Enhanced prediction of total organic carbon in a large complex watershed (the Han River, South Korea) by integrating machine learning with real-time in-situ water quality parameters and fluorescence intensities.

Effective water quality monitoring is critical for protecting aquatic ecosystems and supporting regulatory management. Total organic carbon (TOC), a key indicator of organic pollution, reflects both natural and anthropogenic influences. Building on recent advances in machine learning and optical monitoring, this study developed a machine learning (ML) model that integrates real-time in-situ water quality (WQ) parameters and dissolved organic matter (DOM) fluorescence peaks for TOC prediction across diverse waterbody types and seasons observation from the Han River watershed, South Korea. Five fluorescence peaks (Peaks B, T, A, M, and C) identified from fluorescence excitation-emission matrix (EEM) spectroscopy, along with four in-situ WQ parameters (temperature, pH, dissolved oxygen, and electrical conductivity (EC)), were used as input variables. The XGBoost algorithm outperformed traditional multiple linear regression (MLR), achieving a higher R2 (0.756 vs. 0.460) when using combined WQ and fluorescence inputs, thus demonstrating superior predictive performance. Feature importance analysis using Shapley additive explanations (SHAP) indicated that EC and humic-like fluorescence peaks (A and C) were the most influential predictors. EC dominated TOC prediction in industrial complex rivers, whereas humic-like fluorescence features were more important in non-industrial water bodies. Seasonal analysis further showed that humic-like substances consistently contributed to TOC prediction throughout the year. Overall, this study demonstrates that integrating real-time, field-measurable DOM fluorescence with ML techniques can significantly enhance TOC prediction accuracy and interpretability, providing a practical and scalable approach for addressing regional and seasonal variability in water quality across complex aquatic systems.