Abstract:
Objective Landslides occur frequently in Leiyang city, Hunan province, posing a serious threat to people’s lives, property, and ecological security.
Methods To improve the accuracy of landslide susceptibility assessment, this study took Leiyang city, Hunan province as the study area. An information value (IV)-random forest (RF) coupling model (IV-RF model) was constructed, and a spatially constrained sampling strategy was introduced to optimize the negative sample selection strategy for landslide susceptibility assessment. The three models—IV, RF, and IV-RF—were comparatively analyzed using the ROC curve and the AUC value. Additionally, a composite performance index was proposed to comprehensively assess model performance.
Results The results showed that: 1) the IV-RF coupling model outperformed the single models, with an AUC value of 0.952 and a composite performance index (Accuracy + F1-score + MCC) of 2.593. Landslide points were densely distributed in the very high- and high-susceptibility zones, while very few were found in the very low- and low-susceptibility zones, verifying that the model had high spatial prediction accuracy. 2) The engineering geological rock group factor was identified as one of the most critical assessment factors affecting landslide development in the study area.
Conclusions The IV-RF coupling model effectively integrates the quantitative data interpretation capability of the IV model with the nonlinear identification capability of RF model, which can effectively improve the model’s identification accuracy. The findings of this study can provide a scientific basis for landslide disaster risk prevention and control, soil and water conservation, and territorial spatial planning within the study area.
下载: