應用雷達降雨資料與支撐向量機於洪水預報結果之修正研究

STUDY OF APPLYING RADAR RAINFALL AND SUPPORT VECTOR MACHINE TO CORRECT FLOOD FORCASTS

王靖斈(Jing-Xue Wang) ； 林遠見(Yuan-Chien Lin) ； 石棟鑫(Dong-Sin Shih) ； 吳瑞賢(Ray-Shyan Wu)

《中國土木水利工程學刊》 33卷1期 (2021/03) Pp. 45-54

https://doi.org/10.6652/JoCICHE.202103_33(1).0005

摘要

由於氣象雷達可提供空間降雨資訊，並且較地面雨量站更具有空間變異性，因此本研究利用中央氣象局QPESUMS劇烈天氣監測系統之降雨資料，整合HEC-HMS與WASH123D模式建立一套洪水推估系統，並透過支撐向量機（SVM）對推估值進行修正。結果顯示雷達降雨預測（QPF）所建立之洪水預報雖然能反應出水位的變化，但仍存在誤差，但經支撐向量機修正後，推估水位將可更貼近觀測值；整體來說，使用雷達降雨推估較雨量站降雨資料具有更佳的模擬結果，可提高相關係數約0.07，減少均方根誤差0.1m；若透過支撐向量機修正後，可再提升相關係數約0.08，減少均方根誤差0.09m，並改善峰值誤差約0.2m。

關鍵字

水文模式 ； WASH123D ； 雷達降雨 ； 洪水預報

並列摘要

The weather surveillance radar (WSR) is used to locate precipitation which is more flexible in obtaining spatial and temporal variability than the Rain gauge. Thus, the study employs the rainfall data derived from Quantitative Precipitation Estimation and Segregation Using Multiple Sensor (QPESUMS). By using the data, a flooding forecasting system is established by integrating the HEC-HMS with WASH123D. The Support Vector Machine (SVM) then modifies the errors in the estimation. The result shows that the water level variation can be estimated by the flooding forecasting system established by the Quantitative Precipitation Forecasting (QPF), however, the lack of precision remains. Thus, the SVM modifies the errors in order to improve the accuracy in terms of the water level observation. Overall, using the Weather Surveillance Radar (WSR) obtains more accurate simulation results than the Rain gauge. The correlation coefficient increases about 0.07 and reduces the root-mean-square error around 0.1 m. However, after the modification of SVM, the correlation coefficient increases about 0.08, and reduces the root-mean-square error around 0.09 m and peak-value-error about 0.2 m.

並列關鍵字

hydrological model ； WASH123D ； radar rainfall ； flood forecasting