雷達定量降水估計在河川洪水預報模式之應用

Radar-based Quantitative Precipitation Estimation for Flood Forecast in Rivers

蔡孟原(Meng-Yuan Tsai) ； 許銘熙(Ming-Hsi Hsu) ； 傅金城(Jin-Cheng Fu) ； 林李耀(Lee-Yaw Lin) ； 王安翔(An-Hsiang Wang)

《農業工程學報》 57卷4期 (2011/12) Pp. 49-68

https://doi.org/10.29974/JTAE.201112.0005

摘要

洪水預報為許多非工程防洪方法之重要基礎，但是台灣的地理環境特殊，增加即時洪水預報的困難性。緊急應變作業人員若缺乏適當的資訊，則無法準確發布警戒並採取必要措施。因此，建立有效率且精確的洪水預報模式，可提供洪水資訊以強化決策作業及降低洪災損失。本研究結合QPESUMS系統網格式的雷達定量降水估計，透過類神經網路分別以回饋式神經網路、自組織特徵映射類神經網路及倒傳遞類神經網路等模式建立降雨－水位預報模式，藉以探討雷達雨量在模式預報1－3小時河川水位之模擬成效，並擇優運用於水理預報模式。經由2008年的鳳凰、辛樂克及薔蜜等三場颱洪事件的模擬結果顯示，利用雷達觀測技術可以掌握颱洪期間流域內中小尺度的降雨時間與空間變化，對於水位站之水位預報模式具有良好的模擬效果，當水位預報值回饋至河川洪水數值模式中，除提升洪水預報的可信度外，亦能提供全河系各斷面之預報水位。

關鍵字

降水估計 ； 河川水位預報 ； 暴洪演算 ； 類神經網路

並列摘要

Flood forecasting is one of the basic tools of the non-structural flood mitigation measures. However, the geographical and environmental conditions in Taiwan are the challenge for task of real-lime flood forecasting. The emergency operations authorities are difficult to issue warning and lake proper response if they are lack of flood forecasting information. An efficient and accurate flood forecasting model is developed for generating the flood information to enhance decision-making and reduce flood damage. This study incorporates the QPESUMS (Quantitative Precipitation Estimation and Segregation Using Multiple Sensors) technique and three ANN (Artificial neural networks) models, including RNN (Recurrent neural networks), SOM (Self-organizing maps, and BPNN (Back-propagation neural network s) for the real-time rainfall and river stage forecasting Three Typhoons events in 2008, Typhoon Fung-Wong, Sinlaku and Jangm were taken as the model testing. The ANNs were used to obtain the 1 -3 hours stage forecast at given river sections. Simulation results have good agreement with the observations because small temporal and spatial variations in rainfall within the river basin are well caught by QPESUMS. By feeding the water stage forecasts back into the physical river flood routing models, the water stage along the entire river can be obtained, and the accuracy of stage to recast is improved.

並列關鍵字

Precipitation estimation ； River stage forecasting ； Flash flood routing ； Artificial neural networks