Abstract: Objectives: Aiming at the problems that Complementary Ensemble Empirical Mode Decomposition （ CEEMD） is difficult to effectively separate out the high-frequency deformation signals in the global navigation satellite system （ GNSS） deformation monitoring sequences, as well as the difficulty of removing the vibration noise of the monitoring environment that is interspersed in the low-frequency deformation signals, a filtering and noise reduction method based on the combination of CEEMD, wavelet transform （ WT） , and composite partitioning indexes is presented. Methods: First, CEEMD of the original sequence is performed and the T-value of each Intrinsic Mode Functions （ IMF ） is calculated. Second, the mth layer IMF component corresponding to the first local minima T is searched for, and the first m-1 layers of IMF components are rejected as pure noise. further, according to the magnitude of the T-value, the residual IMF components are classified into retained signals and to-be-processed signals, and the processed signals are secondly noise reduced using the WT method for secondary noise reduction of the signal to be processed, Finally, the processed signal is reconstructed with the retained signal and the trend term to obtain the final noise reduction result. Results: From the simulation experiments with real data experiments on landslides：（ 1）The T-value has better signal retention ability and noise suppression compared to the noise reduction effect of the WT-based, mean standardized absolute moment -based and correlation coefficient-based CEEMD methods, and the method has the best SNR value and RMSE in the four sets of simulated data experiments.（ 2） After denoising with this method, its root mean square errors in E, N and U directions are 1.13 mm, 1.63 mm and 2.22 mm, respectively, which are 21%, 17%, and 12% better than the noise reduction effect of the WT-based, mean standardized absolute moment -based and correlation coefficient-based CEEMD methods, respectively.（ 3） The scheme in this paper has maintained the red warning after the first red warning was issued at 04:00 on 3 October, and compared to the moments when the first three schemes erroneously issued red warnings, it only issued orange warnings, and the change in the displacement rate of this scheme was also the smoothest of the four schemes, with the warning level not dropping to blue.Conclusions: The proposed method can not only retain the relatively high-frequency landslide deformation signals, but also effectively remove the vibration noise interspersed in the lowfrequency deformation signals, and can be applied to the de-noising of monitoring sequences of landslide scenarios, and the filtering method can effectively improve the accuracy of landslide monitoring and early warning, and avoid causing casualties and waste of resources.