基于年径流序列的五种趋势检测方法性能对比  

Comparison among five methods of trend detection for annual runoff series

作　　者：姜瑶 徐宗学[1,2] 王静 JIANG Yao;XU Zongxue;WANG Jing(College of Water Science,Beijing Normal University,Beijing 100875,China;Beijing Key Laboratory of Urban Water Cycle and Sponge City Technology,Beijing 100875,China;Hydrology and Water Resources Survey Bureau,Tibet Autonomous Region,Lhasa 850000,China)

机构地区：[1]北京师范大学水科学研究院,北京100875 [2]城市水循环与海绵城市技术北京市重点实验室,北京100875 [3]西藏自治区水文水资源调查局,西藏拉萨850000

出　　处：《水利学报》2020年第7期845-857,共13页Journal of Hydraulic Engineering

基　　金：国家自然科学基金重大研究计划重点支持项目(91647202);国家自然科学基金青年基金项目(51909003)。

摘　　要：趋势变化是水循环过程的重要变化特性之一,准确识别水文要素变化趋势是认识变化环境下水循环演变规律的基本内容。由于当前趋势检测方法众多,且受多种因素影响,趋势检测结果存在较大的不确定性,很难判断结果的准确性。为提高对水文要素趋势分析方法的认识,研究基于3组变化特征不同且已知成分的人工生成序列,综合对比分析了5种常用趋势检测方法(累积距平法、线性回归法、MK法、EMD法和DWT法)的准确性和可靠性,并以3组不同实测年径流序列为例,进一步评价了不同方法在水文时间序列趋势检测中的性能差异。实例分析表明,累积距平法可用于水文时间序列趋势的初步分析,不宜作为判断依据。MK法与线性回归法所得结果差异不大,对趋势单调且无明显季节或周期变化的水文时间序列具有较高识别精度,适用于演变特征相对简单且趋势变化明显的年径流序列的趋势检测。EMD和DWT法可以准确地识别并分离出序列自身含有的不同尺度波动成分,从而避免序列周期、非线性变化等特征对趋势检测结果的影响,性能相对更优,对于演变机理及变化特征复杂的年径流序列的趋势分析,应优先考虑DWT法,其次为EMD法。研究成果可为水文要素的趋势分析及其方法选择提供参考。Long-term trend is one of major characteristics of hydrological cycle.It is a basic content to ac⁃curately detect the trend of different hydrological variables for understanding the evolution processes of hy⁃drological cycle under changing environment.Presently,there are many trend detection methods,but they are affected by many factors.Thus,the results from different methods are often uncertain and also difficult to judge which one is accurate and reliable.Therefore,the accuracy and reliability of five widely used methods for trend detection(Cumulative anomaly(CA),Linear regression(LR),Mann-Kendall(MK),Empiric mode decomposition(EMD)and Discrete wavelet transfer(DWT))are analyzed based on three ar⁃tificially generated sequences with known components and different change characteristics in this study.Three different observed annual runoff series,which respectively showed typical trend characteristics,were collected and used to further validate the performance and applicability of five methods for hydrological time series.Case studies indicated that the performance of CA method is only applicable for preliminary trend analyses.The results from MK method are similar to those from LR method,while the performance of MK method is better than that of LR method.Both of these two methods have higher accuracy for trend detection of time series with monotonous trend and insignificant periodicity,and is thus suitable for annual runoff with simple variation and large magnitude of the trend.EMD and DWT methods can accurately identi⁃fy and separate the fluctuation components at different scale contained in the sequence itself,so as to avoid the influence of the periodicity,nonlinearity of time series on the trend detection.They are thus rela⁃tively accurate for the trend detection of non-monotonous hydrological time series.Comparatively,the re⁃sults of DWT method are better than that of EMD method.The conclusion can provide reference for the trend detection as well as its method selection for hydrological series.

关 键 词：径流 趋势分析 EMD法 小波分析 MK法 

分 类 号：TV11[水利工程—水文学及水资源]