A multi-task deep neural network reveals inflowing river impacts for predictive lake management

Area of Science:

• Environmental Science
• Water Resource Management
• Artificial Intelligence

Background:

• Lake ecosystems face degradation due to riverine pollution, necessitating effective management strategies.
• Predicting the impact of multiple river inputs on lake water quality is complex and challenging for traditional models.
• Existing methods like mechanistic models and conventional machine learning have limitations in capturing system complexity.

Purpose of the Study:

• To develop an integrated predictive tool for effective environmental management of lake water quality.
• To accurately and simultaneously predict multiple water quality indicators at various lake locations using riverine data.
• To assess the contributions of individual rivers and identify key pollution drivers.

Main Methods:

• Development and application of a multi-task deep neural network (MTDNN).
• Utilizing data from inflowing rivers to predict four key water quality indicators: permanganate index, total phosphorus, total nitrogen, and algal density.
• Comparison of MTDNN performance against established mechanistic and single-task deep learning models.

Main Results:

• The MTDNN model achieved significant improvements in predictive precision, up to 56.3%, compared to existing models.
• The model successfully identified specific river contributions and pinpointed dominant pollution drivers like water temperature and wastewater effluent.
• Scenario-based forecasting indicated that using reclaimed water for lake replenishment is a viable, non-deteriorating strategy.

Conclusions:

• The MTDNN framework provides a powerful, transferable tool for data-driven lake management.
• This approach enables targeted interventions for sustainable water resource protection.
• The study highlights the potential of advanced AI in addressing complex environmental challenges in lake ecosystems.