Regional Flood Frequency Analysis by Hybrid Cluster Analysis and L-moments

Recent studies show that cluster analysis methods are one of the most useful techniques of regionalization of watersheds for regional flood frequency analysis. In this study combination of Ward and K-means clustering algorithms is used for regionalization of gauging stations in Sefidrood and Aras watersheds in order to use advantages and decrease influences of disadvantages of two main categories of cluster analysis methods. Also effects of selection and usage of some geographical, physiographic and meteorological attributes and their combinations on homogeneity of regions formed by cluster analysis is studied. Assessment of homogeneity of final regions and performance of regional flood frequency analysis using L-moments show that combination of longitude, latitude and drainage area as attributes used for regionalization of Sefidrood and Aras watersheds may be the best option to form maximum number of homogeneous regions. Furthermore, Wakeby distribution may be used as regional distribution for heterogeneous regions.