Research On Extreme Value Distribution Of Short-duration Heavy Precipitation In The Yangtze River Basin

The generalized extreme value distribution (GEV) and generalized Pareto distribution (GPD) are used to fit the hour-extreme-precipitation of twelve stations in the Sichuan Basin which is located in the upper reaches of the Yangtze River, with method of parameter estimation of L-moment. Then compare the effects of the two models. It is appropriate to use Hill plot with D*to determine the optimal threshold of GPD, and the selected samples are independent. The probability distribution of hour-extreme-precipitation of each station is all in line with GPD and GEV, but GPD model shows better fitting accuracy than GEV model. The maximum hour-precipitation with a given return period are calculated by the two models for each station, and we find that the estimated results based on GPD are more reliable by the analysis.Based on the hourly precipitation data that from1991to2009(from April to September) over46stations in Hubei, Anhui and Jiangsu provinces, which are in the middle and lower reaches of the Yangtze River, the rain intensity over10mm/h and20mm/h in a year and the spatial distributions of hour-extreme-precipitation with return periods of50and100years are analyzed. Hour-extreme-precipitation of most stations showed an upward trend. Other, hourly precipitation of46stations in these three provinces is all in line with the Gamma distribution in addition to the individual months of the individual stations. For Nanjing, the increase of scale parameter of Gamma distribution of hourly precipitation, which causes the increases of probability of extreme values, has a higher impact than the decrease of shape parameter, which causes the decreases of probability of extreme values. The multi-status Markov chain model is better than the two-status Markov chain model on its simulation of mean number of rain hours during a month spell and the extreme values. The extreme value samples, which are picked out form the precipitation records that simulated by the two models, are both in line with the generalized Pareto distribution. We calculate the hourly extreme precipitation with a return period of50years and100years based on the extreme value samples, and the results shows that extreme values of return period and threshold simulated by the multi-status model is closer to the real observation values, compared with the two-status model. And this proves the high feasibility of hourly extreme precipitation that simulated by the multi-status Markov chain model.