Abstract To address the issue of the complementary ensemble empirical mode decomposition(CEEMD) having to set screening criteria, this paper takes seasonal signals into consideration when using CEEMD to denoise GNSS coordinate time series. The improved method firstly deconstructs GNSS coordinate time series into numerous intrinsic mode function(IMF), calculates their average periods, then utilizes IMFs with an average period of less than 120 days as noise components, while reconstructing the remaining components as signal components. This study applies the method to denoise 227 GNSS vertical coordinate time series stations over mainland China; it compares the results of CEEMD with the continuous mean square error and correlation coefficient methods. The results indicate that the described method does not denoise excessively, whereas the other two methods do. At the stations without excessive noise reduction, the average correction rates of RMS, power law noise, and velocity uncertainty of the GNSS coordinate time series of our methed are 19.13%, 88.29% and 86.46%, which are better than the other two methods.