Causal Geospatial Intelligence Framework for Identifying Environmental Drivers of Landslide Risk in Mountainous Regions

Landslide susceptibility models can effectively predict areas prone to landslides but generally provide limited evidence on whether specific environmental exposures causally influence landslide occurrence. This research proposes an integrated causal-geospatial framework combining predictive modeling with causal effect estimation and spatial heterogeneity analysis for landslide occurrence in Desa Wih Pase, Bener Meriah, Indonesia. An analytical dataset comprising 18,450 30 m × 30 m grid cells from observations in 2025 was analyzed using Random Forest and Logistic Regression for occurrence probability modeling, followed by Causal Forest to derive average treatment effect estimates together with grid-level Conditional Average Treatment Effects (CATE). The Random Forest model reached a discrimination performance of 0.91 in AUC-ROC, exceeding the 0.79 achieved by the Logistic Regression baseline, demonstrating stronger predictive discrimination under the imbalanced landslide dataset. Causal analysis estimated a positive effect of cumulative rainfall, with a 100 mm increase associated with a 0.142 increase in landslide occurrence probability (95% CI: 0.089–0.195). A 10° increase in slope produced an estimated effect of 0.218 (95% CI: 0.161–0.275). Spatial CATE analysis revealed substantial heterogeneity, with rainfall effects ranging from ?0.039 to 0.302 and slope effects from ?0.024 to 0.490 across valid 30 m grid cells. Sensitivity analyses showed that the positive causal effects remained stable across alternative confounder sets, rainfall accumulation windows, and spatial resolutions. The framework distinguishes predictive susceptibility from causal environmental effects and provides a basis for spatially targeted landslide risk reduction and environmental engineering interventions