Prediction and pre-warning of step-like landslide displacement based on deep learning coupled with ICEEMDAN

• The landslide displacement is decomposed and reconstructed to ensure the prediction accuracy. • The step-like landslide displacement prediction model was proposed. • A pre-warning system was developed for step-like landslides. • The impact of landslide displacement prediction error on pre-warning was explored. Characterized by abrupt and creep displacements, prediction and pre-warning of step-like landslides pose significant challenges. To address this issue, we proposed a novel approach that integrated landslide displacement prediction with pre-warning systems. This method involved decomposing the landslide displacement into trend and periodic components to enhance accuracy. A cubic polynomial combined with a Dense Convolutional Network and a Long Short-Term Memory Network was adopted to predict the trend and periodic displacements, which were then combined for the final estimation. Additionally, a pre-warning index system was developed, considering the cumulative landslide displacement effects, including revised tangential angle thresholds, tangential angle variations, and duration considerations. The effectiveness of this model and the pre-warning index were validated through an analysis of the Baishuihe and Bazimen landslides. The results indicated that the proposed model accurately simulated the displacements and provided reliable landslide pre-warning, offering a viable solution for predicting and warning about step-like landslides.