Aquaculture and submerged aquatic vegetation as key drivers of algal blooms in urban lakes: Insights from machine learning modeling

The frequent occurrence of algal blooms poses many threats to aquatic environments and socioeconomic activities. Shallow urban lakes are particularly susceptible to these blooms due to intense anthropogenic disturbances, and identifying the primary factors influencing algal biomass is challenging due to the complex interactions of physical, biogeochemical, and socioeconomic processes. In this study, a machine learning (ML) ensemble approach was utilized to predict chlorophyll- a (Chl- a ) levels in Lake Yangcheng based on a 24-year (2000–2023) dataset. Our results indicate that robust and interpretable predictive modeling relies substantially on appropriate input variable selection, leading to an average prediction improvement of 12 % when using the optimal subset. Among eight ML algorithms tested, tree-based ensemble methods like the extreme gradient boosting model (XGBoost) performed best, showing comparably higher correlations (R 2 = 0.83) and lowest error metrics (RMSE = 6.68, MAE = 4.84) between field observations and model predictions. Furthermore, out of 18 predictive variables, enclosure cage culture area (ECCA) and submerged aquatic vegetation (SAV) emerged as key influencers for predicting Chl- a , alongside traditional factors like chemical oxygen demand (COD Mn ), water temperature (WT), and total phosphorus (TP). The monotonic increasing trends of SHAP values with rising COD Mn , TP and ECCA validated the effectiveness of management strategies implemented in Lake Yangcheng over the past two decades, which have focused on controlling external inputs and restricting aquaculture scale. While the extent of aquaculture has already been brought to a relatively low level, the suppressive effect of dense SAV coverage (>30 %) highlights the importance of expanding vegetation restoration efforts across the lake to further reduce algal biomass and the risk of harmful algal blooms.