中文摘要：

根据阿克苏河流域内出山口水文站1957～2006年的径流实测资料，利用数理统计、Mann—Kendall非参数趋势检验及小波分析等方法对径流年际变化特征进行了分析。研究发现，阿克苏河年径流量具有相对稳定且振荡上升的趋势，年际变化小；径流序列在1993年发生突变，径流量呈现出显著增加趋势，且通过了α=0．05的显著性水平检验；丰、枯水周期变化存在着25a左右的主周期，丰水年发生在20世纪90年代后期，且大量连续出现，最长持续时间达7年，枯水年多以连续2年的形式出现，平水年年数占52％。运用灰色拓扑预测方法对突变前后两个径流时间序列分别进行了模拟和预测，模拟结果的平均相对误差仅为12．06％，且位于同一突变区间的精度达97．7％，预测未来几年阿克苏河年径流量存在上升趋势，但趋势不显著。

英文摘要：

The Aksu River is a main tributary of the Tarim River,whose upper branches are in the territory of Kirghizsta, 132km long. The Aksu River is composed the two tributaries of the Tuoshigan River and the Kunmalike River. The Aksu Basin lies in an arid and semi-arid region of China, and it has typical hydrographical characteristics of inland rivers in the northwest of China. According to gauged discharges in the two representative hydrologic stations of the Shaliguilanke and the Xiehela,this paper uses a statistical method,the Kendall non-parameter tendency test and the wavelet analysis to analyze annual distribution and variation characteristics on the basis of the observations of annual runoff from 1957 to 2006. The result shows that the variation coefficient over years is small and relatively stable ： Cv is 0. 154 and the change range increases by 5.29×10^8m^3/10a. On the one hand, the result reveals there was a sudden change in 1993 in the annual runoff sequence, and after this spot, the amount of runoff volume increased distinctly,which passed significance test （ α= 0.05）. On the other hand, for the alternate time for annual runoff high to low,there is an about 25a main cycle,which was obtained through performing multi-scale transformation of the annual runoff sequence using the wavelet analysis theory. The high runoff year appeared continuously in the late 1990s,with a longest duration of 7 years, while the low runoff year mostly appeared in a cycle of two years. The number of normal runoff year account for 52% of the total years. In order to objectively display this complex wave shape of annual runoff change, this paper introduced the gray topology forecasting method. According to the observations of annual runoff data for 30 years（ 1957 -1986）before the sudden change spot, this paper established a grey topological GM （1,1）model and simulated annual runoff volumes before the sudden spot. The ＇mean relative error of simulated results is very low, only 12.06%. and the fitting accuracy is 97.7%