Applicability of spatial prediction models for landslide susceptibility in land-use zoning instruments: a guideline in a coastal settlement in south-central Chile

Landslide hazard is usually incorporated into land-use planning as susceptibility zoning. Multiple-variable models have been widely used for susceptibility zoning due to their advantage to use different performance techniques to improve their prediction capacity. In Chile, the incorporation of landslide hazard into land-use planning instruments lacks a frame of reference that defines the susceptibility zoning and potential methodologies to be used. To overcome this lack of reference, this study assessed the applicability of GLM and GAM models in the establishment of susceptibility zoning guidelines. The application of both models identified the areas-prone to landslides in enough detail for a scale <1:15.000. The prediction of these zones was improved after spatial k-fold cross-validation, which indicated a slight advantage of the GAM model over the GLM model in susceptibility prediction capacity. The resulting zoning areas for low, moderate, high, and very high susceptibility were very similar for the GLM and GAM models.