Estimating ammonium changes in pilot and full-scale constructed wetlands using kinetic model, linear regression, and...

Area of Science:

• Environmental Engineering
• Wastewater Treatment Technologies
• Nature-Based Solutions

Background:

• Constructed wetlands (CWs) are established nature-based solutions for diverse wastewater treatment.
• Existing CW applications often involve large-scale systems with prolonged operational periods, crucial for real-world efficacy.
• Accurate prediction of effluent ammonium is vital for optimizing CW performance and design.

Purpose of the Study:

• To evaluate the efficacy of kinetic, linear regression (LR), and machine learning (ML) models in estimating effluent ammonium concentrations.
• To compare the predictive performance of various models using data from pilot and full-scale constructed wetlands.
• To identify the most effective modeling approach for practical CW design and management.

Main Methods:

• Data extraction from 24 pilot and full-scale CW studies, encompassing 15 features and 975 data points.
• Implementation and comparison of nine distinct models, including Monod kinetics, simple and multiple linear regression, and advanced ML algorithms (Cubist, Random Forest).
• Statistical evaluation using Root Mean Square Error (RMSE) and coefficient of determination (R-squared) to assess model accuracy.

Main Results:

• Nonlinear ML algorithms significantly outperformed linear models and kinetic approaches in predicting CW effluent ammonium.
• The Monod kinetic model exhibited the poorest performance (RMSE: 41.84 mg/L, R²: 0.34).
• Cubist (RMSE: 12.01 ± 5.38, R²: 0.93 ± 0.07) and Random Forest (RMSE: 15.94 ± 10.69, R²: 0.91 ± 0.08) demonstrated high predictive accuracy.
• The optimized Random Forest model achieved an R² of 0.93 and RMSE of 13.48 mg/L on new data.

Conclusions:

• Machine learning models, particularly Random Forest, offer a robust and efficient method for estimating effluent ammonium in constructed wetlands.
• ML-based predictions provide valuable insights for optimizing the design and operation of pilot and full-scale CW systems.
• This study highlights the potential of ML to enhance the practical application and effectiveness of constructed wetlands in wastewater treatment.