颱風是臺灣水資源的重要來源之一。颱風降雨量隨著其移動路徑、本身特性及降雨地點地域因子的影響，有關因子對颱風降雨之共同影響為非線性。雖然有許多描述降雨機制的物理公式已被發展來推估即時降雨，但成果都不盡理想。
　　本研究使用倒傳遞網路建立颱風降雨量模擬模式與預測模式，將複雜的颱風降雨機制映射在類神經網路架構內，據以推估颱風降雨量。本研究中以颱風資料及測站氣象資料做為類神經網路的輸入變量以模擬降雨量。在模擬模式基礎上，預測模式先將輸入變量與雨量做相關係數分析，對於相關係數較接近0的變量以不同輸入變量組合選取模式中較佳之模式做為類神經網路的輸入變量以預測降雨量。
　　本研究在模擬模式及預測模式中以輸入變量時間上不同分別建立三種模式，模式I輸入變量為t時刻；模式II輸入變量時間為t+1時刻；模式III則為模式I及模式II的所有輸入變量一併納入模擬。另一方面，模擬模式再以變量中有無考慮前1小時雨量細分為模式A及模式B。在預測模式中依據領先時間不同而建置領先1時段、領先3時段及領先N時段。在領先3時段中再建置兩種模式，方法1以領先1時段方法預測未來第1小時雨量，再將預測雨量帶回原本模式做為前1小時雨量持續計算致第3小時雨量，方法2是以t時刻輸入變量直接預測t+3時刻雨量，由預測結果可知以方法1有較好的預測結果。在領先N時段中，N是指颱風事件中海上颱風警報第一報至解除海上颱風警報之間的時間，模式中由於N通常離第1小時相差甚遠，故此模式在輸入變量中未考慮前1小時雨量。在模擬模式中模式A及模式B及預測模式中領先1、3及N時段皆是以模式III有較佳的模擬結果。

Typhoon is one of the most important source of water resources of Taiwan. Typhoon brought abundant rainfall to its covering area. It is generally regarded moving path, characteristics of typhoon, local terrain and meteorological factors of a gage station as major factors affect rainfall depth. There are many approaches were developed to simulate the complicated non-linear process of typhoon rainfall. However, the accuracy of simulation may be improved. It takes typhoon data and weather station data as input of BPN model.

In this study, a typhoon rainfall simulation model and a prediction model were established using the back-propagation artificial neural network model (BPN) to estimate the typhoon rainfall. On the basis of the simulation model, the prediction model firstly analyzes the correlation coefficient (CC) between the input variables and rainfall, For the variables with a CC close to 0, the better model selected from different combinations of input variables is used as the input of the BPN.

In the simulation model and the prediction model, three models are established according to the different time of input variables, mode I input variable time is t; mode II input variable time is t+1; For mode III, all input variables of mode I and mode II are included in the simulation. On the other hand, the variables including and excluding the rainfall before one hour were subdivided into mode A and mode B. In the prediction model, lead time 1 hour, lead time 3 hour and lead time N hour are constructed according to different lead times, among them two different models were constructed in the lead 3 time, method 1 uses the lead time 1 hour method to predict the rainfall of the first hour in the future, and then takes the predicted rainfall back to the original model as the continuous calculation of the rainfall of the third hour in the previous hour, method 2 directly predicted the rainfall at time t+3 with the input variable at time t, the prediction results show that method 1 has better prediction results. In the lead time N hour, N refers to the time between the first report of the typhoon warning at sea and the lifting of the typhoon warning at sea in a typhoon event. The results in the simulation mode that mode A and mode B has better simulation results in mode III. In the prediction model, the mode III at the lead time 1 hour and lead time 3 hour had a better prediction result, while the mode III at the lead time N hour had a better prediction result.

