Water quality parameters-based prediction of dissolved oxygen in estuaries using advanced explainable ensemble...

Area of Science:

• Environmental Science
• Water Quality Management
• Machine Learning Applications

Background:

• Dissolved oxygen (DO) is critical for estuarine and bay health, but prediction is challenged by human activities and complex water quality parameter (WQP) interactions.
• Existing water quality models and statistical methods lack accuracy and fail to clarify the impact mechanisms of WQPs on DO.
• Understanding WQP impacts is crucial for effective hypoxia management in coastal ecosystems.

Purpose of the Study:

• To develop an interpretable ensemble machine learning (EML) framework for accurate DO prediction in six Chinese estuaries.
• To elucidate the impact mechanisms of various WQPs on DO variations.
• To provide insights for improving hypoxia management strategies in coastal rivers.

Main Methods:

• An interpretable ensemble machine learning (EML) framework was developed for DO prediction.
• The framework utilized historical data (November 2020-December 2023) including DO and various WQPs.
• SHapley Additive explanation (SHAP) values were employed to determine feature importance and impact mechanisms.

Main Results:

• The EML framework, particularly the stacking model (SM), demonstrated high accuracy (R²=0.71, RMSE=0.55 in Jilong River), outperforming other models.
• Lagged features of DO (1-3 days) and water temperature (WT) were crucial predictors, with 1-day lagged DO having the most significant impact.
• Factors influencing daily DO varied by river; EC, pH, and TN generally had positive impacts, while WT, NH₃-N, and TP had negative impacts.

Conclusions:

• The developed EML framework offers a robust and interpretable approach for DO prediction and understanding WQP influence.
• Lagged DO and WT are key drivers of DO fluctuations, with specific thresholds and interactions exhibiting spatial heterogeneity.
• The findings provide valuable insights for targeted hypoxia management and ecological health preservation in coastal estuaries.