Regionalization of watersheds based on the concept of rough set

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.