An improved potential landslide hazard points evaluating method considering the heterogeneity of environmental features

Landslide causes great damage to the environment and threatens the safety of people's life and property. The environmental features show heterogeneous changes in an area, but it is rarely used in the evaluation of landslide susceptibility. In this study, cluster analysis and ensemble learning were combined to verify the importance of environmental heterogeneity, and an improved landslide susceptibility mapping method (FCM-RF) was proposed considering the heterogeneity of environmental features. Taking Qichun County, China, as a study area, this paper analyzed the correlation and importance of thirteen landslide causal factors. The landslide and non-landslide samples were used for cluster analysis using k-means and fuzzy C-means, respectively. The study area was divided into several areas, and the type represented by each area was used as one of the input factors of the landslide model. Three methods, random forest, logistic model tree and Bagging, were selected to establish landslide susceptibility models. Finally, receiver operation curve and five statistic indicators (accuracy, precision, recall, specificity and F-measure) were employed to compare and validate the performance of these methods. The results showed that the accuracy of the three models was improved by about 2% after considering the heterogeneity factor and the FCM-RF (AUC = 0.8872) model with better performance was used for landslide susceptibility mapping. Therefore, considering the heterogeneity of environmental features can effectively improve the performance of landslide model, which will be helpful for landslide prediction.

Automated summary

This study proposes an improved landslide susceptibility mapping method (FCM-RF) considering the heterogeneity of environmental features by combining cluster analysis and ensemble learning. The method was applied in Qichun County, China. The results show that the FCM-RF model performs better in landslide susceptibility mapping after considering the heterogeneity factor.