Integrating a hybrid machine learning model with climate projections for landslide susceptibility assessment

Résumé fourni par la source

Landslides represent a major recurring geological hazard triggered by intense precipitation events, which frequently occur in many parts of the world. This natural hazard also frequently occurs in Bangladesh's eastern coastal districts, particularly Chattogram and Cox’s Bazar. Therefore, this hazard is getting recent attention for further study because of increased extreme rainfall events resulting from climate change effects. This study introduces a novel Genetic Algorithm-optimized Multilayer Perceptron Artificial Neural Network (GA-MLP-ANN) model for landslide susceptibility assessment, integrating fourteen geospatial factors and climate projections from the ACCESS-ESM1–5 model under three Shared Socioeconomic Pathways (e.g., SSP1–2.6, SSP2–4.5, and SSP5–8.5) scenarios. The GA-MLP-ANN model demonstrated robust predictive performance, achieving R-squared ( R 2 ) values 0.937 (training) and 0.864 (testing), with corresponding root mean squared error (RMSE) values of 1.775 and 2.606, and mean absolute error (MAE) values of 0.125 and 0.184. In addition, the key driving factors for landslides include geology, elevation, slope, soil texture, land cover, and rainfall, identified by the Random Forest (RF) feature extraction algorithm. The baseline susceptibility map classified 35.82 % of the area as low, 22.51 % as moderate, 13.08 % as high, and 28.60 % as very high susceptibility. This study revealed that moderate and high landslide susceptibility areas are projected to increase by 1 % to 43.89 % and 0.40 % to 7.51 % respectively, by 2030 – 2100 under the SSP1–2.6 and SSP5–8.5 scenarios, with the most pronounced increase under SSP5–8.5. Model validation using the area under the curve (AUC) yielded 0.995 and 0.998 for training and datasets, confirming reliable accuracy. Key high-risk areas include Fatikchhari, Mirsharai, Ukhia, Patia, and Teknaf, with future expansion to Maheshkhali, Hathazari, and Chakaria. This study highlights the value of integrating advanced machine learning with climate projections for improved hazard assessment and risk management in a changing climate.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.