Hybrid Approach Combining Machine Learning and Physics-Based Modelling: A Shirur Landslide Case Study Using SCOOP-3D and Random Forest

This study proposes a hybrid approach that integrates machine learning-based classification with physics-based slope stability modelling to address these challenges. Using the Shirur region in Karnataka, India, a highly vulnerable zone within the Western Ghats, as a case study, a Random Forest Classifier (RFC) was trained using six conditioning factors, which accounted for the geographical diversity of the area and achieved an area under the curve of 0.913 in the receiver operating characteristic curve. Five high-risk zones totalling 42.33 km² were delineated from the LSM and analysed using the SCOOPS-3D model under unsaturated conditions. Soil parameters included cohesion (17 kPa), friction angle (24.16°), and unit weight (22.05 kN/m3), with search volumes ranging from 100 to 10,000 m3. The hybrid approach reduced computation time by 79.6–82.7% compared to full-area modelling across three computing environments. The model showed an accuracy of 88% in the confusion matrix. The zonation-based framework enhances spatial resolution, localizes unstable zones, reduces stability overestimation, and offers a scalable, efficient solution for large-area hazard mapping in data-rich but resource-limited settings.