Condition monitoring of heterogeneous landslide deformation in spatio-temporal domain using advanced graph attention network

Real-time monitoring of landslide deformation patterns is critical for effective hazard forecasting and risk mitigation. Field observations reveal that deformation processes are typically uneven and heterogeneously distributed across slope bodies, creating dynamic uncertainties that challenge prediction models. This research aims to develop an enhanced spatial-temporal monitoring system capable of capturing these complex deformation patterns. In this study, it presents a novel Graph Attention Network (GAT) framework that integrates multi-scale temporal embeddings, adaptive graph learning, and temporal self-attention mechanisms to simultaneously track localized stability variations and global deformation trends across monitoring points. The framework’s key innovation is its ability to detect transitions from homogeneous to heterogeneous deformation states, implementing graph-level rather than traditional node-level alarm systems. Validation using datasets from three landslides in China (Muyubao, Baishuihe, and Shuping) demonstrates that our approach significantly outperforms existing methods in identifying heterogeneous deformation states. This research advances landslide early warning systems by improving the detection of spatially variable deformation patterns, ultimately enhancing risk assessment and mitigation strategies for landslide-prone regions.