A landslide susceptibility modeling method using an optimal parameters-based geographical detector

Abstract

Current methodologies for modeling landslide susceptibility using machine learning algorithms typically involve grading landslide conditioning factors, selecting major factors, and calculating model input data. However, these approaches are often highly subjective and random, requiring sequential, step-by-step calculations that significantly reduce modeling efficiency. To address these limitations, this study proposes a novel landslide susceptibility modeling method based on an Optimal Parameters-based Geographical Detector (OPGD). Leveraging the theory of spatial stratified heterogeneity, the proposed method determines the optimal grading strategy for conditioning factors and ranks the importance of all factors. Furthermore, this study introduces risk detection metrics derived from the OPGD model as inputs to the susceptibility model, enabling simultaneous factor grading and model input data calculation. The effectiveness of the proposed approach is validated through six models based on two machine learning algorithms—Random Forest (RF) and Support Vector Machine (SVM)—across three regions of the Loess Plateau. The results demonstrate that the proposed method achieves exceptional landslide prediction performance, with AUC values consistently close to or exceeding 0.9. Notably, the RF-based model outperforms others, achieving a maximum AUC value of 0.978. The proposed method is not only straightforward to implement but also provides an interpretable modeling process, making it applicable to any region requiring landslide susceptibility analysis. Overall, this study significantly enhances modeling efficiency, bridges the gap between theoretical knowledge and practical applications, and advances the standardization of landslide susceptibility modeling.