Key points are not available for this paper at this time.
Landslides pose a significant risk to the Chittagong Hill Tracts (CHT) of Bangladesh, causing severe socio-economic and environmental impacts that threaten the achievement of sustainable land management. In this study, a landslide susceptibility map for the CHT region was created using two machine learning models: Random Forest (RF) and Maximum Entropy (MaxEnt). A total of 15 landslide conditioning factors were considered, including elevation, slope, rainfall, soil texture, and land cover. A landslide inventory dataset comprising 730 landslide events was used for model training and validation. The results indicate that both models successfully classify landslide prone areas, with MaxEnt identifying 79.12% and RF identifying 78% of the study area as high to very high susceptibility zones. Performance evaluation using the Area Under the Curve (AUC) metric revealed that RF (AUC = 0.93) outperformed compared to MaxEnt (AUC = 0.86), demonstrating superior predictive accuracy. RF also exhibited higher overall accuracy (98%) and precision (99%) compared to MaxEnt (87% and 89%, respectively). Maximum rainfall and elevation were the most influential factors in both models for landslide susppectibilty. These findings provide a critical insight into disaster risk management and policy making in the CHT and directly support SDG 11 (Sustainable Cities and Communities) by improving urban resilience, SDG 13 (Climate Action) by improving adaptation strategies and SDG 15 (Life on the Land) by promoting sustainable land management. By integrating scientific modelling into a global sustainability agenda, the study contributes to the development of risk-informed policies and early warning systems to protect vulnerable communities in the CHT region.
Hossain et al. (Tue,) studied this question.