Monitoring groundwater potential dynamics of north-eastern Bengal Basin in Bangladesh using AHP-Machine learning approaches

Abstract

Groundwater is a vital natural resource that plays a critical role in sustaining agriculture, forest ecosystems, industry, and household uses. However, due to natural and anthropogenic factors, groundwater is facing alarming declines. Therefore, this study aimed to assess the potential groundwater zones (PGWZ) in the north-eastern Bengal Basin of Bangladesh between 1990 and 2021 using satellite images, public and field data pertaining to ten environmental parameters. The study utilized analytical hierarchy process to identify PGWZ and evaluated the effectiveness of machine learning (ML) algorithms (K-nearest neighbors, support vector machine, XGBoost, decision tree, and random forest) for PGWZ classification. The findings indicated a decline in groundwater potential over the decades, which was categorized into five distinct zones based on the relative groundwater potential. The very high PGWZ decreased from 2.19% to 1.3%, and high PGWZ from 34.57% to 28.24%, while there was a sharp increase in the poor status of PGWZ (very low, low, and medium zones) over the same periods. The accuracy and kappa coefficients of the ground data validation for the estimated PGWZ map were 84.34% and 79.61%, respectively. According to accuracy, precision, recall, and f1-score, five ML models are reliable predictors of PGWZ. RF achieved the highest accuracy of 92.33%, while XGBoost achieved an accuracy of 90.31%. Both models demonstrated superior prediction performance for PGWZ based on the normalized leverage factor. The study attributes the alteration of groundwater potential to changes in land use and land covers, increased land surface temperatures, decreased rainfall, and changes in soil erosion in the study region over the three decades. The results of this study offer valuable insights for decision-makers to make informed decisions for the sustainable and responsible management of groundwater resources.