Spatial pattern and driving mechanisms of geohazards in the lower yarlung zangbo suture zone: implications for engineering construction in extreme mountainous environments

This study established a database of 492 hazard points (2013–2022) and developed a PCA-LR hybrid model to address multicollinearity and subjectivity in susceptibility assessment. Principal Component Analysis (PCA) was used for dimensionality reduction, followed by Logistic Regression (LR) for sensitivity-based weight refinement.Results reveal significant spatial clustering.

Notably, 29.5% of hazards occur on gentle valley slopes (0°–13°), identifying these as accumulation zones for high-altitude long-runout events. Although slope aspect showed low variance in PCA, it demonstrated high sensitivity in the LR model; hazard density on windward slopes is 2.3 times higher than on leeward slopes, highlighting critical microclimatic effects. Furthermore, 83% of hazards cluster within 1,000 m of road networks due to the coupling of unloading fractures and fluvial erosion.