Extreme Value Modeling of Precipitation in Case Studies for

Abstract: This paper aims to model extreme rainfall events using 60 years of daily data based on extreme value theory for four cities in China. The purpose is to allow decision makers to make informed decisions and to avoid or at least reduce flood caused damage to life and property. Generalized extreme value distributions are used for fitting monthly and semiannual maxima according to the Block Maxima Approach. Rainfall exceeding an extremely high threshold is modelled by Generalized Pareto distributions.The thresholds are selected based on the analysis of three methods, Hill plot, mean excess plot and Standardized Precipitation Index. Finally, we estimate the parameters for both models and calculate return levels for five different return periods. Statistical tests for stationarity, KPSS and Man-Kendall tests support the study. The results show that GPD has a better fitting performance than GEV. Further, we can determine how often a flood occur in a certain city and during which season, rainy or dry season. For example, for Nanjing it can be neglected that a flood occurs during dry seasons. But for other cities, like Shantou, this might be rare events but with a higher frequency of occurrence