Regionalization of watersheds by combining of self-organizing feature maps and fuzzy C-Means algorithm

Cluster analysis methods are one of the most efficient approaches of regionalization of watersheds for Regional Flood Frequency Analysis (RFFA). Fuzzy regionalization is a kind of regionalization in which each site of interest may be assigned to more than one region simultaneously. In order to perform regionalization، a variety of cluster analysis algorithms named fuzzy clustering are used that fuzzy c-means algorithm is most wellknown of them. Also Self-Organizing Feature Maps (SOFM) are a special class of Artificial Neural Networks (ANN) that has found several applications in the areas of pattern recognition. Capability of this class of networks in areas of pattern recognition and data clustering using their attributes has make some hydrologist interested in testing ability of these maps for regionalization of watersheds in order to perform regional flood frequency of analysis. In this study self-organizing feature maps have been used to determine initial centroids of clusters in fuzzy c-means algorithm for regionalization of Sefidrood watershed. Results of this study showed that this approach has an acceptable performance in formation of homogeneous regions and providing suitable estimates in regional flood frequency analysis using L-moment algorithm in interested watershed. Furthermore it is observed that fuzzy clustering may provide longest reliable flood estimates. Also based on fuzzy clustering validity measures it’s seemed that two or three regions is appropriate number of regions for regional flood frequency analysis in this watershed.