Comparative analysis of SWAT and SWAT coupled with XGBoost model using Optuna hyperparameter optimization for...

Agricultural runoff leading to nitrate (NO₃-N) and orthophosphate (PO₄-P) contamination poses significant environmental and public health risks. This study integrates the Soil and Water Assessment Tool (SWAT) with eXtreme Gradient Boosting (XGBoost), optimized using Optuna hyperparameter tuning, to enhance predictions of nutrient concentrations in water bodies. The hybrid SWAT-XGBoost model combines physical hydrological processes with machine learning for improved accuracy. The optimized model demonstrated significant improvements over SWAT alone, with R² values increasing by up to 50 % and RMSE decreasing by up to 75 % across datasets. These results highlight the enhanced predictive capabilities of the hybrid approach in capturing nutrient transport dynamics. SHAP analysis further identified key factors, such as sediment dynamics and nutrient mineralization, as dominant drivers of contamination, providing actionable insights for effective watershed management. By integrating SHAP analysis, the study identified key processes influencing nutrient transport, such as sediment dynamics and nutrient mineralization, offering deeper insights into pollution pathways. This approach provides a scalable and adaptable framework for improving watershed management, supporting sustainable agricultural practices, and mitigating contamination risks. The findings highlight an approach that combines physical modeling with machine learning to effectively address complex environmental challenges.