Key predisposing factors and susceptibility assessment of landslides along the Yunnan–Tibet traffic corridor, Tibetan plateau: Comparison with the LR, RF, NB, and MLP techniques

Abstract

The Yunnan–Tibet traffic corridor runs through the Three Rivers Region, southeastern Tibetan Plateau, which is characterized by high-relief topography and active tectonics, with favourable conditions for landslides. It is of great significance to identify the key predisposing factors of landslides and to reveal the landslide susceptibility in this area. A total of 2,308 landslides were identified as learning samples through remote sensing interpretation and detailed field surveys, and four machine learning algorithms involving logistic regression (LR), random forest (RF), naïve Bayes (NB) and multilayer perceptron (MLP) were compared to model the landslide susceptibility. Through the multicollinearity test, 13 influential factors were selected as conditioning factors. The area under the curve (AUC) values of LR, RF, NB and MLP models are .788, .918, .785 and .836 respectively, indicating that the four models have good or very good prediction accuracy in landslide susceptibility assessment along the Yunnan–Tibet traffic corridor. In addition, the elevation, slope, rainfall, distance to rivers, and aspect play a major role in landslide development in the study area. The susceptibility zoning map based on the best RF model shows that the areas with high susceptibility and very high susceptibility account for 12.24% and 6.72%, respectively, and are mainly distributed along the Jinsha River, the Lancang River and the G214 highway.