透過您的圖書館登入
IP:52.90.211.141
  • 學位論文

系集降雨預報驅動降雨-逕流模式於長前置時間水庫入流量預報

Long lead-time reservoir inflow forecasting by adapting a rainfall-runoff model with ensemble precipitation forecasts

指導教授 : 林國峰

摘要


台灣位於西太平洋的颱風路徑要衝,每年平均受到三至四場颱風侵襲。伴隨颱風而來的豐沛雨量,產生龐大逕流量,常導致下游洪水宣洩不及而溢淹,造成重大的災損。此時,水庫操作單位得在維護大壩安全與避免洩洪過量引致下游淹水的雙重考量下,進行颱風時期的水庫防洪運轉操作。因此,本研究提出以系集降雨預報驅動降雨-逕流模式之方法,進行石門水庫於颱風時期的長前置時間1至72小時入庫流量預報。主管機關若能於颱風警報發布時,及早預測颱風時期可能之入庫流量變化與總入庫流量,便能據此擬訂有效的颱風時期水庫操作策略,預估颱風過境期間逐階段合理的放水量。 本研究提出之方法分成三部分:第一部分,選定2008年至2011年間十二場颱風事件,結合美國陸軍工兵團發展之HEC-HMS模式與支援向量機,建立一種新型的混合型降雨-逕流模式,本研究命名為HMS-SVM模式。第二部分,使用國家實驗研究院台灣颱風洪水研究中心提供的2012年至2015年間七場颱風事件之定量降雨系集預報,作為混合型降雨-逕流模式之驅動,預報颱風時期的系集入庫流量。第三部分,採用隨機森林整合系集入庫流量預報。 本研究以2015蘇迪勒颱風為例,證實結合HEC-HMS模式與支援向量機的HMS-SVM模式於系集入庫流量預報,相較於單用HEC-HMS模式,能得到誤差範圍較小之系集分布。此外,以隨機森林整合系集入庫流量預報,能得到比系集平均(算術平均法)更準確的預報結果,當前置時間達48小時以上,預報誤差將大幅降低。本研究提出之方法,能於蘇迪勒颱風警發布當下即時提供準確的未來3天入庫流量預報,且累積入流量誤差僅4.03%。操作單位可據此擬訂有效的颱風來臨前預先調節性放水策略以及颱風過境期間防洪運轉策略,在「得到更多防洪空間但增加日後缺水風險」和「忍受洪災危險為保障日後用水無慮」雙重目標考量下取得權衡。

並列摘要


Taiwan is located on the main track of western Pacific typhoons, and approximately three to four typhoons hit Taiwan per year. Typhoons accompanied by heavy rainfall often results in a huge amount of runoff, which causes downstream floods and induces great disasters. Meanwhile, the reservoir operator should assess flood control operations carefully. The dam security and downstream residents are both taken into consideration. In this case, prerelease for real time flood operation is important. Accurate reservoir inflow forecasting with enough lead time helps the reservoir operator to make operational policies in advance during typhoons. For the aforementioned reasons, a new long lead-time reservoir inflow forecasting approach by means of forcing the rainfall-runoff model with ensemble precipitation forecasts is proposed to yield 1- to 72-h ahead reservoir inflow forecasts of the Shihmen reservoir during typhoons. The structure of this study is composed of three parts: First, a novel rainfall-runoff model which combines the HEC-HMS model with support vector machine is proposed. The proposed rainfall-runoff model is calibrated and validated with twelve typhoons during 2008 to 2011. Second, with ensemble quantitative precipitation forecasts from Taiwan Typhoon and Flood Research Institute being the meteorological forcing, the proposed rainfall-runoff model provides ensemble reservoir inflow forecasts for seven typhoons from 2012 to 2015. Third, the study integrates ensemble reservoir inflow forecasts using random forest. Taking Typhoon Soudelor in 2015 for example, results show that coupling HEC-HMS with SVM provides more reasonable ensemble distribution than using only HEC-HMS. Compare with the ensemble mean, reservoir inflow forecasts from random forest have less uncertainty and the advantage of extra lead time, particularly 48 hours to 72 hours. The proposed model could provide accurate 3-day reservoir inflow forecasts immediately after the Typhoon Soudelor warning was issued. The error of cumulative inflow was only 4.03%. According to the proposed approach, the authority may efficiently operate the reservoir and balance a trade-off between ‘gaining more flood buffer for dam security paying the expense of increasing shortage risk’ and ‘ensuring adequate water resources by enduring the potential of flooding damage’.

參考文獻


1. Ahijevych, D., Pinto, J.O., Williams, J.K., Steiner, M., 2016. Probabilistic Forecasts of Mesoscale Convective System Initiation Using the Random Forest Data Mining Technique. Weather and Forecasting 31, 581–599.
2. Amengual, A., Romero, R., Alonso, S., 2008. Hydrometeorological ensemble simulations of flood events over a small basin of Majorca Island, Spain. Quarterly Journal of the Royal Meteorological Society 134, 1221–1242.
3. Anctil, F., Michel, C., Perrin, C., Andréassian, V., 2004. A soil moisture index as an auxiliary ANN input for stream flow forecasting. Journal of Hydrology 286, 155–167.
4. Azmat, M., Laio, F., Poggi, D., 2015. Estimation of Water Resources Availability and Mini-Hydro Productivity in High-Altitude Scarcely-Gauged Watershed. Water Resources Management 29, 5037–5054.
5. Ba, H., Guo, S., Wang, Y., Hong, X., Zhong, Y., Liu, Z., 2017. Improving ANN model performance in runoff forecasting by adding soil moisture input and using data preprocessing techniques. Hydrology Research 49, 744–760.

延伸閱讀