Comparative Evaluation of Statistical and Morphometric Models for Rainfall-Induced Shallow Landslide Susceptibility Mapping: A Case Study Along the Fujian–Guangdong Border

Extreme rainfall can trigger widespread shallow landslides, highlighting the need for reliable susceptibility assessment systems. This study compares an information value–logistic regression (IV-LR) model with a terrain-driven morphometric model in terms of their predictive performance for shallow landslides triggered by an extreme rainfall event along the Fujian–Guangdong border. The IV-LR model integrates 11 environmental predictors and was trained using 3633 landslide samples and an equal number of non-landslide samples, whereas the morphometric model relies on DEM-derived terrain attributes and required no landslide inventory for training. Both models were evaluated using a common spatially separated validation dataset comprising 2273 landslide and 2273 non-landslide samples. The IV-LR model achieved an ROC-AUC of 0.7537, an Earth Mover’s Distance (EMD) of 0.2524, and a Brier score of 0.2336, compared with 0.5225, 0.0084, and 0.4172 for the morphometric model, respectively. Within the highest-ranked 20% of the common assessment area, the IV-LR model captured 53.06% of validation landslides, compared with 23.89% for the morphometric model. These findings support the feasibility of using an IV-LR model for detailed susceptibility assessment, while the morphometric model, which relies only on DEM-derived inputs, shows limited discriminatory performance.