本研究之目的是針對颱風降水所導致的災害，使用1998~2004年 TRMM(Tropical Rainfall Measuring Mission)衛星上的TMI(TRMM Microwave Imager)微波資料及日本宮古島附近十一個島嶼測站之地面觀測降雨量資料，以統計法建立多頻道線性回歸方程式，估算海上颱風之定量降水。首先，研究步驟是於颱風期間運用Ferraro et al.(1994)所發展之SI(Scattering Index)法及Chen and Li(2000)之TC(Threshold Check)法結合而成的CC(Combination Check)法，進行降雨區辨識，區分衛星觀測值為有雨及無雨區，進而將有雨區的資料分别估算出對流及層狀的降雨强度。研究結果顯示，在衛星觀測值分辨為有雨及無雨的整體成功辨識率，2002~2004年分别為99.4%、100%及100%。在定量驗證方面，反演之降雨强度舆島嶼測站降雨觀測值之間相關係數約為0.74，均方根誤差為3.75mm/hr。此外，對弱降水天氣系統而言，衛星反演值有高估的現象；反之，對强降水系統，衛星反演值則有普遍低估的現象，其主要原因為視場(field of view)内降雨分佈小均所造成。因此，本研究採用TRMM衛星上之VIRS(Visible Infrared Scanner)高解析度紅外線頻道，以監督式分類(Supervised Classification)方法去除降雨分佈小均匀的匹配資料，以降低反演誤差。 本研究結果亦與GPROF(Goddard Profiling Algorithm)2A12物理法近地面降雨反演值以及Chen and Li(2000)於梅雨期中尺度對流系統降雨强度回歸式反演值比较，結果顯示皆優於以上雨種方法，顯見GPROF近地面降雨反演值目前僅適合於全球尺度，對於區域性之降水估算仍有不足之處。同時，發現不同季節時期所建立的反演迴歸式，僅適用於該季節的天氣系統使用。因此，對於颱風及梅雨大氣系統必須各自建立其關係式。為了增加衛星降水反演的時間及空間解析度，本研究未來將增加AQUA衛星的AMSR-E(Advanced Microwave Scanning Radiometer-EOS)及NOAA(National Oceanic and Atmospheric Administration)衛星的AMSU(Advanced Microwave Sounding Unit)微波資料，以利提升衛星降水反演的實用性。

This study is to estimate quantitative precipitation over ocean using the microwave data of TMI on board TRMM and eleven island rain gauge near Japan from 1998 to 2004, utilize these make-ups data has been created multiple channels linear regression equation by statistic method. The purpose is to avoid the calamity induced by typhoon. The procedure is at first to classify the rain or no rain by using the ST (scattering Index) method developed by Ferraro (1994) and IC (Threshold Check) method developed by Chen and Li (2000) to be combined as a new CC (Combination Check) method and then separate rain type from convective and stratiform rain. Thus, to estimate the rainfall rate of convective and stratiform respectively. The result of overall successful classification of rain and no rain from 2002 to 2004 are 99.4%, 100% and 100% respectively. The coefficient of correlation was 0.74 between estimated quantitative rainfall rate and ground truth of rain gauge for oceanic validation. The Root-Mean-Square (RMS) was 3.75 mm/hr. Besides, the satellite's rainfall rates estimated is overestimated to weak observed precipitation system. Oppositely, the rainfall rate of satellite is underestimated to strong observed precipitation system. The principal reason is not uniform of precipitation distribution on the field of view. Therefore, this study was used to utilize the high resolution infrared channel of TRMM/VIRS to classify the data and remove the worse make-ups data, so as to reduce the estimated difference. This study will compare with the retrieval rainfall of 2A12 product of GPROF (Goddard Profiling Algorithm) created by physical method and rainfall estimated by Chen and Li (2000) for Mesoscale Convective System (MCS) during Mei-Yu season. The result of this study is better than the 2A12 and Chen and Li (2000) methods, so the GPROF method is only suitable to global region and insufficient for local region. In addition to the different regression equation created by statistic method at various seasons, the result is only independent to various seasons respectively. Therefore, the correlation of regression must be created on typhoon and MCS season respectively. For improving the resolution of spatial and temporal, the retrieval rainfall rate technique will be promoted to utilize AMSR-E (Advanced Microwave Scanning Radiometer-EOS) of AQUA satellite and AMSU (Advanced Microwave Sounding Unit) data of NOAA (National Oceanic and Atmospheric Administration) satellite in the future.