Abstract. We develop a hierarchical, multilevel Bayesian model for reducing uncertainties in local (at-site) and regional (ungauged or short data sites) flood frequency analysis. This model is applied to the annual maximum streamflow of 17 gauged sites in the Huaihe River basin, China. A Generalized Extreme Value (GEV) distribution is considered for each site, and its location and scale parameters depend on the site’s drainage area. We assume the hyper-parameters come from Non-informative (independent, uniform) prior distribution and sample values from posterior distribution by the MCMC method using Gibbs sampling. For comparison, the ordinary GEV fitting by Maximum Likelihood Estimate (MLE) and index flood method fitted by L-moments are also applied. The local simulation results show that for most sites the 95 % credible interval simulated by the Hierarchical Bayesian model are narrower than the at site GEV outputs thus reducing uncertainty. By comparison, the homogeneity assumption of the index flood method often leads to large deviations from the empirical flood frequency curve. Cross validated flood quantiles and associated uncertainty intervals are also derived. These results show that the proposed model can better estimate the flood quantiles and their uncertainty than the index flood method.
Comment on Local and regional flood frequency analysis based on hierarchical Bayesian model: application to annual maximum streamflow for the Huaihe River basin - by Wu et al