Topographic stress proxy as a new causative factor in landslide susceptibility mapping

On the Tibetan Plateau, the frequency of bedrock landslides has significantly increased due to high altitude, high seismic intensity, and high topographic stress, posing severe threats to human safety and infrastructure. Landslide susceptibility assessment (LSA) is an effective tool for preventing and managing landslide hazards. Although the extent of slope failure depends on landscape-scale patterns of bedrock fracturing induced by topographic stress, current LSA models rarely incorporate this factor and fail to adequately link the causative factors considered with the mechanisms of landslide formation. This study introduces a new factor—topographic stress—into existing LSA models to capture the mechanisms underlying landslide formation better. Several popular machine learning models (LGR, MLP, SVM, and RF) were employed to investigate the impact of topographic stress proxies on landslide susceptibility assessments. Results reveal that stress proxies outperform conventional factors across models, with FP recalibrating susceptibility distributions by reducing High/Very High-risk zones by 3.1 ∼ 9.1 % and expanding Low/Very Low areas by 5.3 ∼ 14.9 % in both coseismic and pre-seismic scenarios. Model performance metrics significantly improved: overall accuracy increased by 3 ∼ 6 %, recall by 0.5 ∼ 1.6 %, and AUC values by 2 ∼ 4 %, with RF_FP and SVM_FP achieving peak discriminative power (AUC: 0.898 and 0.858). The topographic stress proxy FP bridges geomechanical processes (stress-fracture-weathering interactions) and data-driven modeling, identifying fracture zones that precondition slopes for failure—particularly near active faults where stress accumulation elevates landslide risk by amplifying fracture density and strain localization. This framework enables proactive hazard mitigation in tectonically active orogens by prioritizing physics-informed stress preconditioning over trigger-centric approaches. Future applications should leverage stress proxies to decode landscape-scale fracture evolution and cascading slope destabilization mechanisms, advancing toward predictive, process-based landslide susceptibility mapping.