Incorporating spatial heterogeneity of environmental variables into landslide susceptibility prediction

Environmental variables, characterized by spatial heterogeneity, exert a significant impact on landslide susceptibility prediction (LSP). Previous studies generally assume that within the entire study area, the impact of each environmental variable on landslide distribution is spatially homogeneous—a perspective that essentially represents a global prediction mode. Herein, we incorporated the spatial heterogeneity of environmental variables into LSP modeling. Building upon the identification of dominant variable combinations of heterogeneous regions using geographically weighted regression, we proposed a zonal prediction method by leveraging random forest and support vector regression. The accuracy of our method (with an average Area Under the Curve (AUC) of 0.818 across all heterogeneous regions) was higher than that of the global prediction mode (AUC = 0.7). Furthermore, we found that the inter-regional distribution of dominant variables may be associated with landslide distribution. Our findings provide insights for LSP research in areas with variations in internal environmental characteristics.