Identification of homogeneous rainfall regions using a genetic algorithm involving multi-criteria decision making techniques

Abstract

Abstract The detection of appropriate homogeneous regions is an important step in regional frequency analysis with the determination of homogeneity depending to a great extent on the type of method used in grouping. So, the study considers a genetic-algorithm-based clustering method to identify homogeneous precipitation regions for 39 gauge stations of the north-eastern region of India. The performance evaluation is done using six cluster validation measures. Further, considering all the six indices together, selection for the optimum cluster is modelled as a multi-criteria decision making (MCDM) problem. Three MCDM methods, namely TOPSIS, WASPAS and VIKOR, are applied to obtain ranked clusters which are then subjected to a heterogeneity test using the L-moments approach. The results suggested the stations to be grouped into three homogeneous regions. Comparison with the k-means method indicated relatively better performance for genetic-algorithm-based clustering. Finally, an L-moment ratio diagram and goodness-of-fit measures were conducted to select regional frequency distributions for the identified homogeneous regions.