台灣擁有豐沛的降雨，但也有致災的可能性，因此高解析度淹水模擬及預報對防災及水資源管理是重要的，但淹水模擬及預報的準確性有很多誤差來源，包括地形資料的準確度、模型的設定、以及降雨資料的不確定性。 本研究針對降雨資料的不確定性，將不同類型及時間解析度的降雨資料應用至3Di 水動力模式，包括理想情境下的降雨、定量降水估計、和MAPLE (McGill Algorithm for Precipitation nowcasting using Lagrangian Extrapolation)的即時預報結果，以此檢視3Di模式對降雨資料的敏感度，並選擇宜蘭的冬山河流域及高雄市區，模擬2024年發生的多個颱風、東北季風造成的淹水事件，以此檢視不同集水區和降雨型態的差異性。過去臺灣用在水文分析的降雨資料大多都是一小時一筆累積降水資料，但當強降雨集中在更短時間例如半小時或十分鐘時，根據實驗結果，可能造成淹水情形更加嚴重，因此更高時間解析度的降雨資料是有存在的必要性。在即時預報方面，MAPLE的即時預報在每小時滾動更新的條件下，淹水歷線趨勢上和使用定量降水估計的模擬相似，但整體淹水面積低估。 此外，淹水模式時常因為真實資料不足或零散而難以驗證，但2024年的凱米颱風在台灣造成多處淹水而有相對充足的資料，因此本研究利用淹水感測器、衛星資料及Emergency Management Information Cloud (EMIC) 作為模式驗證的參考，也顯示了不同資料的優缺點及結果上的差異。根據校驗結果，3Di水動力模式的模擬結果和淹水感測器相比，準確度在不同事件下分布在0.6至0.8，和衛星資料相比，準確度分布在0.75和0.95之間。 ;The accuracy of flood simulations depends on multiple factors, with precipitation data uncertainty being a key influence. Therefore, selecting an appropriate combination of meteorological data and flood models is essential for effective flood management. This study presents a framework that emphasizes the importance of high-temporal-resolution precipitation data by integrating five types of precipitation datasets: designed rainfall, two quantitative precipitation estimation (QPE) products, and the nowcast system MAPLE (McGill Algorithm for Precipitation Nowcasting using Lagrangian Extrapolation). These datasets were used as inputs for the hydrodynamic model 3Di. An idealized experiment revealed that flooding may be underestimated when heavy rainfall occurs within a short duration, particularly less than one hour, if only hourly rainfall data is used, underscoring the necessity of finer temporal resolution in flood forecasting. Six heavy rainfall events from 2024 in northeastern and southwestern Taiwan were analyzed, leveraging improved data availability from these events to validate the flood model. The results demonstrate that higher temporal resolution enables earlier flood detection, which is critical for early warning systems. Additionally, when rainfall intensity increases, the discrepancy between flood extents generated by different datasets becomes more pronounced. Furthermore, MAPLE provides reliable short-term forecasts within one hour, though any forecast errors may be amplified when incorporated into the flood model. These findings highlight the importance of precise precipitation data in flood simulations and the potential challenges associated with forecast uncertainty.
