Abstract: Taking the Baijiabao step-type landslide in the Three Gorges Reservoir Area as a case study, we selected monitoring data at four time scales (daily, weekly, ten-day interval and monthly), decomposed the cumulative displacement into periodic and trend components based on the additive time series model, screened highly sensitive inducing factors via the grey correlation method, and constructed a coupled model of the Ensemble Empirical Mode Decomposition (EEMD) algorithm and Gated Recurrent Unit (GRU) to analyze the prediction accuracy of landslide cumulative displacement. The results show that the prediction accuracy at weekly and ten-day interval scales is significantly higher than that at daily and monthly scales, with the weekly scale yielding the optimal performance. The root mean square error of displacement prediction at the weekly scale is reduced by 56.98%, 64.00%, and 35.51% compared with that at daily, monthly, and ten-day interval scales, respectively, and the mean absolute error is decreased by 61.04%, 69.07%, and 23.72% accordingly. Mechanism analysis reveals that the sampling frequency of the weekly scale matches the 7–15 days lag response period of the landslide to hydrological factors, which can filter out high-frequency noise while retaining key deformation signals. This study confirms that weekly monitoring data are suitable for predicting the deformation trend of the Baijiabao landslide, and the findings provide a scientific basis for the hierarchical and variable-frequency early warning and control of landslide disasters in the Three Gorges Reservoir Area.