Implement of filter to remove the autocorrelation's influence on the Mann-Kendall test: A case in hydrological series

Chiyuan Miao,Jinren Ni

2010-01-01

Abstract:The available water resources are becoming depleted due to the global warming, economic development, ecological degradation and population expansion. In order to optimally manage the water resources, plentiful statistical methods were applied to detect the characteristics of hydrological time series. Mann-Kendall (MK) test is one of the most popular statistical techniques, which can show detailed changing trends and abrupt change for the hydrological time series. However, the autocorrelation of hydrological time series inflates the variance of the estimated mean and decrease effective sample size, and consequently influences the results of the MK test. In order to remove the autocorrelation of time series, the paper firstly described the calculation of MK test and autocorrelation in detail, and put forward the filter processes with 5 steps: (1) detecting the autocorrelation of time series; (2) detrending the time series; (3) calculating the parameters of autoregressive (AR) model; (4) removing AR component and (5) combining post-AR series and "pure trend component". Finally, the paper integrated with a concrete hydrological case. It is found the filter processes are an effective procedure to remove the autocorrelation of time series, and it was still demonstrated the differences of MK results when facing the original hydrological time series and the series after filter processes.