[Investigation of Dissolved Oxygen Drivers in the Chaohu Basin Using Explainable Integrated Machine Learning]

Dissolved oxygen （DO） is a crucial indicator of the health of aquatic ecosystems. Low concentrations of DO can threaten the survival of aquatic life and disrupt the balance of ecosystems. Therefore， accurately identifying and quantifying the factors influencing DO is essential for developing effective water resource management strategies. This study selected the Chaohu Lake Basin as a case study area and utilized integrated machine learning methods and SHAP analysis to identify and explain the key factors affecting DO variations in the region systematically. By integrating models such as Random Forest， LightGBM， and XGBoost， the study demonstrated that a highly accurate predictive model can be constructed （R²=0.94， RMSE=0.62 mg·L^-1， MAE=0.41 mg·L^-1）. Water temperature， ammonia nitrogen （NH₃-N）， and pH were identified as the primary factors affecting DO concentrations， contributing 53.5%， 17.6%， and 9.1% of the effect， respectively. During hypoxic phases， the dominant factors shift， with the importance of water temperature and NH₃-N decreasing by 20.5% and 7.9%， while the significance of pH， relative humidity， and conductivity increases by 7.1%， 3.7%， and 4.8%， respectively. Partial dependence analysis revealed that increasing water temperature and NH₃-N concentration significantly decrease DO levels； a moderate pH facilitates the dissolution of oxygen， whereas extremely acidic or alkaline conditions may negatively impact DO. Under interactive effects， high temperatures， increased total phosphorus， and low atmospheric pressure amplify the negative impact of temperature on DO concentrations. Through these analyses， this study enhances understanding of the dynamics of DO， providing data support and scientific basis for DO monitoring， hypoxia warning， and management in the Chaohu Lake Basin and thereby aiding in the sustainable use and ecological protection of its water resources.