Abstract <p>Seismic records acquired by seismometers inevitably contain substantial noise. Making effective noise suppression is essential for reliable data processing. While traditional filtering techniques (e.g., Butterworth and wavelet filters) are widely used, mode decomposition methods have seen limited application in denoising teleseismic P-wave data. In this study, we enhance and apply Adaptive Variational Mode Decomposition (AVMD) to attenuate both coherent and incoherent noise in teleseismic P-waves. Specifically, we replace the kurtosis criterion with a crosscorrelation criterion to improve noise screening. Since parameter selection strongly influences the performance of AVMD, we derive a general parameter set by applying AVMD to data from multiple stations and events. Using these parameters, AVMD achieves robust denoising results across diverse datasets. Compared with other approaches, AVMD offers simpler parameter tuning and is independent of frequency, and provides superior performance of data with low signal-to-noise ratios, thereby improving seismic monitoring reliability.</p> Highlights <p><UnorderedList Mark="Bullet"> <ItemContent> <p>An improved AVMD method is proposed, using the crosscorrelation coefficient and permutation entropy as criteria to distinguish noise and signal modes.</p> </ItemContent> <ItemContent> <p>Waveform and spectral analyses demonstrate that AVMD effectively suppresses noise and accurately recovers seismic event waveforms.</p> </ItemContent> <ItemContent> <p>Reference threshold values are provided, enabling effective denoising of seismic data across different events and stations.</p> </ItemContent> </UnorderedList></p>