Md. Abdullah-Al Mamun

Accurate groundwater level prediction is critical for sustainable water resource management, particularly in data-scarce regions. This study develops a hybrid ensemble machine learning framework combining multiple algorithms to predict groundwater levels in northern Bangladesh. The framework integrates rigorous cross-validation, hyperparameter optimization, and uncertainty analysis to deliver robust and interpretable predictions across spatially heterogeneous aquifer systems in the Rangpur and Dinajpur districts.

Riverine floodplain communities face escalating risks under climate change, environmental degradation, and socio-economic vulnerability. This study develops a rigorously validated framework for assessing community resilience by integrating multi-criteria decision analysis, principal axis factoring, Monte Carlo–based index construction, deep learning validation, and explainable artificial intelligence. Using 1000 georeferenced observations and 56 multidimensional indicators for 49 unions in the Brahmaputra River Basin, the study constructs a Combined Multidimensional Resilience Index (CMRI). Deep learning models demonstrate high predictive capability (AUC = 0.989), while SHAP analysis identifies hydraulic deficits, ecological degradation, and institutional weaknesses as primary drivers of vulnerability.

This study presents the first integrated explainable and physics-informed AI framework combining ML, deep learning, and Physics-Informed Neural Networks (PINNs) to predict and interpret seasonal water quality dynamics in the Padma River Basin. Using 44 monitoring sites sampled during winter and monsoon seasons, the framework integrates WQI assessment, SHAP and game-theory attribution, probabilistic uncertainty analysis, and spatial autocorrelation. Seasonal variability dominates, with winter low-flow intensifying solute concentrations and monsoon discharge promoting basin-wide dilution and recovery.

Coastal aquifers in southern Bangladesh are increasingly threatened by groundwater overexploitation, saltwater intrusion, and diffuse contamination. This study proposes a hybrid, data-driven, and interpretable modeling framework integrating machine learning, statistical analysis, and spatial assessment to predict nitrate (NO₃⁻) and sulfate (SO₄²⁻) risks in coastal aquifers. The framework identifies dominant hydrogeochemical drivers, evaluates spatial contamination hotspots, and improves interpretability of prediction outcomes to support groundwater risk management.

This study used traditional statistical techniques (PCA, PCoA) together with machine-learning models including SOM, CIT, Ridge Regression, and SHAP to detect and trace heavy metal sources along the northeast coast of Bangladesh. Concentrations of Pb, Cd, Mn, and As exceeded average shale values. SOM indicated natural sources for As, Ni, Fe, and Mn, and shipbreaking industry sources for Cu, Zn, and Pb. SHAP highlighted soil pH, silt, and organic matter as key contributors to HM accumulation.

This study investigates how large-scale climate drivers influence extreme precipitation patterns across Bangladesh, examining spatiotemporal variability and the underlying physical mechanisms. The analysis integrates climate indices, statistical modelling, and data-driven methods to identify dominant drivers and their seasonal and regional signatures. The findings provide critical insights for climate adaptation and disaster risk reduction in one of the world's most flood-vulnerable nations.

This study presents the first integrated probabilistic–spatial framework for assessing radiological risk in Bay of Bengal beach sands. Twenty-two surface sediment samples from Darianagar, Himchari, and Nidania beaches were analyzed using HPGe gamma spectrometry. Activity concentrations of ²²⁶Ra and ²³²Th exceeded global averages by 2.2× and 4.1×. Monte Carlo simulations showed radiological indices exceeding international safety thresholds by 8–23×. GIS-based spatial mapping revealed localized hotspots requiring restricted access and monitoring.

This work investigates Gaussian Process Regression (GPR), Bayesian Ridge Regression (BRR), and Artificial Neural Network (ANN) for predicting groundwater quality in coastal Bangladesh. Optuna-based hyperparameter optimization improves model accuracy. Combined cross-validation and bootstrapping methods build six predictive models. The entropy-based ECWQI is converted to five quality classes. SOM identified four distinct spatial patterns. ANN (CV) and ANN (B) models outperformed all others (R² = 0.971 and 0.969). SO₄²⁻, Cl⁻ and F⁻ played key roles in prediction accuracy.

This work combines geostatistical modeling, SOM, and data-driven algorithms to predict groundwater phosphate (PO₄³⁻) in coastal multi-aquifers of southern Bangladesh. CatBoost, GBM, LSTM, and SVR were evaluated using 380 samples and 15 prediction parameters. CatBoost showed exceptional performance in training (R² = 0.999) and testing (R² = 0.989). Phosphate dissolution and saltwater intrusion, along with phosphorus fertilizers, increase PO₄³⁻ content. Na⁺, K⁺, and Mg²⁺ significantly influenced prediction accuracy.