Landslide susceptibility prediction considering rock integrity and stress state: a case study

Landslide is a major disaster threatening the safety and orderly production of an open-pit mine, so slope stability evaluation is of great significance to the support and monitoring arrangement. Landslide susceptibility mapping (LSM) was widely used in landslide prediction. The former research focused on the algorisms to improve its accuracy, which is relatively complete and left little room for further improvement. In this paper, new factors, including RQD and numerical simulation (NS), are selected to solve the limitation of traditional LSM on the integrity and stress state of the slope. The RQD value was obtained by machine learning and converted into rasters by the ordinary Kriging interpolation method. The slope stress was calculated by the finite difference method and converted into raster data using a program written by Fish language. Based on the information value (INV) method, gradient boosting decision tree (GDBT) was used as the main algorism to generate the LSM-NS. Finally, because LSM-NS contains landslides that have already occurred and those in high susceptibility due to its stress state, commonly used validation methods such as AUROC could no longer be used. Multiple validation methods were applied, such as stress monitoring and UAV tilt photography. The result indicates that the stress increases with crack generating in the high susceptibility area of LSM-NS, where traditional LSM could not predict. Therefore, the addition of RQD and NS could further improve the accuracy using existing algorism. LSM-NS is recommended as the more suitable model for landslide susceptibility assessment in a small area due to its excellent accuracy and efficiency.