River water quality forecasting: a novel LSTM-Transformer approach enhanced by multi-source data

Area of Science:

• Environmental Science
• Water Resource Management
• Data Science

Background:

• Effective water quality prediction is vital for managing water resources and protecting ecosystems.
• Eutrophication in the Beijing-Hangzhou Canal necessitates advanced monitoring and control strategies.

Purpose of the Study:

• To develop a deep learning model for predicting total phosphorus (TP) and total nitrogen (TN) concentrations.
• To enhance water quality prediction accuracy for the Changzhou section of the Beijing-Hangzhou Canal.

Main Methods:

• A hybrid wavelet denoising (WD)-LSTM-Transformer model was developed.
• The model integrates water quality data, land use information, and meteorological factors.
• The SHAP method was employed for model interpretability and variable significance analysis.

Main Results:

• The WD-LSTM-Transformer model achieved high prediction accuracy for TP and TN (R² > 0.9).
• The model demonstrated superior performance compared to four traditional prediction models.
• Significant variables influencing TP and TN fluctuations were identified.

Conclusions:

• The proposed model offers a robust and interpretable approach for water quality prediction.
• This research provides a scientific foundation for identifying pollution sources and improving watershed management.
• The findings support eutrophication prevention and control efforts in critical water bodies.