圖書典藏及影音數位平台

極端事件的雨量推估乃是一具有挑戰性且受重視的議題，而集水區雨量更受氣候變遷、地形、氣流及降雨分布等因素影響，使其推估之精確性往往不高，而導致接續之水文模擬或流域水源管理成效不彰。目前許多的國際性研究計畫皆將焦點著重於極端事件之降雨對洪水相關災情之影響及其對人類之威脅。在台灣，以往主要的雨量觀測來自於地面測站，然地面觀測之空間代表性極低無法呈現雨量在空間中之變異；而近幾年上線應用之雷達觀測系統QPESUMS即能提供較具體之雨量分布趨勢，對於集水區之雨量觀測可有效進一步掌握；此外，就同一降雨系統而言，地面雨量站與雷達則可分別觀測降雨之不同特徵，雷達觀測雖可主動探測降雨系統內部之交互影響，但對其中之水氣交換作用，以及冰雹或雨滴等不同之水氣狀態是否形成降雨則仍有所不及，地面雨量站雖僅能被動蒐集降雨系統落下之降水量，但點觀測量最為精確；因地面雨量站與雷達在觀測降雨上雖各有其限制之處，但亦可分別觀測不同之降雨特徵，故若能有效率統整地面雨量紀錄及雷達觀測雨量，提供高精確度之雨量變化趨勢，且進一步建構集水區降雨-逕流機制以有效預測流域多時刻水文量之變化，並連結資料庫，擷取即時雨量、水位或流量以及QPESUMS雨量資料以建置整合平台，供相關單位後續經營與管理參考。

Estimation of extreme rainfall has been a long-term challenge in Taiwan. The accuracy of rainfall estimation is generally not high enough due to various reasons, such as complex terrains, atmospheric interactions, climate change, or heterogeneous spatial distribution of extreme rainfalls. Traditionally, most rainfall data have been measured by gauges. However, point measurements have difficulty in providing spatial information. Recently, grid-based radar rainfall products become more common and show high potentials in storm rainfall estimation. This 2-year project has taken the Shihmen Reservoir (northern Taiwan), the Dechi Reservoir (mid Taiwan) and the Tsenwen Reservoir (southern Taiwan) into consideration. In the first year of this project, we applied radar rainfall data (Quantitative Precipitation Estimation and Segregation Using Multiple Sensor, QPESUMS) to real-time rainfall forecasting using ANFIS and BPNN neural networks for estimating 1 to 5 hour-ahead inflow at the Dechi and Tsenwen Reservoirs, respectively. The results show that the BPNN is superior to the ANFIS in catching the inflow trend based on real-time rainfall forecasting while the ANFIS performs better at peak values of the inflow forecasting. In addition, we also investigated the online system of automatic real-time rainfall estimation and water level forecasting for the Shihmen Reservoir through storm rainfall events and provided real-time online service by periodically deliver information through e-mail to reservoir authorities. In the second year, we constructed the online system of automatic real-time rainfall estimation with an incorporation of Google Map for these three reservoirs. The online system has been validated well not only by continually dedicating to the accuracy and efficiency of forecasting but also by updating model parameters and maintaining the website in order to keep the system running well. During Typhoons Talim and Saola as well as the storm of June 2012, we demonstrated that the constructed model could firmly download real-time data for facilitating the forecasting process. The forecasting results during T+1 and T+5 were accurate with R2 values ranging from 0.99 to 0.7. Moreover, the fluctuations of the model outputs were obviously smoothed through the updates of the model parameters. A training conference for the staffs of operational units was held in September 2013, which aimed to provide reservoir staffs/managers with online forecasts of rainfall and inflow for an improvement in reservoir operation management. The manuscripts that summarized the results of this study have been submitted to international SCI journals for an exposure of the study achievements to international societies.