An automated machine learning-based framework for predicting groundwater quality with sensor data.

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Groundwater quality monitoring is essential for effective groundwater management.
• Real-time and accurate measurement technologies are crucial for timely interventions.
• Existing methods may lack the precision and speed required for continuous surveillance.

Purpose of the Study:

• To develop an automated framework for predicting ammonia nitrogen (NH3-N) in groundwater.
• To leverage multiparameter sensor data and machine learning for enhanced accuracy.
• To improve the efficiency and reliability of groundwater quality monitoring.

Main Methods:

• Collected sensor data from a carcass burial site, followed by rigorous quality control and laboratory calibration.
• Applied automated machine learning (AutoML) to optimize NH3-N prediction models using key features.
• Validated the framework's predictive performance on diverse hydrogeological datasets.

Main Results:

• Optimized models significantly improved prediction accuracy: R² increased from 0.76 to 0.90.
• Error metrics substantially decreased: RMSE reduced from 0.84 to 0.38, MAE from 0.57 to 0.23.
• External validation confirmed robust performance across different regions (R² = 0.89-0.98).

Conclusions:

• The proposed framework effectively combines calibrated sensor data with automated model selection for reliable groundwater monitoring.
• This approach offers a scalable solution for early detection of contamination in sensitive environments.
• Advanced analytics and automated calibration enhance contamination alerts, supporting proactive groundwater management.