Displacement prediction and failure mechanism analysis of rainfall-induced colluvial landslides

Rainfall is the primary cause of colluvial landslides and constitutes approximately 80% of such events. Colluvial landslides are affected by seasonal rainfall patterns and typically exhibit progressive deformation. The causes of these landslides are complex and their destructive mechanisms cannot be controlled easily, thus resulting in catastrophic events. Accurate prediction of the displacement of rainfall-induced colluvial landslides is crucial for mitigating the associated risks. In this study, we consider the Wufeng landslide as an example to elucidate quantitative correlations between rainfall factors and deformation characteristics via the Spearman correlation analysis. Based on seven years of continuous displacement monitoring data, we develop a rainfall-induced colluvial landslide displacement prediction model using the improved complete integrated empirical mode decomposition with adaptive noise (ICEEMDAN) method and a sparrow search algorithm (SSA)-optimized long short-term memory (LSTM) neural network. Furthermore, we examine the progressive deformation mechanisms of rainfall-induced accumulation landslides based on fluid–solid coupling simulations in FLAC3D, supplemented by field investigations and deformation monitoring. The results indicate that (1) cumulative rainfall over 22 d and effective rainfall over 28 d constitute the primary triggering factors for the Wufeng landslide; (2) the ICEEMDAN-SSA-LSTM hybrid model demonstrates outstanding predictive accuracy for rainfall-induced displacement patterns, particularly in characterizing the correlation between intermittent displacements and rainfall signatures; (3) the pipe network infiltration system in the Wufeng landslide establishes preferential seepage pathways, where coupled fluid–solid interactions between infiltration pressure and anti-sliding resistance generate a distinctive “preferential flow–subduction–resistance” deformation sequence. These findings provide a theoretical basis for enhancing early warning systems for rainfall-induced colluvial landslides and offer a new perspective for analyzing water-related landslide deformations worldwide.