Study On Intermediate-and Long-term Errors In GPS Position Time Series

Continuous GPS (global positioning system) has been become one of the major techniques for crustral deformation monitoring. There are thousands of GPS stations around the world which provide reliable data for studies on plate motion, fault slip, and deformation caused by earthquakes. One of problems in GPS measuerments is spatio-temporal correlated noise in GPS position time series, which contaminates tectonic signals and makes it difficult to separate signals from noises. It is now a frontal subject to extract and to remove those non-tectonic signals. This thesis is an attempt to solve this issue which is of importanance for furher understanding characteristics and origins of noises in GPS data, and would promote the research progress in GPS application fields.This work focuses on the intermediate- and long-term (T>1 day, from 1-day to 10-year time scale) errors in GPS position time series. The most obvious part is the common-mode error (CME) which is the common motion of regional GPS stations in a global reference frame. To learn the characteristics of non-tectonic signals, the time series analysis theory is used to derive the type and magnitude of colored noise, to obtain the amplitudes and phase lags of seasonal variations of GPS stations. Then stable GPS sites are chosen with the criteria for size of colored noise, the amplitude and phase lag of seasonal tem, subsidence or uplift trend, etc. Based upon the correlation relationship of residual position time series of GPS stations, an improved spatial filtering technique is suggested to extract CME of varied spatial scales and transient tectonic activities. Finally the effect of instability of reference frame on CME is assessed.To obtain reliable GPS position time series, this work re-analyzed nearly 11-year continuous GPS data since 1999 from CMONOC (Crustal Motion Observation Network of China) and more than 100 ITRF2005 reference frame stations in a consistent way—using the latest softare (GAMIT/GLOBK v10.3) and models. The rerun results are the major data source of this dissertation. At the same time, GPS position time series produced by SOPAC (Scripps Orbit and Permanent Array Center), JPL (Jet Propulsion Laboratory), and CMONOC data center are also used for comparision of colored noise, seasonal terms, and CME.It is known that CME in GPS position time series under a global frame is one kind of correlated noise. The characteristics of the correlated noise in position time series for GPS fiducial stations of CMONOC is analyzed. The types of colored noises in the CMONOC network are assessed using the maximum likelihood estimator (MLE). Besides flicker noise and random walk noise, more types of noise models were considered, including pow-law noise with fractional spectral index, first-order Gaussian-Markov noise, and band-pass filtering noise. It is found that flicker noise is the primary type for CMONOC network, which is consistent with conclusions from studies in other regions. The magnitude of flicker noise in the spatially filtered GPS potition time series decreases remarkably, showing that the content of CME (common-mode error) is mainly flicker noise. When taking into account correlated noises in estimation, the rate uncertainities increase by over an order of magnitude from white noise estimates, which are usually less than 1 mm/a.Based upon the CMONOC rerun results, this work obtains some new insights into the annual movements of GPS fiducial stations. There are unexplained annual motions at Hailaer (HLAR) and Haerbin (HRBN) stations. The results from several short-baseline GPS station pairs (e.g. at Changchun: CHAN-CHUN, at Kunming: KMIN-KUNM) suggest that local environment factors can cause obvious abnormal movement of GPS stations, either the amplitude or the phase differences of annual terms. Thus, attentions should be paid when interpretating postion results of GPS stations in the case that there are no other continuous GPS stations in the surrounding area. The above GPS-observed annual position variations can be explained at most sites with the loading displacements caused by seasaonl atmospheric pressure and soil moisture variations. However, at southern sites (QION, YONG, and XIAM), Lhasa (LHAS and LHAZ), and Tash Kurghan (TASH), there are still large residual annual amplitudes, suggesting that there are still unknown factors or large mis-modeling errors. At the Lhasa stations, the remaining amplitudes after geophysical mass loading corrections are about 2~3 mm. Results from other nearby continuous GPS stations show that there are negliable phase lag differences between these sites, and the amplitudes at Lhasa seem a bit larger. In the Tibetan plateau and the Himalaya mountain belt, the vertical annual motions are mainly controlled by hygrological factors, showing large phase lag changes.It is found that there are sinusoidal motions with periods around 351/n (n=1,…, 6) days in position time series for CMONOC and IGS (International GNSS Service) sites around China. These periodicities reveal the existence of CME. After spatial filtering of CME was performed, a large sum of corresponding peaks in the power spectrum disappeared. This kind of seasonal terms cannot be explained by surface mass redistributions.A spatial filtering scheme is developed to extract common-mode components (CMC) in continuous GPS positions, corresponding to either the so-called CME or regional tectonic signal. The technique utilizes two weighting factors: 1) correlations between GPS position residuals as distance weights, and 2) areas of Voronoi cells constructed from fiducial GPS sites as azimuthal weights. Comparing to the conventional regional stacking method, this scheme has achieved 5%~15% more reduction of residual RMS (root mean square). And by varying the distance window, it can extract CMC of various spatial scales, such as signals of slow slip events occurring periodically at certain subduction zones. Different from existing spatial filtering algorithms, the correlation-based stacking method presented in this thesis overcomes the spatial scale limit when doing filtering, and does nor need manual interferences.The correlation-based spatial filtering technique is then used to extract CMC for 127 global GPS stations and results are obtained at 96 sites. The magnitude of CMC is found to be correlated with size of flicker noise, i.e., those with larger scatters in CMC are the stations with larger flicker noise. But this relationship does not exist between scatter measurements of CMC and amplitude of annual term. The spatial scale of the major part of CMC is usually larger than 1000 km. It is interesting that the size of CMC is correlated with the density of frame stations, i.e., the CMC are usually smaller in areas with relatively denser GPS stations used in the frame stabilization procedure. This implies that the instability of reference frame definition may account for major part of CMC.In the last section of this thesis, the effect of non-linear reference frame on GPS positioning results is assessed tentatively. It is found that the magnitudes of CMC and flicker noise are correlated with the stability of reference frame definition. It is also shown that applying ATML corrections at the observation level could produce small improvements in reducing the magnitudes of white noises and flicker noises, by reducing the effect of vertical annual motions on frame stabilization.This work has improved the understanding of non-tectonic signals in GPS position time series, and satisfying results in data filtering and non-tectonic noise reduction have been obtained. Nevertheless, only limited analyses on GPS time series errors have been performed, and many kinds of noises in GPS data aquisation and processing remain to be further explored. More work need to be done in this field for deeper understanding of GPS signals and noise removal, so that to promote research of tectonic deformation.