Analyzing streamflow extremes in the upper Ürümqi River with the generalized Pareto distribution

The generalized Pareto distribution (GPD), as one of the most important distributions in statistical theory of extreme value, is often used to explain the probability of extreme events in nature, through setting up a model with the observation points exceeding the threshold. This paper uses GPD model to fit the distributions of the relatively large and small monthly average discharge of Urumqi River in Northwest China and gives the detailed steps of this method. Firstly, the Mean Excess, shape parameter and modified scale parameter plots are applied to determine the thresholds; then the parameters of GPD are estimated by the maximum likelihood method; next the models are diagnosed by the probability and quantile plots; finally the return levels and the corresponding 95 % confidence intervals of the discharge are calculated by the maximum likelihood and profile likelihood methods, respectively. The results show that the return levels of the maximum monthly average discharge with the return periods 10, 25, 50 and 100 years are 35.4, 39.9, 43.2 and 46.3 m3 s−1, respectively, and the return levels of the minimum monthly average discharge with the return periods 10, 25, 50 and 100 years are 0.60, 0.43, 0.30 and 0.18 m3 s−1, respectively. Some comparisons are also made between the generalized extreme value (GEV) and GPD models. The results of these two models are close to each other while the GPD model should be more reliable because it can make use of more information than the GEV model. This paper proposes a complete framework for modeling hydrological data by GPD model.