Identification and distribution characteristics of potential freezethaw-induced landslides using remote sensing and deep learning: Insights from Qinghai-Tibet Engineering Corridor

Abstract

Potential freezethaw-induced landslides pose a significant threat to regional ecosystems and the stability of engineering structures. Understanding their spatial distribution and influencing factors is critical for future infrastructure planning on the Qinghai-Tibet Plateau. However, the distribution characteristics, driving factors, and influence mechanisms of such landslides remain poorly understood. In this study, we employed deep learning techniques and remote sensing imagery to identify potential freezethaw-induced landslide hazards in the permafrost zone along the Qinghai-Tibet Engineering Corridor, and to investigate the environmental variables influencing their distribution. The results indicate that: (1) A total of 131 potential freezethaw-induced landslides were identified, primarily located on the shaded sides of gentle slopes at elevations between 4500 and 5100 m, where near-surface temperatures range from \(-4^\circ\)C to \(-1^\circ\)C. In addition, approximately 73\(\%\) of these hazards are distributed in areas with an average annual precipitation of 300 to 330 mm. (2) Rainfall and slope gradient exert strong control over the spatial distribution of these hazards. Rainwater infiltration increases soil moisture in the permafrost layer and reduces the shear strength of the geomaterials. Moreover, precipitation can accelerate the thawing of the active layer and ground ice, potentially triggering sliding or subsidence on slopes of susceptible gradients. These findings provide important guidance for future infrastructure planning, resource development, and environmentally sustainable development on the Qinghai-Tibet Plateau.