整合遙測資訊於山區雨量推估

Integrating Multiple Information for Precipitation Estimating in Mountainous Area

10.6574/JPRS.2013.17(1).2

張斐章(Fi-John Chang)；蔡孟蓉(Meng-Jung Tsai)；江衍銘(Yen-Ming Chiang)；謝明昌(Ming-Chang Shieh)

資料融合 ； 類神經網路 ； 定量降雨預報 ； data integration ； artificial neural network ； quantitative precipitation forecasting

航測及遙測學刊

17卷1期（2013 / 03 / 01）

17 - 30

繁體中文

近年來遙測資訊相繼應用於降雨預報，其主要優點在於有效觀察大範圍降雨在時空之變化。本研究首先蒐集QPESUMS系統之雷達雨量產品、PERSIANN-CCS衛星觀測系統之雨量產品以及地面雨量站觀測紀錄；接著以遺傳演算法融合地面雨量、雷達及衛星影像推估雨量三種資訊；為比較融合雨量之有效性，本研究再以ANFIS架構三種定量降雨預報模式分別為：(1)三種未融合資訊(2)二種融合資訊(3)三種融合資訊，預測未來一小時降雨，由研究結果可知三種融合資訊之定量降雨預報系統為最佳，模式測試部分相關係數、RMSE及MAE分別為0.88、3.88及2.39且RMSE在t+1時刻有28%之改善率。

Simulation of extreme rainfall-runoff events is the key issue for flood mitigation. The accuracy of flood forecasting driven by models is usually dependent on whether the upstream precipitation information is sufficient or not. In the past, such information was provided by ground measurements. The development of remotely sensed technology enables researchers to realize the spatial distribution of rainfall. Remote sensing data provide more useful information than ground measurements. The GA was applied to merging different precipitation products through various input combinations. Finally, the ANFIS was conducted to build three quantitative precipitation forecast models by feeding different input combinations, which are1) three precipitation products, 2) merged precipitation generated by QPESUMS and ground measurements, and 3) merged precipitation generated by PERSIANN-CCS, QPESUMS and ground measurements, respectively. The improvement rate of model 3) over model 1) for quantitative precipitation forecast is 28% at t+1.The results show the model merging ground measurements, QPESUMS and PERSIANN-CCS produced the best precision for quantitative precipitation forecast.