Pollution loads in the middle-lower Yangtze river by coupling water quality models with machine learning

Area of Science:

• Environmental Science
• Water Quality Modeling
• Machine Learning Applications

Background:

• Pollution control in the Yangtze River is critical but challenged by data limitations.
• Previous models underestimated the impact of human activities on river pollution.

Purpose of the Study:

• To develop a coupled hydrodynamic-based water quality (HWQ) and machine learning (ML) model.
• To accurately quantify daily pollution loads (COD, TP) and identify their sources in the Yangtze River.
• To assess the contribution of anthropogenic activities to river pollution.

Main Methods:

• Coupling a hydrodynamic-based water quality (HWQ) model with an attention-based Gated Recurrent Unit (GRU) machine learning model.
• Analyzing pollution data from 2014-2018 for the Middle-Lower Yangtze River.
• Utilizing attention weights to determine the drivers of pollution sources.

Main Results:

• The coupled HWQ-ML model showed superior performance (KGE 0.77-0.91 for COD, 0.47-0.64 for TP) compared to standalone ML.
• Lateral anthropogenic discharges were the dominant source of COD (66% at Hankou, 69% at Datong) and TP (35% at Hankou, 42% at Datong).
• Temperature, date, and precipitation were key drivers of anthropogenic pollution.

Conclusions:

• The synergistic HWQ-ML model effectively deciphers pollution dynamics and sources in the Yangtze River.
• Anthropogenic activities significantly contribute to Yangtze River pollution, influenced by seasonal and weather factors.
• This approach offers valuable insights for effective pollution management and environmental protection strategies.