DEVELOPMENT OF REAL-TIME FLOOD PREDICTION UTILIZING THE RRI MODEL WITH A PARTICLE FILTER

Abstract

 The main objective of this study was the improvement of flood prediction. We combined a particle filter (PF), a type of data assimilation technology, with the RRI Model, and tested this model for effectiveness in flood prediction. The state space of the filtering target was water depth on the slope in the RRI Model, and the water depth was sequentially assimilated to the corresponding observed water level. The method was applied to the heavy rain in the northern Kyushu on July 2017 in Kagetsu River basin. The results of simulation under the condition of the perfect forecasted rainfall, found that the RRI Model with the particle filter achieved a better reproducibility of the water level predicted for the next 3 hours by evaluating likelihood by RMSE. Moreover, as a result of comparing PF with Non-PF using the forecasted rainfall, the prediction of water level was found to improve regardless of the length of lead-time; in particular, the water level predicted for the next 1 - 3 hours was improved 0.24 meters or lower. In conclusion, the present study has demonstrated the effectiveness of the RRI Model with PF in real-time flood prediction.