中文摘要：

采用北江流域10个水文站点1959-2005年日流量数据,运用滑动秩和检验法（Mann-Whitney U test）和AMOC检验法确定样本最佳突变点,采用Mann-Kendall（MK）检验法检测时间趋势性,最后结合广义可加模型（GAMLSS）分别构建洪水量级和频率与影响因子的非平稳性模型.研究表明：1）北江流域年及季节洪峰流量普遍呈下降或显著下降趋势,在1990年发生突变,但年最大洪峰无显著变化趋势;2）1990年之后,最大三场洪水及重现期大于10年的洪水事件多集中发生,且洪水发生频率、量级及峰现时间均发生较大改变;3）LOGNO分布为年最大洪峰流量序列最优极值分布,计算出非平稳性条件下北江流域4个站点百年一遇设计洪水流量值,并分别以时间（Model 1）和气候指标（Model 2）为解释变量对洪水发生次数进行模拟分析,充分反映洪水发生次数的随机过程,为区域防洪减灾提供理论依据.

英文摘要：

Flooding behaviors in the North River basin,the important tributary of the Pearl River basin,were thoroughly investigated by using Mann-Whitney U test and AMOC based on daily streamflow data covering aperiod of 1959-2005 from 10hydrological stations.In addition,GAMLSS（generalized additive models for location scale and shap）was used to develop nonstationary relations between magnitude,frequency and influencing factors.The results indicate that：1）annual and seasonal peak flood flow are in generally decreasing or significant decreasing trends.Besides,no significant trends can be identified for annual maximum streamflow after the change point,i.e.1990s;2）after 1990 s,evident clustering phenomenon can be observed for the largest three flood events and flood event with return periods longer than 10 years.Moreover,significant alterations can be detected in changes of frequency,magnitude and timing of flooding regimes;3）LOGNO is the right distribution model that can well describes changes of annual maximum streamflow.Furthermore,flood flow with return period of 100 years was computed with considerationof nonstationarity.In addition,modelling was done for occurrence rates of floods using time（Model 1）and climate indices（Model 2）as covariate variables.It was corroborated that GAMLSS is the best model for modelling of randomness of flooding behaviors,providing theoretical grounds for regional human mitigation to flood hazards in the North River basin.