Highlights What are the main findings? Variations in hazard-prone environments dominate the spatial heterogeneity of multi-hazard distribution. Thermal hazard susceptibility is expected to increase greatly by the end of the century due to permafrost degradation. What is the implication of the main findings? Segmented assessment can effectively improve evaluation accuracy and model interpretability. Thermal hazards exhibit significant sensitivity to climate change, while gravity hazards do not.Highlights What are the main findings? Variations in hazard-prone environments dominate the spatial heterogeneity of multi-hazard distribution. Thermal hazard susceptibility is expected to increase greatly by the end of the century due to permafrost degradation. What is the implication of the main findings? Segmented assessment can effectively improve evaluation accuracy and model interpretability. Thermal hazards exhibit significant sensitivity to climate change, while gravity hazards do not.Abstract With climate change, the Qinghai-Tibet Highway (QTH) is facing increasingly severe risks of natural hazards, posing a significant threat to its normal operation. However, the types, distribution, and future risks of hazards along the QTH are still unclear. In this study, we established an inventory of multi-hazards along the QTH by remote sensing interpretation and field validation, including landslides, debris flows, thaw slumps, and thermokarst lakes. The QTH was segmented into three sections based on hazard distribution and environmental factors. Susceptibility modelling was performed for each hazard within each using machine learning models, followed by further evaluation of hazard susceptibility under future climate change scenarios. The results show that, at present, approximately 15.50% of the area along the QTH exhibits high susceptibility to multi-hazards, with this proportion projected to increase to 20.85% and 23.32% under the representative concentration pathways (RCP) 4.5 and RCP 8.5 distant future scenarios, respectively. Variations in hazard-prone environments dominate the spatial heterogeneity of multi-hazard distribution. Gravity hazards demonstrate limited sensitivity to climate change, whereas thermal hazards exhibit a more pronounced response. Our geomorphology-based segmented assessment framework effectively enhances evaluation accuracy and model interpretability. The results can provide critical insights for the operation, maintenance, and hazard risk management of the QTH.

Soil erosion on highway side-slope has been recognized as a cause of environmental damage and a potential threat to road embankments in the high-altitude permafrost regions. To assess the risk to roads and to protect them effectively, it is crucial to clarify the mechanisms governing roadside erosion. However, the cold climate and extremely vulnerable environment under permafrost conditions may result in a unique process of roadside erosion, which differs from the results of current studies conducted at lower altitudes. In this study, a field survey was conducted to investigate side-slope rill erosion along the permafrost of a highway on the Qinghai-Tibet Plateau of China. Variations in erosion rates have been revealed, and intense erosion risks (with an average erosion rate of 13.05 kg/m2/a) have been identified on the northern side of the Tanggula Mountains. In the case of individual rills, the detailed rill morphology data indicate that the rill heads are generally close to the slope top and that erosion predominantly occurs in the upper parts of highway slopes, as they are affected by road surface runoff. In the road segment scale, the Pearson correlation and principal component analysis results revealed that the protective effect of vegetation, which was influenced by precipitation, was greater than the erosive effect of precipitation on roadside erosion. A random forest model was then adopted to quantify the importance of influencing factors, and the slope gradient was identified as the most significant factor, with a value of 0.474. Accordingly, the integrated slope and slope length index (L0.5S2) proved to be a reliable predictor, and a comprehensive model was built for highway side-slope rill erosion prediction (model efficiency = 0.802). These results could be helpful for highway side-slope conservation and ecological risk prediction in alpine permafrost areas.