LIU Li-xia, LI Gui-rong, LIU Dong-ya, LI Kui, ZHANG Qing-yuan, Reyilai Yiligongmu
GNSS coordinate time series is the key data for crustal deformation monitoring and seismic research, but its accuracy is seriously disturbed by complex noises such as multipath and atmospheric delay. Noise reduction processing has become a necessary prerequisite for high-precision applications. In recent years, noise reduction methods have developed from traditional empirical mode decomposition (EMD) and wavelet transform to variational mode decomposition (VMD), local mean decomposition (LMD), empirical wavelet transform (EWT) and other adaptive methods, and combined with particle swarm optimization, singular spectrum analysis (SSA), independent component analysis (ICA) and other algorithms to form a variety of hybrid strategies. The principles, advantages and limitations of the above methods are systematically sorted out. Through three typical application cases of post-earthquake deformation monitoring, multipath correction and long-term trend extraction, the performance differences of different methods in actual scenarios are compared and analyzed, and then a comprehensive selection strategy based on accuracy, noise type and efficiency is proposed. Studies have shown that existing methods still face challenges such as different evaluation criteria, insufficient real-time processing capabilities, and poor adaptability to special geographical environments. In the future, we should focus on the development of end-to-end noise reduction driven by deep learning, joint filtering under the constraint of multi-source data fusion, and intelligent automatic processing platform to further improve the availability and reliability of GNSS time series.
