Separation of Sources of Seasonal Uplift in China Using Independent Component Analysis of GNSS Time Series

Abstract

With the improvement of Global Navigation Satellite System (GNSS) observation accuracy and the establishment of large continuously operating networks, long GNSS time series are now widely used to understand a range of Earth deformation processes. The continuously operating stations of the Crustal Movement Observation Network of China capture deformation signals due to time-dependent tectonic, nontectonic mass loading, and potentially unknown geophysical processes. In order to separate and recover these underlying sources accurately and effectively, we apply the independent component analysis (ICA) to decompose the observed time series of vertical displacements. Then, we compare these signals with those predicted from independently developed geophysical process models of atmospheric, nontidal ocean, snow, soil moisture mass loading, and the Land Surface Discharge Model, as well as with Gravity Recovery and Climate Experiment observations. For comparison, we also perform the principal component analysis decomposition of time series and find that the ICA achieves a more consistent representation of multiple geophysical contributors to annual vertical GNSS displacements. ICA can decompose the long-term trend and different seasonal and multiannual signals that closely correspond to the independently derived mass loading models. We find that independent contributions from atmospheric, soil moisture, and snow mass loading can be resolved from the GNSS data. Discrepancies are likely due to the correlated nature of some of the loading processes and unmodeled contributions from groundwater and surface water changes in South Central China and the Ganges Basin.

Independent Component Extraction from the Incomplete Coordinate Time Series of Regional GNSS Networks

“…In past decades, high-accuracy coordinate time series of global navigation satellite system (GNSS) stations have been widely used for monitoring seismic, coseismic displacements [ 1 , 2 ], and regional tectonic deformation [ 3 ]. The deformation signals, such as trend, annual and semiannual signals, as well as transit signals, can be detected from the GNSS time series, which contain abundant information from different sources, including tectonic and non-tectonic processes, such as the mass loading variations of snow, atmosphere and soil moisture [ 4 , 5 , 6 , 7 ]. The trend, annual and semiannual signals are usually estimated by the least-squares fitting.…”

“…The other spatiotemporal signals are more effectively extracted and analyzed with some classic signal analysis methods, such as wavelet analysis (WA) [ 8 , 9 , 10 ], Kalman filter (KF) [ 11 , 12 ], empirical orthogonal function (EOF) [ 13 ], singular spectrum analysis (SSA) [ 14 , 15 ], and principal component analysis (PCA) [ 16 , 17 , 18 , 19 ]. Among these methods, PCA is one of the data-driven multivariate approaches based on second-order statistics (variance and covariance) and isolates the underlying sources without any prior knowledge [ 7 ], which implicitly assumes that a GNSS time series is polluted only by multivariate Gaussian noise. Nevertheless, previous studies demonstrated that a GNSS time series usually contains colored noise too [ 20 , 21 ].…”

“…Since PCA decomposition is based on the maximization of the variance of decomposed components, thus PCA works efficiently if only a single source exists in the GNSS time series. When multiple competing sources exist, however, the dominant components determined by PCA are probably the mixture from different physical sources [ 7 , 22 ].…”

Signal Extraction from GNSS Position Time Series Using Weighted Wavelet Analysis

“…Due to time variable characteristics of external environments and geophysical effects, GNSS position time series show significant seasonal changes, especially for the vertical component [8]. The main causes of seasonal oscillations are attributed to gravitational excitation, soil moisture and environmental loading variations [9][10][11]. If seasonal signals are not well characterized or there is non-random noise in the position time series, the velocity of a permanent station and its uncertainty will be misestimated [12,13].…”

Modeling Seasonal Variations in Vertical GPS Coordinate Time Series Using Independent Component Analysis and Varying Coefficient Regression

“…Gruszczynska et al [ 30 ] used multichannel singular spectrum analysis (MSSA) to model the common seasonal signals in GPS observations in different regions around the world and without influencing the high frequency part of the spectra. Some other data-driven mathematical methods have been proposed to be used in modeling the geophysical signals from regional geodetic time series, e.g., principle component analysis (PCA) and independent component analysis (ICA), which have been widely used in smoothing and modeling GPS time series [ 15 , 23 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ]. ICA has especially been widely used in the separation of potential sources of seasonal signals in vertical GPS coordinate time series, and the main independent components (ICs) were thought to be consistent with some of the geophysical signals [ 32 , 33 , 34 ].…”