长江流域夏季极端降水非平稳极值建模及其趋势分析

Non-Stationary Modeling and Trend Analysis of Extreme Precipitation in Summer over Yangtze River Basin

使用国家气象信息中心位于长江流域共124个有效站点的降水资料,引入随时间变化的动态阈值,建立超门限广义帕累托分布模型,利用非平稳极值模型分析长江流域近57年(1961~2017年)夏季极端降水的时空特征及其趋势分析,并给出了重现期趋势变化情况。该分析对水源管理、水利建设以及城市设计等都具有重要的意义。1) 相对于平稳极值模型,非平稳极值模型能够更好地刻画极端降水;2) 给出了非平稳极值模型下极值模型参数分布情况,结果显示:长江流域尺度参数的趋势存在局部性,大部分地区趋势为正,形状参数呈现出正负相间的分布特征;3) 最后计算了20年、50年、100年一遇重现期趋势变化率,结果显示:不同重现期的降水量分布相似,重现期越长,极端降水重现水平最大的地方越多,重现水平达到了20 mm/decade以上;重现期均值也显示:相较平稳极值模型,非平稳极值模型极端降水的100年重现期所显示的重现水平更大,大的重现水平的区域范围增加。

Using the precipitation data of 124 effective stations located in the Yangtze River Basin by the National Meteorological Information Center, the dynamic threshold varying with time is introduced, and the over threshold generalized Pareto distribution model is established. The temporal and spatial characteristics and trend analysis of summer extreme precipitation in the Yangtze River Basin in recent 57 years (1961~2017) are analyzed by using the non-stationary extreme value model, and the trend change of return period is given. The analysis is of great significance to water source management, water conservancy construction and urban design. 1) Compared with the stationary extreme value model, the non-stationary extreme value model can better describe extreme precipitation;2) The results show that the trend of scale parameters in the Yangtze River Basin is local, the trend is positive in most areas, and the shape parameters show the distribution characteristics of positive and negative phases;3) Finally, the trend change rates of 20-year, 50-year and 100-year return periods are calculated. The results show that the precipitation distribution in different return periods is similar. The longer the return period, the more places with the largest return level of extreme precipitation, and the return level reaches more than 20 mm/decade;the mean value of return period also shows that compared with the stationary extreme value model, the 100-year return period of extreme precipitation in the non-stationary extreme value model shows a larger return level, and the regional range of large return level increases.