Abstract: There is not only white noise in GPS position time series, but also a lot of flicker noise. If only the covariance matrix of pure white noise is used to estimate the velocity, the velocity accuracy will be seriously overestimated. Spectrum analysis and maximum likelihood estimation methods are used to analyze the 20-year GPS coordinate time series of 264 GPS continuous stations of the Chinese Continental Crustal Movement Observation Network, and the velocity accuracy of each station calculated. The results show that there are different noise characteristics in E, N and U directions. The spectral exponents estimated by spectral analysis are higher than those estimated by maximum likelihood, but they are all around −1, which indicates that the main noise model is white noise plus flicker noise. The velocity accuracy calculated by different noise models is quite different. In the east, the north, and the vertical directions, the velocity accuracy calculated by the optimal noise model is, respectively, (11.5±2.1), (12.9±2.9), and (14.8±3.7) times that of the white noise model only. The velocity between maximum likelihood estimation method and FOGMEx method is not very different. In the east, the north, and the vertical directions, the velocity accuracy calculated by the maximum likelihood estimation method is (2.8±1.5), (1.5±0.7), and (3.5±2.8) times that of the FOGMEx method, respectively.

Key words: GPS, time series, noise model, flicker noise