GIS-based spatial prediction of landslide using road factors and random forest for Sichuan-Tibet Highway

Accurate evaluation of landslide susceptibility is very important to ensure the safe operation of mountain highways. The Sichuan-Tibet Highway, which traverses the east of the Tibetan Plateau, frequently encounters natural hazards. Previous studies generally use statistical methods to analyze the hazards along the Sichuan-Tibet Highway. In this research, we present two road factors, namely aspect to road and road profile to increase the accuracy of landslide susceptibility mapping by considering the influence of landslide movement direction on road. First, the aspect to road, which represents the impact of different landslide movement directions on the highway, was extracted by combining road direction with mountain aspect. Then, the road profile, which reflects the subgrade structure between the road and surrounding mountains, was extracted according to the terrain data. Finally, the landslide susceptibility maps were produced based on the random forest (RF) method by using 473 landslides and 10 conditioning factors, including road factors (aspect to road, road profile) and primitive factors (slope, aspect, curvature, relief amplitude, peak ground acceleration, crustal movement velocity, faults, rainfall). The area under the receiver operating characteristic curve (AUC) and the Gini importance were used to evaluate the performance of proposed road factors. The AUC values on two groups that add road factors and only use primitive factors were 0.8517 and 0.8243, respectively. The Gini importance indicated that road profile (0.123) and aspect to road (0.116) have a significant contribution to landslides compared with the primitive factors. The results of multi-collinearity analysis and frequency ratio confirmed the suitability of the road factors for predicting hazards along the highway.

Automated summary

The research introduced road factors of aspect to road and road profile to improve the accuracy of landslide susceptibility mapping by considering the influence of landslide movement direction on road. Using the random forest method, landslide susceptibility maps were generated, and the performance of road factors was evaluated using AUC and Gini importance. The results showed that road profile and aspect to road made significant contributions to predicting landslides along the highway compared to primitive factors.