Assessment of soil salinity using artificial intelligence and Sentinel 2 in the Mekong Delta

This research aims to assess soil salinity area distribution in the delta’s Vietnamese sections, using artificial intelligence and Sentinel 2B. Seventy-six soil samples were obtained from July and August in 2024, and conditioning variables extract by Sentinel 2B, which were applied as input data for the artificial intelligence models. This database was divised by two parts: 70% data to construit the model and 30% for validating. After the construction of the models (namely XGBoost (XGB) and the hybrid models XGBoost-Grey Wolf Optimizer (XGB-GWO), XGBoost-Zebra Optimisation Algorithm (XGB-ZOA), XGBoost-Walrus Optimization Algorithm (XGB-WaOA), XGBoost-Tasmanian Devil Optimization (XGB-TDO), XGBoost-Northern Goshawk Optimization (XGB-NGO) the statistical indices root-mean-square error (RMSE), coefficient of determination (R²), and mean absolute error (MAE) were used to assess the models’ prediction ability. The outcome showed that all six models were effective in assessing surface soil’s salinity distribution with the R² value more 0.7. The XGB-TDO model exhibited superior performance, with an R² of 0.919, followed by the XGB-GWO and WGB-NGO model (0.918), XGB-WaOA model (0.912), XGB-ZOA (0.907), and XGB (0.796). The results showed that about 962 km² of the study area is located in the non-saline zone, 1718 km² is located in the slightly saline zone and 600 km² is located in the saline zone. This study highlights the powerful ability of artificial intelligence to assess the distribution of salinity of surface soil in the delta. In addition, the electrical conductivity (EC) value is higher near the coasts and river mouths, where mangrove forests grow and many aquaculture ponds are distributed.