颱風期間即時淹水地圖預報之研究

Effective Real-Time Forecasting of Inundation Maps During Typhoons

王志煌(Jhih-Huang Wang) ； 鍾秉宸(Bing-Chen Jhong) ； 林國峰(Gwo-Fong Lin)

《農業工程學報》 62卷4期 (2016/12) Pp. 69-86

https://doi.org/10.29974/JTAE.201612_62(4).0005

摘要

颱風期間之降雨常造成淹水災害而導致人命及經濟財產的損失，對於淹水預警及減災上，有效及準確的淹水預報是必須的。本研究提出一有效即時淹水預報模式產生未來1到3小時之淹水地圖，提出之模式分為三個階段：決定預報淹水點、點預報及空間延展。首先，選擇7-11便利商店位置作為預報淹水點，再對於各淹水預報點分別以支援向量機(support vector machine, SVM)建置點預報模式產生預報淹水度，另外，本研究也比較以SVM與倒傳遞類神經網路(back-propagation nerural network, BPN)為基礎的淹水預報模式之預報表現能力。最後依據點預報模式之預報淹水深度及地理資訊，並以SVM建置空間預報模式。本研究以嘉義市作為FLO-2D之淹水模擬，並證明提出之模式能有效改善預報的能力。結果顯示在預報未來1至3小時淹水深度之RMSE及CC值表現上，SVM有54%到100%的淹水點預報表現優於BPN，且提出模式淹水模式有76%到100%的淹水點預報表現優於SVM。表示所提出之淹水預報模式可有效改善預報的能力。此外，提出淹水預報模式可準確提供未來1到3小時之淹水地圖。總結而言，本研究提出之模式對於淹水預警及減災方面有相當大的幫助。

關鍵字

淹水預報 ； 淹水地圖 ； 支援向量機 ； 倒傳遞類神經網路

並列摘要

The inundation resulting from typhoon rainfall frequently causes loss of life and serious economic damage. Efficient and accurate forecasts of inundation depth are necessary for inundation warning and mitigation. In this study, an effective real-time forecasting approach is proposed to yield 1- to 3-h lead-time inundation maps during typhoon periods. The proposed model is composed of three steps: determination of inundation points, point forecasting and spatial expansion. First, 7-Eleven stores are determined as inundation points for the point forecasting. Second, the support vector machine (SVM) is used as the computational method to develop a point forecasting model to yield inundation forecasts for each inundation point. In addition, the SVM-based model is compared with an existing model based on the back-propagation network (BPN) to show the improvement in point forecasting performance. Third, based on the forecasted depths and the geographic information, the point forecasts are expanded to the spatial forecasts using the proposed spatial expansion model. An application to Chiayi City is conducted to demonstrate the superiority of the proposed model. The results show that the percentage of the number of inundation points for which the SVM-based model performs better than the BPN-based model is from 50 % to 100% for 1- to 3-h lead-time forecasting. The percentage of the number of inundation points for which proposed model performs better than SVM-based mode is from 76 % to 100% for 1- to 3-h lead-time forecasting. The proposed model effectively improves the forecasting performance. Moreover, the proposed model can provide accurate inundation maps for 1- to 3-h lead-times. In conclusion, the proposed forecasting model is expected to be useful to support inundation warning and mitigation.

並列關鍵字

Inundation forecasting ； Inundation map ； Support vector machine ； Backpropagation network