[G04-4] Earth rotation and geodynamics IV

[G04-4-03] Modelling of Time-Varying Seasonal Signals in GNSS Time Series

^○Anna Klos¹, Machiel Bos², Janusz Bogusz¹ (1.Military University of Technology, Faculty of Civil Engineering and Geodesy, Warsaw, Poland, 2.University of Beira Interior, Instituto D. Luis, R. Marques d'Avila e Boloma, Portugal)

Most GNSS (Global Navigation Satellite System) coordinate time series contain oscillations of annual and semi-annual periods that are routinely modelled by two periodic signals with constant amplitudes. In reality, the amplitudes of these seasonal signals varies slightly over time. The stochastic properties of the GNSS coordinate time series are well described by power-law noise. Subtracting of the time varying seasonal signals is crucial, since each residual periodicity which is still left in the data naturally will make it appear to be autocorrelated. In this research, we employed 174 IGS (International GNSS Service) stations processed by JPL (Jet Propulsion Laboratory) and showed that the mean variation of the annual amplitude has a standard deviation of 0.8 mm and that for c.a. 15% of stations these variations are larger than 1.0 mm. We performed a quantitative comparison of Wavelet Decomposition (WD), Singular Spectrum Analysis (SSA), Chebyshev Polynomial (CP) or Kalman Filter (KF). Using synthetic time series with realistic power-law noise we showed that ignoring the variations in the amplitude of the seasonal signal results in too large estimates of the spectral index κ of 0.1-0.2 (bias towards flicker noise) and an overestimation of the noise amplitude of about 0.4-1.0 mm/yrκ/4. When the flicker noise amplitude is very low relative to the size of the variations in the seasonal signal, estimating a constant seasonal signal performs worse than any of the methods that try to model varying seasonal signal and can produce an estimated trend error that is overestimated by 0.03 mm/yr. When the noise level is average, the varying seasonal signals can no longer be estimated accurately and one can continue to estimate only a constant seasonal signal. For the JPL-derived GPS data, we noticed a decrease in spectral indices for all coordinates of -0.14 in average and a decrease of trend error of 0.05 mm/yr when time-varying seasonal curves were modelled.