Spatiotemporal Filtering of GNSS Coordinate Time Series Based on Variational Bayesian Independent Component Analysis

Denoising and refining GNSS coordinate time series can significantly increase the application level since they contain rich constructive and non-constructive information and are combined with noise due to observation conditions and other factors. To reduce the influence of common mode errors, we used the variational Bayesian independent component analysis (vbICA) technique to filter the GNSS coordinate time series of 12 stations in the Sichuan and Yunnan areas from 2011 to 2015. The experimental findings reveal that root mean square of coordinate residual series in the N, E, and U directions of the selected 12 stations were decreased by 26.5%, 23.7%, and 39.1% respectively, after vbICA filtering. The station velocities were estimated from the coordinate time series before and after filtering, and velocity uncertainties in the N, E, and U directions were reduced by 40.71%, 41.2%, and 52.34%, respectively. The results show that the vbICA method is effective for filtering coordinate time series. Meanwhile, the vbICA and PCA filtering algorithms are compared, and the former outperforms the latter on average.