本研究為驗證並改善洪水水位預警模式之預報結果,分別將 2019 年的 4 場降雨事件之 QPF 雷達降雨預報資料以 3 種不同方式計算 QPF 之降雨量,並輸入至模式中,以評估其優劣。結果顯示,在單峰 形降雨事件中以預測平均計算 QPF 降雨量輸入模式模擬水位可以有 效提升只使用單一時刻之 QPF 降雨量模擬之準確性,並且降低洪峰 誤差,其他降雨型態下對水位預測的精準度提升不明顯,但可以減少 單一時刻極值造成的影響。此外評估即時校正(Realtime Correct,RT)對 模式質量守恆之影響性,模式經 RT 即時校正後河道總水量改變幅度 約為+0.02%~0.26%,總流量的改變量約為 16%~42%,由於率定曲線 本身亦存在許多不確定性,此改變量應屬可接受的範圍。故應用 RT 對模式預測水位進行即時校正具有一定的可信度,且對模式的平衡不 會造成過大的衝擊。;In order to verify and improve the forecast results of the flood warning system,the QPF radar rainfall forecast data of four rainfall event in 2019 were used to estimate the rainfall in three different ways and was input to the model to evaluate their pros and cons.The results show that in a singlepeak rainfall event, using average of forecast to estimate the QPF rainfall in model can effectively improve the accuracy of the forecasting that only uses single prediction value in model, and reduces the flood peak error.The increase in accuracy of water level prediction of other rainfall types is not significant, but it can reduce the impact of extreme values at a single prediction value. In addition, the impact of Realtime Correct (RT) on the conservation of model quality is evaluated. After the model is instantly corrected by RT, the amount of change in the total water volume of the river channel is about +0.02%~0.26%, and the difference of total flow is about 16%~42%. Therefore, the application of RT to real-time correction of the model′s predicted water level has a certain degree of credibility.