Research On The GPS Coordinate Time Series Analysis

Over the past two decades,with the continuous improvement of the observation precision of the space-geodetic techniques,abundant,high precision and globally covered GPS coordinate time series has been accumulated.These data have been providing an important data support and useful geometric constraints for studying various types of geophysical phenomena on various spatiotemporal scales.However,the GPS coordinate time series contains various “signals” and “noises” with different sources and properties reflecting in horizontal and vertical components,therefore,it is a fundamental and important task to extract the correct “signal” from the raw time series and accurately identify and eliminate the dominant sources of the errors,then to obtain high quality,sub-centimeter accurate position results.This topic has always been one of the hot spots and key issue in earth science.On the other hand,analyzing and comparing the GPS coordinate time series with the results from other geodetic techniques and/or geophysical models we may more reliably identify technique-specific systematic or geophysical model errors and further improve the data processing accuracy.The models and methods related to the analysis of GPS coordinate time series are studied in detail.It covers the establishment of terrestrial reference frame,the time-frequency characteristics of the colored noise,the quality control of GPS coordinate time series,spatiotemporal filtering of regional GPS network in China,the trend and seasonal signal estimation of height time series as well as the GPS coordinate time series decomposition using network-based Kalman filter.The main work and contributions of the dissertation are as follows: 1.The current three different methods for establishing terrestrial reference frame are systematically summarized and analyzed.The basic theories,procedures and key technologies of the International Terrestrial Reference Frame(ITRF)are discussed.The mathematical formula used in the generalized system constraint for reference frame transformation is theoretically deduced.At the same time,the establishment of ITRF2014 and its challenges are described.2.The time-frequency characteristics of several noise models including white noise,power-law noise,autoregressive fractionally integrated moving average process,generalized Gauss-Markov process and first order Gauss-Markov process are analyzed in detail.The formulae for constructing covariance matrix and power spectral density for each noise model are deduced,respectively.3.The methods for the quality control of GPS coordinate time series are systematically studied.Two data-driven interpolation methods,which are RegEM and DINEOF,are investigated,respectively.The performance of these two methods under different magnitudes of colored noise are compared and the RegEM method is recommended.An algorithm combined L1-norm estimation and interquartile range(IQR)statistics,namely L1_Mod IQR,is proposed to detect and identify outliers,and its results are compared with that of other traditional methods,such as “3” method.In terms of offset detection,considering that the GPS coordinate time series contains colored noises,a new algorithm named COL-STARS was proposed and its performance is evaluated using real GPS coordinate time series.4.Aiming at the shortcoming of spatiotemporal filtering using principal component analysis(PCA),a method based on independent component analysis(ICA)is proposed,which is applied to Crustal Movement Observation Network of China(CMONOC)II and is compared with PCA and region stacking methods.The results show that PCA leads to “artificial” spatial features,while the ICA can extract common mode error reliably.After identifying and discarding the abnormal stations,it is shown that the results of spatiotemporal filtering based on ICA show that the spatial distribution of common mode error for CMONOC II is not uniform,but have an obviously symmetry east-west and south-north distribution.5.A two-step estimation procedure based on the local weighted regression seasonal-trend decomposition(STL)method is proposed to analyze the seasonal signals and long-term trend of height time series of 10 IGS stations in China.The seasonal signal is firstly extracted by using STL and then removed from the original time series.Secondly,the long-term trend is estimated for deseasonal time series using maximum likelihood estimation(MLE).The estimated velocity for each site is compared with that of Scripps Orbit and Permanent Array Center(SOPAC).The results show that the velocity precision based on the two-step strategy is higher than that of SOPAC.6.A network-based Kalman filtering method,namely NETKF,is proposed to decompose GPS coordinate time series into trend,seasonal signals,and colored noise and residuals.As each structural component is reformatted in state-space model and allows to change over time,therefore the local variations of each component can be captured.The NETKF is applied to 10 GPS stations from CMONOC II and the results are compared with that of the ordinary MLE method with singlestation,single-component strategy.The results show that the NETKF can extract all structured variables successfully,and the estimated long-term velocities are consistent at the 95% confidence level(2)in horizontal components.However,there are still some differences in vertical component.