Nonlinear Time-varying Study Of GPS Height Time Series Of National CORS Stations

The analysis about coordinate time series for domestic and international GNSS continuous operational reference system show that, the series in the horizontal direction and vertical direction mainly appear to be linear, but due to the variety of geophysical factors it also includes complex periodic movement and non-linear trend in the vertical direction. As a frame of CGCS2000coordinate system, the research about non-linear time-varying and noise characteristics in height time series is the important basis of building high-precision height velocity field model and maintaining the up-to-date state and dynamics state of height component in national geocentric reference system.This paper researches the height time series of national CORS stations which maintaining China’s CGCS2000coordinate system. First, it starts from calculating the original coordinates from GPS base stations, and then adopts robust estimation theory to position and delete gross errors in GPS height time series, also interpolates missing data points and homogenize the ranges of series. It provides the available data for the next analysis about frequency characteristics and its noise. Second, it analyses main period characteristics in series by using traditional robust spectral estimation method. Third, it applies adaptive time series time-frequency analysis, which can not only identify non-linear trend of series, but also detect seasonal and long periodic oscillation characteristics in base station. Then color noise is separated from GPS height time series. At last, it analyses height movement and noise characteristics under different kinds of noise models and the influence of surface mass loading on the height component in the base station.Using GAMIT/GLOBK(v10.35) software and absolute antenna phase center model the the paper recalculates the data from34CORS stations for nearly11years since1999to2010. It provides the sources of data for the next analysis.It compares two robust estimation methods to position and delete gross errors in GPS height time series. One is robust least-square, the other is IQR criterion. It turns out that the former is much better. According to the interpolation problem after eliminating gross errors in series, the interpolation method of using orthogonal polynomials to regress least squares is proposed. The experimental results show that, when there are only a few continuous missing points, its effects equals to3rd spline interpolation method, but when missing value points are more than6-10,3rd spline interpolation effect is distorted. The fitting results still ensure the overall movement trend under the condition that continuous missing values are less than two months when using orthogonal polynomials.Combining robust estimation and traditional spectrum analysis it proposes the robust spectrum estimation, by which the main period characteristics of national CORS station height time series are exhaustively analyzed. The results show that, the indirect method in power spectrum estimation based on Wiener-Khintchine laws behaves much better than periodogram method and welch method. But the problem that the resolution of traditional spectrum analysis is low and maybe there is spectrum leakage makes other periods produce the bad results except the first period.By introducing Hilbert-Huang transformation(HHT) technology into GPS height time series signal analysis, this paper puts forward the concept of non-linear trend of GPS height time series, and then indentifies and separeates non-linear trend from the series with the contribution rate of variance. In general, the traditional analysis of GPS height time series uses one-year or half-year period function to fit model. However, this article adopts EMD and EEMD decomposition methods. This two methods not only detect periodic oscillation characteristics in series, but also find approximately seasonal, annual and biennial period movement characteristics, where anniversary period is not obvious. The frequency of IMF component transformed by HHT is not a constant, but behaves like a curve. It changes rapidly near high frequency, and becomes gently near low frequency.This paper proposes overall experience model decomposition technique to extract the color noise of GPS height time series. The traditional strategy takes residual in periodic function fitting as color noise contained in series. In fact, the precise year or half year period does not exist in GPS time series, which leads to the effect of annual signal still being in the fitting residual, so the method of using this noise characteristics needs further consideration. This paper puts forward a new idea for noise characteristics estimation by combining colored noise separated from HHT with maximum likelihood estimator.At the end of the paper, it estimates noise content of height time series signal from national33CORS base stations, and considers base stations linear motion rate and its error estimation under three models which are respectively white noise, white noise and power spectrum noise model, white noise and flicker noise and random walk noise model. At last it quantitatively analyzes the influence of changes in surface mass loading (atmospheric loading, non-tidal ocean mass loading, soil moisture and snow depth)on base station height.