<p>Landslides are among the most prevalent and hazardous geological disasters worldwide. In particular, their frequency and severity are significant in China. Traditional landslide prediction models often struggle to address the phased and abrupt displacement patterns of step-type landslides. These landslides undergo significant deformation during the rainy season, while exhibiting relatively mild changes during non-rainy periods. This seasonal variability makes traditional static models inadequate for capturing complex, nonlinear dynamic processes. To overcome these limitations, this study proposes a novel landslide displacement prediction method based on the VMD-Mamba model. The Variational Mode Decomposition (VMD) algorithm decomposes the original displacement data into trend and periodic components, effectively capturing trend and periodic variations. The Mamba model uses the time-series information in the data to build a predictive model. This integrated approach enhances the model's ability to predict sudden displacement changes and offers a robust solution for step-type landslide prediction, addressing the challenges posed by nonlinear and dynamic processes in landslide monitoring. The results indicate that the proposed model achieves a landslide displacement prediction fit (R²) greater than 0.97. During periods of rapid deformation, its predictive performance significantly surpasses other intelligent forecasting methods such as IPSO-LSTM model, CNN-LSTM model and transformer models. The main conclusion of this study is that the VMD-Mamba model effectively captures the dynamic nonlinear characteristics of landslide displacement, offering a novel approach to improving landslide prediction accuracy. This advancement holds significant potential for application in landslide early warning and disaster prevention, providing a valuable tool for landslide risk management on a global scale.</p>