Predicting Heavy Metal Concentrations in Shallow Aquifer Systems Based on Low-Cost Physiochemical Parameters Using...

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Computational Hydrogeology

Background:

• Ex-situ monitoring of heavy metals in water is time-consuming and lab-intensive.
• Existing AI models predict heavy metals but lack low-cost feature predictability and explainability assessments.
• Rapid response to groundwater pollution events is hindered by traditional monitoring methods.

Purpose of the Study:

• To develop a reliable and explainable framework for predicting heavy metals in groundwater.
• To identify effective models and feature sets for heavy metal prediction.
• To assess model and feature selection uncertainty, predictive uncertainty, and model interpretability.

Main Methods:

• An integrated assessment framework was developed with four steps: model selection uncertainty, feature selection uncertainty, predictive uncertainty, and model interpretability.
• Random Forest (RF) was evaluated as a predictive model.
• Quick-measure parameters were investigated as predictors for heavy metals.

Main Results:

• Random Forest emerged as the most suitable model for predicting heavy metals.
• Quick-measure parameters were identified as effective predictors for arsenic (As), iron (Fe), and manganese (Mn).
• Arsenic prediction was linked to nutrients and spatial distribution; Fe and Mn were associated with spatial distribution and salinity.

Conclusions:

• The proposed framework provides a reliable and explainable approach for heavy metal prediction in groundwater.
• The study highlights the potential of using readily available parameters for predicting key heavy metals.
• Further research is suggested to enhance prediction accuracy and model interpretability.