A deep learning-IoT integrated framework for real-time groundwater quality monitoring and prediction in urban aquifers

Bommi Rammohan, Pachaivannan Partheeban, Sundarambal Balaraman, Ferdin Joe John Joseph,
A deep learning-IoT integrated framework for real-time groundwater quality monitoring and prediction in urban aquifers,
Journal of Water Process Engineering,
Volume 80,
2025,
109052,
ISSN 2214-7144,
https://doi.org/10.1016/j.jwpe.2025.109052.
(https://www.sciencedirect.com/science/article/pii/S2214714425021257)
Abstract: Water is one of the precious human needs worldwide. It is available as surface water and groundwater. Globally, including in India, excessive urban groundwater pumping has resulted in depletion, seawater intrusion, and deteriorating water quality. This study aims to collect real-time groundwater quality data using newly designed IoT devices equipped with sensors, controllers, and GPS, while also developing deep learning models for groundwater quality prediction. This research examines groundwater quality parameters, including well water depth, temperature, turbidity, dissolved oxygen, total dissolved solids, and pH. Various deep learning models are employed to predict groundwater quality by exploring different permutations of optimizers and activation functions. In this research, 13 wells of real-time data were collected and used to predict the groundwater quality. The analysis confirms that excessive pumping directly impairs groundwater quality and identifies critical thresholds (turbidity >150 NTU, TDS >1000 ppm, pH beyond 6.5–8.5) necessitating regulatory intervention. It was found that there is a good correlation between water level in the well and pH and turbidity, and a lesser correlation with pH values. Well 2 demonstrated the best overall accuracy across all parameters. The visualization aids in understanding how different functions and optimizers influence model performance. The analysis highlights the effectiveness of different optimizer-activation function combinations for improving model accuracy, with Well 2 achieving near-perfect predictive accuracy (R2 ≈ 1.0 for TDS and turbidity) using an Adamax-ReLU-Sigmoid configuration. Further, excessive pumping leads to a reduction in groundwater quality. This research lays the groundwork for informed policy decisions on groundwater quality control and measures to mitigate excessive pumping.
Keywords: Groundwater quality prediction; Internet of things; Real-time monitoring; Deep learning; Prediction models; Optimizer