Eutrophication, driven by nutrient overload, monitored through Chlorophyll-a (Chla) levels in inland and coastal waters. Traditional in-situ methods faced challenges such as time consumption, labor intensity, and spatial-temporal limitations. Remote sensing and machine learning offered solutions but faced challenges due to insufficient in-situ data, diverse water characteristics, and model transferability. Existing global models emphasized data quantity over quality, lacking in comprehensive analysis of relationships between water quality parameters and remote sensing features. This study aimed to enhance global Chla prediction by improving data quality and identifying key features using Earth Observation (EO) data. Two feature groups were analyzed: Group 1 (reflectance from single bands and band ratio indices) and Group 2 (reflectance from single bands and mathematical transformations of bands). Machine learning models, including Random Forest (RF), Least Squares Boosting (LSBoost), Support Vector Regression (SVR), and Gaussian Process Regression (GPR), were evaluated for overall, cross-validation, and transfer-validation performances. GPR performed best overall (R² = 0.95, RMSE = 2.82 µg/L in Group 2), while SVR was weakest. Transfer validation with the external lake showed RF (R² = 0.73, RMSE = 12.39 µg/L) and LSBoost had the strongest transferability. Spatial-temporal predictions of the transferred validated lake for 2023–2024 demonstrated reliable and consistent Chla distribution patterns, capturing seasonal variations. The present study highlights the potential of the proposed framework for global Chla monitoring in inland waters, also, emphasizing the potential in areas outside the training dataset.
How to cite: Moe, A.C., K.C. Saddi, R. Zhuang, D. Miglino, J.A. Saavedra Navarro, and S. Manfreda, Global-Scale Chlorophyll-A Monitoring for Inland Lake Water Quality Framework: Advancements, Machine Learning Models, and Transferability Challenges. SSRN Preprints, 2025. [pdf]