Extracting The Trend And Seasonal Component Information Of GNSS Coordinate Time Series Based On Improved Singular Spectrum Analysis Method

After the first GNSS continuous observation station was established in January 20,1991,more GNSS continuous observation stations have been established in the world.Along with the progress of the GNSS measurement technology and the improvement of the model accuracy in data processing,over the world has accumulated more than 20 years of high precision GNSS continuous observation data.This provides important data support for studying geophysical phenomena at different time and space scales,such as earth rotation,regional deformation,post glacial rebound,seismic deformation monitoring,fault sliding,and the global plate tectonic movement,etc.Data preprocessing is the first step in the analysis of the GNSS coordinate time series,its main content consists of three parts:gross error detection and elimination,default data interpolation and offset correction in GNSS observations.When the wavelet analysis and power spectrum analysis are used to analyze GNSS station the time series,the GNSS coordinate time series is required to remove the linear trend and to zero the mean.Firstly,the wavelet analysis is used to analyze the seasonal component extracted from the GNSS coordinate time series in the time domain.Secondly,the power spectrum analysis is used to analyze the GNSS observation data,it was found that the spectral energy at the low frequency was large and the spectrum was tilted(the slope tend to-1),this shows that the noise contains flicker noise;but with the increase of frequency,the the spectral energy at high frequency gradually decreased;the spectrum tends to be gentle at high frequencies(the slope tend to 0),the noise characteristic is white noise;this is in line with the best noise model for GNSS station time series which is the combination of white noise and flicker noise spectral noise models.At the same time,the power spectrum analysis results show that the GNSS coordinate time series contains seasonal component with frequency(cpy)close to 1.0 and 2.0.Due to the influence of the systematic error associated with GNSS observation techniques and various geophysical effects,many GNSS time series may contain offsets,nonlinear trends,and seasonal signals with time-varying amplitudes.How to separate the above information and some other information which is not clear from the GNSS time series is a hot spot in the research of time series.There are limitations when using traditional parametric models to solve these complex problems.The singular spectrum analysis(SSA)is a nonparametric adaptive analysis method from the time series itself,it can extract the useful information from the time series of the GNSS site which is disturbed by the noise without any prior information of any geophysical phenomena in the original data.But the traditional singular spectrum analysis method has the disadvantages of phase shift phenomenon.In view of the above,an improved singular spectrum analysis(SSA-PD)method for fitting GNSS time series is proposed.Simulation results show that the RMS(root-mean-square)difference between the reconstructed and simulated signals is less than 1.8 mm;Comparing SSA-PD with wavelet analysis method using real IGS site data,the result indicates SSA-PD is better than wavelet analysis method in extracting the annual/semiannual components from the GNSS coordinate time series.Finally,we analyze the trend component and seasonal component extracted by the improved singular spectrum analysis method from the GNSS coordinate time series,the results show that the time-frequency characteristics of GNSS coordinate time series show significant regional characteristics.And the factors of the formation are analyzed qualitatively.