Integrated machine learning and remote sensing for groundwater potential mapping in the Mekong Delta in Vietnam

Abstract

Evaluating groundwater potential is critical for the socioeconomic development of Vietnam. This research aims to assess the underground water potential in the country’s Mekong Delta using the machine learning (ML) such as support vector machines (SVM), CatBoost (CB), K-nearest neighbors (KNN), random forest (RF) and AdaBoost (ADB). The problem of exploitation of groundwater resources in the delta is aggravated due to global warming and growth of population. In total, 146 groundwater points and 14 drivers (namely elevation, aspect, curvature, slope distance to river and river density, land use, normalized difference built-up index, flow accumulation, rainfall, soil type, normalized difference vegetation index, stream power index, terrain roughness index, and topographic wetness index) were used to assess groundwater potential. Each proposed model was evaluated utilizing area under curve (AUC), root mean square error, coefficient of determination (R2), and mean absolute error. The findings showed that the RF outperformed the others in building of a groundwater potential map. In which, AUC value was estimated at 0.99 and R2 value was estimated at 0.63 then came CB (AUC = 0.98, R2 = 0.56), ADB (AUC = 0.92, R2 = 0.50), SVM (AUC = 0.91, R2 = 0.57), and KNN (AUC = 0.75, R2 = 0.45). The results illustrate the power of ML in assessing groundwater potential and can support decision makers, planners, and local authorities responsible for sustainable groundwater planning in the Mekong Delta and beyond.