Typhoon Rainfall and Inundation Forecasting Using Support Vector Machines

• Main Authors: Bing-Chen Jhong, 鍾秉宸
• Other Authors: Gwo-Fong Lin
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/45998276546414966796

博士 === 國立臺灣大學 === 土木工程學研究所 === 103 === Heavy rainfall caused by typhoons frequently result in inundation which frequently leads to loss of human life and property. Typhoon rainfall and inundation forecasting are very important issues in early warning systems. In this thesis, effective rainfall and inundation forecasting models based on the support vector machine (SVM) are proposed. However, the traditional models were established using the trial and error method, which requires much time. Moreover, the conventional SVM-based models are used to produce point forecasts rather than regional forecasts. In this thesis, effective approaches are established to construct forecasting models in rainfall and inundation forecasting. Two parts are conducted herein to demonstrate the superiority of the proposed models. In the first part of the thesis, a typhoon rainfall forecasting model is proposed to yield 1- to 6-h ahead forecasts of hourly rainfall. First, an input optimization step integrating multi-objective genetic algorithm with SVM is developed to identify the optimal input combinations. Second, based on the forecasted rainfall of each station, the spatial characteristics of the rainfall process are obtained by spatial interpolation. An actual application to the Tsengwen River basin is conducted to demonstrate the advantage of the proposed model. The results show that the proposed model effectively improves the forecasting performance and decreases the negative impact of increasing forecast lead time. In the second part of the thesis, an effective forecasting model is proposed to yield 1- to 6-h lead time inundation maps for early warning system during typhoons. First, 7-Eleven stores are determined as inundation points for point forecasting. Second, a point forecasting module on the basis of the SVM is developed to yield 1- to 6-h lead time inundation forecasts at each inundation point. Finally, according to the point forecasting results and geographic information, the point forecasts are expanded to the spatial forecasts using the proposed spatial expansion module. An application to Chiayi City, Taiwan, is conducted to demonstrate the superiority of the proposed forecasting model. The results indicate that the proposed model effectively improves the forecasting performance and decreases the negative impact of increasing forecast lead time. Moreover, the proposed model is capable of providing accurate inundation maps for 1- to 6-h lead times. In conclusion, the proposed modeling technique is recommended as an alternative to the conventional model to support the disaster warning systems.