关键词: 趋势, 突变, 秩检验, 小波分析, 水文气象序列

Abstract: Some of the characteristics that complicate the analysis of hydroclimatic time series are the errors or changes in instruments, the new methods of data collection, the moved station location, non-normal distributions of data, missing values, values below the limit of detection, serial correlation, etc. In the face of the complications listed above for the trend analysis and abrupt change detection of series data, almost all techniques used in common are presented here, which include traditional parametric statistical techniques, non-parametric rank test, and wavelet analysis. First, all of the technical details needed to apply these methods are listed and proposed in sequence. Some of the theoretical basis for different alternative techniques are discussed and provided, about motivation for their use and quantitative measures for their performance by comparing with each other. The validity of parametric statistical techniques' estimates are based on underlying assumptions of normality and homogeneity, which are not met conditions by real climate data sometimes. The powerful nonparametric techniques do not need the data be distributed normally and have a low sensitivity to some missed data in the time series data. Then, the application and practical advantages/disadvantages of these alternative techniques are demonstrated by analyzing an example of annual mean atmospheric temperature data set with serious consideration of these techniques. The existence of serial correlation has a strong impact on the trend and abrupt change of a series data whether using parametric statistical method or non-parametric rank test. It was shown that removal of a positive serial correlation component from time series by pre-whitening resulted in a reduction in the magnitude of the existing trend. The variance of the time series data length alters the test result of abrupt change, and the most slightly shift of abrupt chang was calculated by Reverse Spearman' s rho test in contrast to other different time sequence methods. Finally, some recommendations on the analysis of hydroclimatic data sets using properly and efficiently techniques are presented. In order to confirm accurately the changepoint of time series data, different tests, which must be based on different theories other than analog principles, should be used by contrast to identify the point. Compared with parametric statistical and non-parametric rank methods only for the identification of a single changepoint, wavelet transforms can detect multiple changepoints without underlying assumptions which are sometimes not met by real hydroclimate data.

Key words: trend analysis, abrupt change, rank test, wavelet analysis, hydroclimatic time series