<p>Existing approaches to forecasting landslide displacement primarily focus on time series predictions of surface displacement, while neglecting analysis and prediction of displacement trend changes within the landslide’s internal structure. This limits the ability to comprehensively reflect landslide deep movement. Therefore, this paper presents a spatial–temporal prediction model to forecast deep displacement using shallow displacement data. Firstly, the cumulative displacement data of landslides are decomposed into high- and low-frequency displacements and fluctuating displacements by variational modal decomposition (VMD). Secondly, the long and short-term memory network model (LSTM) is optimized using the grey wolf optimization algorithm (GWO), incorporating the spatial correlation of the shallow displacement and the temporal correlation of the historical measurements as feature inputs. The displacement components are modeled and predicted, respectively. Finally, the high- and low-frequency displacement results are reconstructed to obtain the final predicted cumulative landslide displacement. The model presented in this paper demonstrates high predictive accuracy at all three measurement points of a reservoir landslide located in the upper reaches of the Yellow River, with <i>R</i>^<i>2</i> exceeding 0.96. The proposed model outperforms SVR, RF, BiLSTM, and CNN models in prediction accuracy. The results show that the method can be used for deep displacement prediction in landslides, revealing the spatio-temporal characteristics of the landslide interior, and offering valuable insights for further research on internal spatial displacement in landslides.</p>