Prediction of riverbank erosion susceptibility areas of the Vietnam Mekong Delta using machine learning and remote sensing in the context of geographical studies

The objective of this study is to assess and observe riverbank erosion along the Mekong River in Vietnam using machine learning algorithms (Adaboost (ADB), Catboost (CB), Gradient boost (GB), and Random forest (RF)) and remote sensing. Landsat OLI images from 2010, 2015 and 2024 were used to detect erosion points on riverbanks. These points, combined with 16 conditioning factors grouped into four categories (topography, climate, hydrology, and human activities), are the input data for machine learning models to predict riverbank erosion susceptibility. Various statistical indices were used to evaluate the proposed models, namely Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Area Under the Curve (AUC), and Coefficient of Determination (R2). The results showed that the GB model performed better with an AUC value of 0.85, followed by ADB with an AUC value of 0.84, RF with an AUC value of 0.83 and CB with an AUC value of 0.82. The results showed that bank erosion is more serious in the provinces of An Giang, Dong Thap concentrated along the main branches of the Tien and Hau rivers. This area is highly affected by the main flow, where the high discharge of water and high flow velocities have increased the pressure on the banks of the river. Specifically, approximately 1337 km2 of the area was classified as having high to very high river bank erosion according to the RF model; approximately 1272 km2 according to the GB model; 1000 km2 according to the ADB model; and 600 km2 according to the CB model.