，将CPU时间降低到几十秒。同时，研制了以JMA再分析雷达降水为输入的Yodogawa实时洪水预报系统样机。

The basic research regarding the parallelization of a flood model and real-time flood forecasting system has been carried out in the framework of this project. First, a Yodogawa river distributed rainfall-runoff/flood-inundation (DRR/FI) model is developed. The resolution of the Yodogawa DRR/FI model is 250m (427 times 565=241255 nodes). The CPU time was around 5 min and 30 sec. for 1 hour simulation in the beginning of the research even if a fastest commercial CPU Intel Xeon is used. However, this CPU time is reduced down to 1 min. and 30 sec. by the Open MP thread parallelization with 14 cores of the Intel Xeon. Afterwards, through the cooperation with the SPIRE (Strategic Programs for Innovative Research), the Yodogawa DRR/FI model is transferred to the K supercomputer, then the CPU time is reduced down to several tens of seconds. On the same time, a prototype of the Yodogawa real time flood forecasting system with the input of 1km JMA reanalysis radar rainfall is developed.