Landslide susceptibility based on statistical and machine learning methods: a case study of western Tibet plateau

The evaluation of landslide exposure plays a crucial role in estimating the risks associated with landslides and debris flows in a specific region, providing valuable insights for effective prevention and mitigation of geological hazards. The western Tibetan Plateau was chosen for this study from human interferences, and then this paper can obtain the comparison between the statistical and machine learning methods. Seven landslide factors were applied for the landslide susceptibility maps, including the slope, aspect, lithology, distance to faults, distance to rivers, distance to roads and elevation. In this study, the Information Value Model (IVM) and weight of evidence method were employed in conjunction with Logistic Regression (LR) and Multi-Layer Perceptron (MLP), utilizing IVM-LR, WOE-LR, IVM-MLP, and WOE-MLP approaches, to assess landslide hazards. The study area was divided into five hazard grades, namely very high, high, moderate, low, and very low, based on the generated susceptibility maps. The credibility level of all susceptibility maps produced by the models exceeded 85%, as revealed by a comparative analysis of Receiver Operating Characteristic (ROC) curves. Notably, the IVM-LR model exhibited superior performance in assessing landslide susceptibility in the study area.