(1)王時鼎(1983)，「臺灣颱風降雨特性之研究」，行政院國家科學委員會防災科技研究報告。
(2)王時鼎、顏清連、陳泰然、謝信良(1986)，「臺灣颱風降雨特性及其預報研究(三)」，行政院國家科學委員會防災科技研究報告，研究報告74-51號，第57～90頁。
(3)王如意、謝龍生、王鵬瑞(1998)，「八掌溪流域降雨-逕流預報模式串聯應用之研究」，臺灣水利報，第46卷，第1期，P12-31。
(4)王珍貴(2007)，「應用類神經網路模式模擬颱風降雨量」，國立成功大學水利及海洋工程學系研究所碩士論文。
(5)李清勝、饒心誠、謝宜桓(2018)，「康芮颱風(2013)侵台期間臺灣西南部強降雨事件之研究」，大氣科學 46.2: 第149～171頁。
(6)林柏承(2000)，「應用類神經網路於颱風降雨量的推估」，成功大學水利及海洋工程學系研究所碩士論文。
(7)陳正斌(2004)，「應用模糊理論於颱風降雨量之推估」，成功大學水利及海洋工程學系研究所碩士論文。
(8)陳昶憲、鍾侑達、梁家瑋、王晉倫(2006)，「颱風早期降雨預測」，中華水土保持學報，37(2)，第201～208頁。
(9)曹嘉宏、洪景山(2011)，「系集模式颱風定量降水：個案研究」，建國百年天氣分析預報與地震測報研討會，第278～282頁。
(10)黃明宏(2009)，「格網型類神經網路颱風降雨量模擬模式」，國立成功大學水利及海洋工程學系研究所碩士論文。
(11)鄭連焜、陳秀娟(1998)，「颱風模式法推估PMP之進一步研究」，中興工程顧問股份有限公司水利工程部專案研究報告。
(12)蔡孝忠(2000)，「颱風定量降雨氣候預報模式改進之研究」，國立臺灣大學土木工程研究所碩士論文。
(13)羅竣文(2010)，「應用類神經網路於颱風降雨量即時預報之研究」，國立臺灣大學土木工程學系研究所碩士論文。
(14)Chang, C. P., Yeh, T. C., & Chen, J. M. (1993). Effects of terrain on the surface structure of typhoons over Taiwan. Monthly weather review, 121(3), 734-752.
(15)Chiang, Y. M., Chang, L. C., & Chang, F. J. (2004). Comparison of static-feedforward and dynamic-feedback neural networks for rainfall–runoff modeling. Journal of hydrology, 290(3-4), 297-311.
(16)Chien, F. C., & Kuo, H. C. (2011). On the extreme rainfall of Typhoon Morakot (2009). Journal of Geophysical Research: Atmospheres, 116(D5).
(17)Hong, J. S., Fong, C. T., Hsiao, L. F., Yu, Y. C., & Tzeng, C. Y. (2015). Ensemble typhoon quantitative precipitation forecasts model in Taiwan. Weather and Forecasting, 30(1), 217-237.
(18)Joseph, J., & Ratheesh, T. K. (2013). Rainfall prediction using data mining techniques. International Journal of Computer Applications, 83(8).
(19)Ko, C. M., Jeong, Y. Y., Lee, Y. M., & Kim, B. S. (2020). The Development of a Quantitative Precipitation Forecast Correction Technique Based on Machine Learning for Hydrological Applications. Atmosphere, 11(1), 111.
(20)Lin, Y. L., Chen, S. H., & Liu, L. (2016). Orographic influence on basic flow and cyclone circulation and their impacts on track deflection of an idealized tropical cyclone. Journal of the Atmospheric Sciences, 73(10), 3951-3974.
(21)Ou, S. H., Liau, J. M., Hsu, T. W., & Tzang, S. Y. (2002). Simulating typhoon waves by SWAN wave model in coastal waters of Taiwan. Ocean Engineering, 29(8), 947-971.
(22)Yu, P. S., Yang, T. C., Chen, S. Y., Kuo, C. M., & Tseng, H. W. (2017). Comparison of random forests and support vector machine for real-time radar-derived rainfall forecasting. Journal of Hydrology, 552, 92-104.