A Virtual Sensing Concept for Nitrogen and Phosphorus Monitoring Using Machine Learning Techniques.

Area of Science:

• Environmental Science
• Water Quality Monitoring
• Machine Learning Applications

Background:

• Harmful cyanobacterial blooms (HCB) pose risks to drinking water treatment and human health.
• Effective eutrophication and HCB management requires continuous monitoring of nitrogen (N) and phosphorus (P).
• High-frequency water quality monitoring is often cost-prohibitive.

Purpose of the Study:

• To develop and evaluate machine learning-based virtual sensors for predicting N and P concentrations.
• To identify optimal machine learning models, data scaling techniques, and imputation methods for water quality prediction.
• To assess the performance of virtual sensors in contrasting rural and urban catchments.

Main Methods:

• Employed six machine learning algorithms: random forest, extremely randomized trees (ET), extreme gradient boosting, k-nearest neighbors, light gradient boosting machine, and bagging regressor.
• Investigated the impact of data scaling (MinMax scaler) and missing value imputation (multivariate imputer).
• Utilized Shapley additive explanations for feature importance ranking.

Main Results:

• Extremely randomized trees (ET) demonstrated the best predictive performance.
• MinMax scaler and multivariate imputer were identified as the optimal scaler and imputer, respectively.
• Achieved high predictive performance with R² values of 97% in a rural catchment and 82% in an urban catchment.

Conclusions:

• Machine learning virtual sensors provide a viable and cost-effective alternative for high-frequency water quality monitoring.
• The ET model, combined with MinMax scaling and multivariate imputation, offers a robust approach for predicting N and P.
• These virtual sensors can aid catchment managers in controlling eutrophication and mitigating harmful cyanobacterial blooms.