Hyperparameter-optimized deep learning for next-day pH forecasting in distributed water quality monitoring systems

Area of Science:

• Environmental Science
• Data Science
• Water Resource Management

Background:

• Accurate short-horizon water quality forecasting is vital for proactive environmental monitoring and intervention.
• Next-day pH prediction is critical as pH indicates aquatic chemical balance, pollutant behavior, and ecosystem health.

Purpose of the Study:

• To develop and evaluate a spatio-temporal framework for next-day pH prediction using a two-stage learning approach.
• To benchmark baseline models and refine the strongest performer through metaheuristic hyperparameter optimization.

Main Methods:

• Utilized a two-stage learning framework combining preprocessing, comparative benchmarking, and optimizer-guided refinement.
• Employed RegNet as the predictive model and the Ocotillo Optimization Algorithm (OCOA) for hyperparameter optimization, comparing it against other optimizers.
• Analyzed the public Water Quality Prediction dataset from Georgia, USA, focusing on spatio-temporal dependencies.

Main Results:

• RegNet demonstrated the strongest untuned performance with an RMSE of 0.0285 in the baseline stage.
• OCOA + RegNet achieved the lowest RMSE across independent runs, outperforming other optimizer-assisted RegNet variants.
• The study confirmed the benefits of model selection and rigorous hyperparameter optimization for pH forecasting.

Conclusions:

• The proposed OCOA-RegNet workflow provides an effective approach for next-day pH forecasting in distributed water quality monitoring.
• This method supports timely, data-driven interventions in environmental management.
• The findings highlight the importance of advanced machine learning techniques for environmental monitoring systems.