Real-time prediction of river chloride concentration using ensemble learning

Area of Science:

• Environmental Science
• Water Resource Management
• Artificial Intelligence in Environmental Monitoring

Background:

• Real-time river chloride prediction is crucial for effective chloride control and watershed management.
• Existing models may have limitations in accurately predicting chloride concentrations, especially extreme values.

Purpose of the Study:

• To develop and evaluate an ensemble learning model for accurate real-time river chloride prediction.
• To improve upon the performance of individual artificial neural network (multi-layer perceptron, MLP) and statistical inference (stepwise-cluster analysis, SCA) models.

Main Methods:

• Developed an artificial neural network model (multi-layer perceptron, MLP).
• Developed a statistical inference model (stepwise-cluster analysis, SCA).
• Proposed and tested an ensemble learning model (MLP-SCA) combining MLP and SCA for improved accuracy.

Main Results:

• The ensemble model MLP-SCA demonstrated superior performance (RMSE: 11.58 mg/L, MAPE: 27.55%, NSE: 0.90, R²: 0.90) compared to individual models.
• MLP-SCA showed enhanced capability in predicting extremely high chloride concentrations (above 150 mg/L) with RMSE of 9.88 mg/L and MAPE of 4.40%.
• The model effectively utilizes common environmental data: conductivity, water temperature, river flow rate, and rainfall.

Conclusions:

• The MLP-SCA model offers reliable and accurate real-time river chloride prediction, particularly for extreme events.
• This novel ensemble approach can supplement water quality monitoring and support watershed management strategies.
• The MLP-SCA model's success suggests its potential applicability to chloride prediction in other river systems.