Sediment-associated ammonium-N increased benthic chlorophyll-a in streams: evidence from SHAP integrated machine learning analysis

Predicting benthic chlorophyll-a (Chl-a) is crucial for assessing freshwater ecosystem health, yet there is little information about the input of sediment data in predicting benthic Chl-a. We address this gap by evaluating the drivers of benthic Chl-a using an integrated dataset (n = 135) encompassing water chemistry, stream physical characteristics, and sediment variables from the Oujiang River basin, China. We compared four machine learning (ML) models: extreme gradient boosted trees (XGBoost), random forest (RF), support vector machine (SVM), and CatBoost. XGBoost demonstrated superior predictive performance (RMSE = 1.7, MAE = 1.2, R² = 0.97) compared to RF (RMSE = 2.9, R² = 0.91), SVM (RMSE = 2.8, R² = 0.90), and CatBoost (RMSE = 3.1, R² = 0.90), and was selected for further analysis. SHAP (SHapley Additive exPlanations) analysis revealed that sediment-associated ammonium nitrogen (NH₄-N_sed) was the most influential predictor (SHAP value ≈ 0.28), followed by water column ammonium-N (NH₄-N; SHAP value ≈ 0.04) and sediment total phosphorus (TP_sed; SHAP value ≈ 0.03). The prominence of sediment variables highlights their critical role in providing stable habitats and nutrient sources for benthic algae. Furthermore, two-way partial dependence plot analysis indicated that wider stream sections and higher NH₄-N_sed concentrations enhance N levels, likely due to increased sediment–water exchange and nutrient release facilitated by reduced flow velocity. This study highlights the significant influence of NH₄-N on benthic Chl-a dynamics, offering new insights into the ecological processes governing algal growth. Incorporating sediment variables significantly improves predictive model accuracy and provides practical implications for freshwater ecosystem management, emphasizing the need to consider sediment characteristics in monitoring and conservation strategies.