Enhancing Landslide Susceptibility Mapping by Integrating Neighboring Information in Slope Units: A Spatial Logistic Regression

Landslide susceptibility mapping (LSM) is a vital tool for proactive disaster mitigation. Although numerous studies utilize slope units (SUs) for LSM, the limited integration of adjacency information, including spatial autocorrelation, often reduces predictive accuracy. In this study, GRASS GIS was utilized to generate slope units, and a spatial logistic regression (SLR) model was developed to incorporate the adjacency information of the slope units to predict the landslide susceptibility. Then, the spatial stratification heterogeneity patterns of landslide susceptibility were analyzed using GeoDetector. The results showed that the SLR model achieved an area under the curve (AUC) of 0.89, a notable improvement of 0.26 compared to the traditional logistic regression (LR) model that does not incorporate adjacency information. This indicates that incorporating adjacency information effectively enhances LSM accuracy by mitigating spatial autocorrelation. Furthermore, lithology, PGV, and distance to the epicenter were identified as the primary factors contributing to the formation of the spatial stratification heterogeneity of landslide susceptibility. Among these, the interaction between lithology and PGV exhibits the strongest nonlinear enhancement. By integrating both mapping units and their adjacency information, this study provides a novel approach to improving the predictive accuracy of LSM. Moreover, by analyzing the driving factors of spatial stratification heterogeneity in landslide susceptibility maps, the study advances the practical utility of LSM for disaster management and mitigation.