Bivariate Flood Frequency Analysis of Krishna River using Copula Theory

Copula theory has received attention in the field of hydrology. Copula function is used toderive the multivariate distribution of variables. Using copula have an advantage thatmarginal distribution of independent variables can be of any form and the variables can becorrelated. Flood frequency analysis (FFA) help us to quantify the risk associated with flood.In this study copula theory is used for flood frequency analysis of Krishna River in India.Four stations (i.e., Kurundwad, Huvinhedigi, K. Agrharam, and Wadenpally) was selected onKrishna river basin. Peak over threshold method (POT-method) was used to select theindependent events for analysis. Using methodology provided in Flood Estimation Handbook(FEH), Volume and Duration data is extracted from the selected events. The jointdependence structure of flood variables is derived, for frequency analysis of Peak Flow (P),Flood Volume (V), and Flood Duration (D). Best fit marginal distributions of these floodvariables are determined using five parametric (Normal, Exponential, Extreme value,Lognormal, and Gamma distribution) and one non-parametric (Kernel distribution)probability distributions. Kolmogorov-Smirnov Anderson-Darling test was performed tofind out the best fit distribution for flood variables. For modelling of the joint dependencestructure of peak flow-volume (P-V), flood volume-duration (V-D), peak flow-duration (P-D), five Archimedean family of copulas, namely Independence, Clayton, Frank, Gumbel-Hougaard, and Ali-Mikhail-Haq Copulas are evaluated. Goodness-of-fit (GOF) test usingRosenblatts probability integral transformation was used to find out the best fitted copula forbivariate models. Clayton copula has been identified as the best fitted copula for all thebivariate models considered. Clayton copula function is used to obtain conditional returnperiods, Conditional return periods of flood characteristics can be useful for risk based designof water resource projects.