Performance of Na?ve Bayes Tree with ensemble learner techniques for groundwater potential mapping

Water supply is a key challenge and priority for achieving sustainable development goals in many countries. Recognizing areas with groundwater potential is crucial in addressing this challenge. In this study, the objective was to model and predict the potential of groundwater availability in the Central Highlands of Vietnam utilizing an ensemble modeling approach that combined the Bagging (B), Decorate (D), and MultiBoost (MBAB) techniques with Naive Bayes Tree (NBT) and developed three ensemble models: B-NBT, D-NBT, and MBAB-NBT. We applied the models to a geospatial dataset consisting of 501 wells data and twelve explanatory variables, i.e., rainfall, land use/cover (LULC), elevation, river density, fault density, flow accumulation, aspect, topographic wetness index (TWI), deep division, geology, slope, and curvature. We then evaluated the models using various criteria, including the area under the receiver operating characteristic curve (AUC), accuracy sensitivity, specificity, and Kappa. According to the results, the ensemble models exhibited better performance than the single NBT model in both fitting with the training dataset and predictive accuracy. The MBAB-NBT model demonstrated superior perform (accuracy = 69.33%, sensitivity = 70%, specificity = 67%, RMSE = 0.45, and AUC = 0.741%), followed by the B-NBT model (accuracy = 64.67%, sensitivity = 67%, specificity = 60%, RMSE = 0.46, and AUC = 0.732%). The most relevant variables for groundwater potential were land use/land cover, rainfall, flow accumulation, fault density, elevation, river density, and topographic wetness index. The information provided by this study can assist in recognizing regions with a high potential for groundwater availability, which can be crucial in determining suitable locations for the development of irrigation systems, industrial facilities, and residential areas.