研究目的為建立一套分析預測颱風事件水庫集水區逕流反應特性之分散式水文模式。研究使用資料來源分為雨量及流量兩部分，臺灣地區之降雨分布資料來自於中央氣象局之自記雨量站，而研究區域（水庫集水區）之雨量站則來自於水利屬及水庫管理局，水庫入流量資料則來自各水庫管理局。 本研究以自行開發之分散式水文模式模擬颱風期間翡翠與石門水庫入流量，利用賀伯(1996)、瑞伯(1998)及納莉(2001)等三場颱風事件進行模式檢定與驗證，並分析北勢溪主流及部分支流在颱風事件下的特性。研究發現颱風事件下模式所估計水庫入流量，對使用不同空間分布特性降雨資料之模擬結果差異有限，但若需要對水庫上游集水區逕流情形進行分析時，則建議採用空間分布降雨資料進行模擬。分析降雨資料時發現集水區雨量站數量會決定總雨量推估值的精確性，雨量站數愈多所計算之總雨量的標準差愈小。目前使用小時解析度之雨量資料，所發展模式對於短時間生成洪峰反應敏感性較差，分析結果顯示使用時間解析度較佳的降雨資料，能改善模式對短時間生成洪峰量的預測能力。 The purpose of this study is to apply a distributed hydrologic model to simulate typhoon floods of reservoir watersheds in Taiwan. Observed precipitation and reservoir inflows are used to drive the model and to calibrate model’s parameters. Island wide rainfall patterns are analyzed from the data of the CWB (Central Weather Bureau) auto-recorded rain gages. Rainfall data and reservoir inflows of studied areas are provided by the WRA (Water Resource Agency) and the reservoir administration agencies. A newly developed distributed hydrologic model is employed to simulated reservoir inflows of Feitsui and Shihmen reservoirs in typhoon events. Three typhoon events, Herb (1996), Zeb (1998), and Nari (2001) are applied for model calibration and validation. Simulated results are further retrieved to analyze the upstream runoff characteristics of the Beishi creek. Three important conclusions are obtained in this study. First, different spatial rainfall patters did not significantly affect the reservoir inflows simulated by the distributed hydrologic model developed in this study. However, the runoff characteristics of upstream sub-basins can only be seen with the input of distributed rainfalls and the simulations of distributed hydrologic model. Second, the more raingages used for estimating basin total rainfalls, the less deviation can be obtained. Finally, the developed model is not sensitive to quick runoff generated by intensive rainfall in short periods due to the current rainfall resolution of one hour. To improve the capability of the model on predicting typhoon floods, it is suggested to use a higher temporal resolution of rainfall data in future studies.