Author(s): Takehiko Ito; Koyo Ota; Yasuo Nihei

Linked Author(s): Takehiko Ito, Yasuo Nihei

Keywords: Hazardous water level; Lv4 rainfall index; Flood forecast; Ensemble forecast rainfall; MEPS

Abstract: In Japan, heavy rains and floods caused enormous damage frequently. It is necessary to develop and implement flood forecasting system in order to protect people’s living and assets in the basin. As one of components to do flood forecasting, it is important to set a rainfall index that predict the occurrence of flooding directly for early evacuation of the residents. In this study, we present a new rainfall index equivalent to the hazardous water level, called as Level 4 (Lv4) rainfall index which is obtained from observed rainfall and water level during past floods for each 109 river basin systems in Japan. The Lv4 rainfall index is the cumulative rainfall over 24 hours and averaged over the catchment area, which means that if the 24-h rainfall exceeds the Lv4 rainfall index in each river basin system, flooding may occur somewhere in the basin. Then, we propose a new flood forecast system to predict the occurrence of flooding directly and probabilistically using the Lv4 rainfall index and ensemble forecast rainfall data. Here, we used the forecast rainfall consisting of 21 ensemble members by Meso-scale Ensemble Prediction System (MEPS) operated by the Japan Meteorological Agency. As a result of the constructed prediction system, flood hazard probability is calculated every 6 hours based on the total number of ensemble members that exceed the Lv4 rainfall index. We applied the present system to the heavy rainfall events in July 2020 in Kyushu region Japan and evaluated the validity of the Lv4 rainfall index and the accuracy of the present system by comparing the prediction results of the flood hazard probability with the actual river water level. The results indicated that the rates of concordance, oversight and missing of the present system were 27.8, 0.76 and 5.46%, respectively when we select the optimal Lv4 rainfall index. The median of lead time of all 15 flood events at 20 river basins was 19 hours. These results suggest the fundamental accuracy and performance of the present system. In particular, enormous flooding have actually occurred in the Chikugo and the Kuma Rivers, and it has been confirmed that the flood hazard probability increases over time until the actual water level exceeds the hazardous water level at these rivers. This prediction system depends only on the setting of the Lv4 rainfall index and the accuracy of the ensemble forecast rainfall data. In this study, it was confirmed that the accuracy of the forecast rainfall data is generally high, so the optimal setting of the Lv4 rainfall index is required to improve the accuracy more.

DOI: https://doi.org/10.3850/IAHR-39WC2521711920221429

Year: 2022