Abstract: [Objective] To address the multi-scale coupling of reservoir-bank landslide displacement, the empirical selection of the VMD penalty factor, and the limited ability of commonly used time-series models to capture periodic deformation characteristics, a landslide displacement prediction method combining sparrow search algorithm-optimized variational mode decomposition (SSA-VMD) and a temporal convolutional network (TCN) is proposed. [Methods] The Huangtupo Linjiang No. 1 Landslide in the Three Gorges Reservoir Area was selected as the study site. SSA was used to optimize the VMD penalty factor, with the number of modes fixed at three. Based on their center frequencies in ascending order, the cumulative displacement was decomposed into trend, periodic, and random components. Reservoir water level, rainfall, and their derived factors were screened using grey relational analysis and the variance inflation factor. A direct multi-step forecasting strategy was adopted to compare the performance of CNN, LSTM, Transformer, and TCN in predicting the periodic component. The trend component was predicted using a rolling polynomial model, and the predicted components were subsequently combined to reconstruct the total landslide displacement. [Results] Prediction-horizon sensitivity analysis indicated that a 7-day horizon achieved an appropriate balance between prediction accuracy and early-warning timeliness. For periodic-component prediction, the TCN achieved an average coefficient of determination (R²) of 0.9563 and an average root mean square error (RMSE) of 0.6431 mm, outperforming the other three models overall. The R² values for trend-component prediction ranged from 0.9968 to 0.9996. For total-displacement prediction, R² ranged from 0.9513 to 0.9940, while RMSE ranged from 0.6749 to 1.0076 mm. [Conclusions] SSA-VMD effectively separates the displacement sequence into components at different temporal scales, and the TCN exhibits a strong ability to characterize the nonlinear temporal features of the periodic component. The proposed method provides a useful reference for short-term displacement forecasting and early warning of reservoir-bank landslides.