Predicting future groundwater levels in Bangladesh under climate change using boosting algorithms

ABSTRACT Initially, 26 parameters were assessed, and key factors were determined using collinearity assessments and the variance inflation factor (VIF). Precipitation and land use and land cover were identified as the primary determinants of groundwater levels in Bangladesh. The dataset consists of 2,145 data points gathered from 2023 to 2024. Five advanced boosting algorithms, such as adaptive boosting, categorical boosting, gradient boosted decision trees, light gradient boosting machine (LightGBM), and extreme gradient boosting, were evaluated, whereas LightGBM exhibited the highest performance (AUC: 0.92, Accuracy: 0.85), attributed to its efficacy in managing extensive datasets and class imbalances. The model predicts that low groundwater potential zones will represent 1.33–11.70% of the total land area in Bangladesh, whereas high-potential zones will comprise 29.43–37.65%. The Shared Socioeconomic Pathways (SSPs) framework was utilized to forecast future groundwater availability under scenarios SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, with precipitation data projected for the years 2040, 2060, 2080, and 2100. The estimates indicate significant spatiotemporal variations in groundwater potential, with high-potential regions spanning 13.84–80.59%, while low-potential regions vary from 0.00 to 22.35%. Some places see increased precipitation and enhanced groundwater recharge, while others are vulnerable to groundwater scarcity due to consistently low precipitation.