基于四种概率分布模型的广东省极端气温与降水事件概率特征分析

Analyze on the Best-Fit Probability Distribution of Extreme Temperature and Precipitation Events in Guangdong Province based on Four Models

利用广东省25个气象站点1951～2010年气温与降水数据,选取出每年的最高、最低气温和最大日降水量,将三者定义为极端高温、低温和降水事件.采用GEV、Gamma、GL和GP等四种理论分布模型及其对数分布,分别对极端高温、低温和降水事件进行极值拟合,以探究它们的最优分布.结果表明,广东省极端气温与降水事件各个站点拟合的最优分布并不唯一.GEV、Gamma和GL三种分布模型均能较好地对它们拟合,而GP分布及其对数分布的拟合效果并不理想,不适合用于年极值的拟合.除GP分布模型外,其它三种模型的对数分布对极端高温和降水事件的拟合效果均优于原分布.在参数估计方法上,数值积分双权函数法为极端高温和降水事件概率分布拟合的最优估参方法,而极端低温事件概率分布拟合的最优估参方法则为常规矩修正法.

Annual highest temperature, lowest temperature and maximum daily precipitation were defined as the extreme high temperature, low temperature and precipitation events, according to the data of 25 meteorological stations in Guangdong province from 1951 to 2010. In order to find the best-fit distribution, four theoretical distribution models, GEV, Gamma, GL and GP, were applied to the fitting of the three extreme events. Expecially, their logarithmic distributions, LGEV, LGamma, LGL and LGP, were also applied to the armual highest temperature and maximum daily precipitation. The results indicated that the best-fit distributions of the three extreme events in different stations are not unique. As a whole, GEV, Gamma and GL model all are well on their fitting, but GP and its logarithmic distribution are not well, which can conclude that the GP model is not suitable for the fitting of annual extreme value. Except the GP model, the goodness of fit of the logarithmic distributions of other three models is all better than their original distributions. On the whole, numerical integration double weighted function （NIDWF） method is the best parameter extimation methods for the fitting of annual highest temperature and maximum daily precipitation, and moments repair method （MMR） is the best of annual lowest temperature.