Landslide Susceptibility Evaluation Based on Coupling of GBDT-LR Model and Information Model

Abstract

[Objective] The accurate and rapid landslide susceptibility zoning method were studied in order to provide a reference for regional safety monitoring, and provide a scientific basis for the government to control landslide disasters. [Methods] The study was conducted in the Guichi District of Chizhou City, Anhui Province. The coupled model of gradient boosting decision tree-logistic regression (GBDT-LR) and an information value (I) model was used to determine the evaluation of regional landslide susceptibility. The model learns from the original samples and combines them to generate new simulation samples in order to enhance the fitting ability of the model to evaluate landslide susceptibility. The Borderline-Smote algorithm was used to solve the problem of sample data asymmetry. The slope unit divided by r.slopeunits software was selected as the minimum evaluation unit, and a total of 10 evaluation factors were selected: slope gradient, slope aspect, terrain curvature, profile curvature, plane curvature, topographic wetness index (TWI), topographic relief, normalized difference vegetation index (NDVI), distance from fault, and distance from river. The landslide susceptibility model was evaluated from three aspects: frequency ratio, density of landslide disaster points and hidden danger points, and the receiver operating characteristic (ROC) curve. [Results] The experimental results showed that the frequency ratio of the coupled model I-GBDT-LR was 10%, 13%, and 7% greater than that of the I, LR, and I-LR models, respectively. The density of landslide disaster points and hidden danger points in the high risk area increased by about 9, 11, and 7, respectively, and the ROC accuracy increased by about 10%, 9%, and 5%, respectively. [Conclusion] The accuracy of the coupled model was higher than that of the single model, and the accuracy of the coupled model proposed was higher than that of the I-LR coupled model, which provides an effective and new evaluation method for landslide susceptibility evaluation.

Keywords

• landslide susceptibility
• information value
• logistic regression
• gradient boosting decision tree-logistic regression (gbdt-lr)
• chizhou city of anhui province