In this study, an algorithm inspired by some concepts of the rough set theory is proposed for regionalization of watersheds. The algorithm includes a clustering step and a classification step and utilizes canonical correlation analysis and cluster analysis methods. The proposed algorithm can use flood-related features to form feature vectors for gauged watersheds and also, it can be applied to an area including both gauged and ungauged watersheds. The results of applying the method to the basin Sefidrud in Iran show that when all the watersheds in the study area are considered as gauged, the proposed algorithm clearly provides more suitable results in comparison with a common cluster analysis method in terms of the number of watersheds assigned to the homogeneous regions. Also, by performing a leave-one-out procedure to consider each watershed as ungauged in one turn, the ability of the proposed algorithm in simultaneous regionalization of gauged and ungauged watersheds was examined. According to the results, for the number of regions 2, 3, 4, and 5, the proposed algorithm outperforms the common clustering algorithm used for regionalization in terms of the number of watersheds assigned to homogeneous regions.

在这项研究中，提出了一种受粗糙集理论的某些概念启发的算法，用于分水岭的区域化。该算法包括聚类步骤和分类步骤，并利用规范相关分析和聚类分析方法。所提出的算法可以使用与洪水有关的特征来形成测距流域的特征向量，并且可以应用于包括测距流域和未测流域的区域。将这种方法应用于伊朗塞夫德鲁德盆地的结果表明，当研究区域中的所有集水区都经过测量时，与常规聚类分析方法相比，在集水区数量上，所提出的算法显然提供了更合适的结果分配给同质区域。也，通过执行“留一法”程序，将每个集水区视为一圈未注水，对拟议算法在测量和未加注水集水区同时分区中的能力进行了检验。根据结果​​，就区域2、3、4和5的数量而言，该算法在分配给同质区域的分水岭数量上胜过用于区域化的通用聚类算法。