Bivariate regional drought frequency analysis using multivariate approaches: a case study in southwestern Iran

Abstract

Bivariate approaches in Regional Frequency Analysis (RFA) address two issues: first, to evaluate the homogeneity of regions, and second, to estimate the joint return periods. This study was conducted to investigate the joint return period of a severe historical drought in southwestern Iran. Fifty-nine rain gauges were first clustered into three, four, and five regions using the fuzzy c-means clustering (FCM) algorithm. Then bivariate discordancy and homogeneity tests were applied to adjust the initial clusters. The results showed that only in the case of three clusters were all the regions homogeneous. Therefore, it can be inferred that combining clustering analysis and discordancy test is insufficient to form homogeneous regions. Finally, the joint return period, by choosing Generalized Logistic and Wakeby as marginal distributions and Clayton as a copula, was estimated for all the sites in the three regions. Since no three-parameter distribution function fitted well to the variable severity, the bivariate homogeneity index does not necessarily attest to region homogeneity regarding the marginal distribution functions. It is also deduced that sites with higher mean annual precipiataion (MAP) and, correspondingly, higher elevation are more likely to experience shorter return periods of same drought events, in contrast to sites with lower MAP or lower elevation. HIGHLIGHTS The estimation of return periods for extreme events relies on the availability of sufficient data.; Generalized Logistic and Wakeby are chosen as marginal distributions, and Clayton is selected as a copula.; Sites with higher mean annual precipitation (MAP) and higher elevation are found to be more likely to experience shorter return periods of the same drought events.;

Keywords

• copula
• clustering analysis
• drought
• homogeneity test
• l-comoments
• reginalization