Mapping groundwater potentiality by using hybrid machine learning models under the scenario of climate variability: a national level study of Bangladesh

A severe threat to natural resources and human livelihood is groundwater scarcity. Therefore, mapping groundwater potentiality (GWP) is necessary for future resource management. In this article, a framework for conducting ensemble modeling is introduced. This framework is used to map GWP at the national level under the scenario of climatic variability. Thirteen elements linked to topography, geology, hydrology, and land cover, as well as six climatic indicators based on historical time series data, were used to map the GWP. This study has used three conventional machine learning algorithms (< MLAs), such as logistic model tree, logistic regression, and artificial neural network and five ensemble models by combining standalone models with random forest under stacking framework to produce GWP map. Using the empirical and binormal receiver operating characteristic curves, the GWP mapping has been validated. Result shows that Bangladesh's major rivers run along the high GWP zones in the country's southern and central regions. In addition, the validation using the area under curve (AUC) of ROC curve demonstrates that the stacking model which combined all three MLAs outperformed other models (AUC: 0.971). The findings of this study may help the authorities and stakeholders to formulate the adequate groundwater management plans at national level. In addition, the suggested method might be applied to map GWP on a broader scale in additional nations as well as at the continental level.