Several hydrological events, such as floods and droughts, are described through a number of characteristics. In general, these characteristics are stochastically dependent and should be jointly considered in a multivariate framework. Therefore, univariate frequency analysis (FA) provides only a partial assessment of the true probabilities of occurrence. Regional FA aims to transfer information from gauged sites to target sites. In the present presentation we propose a procedure for regional flood FA in a multivariate framework. The proposed procedure consists in providing a multivariate version of the index-flood model. The model is flexible for designers where several scenarios associated to the same risk are given. In addition, the model is general where the univariate version, for each variable, is a special case corresponding to an extreme scenario. In order to evaluate the performance of the model, a simulation study is carried out. The obtained results show that the univariate results are incomplete since they ignore the dependence structure of the events. The univariate estimated values are given by the multivariate procedure directly and with equivalent accuracy. It is also shown that the behaviour of the multivariate procedure is similar to the univariate one, i.e. the model performs better when the regional homogeneity is high and the impacts of the region size and the record length at the gauged sites are not significant.

Keywords: Regional frequency analysis; Floods; Multivariate analysis; Index-flood model

Biography: Taha Ouarda is professor for statistical hydrometeorology at the INRS. He has a PhD from Colorado State University in civil engineering. He is the Chairman of the Canada research Chair on the estimation of hydrometeorological variables. His main interest is in the different aspect of hydrological frequency analysis as well as modeling of climate changes impacts. He published more than 140 papers in different scientific journals.