藉由Kidder et al. (2005) 提出之TRaP 方法，將衛星反演之降雨分佈平移，可迅速估算熱帶氣旋未來24小時內可能帶來的累積降雨，然而實際應用上TRaP並不適用於地形較複雜的台灣。Liu et al. (2011) 修正TRaP方法，納入地形因素，並根據測站降雨資料重新分配颱風所帶來之降雨量，以改善TRaP 方法，此法稱為I-TRaP。 前人研究發現颱風強度隨時間的改變、預報路徑的誤差程度和衛星反演雨量的準確性皆會造成I-TRaP預報的誤差。為納入這些不確定性因素，期望藉由多個系集成員預報，以彌補單一成員預報的不足。本研究使用多衛星反演降雨產品GSMaP進行I-TRaP颱風降雨潛勢預報(本文中稱為Ensemble I-TRaP)，選用GSMaP原因在於其高時間解析度特性可以提供多筆降雨資料。另外，為考慮颱風降雨型態距颱風中心的遠近而有不同，因此將其分為「內圈」環流雨和「外圈」地形雨，本研究另建立一組降雨預報模式(稱為Ensemble I-TRaP B)，先分別估算兩形態的降雨再依權重結合，最後再比較此兩種方法與I-TRaP的24小時累積降雨預報結果，另外本研究也探討此模式於短時間(6、12、18小時)和長時間(24、36、48小時)累積降雨的預報能力。 使用Ensemble I-TRaP和Ensemble I-TRaP B估算2001-2012年間77個颱風個案之24小時累積降雨，再和測站觀測資料比較，此兩種系集模式都可將相關係數由0.53提升至0.62，均方根誤差由81.68 mm分別降低至64.05 mm和63.76 mm，減少24%的均方根誤差，改善程度顯著。而在不同預報時間的降雨趨勢方面，此方法對於短時間的降雨較難掌握，但在長時間的累積降雨則有較好的表現。兩種方法的整體統計結果顯示加入系集成員可以提升I-TRaP預報能力，且加入降雨型態可再提升其定量降雨的精準度。 ;Weather satellite observations are used widely for the quantitative precipitation forecast (QPF) of typhoon rainfall because they can provide relevant atmospheric parameters over both the ocean and land. Microwave observations by satellites have become the main data in forecasting tropical rainfall potential (TRaP) in a 24-h period (Kidder et al., 2005). The improved Tropical Rainfall Potential (I-TRaP) technique presented by Liu et al. in 2011 is a useful method for typhoon quantitative precipitation estimation and a powerful tool for rain-band monitoring before the typhoon makes landfall in Taiwan. However, the method only provides single prediction which may pose a difficulty when using single sensor or time segment data. To smooth the random error made by single forecast and quantify the uncertainties in prediction, this study seeks to adopt ensemble forecasts to help to provide more reliable predictions. In other words, the goal of this study is to construct a new ensemble I-TRaP technique for typhoon rainfall potential. Besides that, to consider about different rainfall types within a typhoon, the rain-band of a typhoon is separated into two parts: inner rain-band (circulation affected rainfall) and outer rain-band (terrain affected rainfall). This study again constructed another ensemble potential model called the Ensemble I-TRaP B model (“B” stands for “bi-types”), to consider the two rainfall types. Then, the results including that the performances of both ensemble methods in 24-h QPE and the ability of forecasts for different time periods are investigated. There are 77 typhoons from 2001 to 2012 used for long term statistics. Comparing to the I-TRaP model, the ensemble technique (i.e. Ensemble I-TRaP model), and the new model which additionally considering the two rainfall types (i.e. Ensemble I-TRaP B models) can both promote the correlation coefficient from 0.53 to 0.62, and decrease root-mean-square from 81.68 mm to 64.05 mm and to 63.76 mm respectively. It shows that this ensemble technique is useful for improving the rainfall pattern estimation in short accumulated periods, and moreover, it did better forecasting in long periods with a higher correlation coefficient. The results suggest that using the ensemble technique may improve I-TRaP, and considering rainfall types can again promote the rainfall amount prediction.
