Objectives The complex geological conditions in the mountainous areas of western China, with strong internal and external dynamics effects, make catastrophic landslides frequent. The analysis of landslide susceptibility has become a necessary means for scientific early warning and active prevention before disasters.In the traditional landslide susceptibility analysis method, the general calculation accuracy of the single knowledge-driven model is limited, and the weight of the impact factor is highly subjective. The data-driven model also relies too much on the quality and quantity of sample data, and the heterogeneity of the landslide disaster environment is prominent.

Methods In order to overcome the problems of limited quantity and quality of sample data and large differences in landslide disaster environment, we propose a regional landslide susceptibility method that couples the contribution weight of landslide disaster environmental factors and heuristic knowledge fuzzy logic model. The proposed method uses spatial statistical indicators such as the historical landslide frequency ratio and the information entropy weight of the landslide disaster environmental factors to explicitly describe the contribution and spatial distribution characteristics of the landslide disaster environmental factors, which measures the constraint relationship and the mapping structure between multi-factors and landslides, and realizes multi-factor coupling regional landslide susceptibility.

Results The experiment selects the disaster-prone areas in Fengjie, Chongqing for verification and evaluation; The results show that the proposed method has a more uniform and reasonable partition area, with an area under curve(AUC) value of 0.854, and the best prediction accuracy, than single information value(IV) model and information value and logistic regression(IVLR) model, which ensures the reliability and accuracy of the method.

Conclusions The proposed method overcomes the strict requirements of landslide susceptibility analysis on the number of historical observation samples, improves the accuracy of landslide susceptibility analysis through a hierarchical stacking strategy, and provides reliable technical support for the susceptibility analysis of large-scale.