圖書典藏及影音數位平台

本署95年建立之淹水預警系統，係應用淹水潛勢圖與水文資料進行淹水預警，近兩年因各地區淹水潛勢圖已陸續完成，為擴大應用範圍，並持續研發與維護，以提升淹水預警系統之準確度及實用性，本計畫規劃2年期程完成淹水預警系統之更新工作，第一年完成淹水預警方法評估，第二年完成淹水預警系統更新。

In the first year, this project had built back-propagation neural network (BPNN) and the adaptive network-based fuzzy inference system (ANFIS) to forecast the one- to six-hour-ahead flood depths of many security spots in Yilan, Yunlin and Pingtung Counties. The results of the first year project showed that the BPNN models perform superior to the ANFIS model. The correct percentages of forecasting flood depths were above 70% at all security spots of these three Counties. The rainfall hyetograph comparative method was to investigate the linear relationship of rainfall characteristics from10 different rainfall hyetographs and to evaluate the linear model of rainfall hyetograph factors. For the effect of the hydrological and geomorphologic factors, different tidal waves, surface features and number of pumps were designed to investigate how variation of the inundation area and to build the curves of the inundation areas and volumes versus tidal waves and the number of pumps. In the second year, this project extended and applied the methodologies proposed in the first year to other Counties in Taiwan. In the rain hyetograph forecasting, Typhoon Rainfall Climatology Model (TRCM) was integrated into BNNs to forecast six-hour-ahead rainfall. The flood depth forecasting models have been established in all other Counties. Again, the results showed that the BPNN models perform superior to the ANFIS model. The correct percentages of forecasting flood depths were above 70% and 90% at the security spots that the flood inundation data were generated by SINOTOPO and two-dimensional non-inertial overland flow simulation model, respectively. In the effect of hydrological factors, four rainfall characteristics of rainfall hyetograph were investigated in Yilan and Yunlin. Both rainfall characteristics and TRCM are integrated into the BPNN model to calculate the increment of inundation area. Tidal waves, drainage, other hydrological and geomorphologic factors were also considered in this project. The various potential flood inundation maps were simulated under different kinds of scenarios that consisted of different tide levels (or drainage percentage) and different amount of designed rainfall. Then, find the formulas of factor-rainfall-inundation relations to calculate inundation areas in different tide levels (or drainage percentage). The results showed that the effects of subsidence and the number of pumps were not significant. Hence, these two factors can be ignored. This project integrated the aforementioned models into the on-line flood inundation warning system to forecast the flood depths of all security spots of all Counties in Taiwan, to calculate the amount of inundated area under different rainfall hyetograph, hydrological or topographical factors in Yilan, Yunlin and Pingtung Counties.