本研究主要目的在於應用主成分分析(principal componentanalysis)、自組織映射圖(self-organizing map, SOM)網路、指數洪水法(index flood method)與克利金法(kriging)推估台灣地區未設測站地點不同重現期距之頻率降雨量，此方法可分為三個階段：分類階段、指派階段及建立階段。首先，利用主成分分析針對台灣地區127 個雨量站之年最大一日雨量進行分析，總共擷取9 個主要成分，再加上雨量測站之地文因子，以SOM 網路進行群集分析。由SOM 網路所得之二維密度圖可知，全台雨量測站可被分成17 群，亦即台灣地區可劃分成17 個均一性區域。根據未設測站地點之位置位於哪個均一性區域，即可知其屬於哪個群集，再應用指數洪水法與克利金法推估該地點之平均年最大一日雨量，以進行該地點之年最大一日雨
量頻率分析。最後，經由交互驗證之結果顯示，不同重現期距之雨量估計平均誤差皆低於25%，因此建議可將此方法用於未設測站地點之雨量頻率分析。

The main purpose of the study aims to investigate the regional frequency for ungauged sites using principal components analysis,self-organizing map(SOM), index flood method and kriging method.The proposed approach is composed of three steps: classification,assignment, and construction. First, the 127 annual maximum daily rainfall data in Taiwan are available. Principal component analysis is applied to obtain the first nine principal components. Based on the transformed data resulting from principal components analysis and the geographic characters of the gauges, the SOM is used to group the rain gauges into specific clusters. A two-dimensional feature map indicates that the rain gauges can be grouped into 17 clusters. That is, the 17 homogeneous regions for regional frequency analysis can be delineated.One can determine which ungauged site belongs to which region according to the location of the site. The index flood method and kriging method then are applied to investigate the regional frequency for ungauged sites. The rainfall magnitudes with various return periods for 17 gauged sites are estimated. Finally, the results of cross-validation tests show that the values of mean absolute percent error (MAPE) for allregions are less than 25%. Hence, the proposed approach is recommended as an alternative to perform the regional rainfall frequency for ungauged sites.