Landslide susceptibility mapping using Frequency Ratio (FR) and Random Forest (RF) model in the Bandarban Hill District, Bangladesh

The research aims to investigate the connections between landslide inventory and causative factors for Landslide Susceptibility Mapping (LSM) in the Bandarban Hill District. The investigation used 14 causative factors (resampling with 30 m resolution) drawn from prior literature and data availability to implement Frequency Ratio (FR), Random Forest (RF), and ensemble FR-RF approaches for generating LSM. An inventory map of 237 landslide sites with equal numbers of non-landslide incidents was randomly split into model training (75%) and validation (25%), utilizing historical landslide data, Google Earth, and a field survey. Also, the study evaluated the performance of three LSM maps using statistical metrics like accuracy, recall, precision, F1 score, and ROC curve, revealing a high accuracy rate exceeding 85%. The ensemble FR-RF approach map indicates five susceptible zones: very low (16.74%), low (23.97%), moderate (27.44%), high (11.97%), and very high (19.88%). Regional analysis reveals that the Bandarban Sadar and Lama Upazilas are very highly susceptible to landslides, which is significantly influenced by factors like NDVI, geology, road distance, rainfall, and slope. The study aids local authorities and stakeholders in land use planning, evaluates potential loss and damage, and devises prevention and mitigation strategies through enhanced drainage systems, slope stabilization, and structural fortifications.