Identification and assessment of avalanche hazards in Aerxiangou section of Duku expressway in TianShan mountainous region based on unmanned aerial vehicle photography

Abstract

In this study, avalanches in the Aerxiangou section of the Duku Expressway in the Tianshan Mountain area of Xinjiang were taken as the research object, and 92 avalanches were accurately identified through onsite research. A high-resolution three-dimensional model was established by collecting images from unmanned aerial vehicles for an in-depth understanding of the avalanche danger of the region, according to the sample set selection of different uses of machine learning support vector machines to establish the S1-RBFKSVM, S1-PKSVM, S2-RBFKSVM, and S2-PKSVM avalanche susceptibility coupling models. On the basis of the avalanche point susceptibility, the impact velocity, impact force, avalanche volume, and throw distance constitute the hazard evaluation system. The study results revealed that slopes in the range of 26.6°–46.9° are more prone to avalanches, and sample set 2 improved the accuracy by approximately 30% compared with sample set 1 trained in the avalanche susceptibility model. Principal component analysis revealed a total of 16 high-risk avalanches, which were distributed mainly on the southern side of the route. This study provides data support for avalanche simulations as well as early warning and prevention and provides theoretical and methodological guidance for the construction and operation of the Duku Expressway.