A Prototype Geo-Digital Twin Framework for Scenario-Based Landslide Susceptibility Assessment Using Machine Learning

Landslides are among the most destructive natural hazards in mountainous and environmentally sensitive regions, causing significant damage to infrastructure, ecosystems, and human life. This study proposes a framework as a prototype Geo Digital Twin for landslide susceptibility assessment and scenario analysis in Eastern Kentucky, USA. A Random Forest model was developed using topographic, hydrological, geological, and environmental conditioning factors, including aspect, elevation, lithology, distance to streams, and NDVI. The model achieved strong predictive performance, with an accuracy of 0.87, an F1 score of 0.85, and a ROC–AUC of 0.94. The trained model was integrated into a service-oriented prototype Geo Digital Twin platform implemented with Django and Mapbox, enabling interactive spatial visualization and scenario simulation. In particular, vegetation change scenarios were evaluated by modifying NDVI values to examine their influence on landslide susceptibility. To improve the interpretation of these scenario outcomes, a fuzzy inference system was introduced by combining changes in predicted susceptibility (ΔP) with NDVI values into a Fuzzy Impact Index (FII). The fuzzy analysis indicates that vegetation reduction can substantially alter slope responses depending on baseline environmental conditions. The results demonstrate that the proposed framework not only provides accurate susceptibility mapping but also facilitates explainable scenario analysis to support geospatial decision-making regarding landslides.