Remote Sensing-Based Detection and Prediction of Rainfall–Flood–Landslide Cascades in Eastern Kentucky

Eastern Kentucky is highly susceptible to flooding and landslides due to its steep terrain and narrow river valleys. Extreme rainfall can trigger flooding and destabilize hillslopes, creating cascading hazards that threaten communities and infrastructure. This study evaluates the relationships between natural and anthropogenic landscape characteristics and landslide susceptibility while comparing the predictive performance of statistical and machine learning approaches. Topographic, hydrologic, geologic, land cover, and anthropogenic variables were compiled from satellite-derived products and publicly available geospatial datasets for four major watersheds in Eastern Kentucky. The analysis covered August 2016 through August 2026, and a statewide landslide inventory was used to develop and evaluate three predictive models: binary logistic regression, Random Forest, and Extreme Gradient Boosting (XGBoost). Model performance was assessed using receiver operating characteristic curves and the area under the curve (AUC). All three models demonstrated strong discriminatory performance, with AUC values of 0.9315 for logistic regression, 0.9596 for Random Forest, and 0.9495 for XGBoost. Random Forest achieved the highest AUC, although the improvement over logistic regression was marginal. These findings suggest that a comparatively interpretable statistical model can provide strong predictive performance without the additional complexity of more advanced machine learning approaches. This research provides a foundation for identifying landscape characteristics associated with landslide susceptibility in Eastern Kentucky and demonstrates the potential value of integrating geospatial data and predictive modeling to support regional natural hazard assessment.