Performance Evaluation of the Robust Discordancy Measure in L-moments Approach

Detection of discordant sites in a region is a very important issue in determining the homogeneous region which has the same statistical distribution. One of the powerful measures for this purpose is the discordancy measure in the L-moment approach. This measure, which is based on the mean and covariance of the data of all sites, sometimes cannot identify correctly the discordant sites. In order to improve this problem, the robust discordancy measure can be used based on the minimum covariance determinant estimator, which is less sensitive to outlying data. In order to analyze the regional frequency of droughts in the East Azarbaijan province, the classic and robust discordancy measures were utilized in this study. The results of two measures showed thsuperiorityof therobust discordancy measure against the classic one. In addition, after application of the cluster analysis and the heterogeneity measure, the study area was divided to three sub regions. Finally, the results of the goodness-of-fit measure revealed that the Pearson type 3 was the best regional distribution for all sub-regions.