中文摘要：

利用Mann-Kendall（MK）方法进行水文序列趋势检验时，去趋势预置白（TFPW）作为处理水文序列自相关性影响的重要方法之一，其处理过程的合理性及方法的适用性在变化环境下备受关注。本文以三种具有典型趋势特征的实测年径流资料为研究对象，开展TFPW-MK检验（6种去趋势方法并2种预置白方法），并与传统MK法及改进MK法进行对比，探讨不同TFPW方法在变化环境下的适用性。结果显示不同TFPW方法对典型趋势径流序列的MK检验结果影响不一；在AR（1）预置白处理时，不同去趋势方法表现较为接近；而在去除AR（n）（n表示所有显著自相关阶数）时，仅直线滑动平均法（LMA）法和直线滑动回归平均法（LRMA）法表现较为合理，传统TFPW方法出现失真现象。综合对比表明基于LRMA并AR（n）的TFPW方法表现最优，而传统TFPW法仅在AR（1）条件下可获得较为稳定的趋势检验值。

英文摘要：

Trend-free pre-whitening （TFPW） is a key method for removing the effect of autocorrelation when a non-parametric Mann-KendaU （MK） trend test is used in anatysis of river runoff series. However, its rationality and applicability are questioned in the cases that some hydrological changes occur in a changing environment, thus catching researchers＇ increasingly attention. In this study, we adopted a hybrid TFPW for data-preprocessing prior to applying the MK trend test to evaluate its performance in analysis of the runoff series featured with hydrological changes, focusing on six detrending measures, i.e. the slope method （SM）, EMD method, first order differential method （FD）, log-linear detrending method （LLD）, 5-point linear moving average method （LMA）, and linear regression moving average method （LRMA）. This approach is coupled with two pre-whitening methods, one for removing autocorrelation AR（1） only, and the other for removing all the AR（n）s where n stands for all the lag orders of significance. A case study was taken testing the runoff series gauged at the three gauge stations ofDingjiagou, Shenmu and Suide in northern Shaanxi that represent three types of trend changes in the series respectively. These calculations, as results of different hybrid TFPW-MK testing, were compared with those by conventional and modified MK methods. The comparison shows that different hybrid TFPWs give different MK test results of the runoff series. All the detrending methods are good and similar when only AR（1） is removed, while when all the AR（n）s are removed, only LMA and LRMA work satisfactorily and even the SM, if coupled with the conventional TFPW, works poorly deviating far away from the real case. This indicates that among all the TFPW methods, LRMA with AR（n） being retained is the best owing to less autocorrelation information lost in its detrending process and more autocorrelation removed in its pre-whitening. In addition, the conventional TFPW-MK, or SM coupled with the