臺灣地處西太平洋亞熱帶地區,每年平均受到3.6次颱風侵襲,颱風災害損失平均每年約為新臺幣170億元,其中颱風豪雨造成河岸地區與低窪地區的洪水及淹水災害損失最為嚴重,颱風過境期間河川洪水位受到暴雨影響水位暴漲,加劇排水之困難度甚至溢流城鄉地區造成淹水。倘若能夠準確預測洪水位的變化,提供給災害應變單位作為決策支援,就能夠增加災害發生前的應變時間,將災害損失降到最低。 近年來臺灣加強在觀測降雨的研發與應用方面,隨著中央氣象局與水利署合作建置完成都卜勒雷達觀測網,引進美國劇烈風暴實驗室的多重觀測工具之定量降水估計與分類技術(Quantitative Precipitation Estimation and Segregation Using Multiple Sensors, QPESUMS)後,降雨的即時估計技術已日趨成熟。目前,定量降水估計(Quantitative Precipitation Estimation, QPE)技術已可以提供一高解析度的降雨資訊,故本研究為整合氣象雷達降雨、地面觀測雨量及河川洪水位等資訊,利用類神經網路建立簡單快速的洪水位預報模式,最後結合一維變量流動力波模式,建立一套淡水河系之河川洪水預報模式,以做為發布洪水警報、淹水疏散及防救災應變措施之參考。 本研究以2008年的鳳凰(Fung-Wong)、辛樂克(Sinlaku)與薔蜜 (Jangm)等3場颱風事件進行模式測試。由模擬結果顯示,因為雷達定量降水估計提供一高解析度降雨資訊,可以有效提升流域內水位站洪水位預報的準確性,再藉由精準的定點水位預報結果,使的預報水位剖線更能趨近於觀測值,克服預報水位偏離問題,故本研究的成果確實可在颱風期間提供更為合理及準確的河川洪水資訊。
Taiwan, located on subtropical zone of west Pacific, encounters typhoons around 3.6 times annually in average. The typhoons disaster caused about 17 billions NT dollars loss per year. The most part of loss in typhoons is the flooding in low-lying areas by the torrential rain. Due to the torrential rain, the sudden rise of flood stage will not only destroy water conservancy facilities, but also threaten the life and property of the residents near the riverbank. If the variation trend of flood stage precisely forecast in advance, the flood management agency will take proper actions for response and mitigation before the disaster happens. In recent years, Taiwan made an effort on improving the technique of real-time rainfall observations. After the completion and applications of the Doppler Radar Networks developed from Center Weather Bureau and Water Resources Agency with the cooperation of U.S. National Severe Storm Laboratory, QPESUMS (Quantitative Precipitation Estimation and Segregation Using Multiple Sensors) makes the technique of real-time spatial rainfall estimation better. At present, QPE (Quantitative Precipitation Estimation) is capable of providing the information of high-resolution rainfall. In order to integrate the information of weather radar rainfall, rainfall-gauges data and flood stage, this research uses ANN (Artificial Neural Networks) to establish a set of simple and fast flood stage forecasting and combine 1-D gradually-varied unsteady flow with feedback routing. Then a set of river flood forecasting model was developed to provide accurate and detailed flood information for Tanshui Basin in typhoon period. The above will be taken as reference for announcing flood alert, evacuation and actions for response and mitigation before the disaster happens. In present research, Typhoons Fung-Wong, Sinlaku and Jangm in 2008 are taken as the model testing. The result shows that the application of QPE makes the stage forecasting closer to observations in river flood routing model and provides reasonable and accurate results for river flood mitigation.