Abstract Groundwater is among the most critical natural resources used for domestic, agricultural, and industrial purposes, especially in densely populated areas like West Bengal, India. This study explores a new integrated framework to predict groundwater level and groundwater quality in the North 24 Parganas district. Historical data from the CGWB, WRIS, and IMD were accompanied with IoT sensor data to capture the spatio-temporal dynamics of groundwater used in this study. Groundwater quality was evaluated by calculating a Weighted Arithmetic Water Quality Index. A comparative predictive analysis was performed by using traditional ML methods, Support Vector Machine and Random Forest with deep learning approaches Convolution Neural Network and Long Short-Term Memory. The discoveries revealed that LSTM outperformed all other algorithms for groundwater level with predictions of highest R² = 0.98 and CNN-based spatial classifiers provide the greatest accuracy 88% for quality measurement. The main contribution of this research is the incorporation of IoT sensing with ML-based predictive analytics into a scalable decision-support framework that can dramatically improve sustainable groundwater management in fragile areas. This research bridges technological innovation with sustainability imperatives, thus creates a roadmap where digital water governance directly advances multiple sustainable development goals, while safeguarding public health and agricultural security. Graphical Abstract Highlights <UnorderedList Mark="Bullet"> <ItemContent> A novel IoT–ML framework integrates real-time groundwater sensing with machine learning and deep learning for predicting level and quality in North 24 Parganas, West Bengal. </ItemContent> <ItemContent> Hybrid LSTM achieved the best temporal accuracy (R² = 0.98), while CNN yielded the highest spatial classification (AUC = 0.89). </ItemContent> <ItemContent> Weighted Water Quality Index revealed arsenic and iron as major contaminants. </ItemContent> <ItemContent> IoT deployment achieved < 5 s latency, enabling real-time monitoring and alerts. </ItemContent> <ItemContent> The framework supports SDG-6 goals via sustainable, data-driven groundwater management. </ItemContent> </UnorderedList>

Spatio-Temporal Prediction of Groundwater Level and Quality using IoT Sensing and Hybrid Machine Learning Approaches

期刊论文
发表日期：

2026年2月7日

查看全文

Bijaya Banerjee,
Payal Bose,
Rana Basu

摘要

Abstract

Groundwater is among the most critical natural resources used for domestic, agricultural, and industrial purposes, especially in densely populated areas like West Bengal, India. This study explores a new integrated framework to predict groundwater level and groundwater quality in the North 24 Parganas district. Historical data from the CGWB, WRIS, and IMD were accompanied with IoT sensor data to capture the spatio-temporal dynamics of groundwater used in this study. Groundwater quality was evaluated by calculating a Weighted Arithmetic Water Quality Index. A comparative predictive analysis was performed by using traditional ML methods, Support Vector Machine and Random Forest with deep learning approaches Convolution Neural Network and Long Short-Term Memory. The discoveries revealed that LSTM outperformed all other algorithms for groundwater level with predictions of highest R² = 0.98 and CNN-based spatial classifiers provide the greatest accuracy 88% for quality measurement. The main contribution of this research is the incorporation of IoT sensing with ML-based predictive analytics into a scalable decision-support framework that can dramatically improve sustainable groundwater management in fragile areas. This research bridges technological innovation with sustainability imperatives, thus creates a roadmap where digital water governance directly advances multiple sustainable development goals, while safeguarding public health and agricultural security.

Graphical Abstract

Highlights

A novel IoT–ML framework integrates real-time groundwater sensing with machine learning and deep learning for predicting level and quality in North 24 Parganas, West Bengal.

Hybrid LSTM achieved the best temporal accuracy (R² = 0.98), while CNN yielded the highest spatial classification (AUC = 0.89).

Weighted Water Quality Index revealed arsenic and iron as major contaminants.

IoT deployment achieved < 5 s latency, enabling real-time monitoring and alerts.

The framework supports SDG-6 goals via sustainable, data-driven groundwater management.