Predicting urban landslides in the hilly regions of Bangladesh leveraging a hybrid machine learning model and CMIP6 climate projections

Landslides pose significant risks to infrastructure and human lives in cities, exacerbated by climate change. Therefore, a reliable predictive landslide model is crucial for mitigation, especially in resource-limited nations. This study employs hybrid machine learning (ML) techniques and climate projections to predict landslides in the Chattogram development area (CDA) of Bangladesh – a rapidly growing urban city in Bangladesh. The model was trained using diverse geospatial parameters including topographical, hydrological, soil, and geological parameters, along with an updated landslide inventory, enabling spatially explicit predictions of landslide susceptibility. To incorporate future climate scenarios, we utilized the Coupled Model Intercomparison Project Phase 6 (CMIP6) Global Climate Model (GCM), projecting climate impacts under SSP1-2.6 and SSP5-8.5 scenarios for the periods of 2021–2040, 2041–2060, 2061–2080, and 2081–2100, respectively. These scenarios reflect different pathways of greenhouse gas emissions, providing a range of possible future climate conditions. We tested six ML classifiers: random forest (RF), extra trees (ExT), support vector machine (SVM), logistic regression (LR), Bernoulli Naïve Bayes (bNB), and K-nearest neighbor (KNN). Each base model demonstrated high accuracy (>90 %) but combining them improved both accuracy and computational efficiency. The LR-bNB hybrid model outperformed all others, effectively mapping landslide susceptibility in the study area for the current timeframe and future projections. Our results revealed significant variability in landslide-prone areas across the area, with 12 % of the region categorized as high to very high risk, a figure that slightly rises with predicted increased rainfall due to climate change. The present study demonstrates the efficacy of a hybrid ML model for nowcasting as well as forecasting landslide susceptibility under future climate scenarios. These findings offer valuable insights for proactive risk management and infrastructure planning in the CDA, helping to safeguard communities and improve resilience against future landslide events.