A new framework for landslide susceptibility mapping in contiguous impoverished areas using machine learning and catastrophe theory.

Landslides are among the most frequent and dangerous geological disasters worldwide, making accurate landslide susceptibility mapping (LSM) crucial for effective disaster prevention. This study introduces a novel LSM framework by integrating random forest (RF), support vector machine (SVM), and catastrophe theory (CT), and applies it to the contiguous impoverished areas of Liangshan, Sichuan. First, we selected 12 factors representing both internal environmental and external triggering conditions to assess landslide susceptibility. The frequency ratio method was used to assess the correlation between historical landslides and these factors. Second, CT was integrated into the RF- and SVM-based LSM models, resulting in two integrated models (RF-CT and SVM-CT) for generating LSM in the region. Finally, the receiver operating characteristic curve was used to evaluate and compare the accuracy of the methods. The results show that the RF-CT and SVM-CT frameworks performed well, with a 10% improvement in the success rate (0.899 for RF-CT and 0.873 for SVM-CT), and a 5% improvement in the prediction rate (0.783 for RF-CT and 0.775 for SVM-CT) compared with the individual RF and SVM models. These findings provide valuable insights for disaster prevention, poverty alleviation, and sustainable development in the study area.