Landslide susceptibility modeling using data-driven weights-of-evidence based random forest and radial basis function network
- 1Xi’an University of Science and Technology, College of Geology and Environment, China (19209071022@stu.xust.edu.cn)
- 2Xi’an University of Science and Technology, College of Geology and Environment, China (chenwei0930@xust.edu.cn)
Xiaojin County, Sichuan Province, China was selected as the study area of this paper, and twelve conditioning factors were determined according to the literature review. The spatial correlation between landslide and conditioning factors is analyzed using the weights-of-evidence (WoE) model, and the landslide susceptibility in Xiaojin county is predicted. The landslide susceptibility in this region was mainly assessment by WoE based random forest (RF) model. The radial basis function network (RBFNetwork) model was also exploited to map landslide susceptibility with the identical datasets. Finally, the landslide susceptibility maps were produced, and the comprehensive performance of the three models was quantitatively evaluated and compared by the receiver operating characteristic (ROC) curves and area under curve (AUC) values. The results show that the three models are suitable for landslide susceptibility evaluation in the study area, and the evaluation effect of the WoE model is better than that of the RF and RBF network models. More concretely, the goodness-of-fit values of the WoE, RF and RBFNetwork models in the training dataset are 0.899, 0.880 and 0.866, respectively. In terms of prediction accuracy, AUC values are 0.892, 0.874 and 0.863 respectively. Additionally, mean decrease accuracy (MDA) and means decrease Gini (MDG) are used to quantify the importance of landslide conditioning factors. Elevation, soil, distance to roads and distance to rivers are considered as the most important conditioning factors in landslide susceptibility modeling. Consequently, the study achievements in this paper have reference significance on the development and exploitation of land resources in Xiaojin County.
How to cite: Zhao, X. and Chen, W.: Landslide susceptibility modeling using data-driven weights-of-evidence based random forest and radial basis function network, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-7006, https://doi.org/10.5194/egusphere-egu24-7006, 2024.
Comments on the supplementary material
AC: Author Comment | CC: Community Comment | Report abuse