Abstract: Objectives: The Global Navigation Satellite System (GNSS) can obtain instantaneous dynamic ground displacement. However, GNSS-derived coseismic displacement time series are easily affected by random noise and non-stationary noise, which restricts the stability of seismic phase identification and rapid earthquake response. To address the inherent conflict between noise suppression and seismic waveform fidelity in single-domain denoising methods, this study proposes a Multi-Domain Adaptive Collaborative Denoising method, termed MD-ACD, to improve the stability of high-rate GNSS coseismic displacement series in seismic dynamic response identification. Methods: The proposed method constructs a cascaded framework from the frequency domain to the wavelet multi-scale domain. Frequency-domain spectral subtraction is first introduced to suppress broadband noise. Hampel filtering is then used to reduce local outliers and impulsive disturbances. Savitzky-Golay filtering is applied to separate the trend component from the residual component, and wavelet soft-threshold denoising is further performed on the residual series. A comprehensive quality evaluation criterion is incorporated to adaptively optimize the key parameters, thereby reducing the risk of insufficient denoising or local over-smoothing caused by fixed parameter settings. Finally, zero-phase low-pass filtering is adopted to suppress boundary effects and remaining high-frequency disturbances. Results: In the 120 s preearthquake window, the average standard deviations (STD) of the raw displacement series in the E, N and U components are 2.59 mm, 4.61 mm and 14.28 mm, respectively. After MD-ACD processing, they decrease to 1.45 mm, 3.13 mm and 9.14 mm, corresponding to reductions of 44.0%, 32.1% and 36.0%, respectively. In terms of computational efficiency, the average processing time of MD-ACD for a complete 2 h single-station single-component series is 0.104 s, indicating a relatively low computational cost under the experimental environment of this study. Comparison with the collocated strong-motion record shows that the proposed method can effectively suppress middle- and high-frequency noise while preserving the main seismic motion and the dynamic response characteristics near the P-wave arrival. Conclusions: The proposed MD-ACD method improves the stability of high-rate GNSS coseismic displacement series while maintaining the main seismic waveform characteristics, and can provide a methodological reference for high-quality displacement data processing in subsequent earthquake rapid response and source mechanism inversion applications.