Micro-macro spatiotemporal multi-graph network model for landslide displacement prediction

The precise prediction of landslide displacement is crucial for effective geological disaster prevention and management. Existing models predominantly focus on temporal prediction, often neglecting the intricate spatiotemporal deformation characteristics of landslides. To address this gap, this study proposed a micro-macro spatiotemporal multi-graph network model (MM-STMGN) to analyze landslide deformation from micro-macro perspectives, enhancing spatiotemporal fusion prediction performance. The model extracts data on internal seepage in order to enhance the dataset. This involves integrating multiple heterogeneous spatiotemporal datasets that combine external influencing factors and spatiotemporal information on landslide displacement. By leveraging multiple graphs, it effectively captured the diversity of micro-scale and regional macro-scale spatial characteristics of landslide deformation, including spatial adjacency and deformation pattern correlations. Hierarchical Graph Neural Networks (HGNNs) and spatial attention networks were employed to adaptively process these micro-macro spatial features, while the Temporal Fusion Transformer (TFT) dynamically captured global and local temporal dependencies of landslide displacement. The micro-macro fusion module further processed aforementioned multiple heterogeneous datasets, achieving accurate prediction of landslide displacement within complex multidimensional spatiotemporal relationships. Applied to a landslide in the Three Gorges Reservoir area, MM-STMGN outperformed MLP, LSTM, and ST-GCN models across multiple evaluation metrics (MAE, MAPE, RMSE, R2) and various predictive performance aspects. Ablation experiments indicate that incorporating micro-macro deformation features and seepage factors can significantly enhance prediction performance of landslide displacement. The research findings provide a reliable and advanced approach for landslide disaster prevention and mitigation.