本研究之目的是應用區域化(regionalization)方法建立台灣地區未設測站地點之設計雨型。首先，選用全台灣地區174 個雨量測站，以主成分分析(principal component analysis)針對所有測站之設計雨型進行分析，總共擷取累積解釋百分比達90%以上之5 個主要成分， 再加上雨量測站之三項地文因子作為自組織映射圖(self-organizing map, SOM)網路之輸入項，以SOM 網路進行群集分析。由SOM 網路之二維密度圖可知，174 個雨量測站可分為19 個群
集，亦即可將台灣地區劃分成19 個均一性區域。最後，由未設測站地點之所在位置，可根據均一性區域圖找出其所屬之特定群集，再應用反距離權重法(inverse distance weighting)之理論建立未設測站地點之設計雨型，並與最近測站法及SOM 基礎分析法之估計結果做比
較。經由交互驗證之結果顯示，本研究所提出未設測站地點設計雨型建立之方法，其估計值具有較佳之準確性及穩定性。

The purpose of the paper aims to construct design hyetographs for ungauged sites in Taiwan using regionalization approach. First,principal components analysis (PCA) is applied to obtain the principal components of the design hyetographs for the 174 rain gauges in Taiwan.It is found that the first five principal components explain over 90% of the information. Based on the transformed data resulting from PCA and the
three geographic characters of the gauges, the self-organizing map (SOM)
is used to group the rain gauges into specific clusters. According to the
two-dimensional density map resulting from SOM, it is observed that the
rain gauges can be grouped into 19 clusters. That is, the 19 homogeneous regions for these clusters can be delineated. An ungauged site is then assigned to a proper homogeneous region according to the homogeneous regions map. Finally, the design hyetographs for ungauged sites is constructed by using inverse distance weighting method.For comparison with the proposed approach, the closest-site approach and the SOM-based approach have also been executed in this paper.Moreover, cross-validation tests have been performed to examine the stability and the accuracy of these approaches. The results provide evidence to support that the proposed approach is a robust and efficient way for improving the accuracy of design hyetographs for ungauged sites.