- 期刊
台灣北部地區最大三十分鐘降雨強度之分析及預測
Analysis and Prediction of the Maximum Thirty-Minute Rainfall Intensity in Northern Taiwan
摘要
最大三十分鐘降雨強度(I30)在預測土壤沖蝕率時，扮演著非常重要的角色。本文蒐集台灣北部地區主要氣象站自1961至1990年共三十年之雨量資料，參照Wischmeier and Simth (1978)之定義篩選出有效降雨事件，分析最大三十分鐘降雨強度與平均降雨強度、降雨延時以及累積雨量之關係。結果顯示，台灣北部地區有效降雨事件之平均降雨強度大於3(mm/hr)，平均最大三十分鐘降雨強度則大於10(mm/hr)，而最大三十分鐘降雨強度與平均降雨強度之關係較佳，相關係數介於0.702至0.789之間，累積雨量次之，與降雨延時之相關性最差，呈現負的相關性；然而，以平均降雨強為回歸變數所建立之回歸公式，不若以累積雨量與降雨延時所建立之公式，有較佳的推估能力。此外，本研究運用倒傳遞類神經網路，建構一個最大三十分鐘降雨強度推估模式；結果亦顯示，具有三層架構之倒傳遞類神經網路有較佳之推估能力，其推估之誤差均方根(RMSE)達4.53。
並列摘要
The maximum thirty-minute rainfall intensity (I30) plays an important role in predicting soil erosion rate. In this study, thirty years of rainfall data from 1961 to 1990, were collected from three weather stations in northern Taiwan, and effective rainfall events were selected based on the definition by Wischmeier and Simth (1978). Then the relationships among the I30, duration of rainfall, cumulative rainfall amount and mean rainfall intensity were analyzed. It was found that the mean rainfall intensity of effective rainfall events was greater than 3 mm/hr, the average of I30 was greater than 10 mm/hr, and there was a good relationship between I30 and mean rainfall intensity, the correlation coefficient ranged from 0.702 to 0.789. However, the equation established to predict the I30 by using cumulative rainfall amount and mean rainfall intensity as the regression variables was found to be better than that by using mean rainfall intensity. Besides, a model was established to predict the I30 by using Back-Propagation Artificial Neural Network (BPN). And the result showed that BPN with three layers has a better performance with root mean square error (RMSE) of 4.53 for predicting I30.