Landslide susceptibility assessment along Sheringal Road, Upper Dir Northern Pakistan using AHP and bivariate statistical models

Abstract

Landslides pose a persistent threat to transportation infrastructure and mountain communities in northern Pakistan; however, detailed corridor-scale susceptibility information remains limited for Upper Dir District. This study develops and comparatively evaluates landslide susceptibility maps along the Sheringal–Kumrat Road using the Analytical Hierarchy Process (AHP), Frequency Ratio (FR), and Statistical Index (SI) models within a consistent geospatial framework. A multi-temporal inventory comprising 90 landslides was prepared using satellite-image interpretation, documentary information, and field observations. An equal number of non-landslide locations was selected, which were divided into 75% training and 25% testing subsets. Eight landslide-conditioning factors slope, elevation, aspect, curvature, profile curvature, lithology, drainage density, and relative relief were prepared at a spatial resolution of 12.5 m. Spearman’s rank correlation and Variance Inflation Factor (VIF) were used to assess predictor redundancy and multicollinearity; VIF values ranged from 1.01 to 2.43, indicating no problematic multicollinearity among the selected factors. The resulting susceptibility indices were classified into five levels ranging from very low to very high susceptibility and evaluated using an independent testing subset. The SI model achieved the highest ROC–AUC (0.903), followed by FR (0.881) and AHP (0.847), demonstrating excellent discrimination for SI and good discrimination for FR and AHP. Spatially, the three models consistently identified the northwestern part of the corridor as the principal susceptibility hotspot, where steep and highly dissected terrain, dense drainage networks, high relative relief, and comparatively weak lithological units coincide. Notably, the SI model classified only 12.2% of the study area as very high susceptibility while capturing 57.7% of the mapped landslides within this class. These quantitative and spatial results demonstrate that the modelling framework can provide an effective screening tool for prioritizing slope monitoring, drainage improvement, detailed geotechnical investigation, and road-maintenance interventions in data-constrained mountainous environments.