Dynamic landslide susceptibility mapping using swarm intelligence and multi-temporal InSAR

Accurate landslide susceptibility mapping is key to effectively mitigating risks in mining-affected environments; however, most existing approaches rely on static assumptions and single-dimension optimization, limiting their ability to capture evolving slope instability. To overcome this limitation, the present study proposed a hybrid framework combining swarm intelligence-optimized machine learning with multi-temporal InSAR technology for dynamic landslide susceptibility mapping. Static landslide susceptibility maps (LSMs) were first produced utilizing SI-MaxEnt-ANN models, in which MaxEnt was employed for conditioning factor screening, discretization, and non-landslide sample extraction to improve input data quality, while three swarm intelligence algorithms, including Honey Badger Algorithm, Sailfish Optimizer, and Gray Wolf Optimizer, were used to optimize ANN parameters and enhance global search efficiency and predictive robustness. Subsequently, surface deformation rates derived from Stacking-InSAR were incorporated through an update matrix, enabling LSM from static patterns to dynamical representations. Evaluation results show that the HBA-MaxEnt-ANN model outperformed the conventional ANN by 11.3% (AUC = 0.959). Interpretability analysis identified altitude, distance to roads, distance to goafs, STI, lithology, and slope as the dominant controlling factors. Validation across three representative areas demonstrates that the dynamic LSM effectively captures active landslide processes. The proposed framework aids hazard governance and sustainable development for resource-based cities.