Feature importances from XGBoost model.

Landslides threaten sustainable development through economic and human losses. This study integrates machine learning methods to construct susceptibility maps, including topographic-hydrological indicators, to improve the inclusion of earthflow landslides. Furthermore, we aim to find relationships between landslide susceptibility and social lag using Copula models and SHAP values. Results reveal differentiated dependence across different partitions. Specifically, we found regime-specific co-occurrences of high social lag and high landslide susceptibility areas in steep, deprived areas, contrasting resilient affluent zones. Educational deprivation emerges as the top vulnerability factor, followed by healthcare access, overcrowding, and housing deficits. Highlighting spatial inequities, the analysis advocates targeted interventions blending slope stabilization and social policies.