基于GPD模型的黑河出山径流极值变化分析

以黑河出山径流为研究对象, 利用黑河上游1944-2010年67 a的实测月均径流资料, 采用广义Pareto分布模型（GPD模型）, 对黑河未来出山月均径流的变化进行了预测。结果表明：运用超阈值的观测资料建模的GPD模型, 能够达到信息量使用的最大化, 对黑河极值变化的预测精度较高; 黑河出山径流重现期为25 a、50 a、100 a和150 a的月均径流量极大值分别为219.8 m³?s^-1、235.3 m³?s^-1、 249.5m³?s^-1和257.4 m³?s^-1; 重现期为25 a、50 a、100 a和150 a的月均径流量极小值分别为9.0 m³?s^-1、8.8 m³?s^-1、8.6 m³?s^-1和8.5 m³?s^-1;在当前“暖湿”的气候背景下, 黑河出山径流变化不大。

Variation of Extreme Runoff Volume at Debouchure of the Heihe River Based on GPD Model

In this paper, the Generalized Pareto Distribution (GPD) model was used to predict the extreme values of average monthly runoff volume at debouchure of the Heihe River in different return periods. The return levels of extreme average monthly runoff volume in both flood season and dry season were analyzed based on the average monthly runoff volume data collected at Yingluoxia Hydrological Station from 1944 to 2010. In order to make the use of GPD model to predict the average minimum monthly runoff volume in dry season, the opposite number of the average monthly runoff volume below 20 m3•s-1 was taken to transform the minimum into the maximum. Under both cases, a proper threshold was selected by the mean residual life plot and by observing the change of modified scale parameter and shape parameter against thresholds. The maximum likelihood method was used to estimate the scale parameter and the shape parameter in GPD model. The different return levels of the extreme average monthly runoff volume of the Heihe River were calculated based on the fitted GPD model. The results showed that the average maximum monthly runoff volumes at the 25 year, 50 year, 100 year and 150 year return levels were 219.8 m3•s-1, 235.3 m3•s-1, 249.5 m3•s-1 and 257.4 m3•s-1, and the average minimum monthly runoff volumes at corresponding return levels were 9.0 m3•s-1 , 8.8 m3•s-1, 8.6 m3•s-1 and 8.5 m3•s-1, respectively. The results of the average minimum monthly runoff volume in dry season suggested that the cessation probability of the Heihe River was extremely small under current climate change. Moreover, the 95% confidence interval at the 100 year return level of the extreme average maximum monthly runoff volume was ［228.3, 303.1］ in flood season, and the estimation of the 100 year at return level of the average monthly runoff volume in dry season was very close to the average minimum monthly runoff volume observed by a deviation of 0.08 m3•s-1. The results indicated that the GPD model has a high accuracy in estimating the return levels of extreme values of average monthly runoff volume in the mountain areas of the Heihe River Basin.