Frequency analyses are a basis for designing discharge estimations. Univariate flood frequency analyses are
usually applied in hydrological practice. Hydrological processes are multivariate, however multivariate
analyses are needed. Copula function can be used for multivariate modelling. Classical univariate flood
frequency analyses are a precondition for the copula analyses. Flood frequency analyses were made on the
annual maximum series data from gauging station Litija on the Sava River. Three copulas from the
Archimedean family were used; parameters were estimated with method of moments (based on the Kendall
correlation coefficient). Some joint return periods were calculated and compared with the univarite return
periods. Differences between return periods were not negligible. In the case of a flood event in 1990, which
was the largest in the observed period, TAND was 92 years and TOR was 17 years. Univariate return periods lay between these two values. Statistical and graphical performance measures were used to choose the best fit
copula function. Gumbel-Hougaard copula gave better results than Clayton and Frank copulas.
|