Ensemble machine learning and landsat observations reveal seasonal and spatial dynamics of water quality in a river-influenced estuarine system

Monitoring coastal water quality is essential for understanding ecosystem health and disentangling the combined effects of natural variability and human pressures in optically complex estuarine environments. However, conventional empirical algorithms often struggle to capture nonlinear bio-optical relationships, limiting their applicability for long-term and spatially extensive monitoring. In this study, we integrate high-resolution field observations collected using an Autonomous Surface Vessel (ASV) with Landsat-derived remote sensing reflectance (Rrs) band combinations to develop and evaluate machine learning (ML) models for estimating chlorophyll-a (Chla), colored dissolved organic matter (CDOM), and turbidity in the western Mississippi Sound (WMS). Ensemble-based algorithms, particularly Extreme Gradient Boosting (XGB) and Random Forest (RF), demonstrated superior predictive skill and robustness compared to other tested models. XGB achieved the highest accuracy for Chla (test R 2 = 0.957, MAE = 0.226 μg/L) and CDOM (test R 2 = 0.978, MAE = 0.708 ppb), while RF performed best for turbidity (test R 2 = 0.965, MAE = 0.220 NTU). Leveraging the best-performing models, we generated a decade-long (2014–2024) spatially continuous time series of Chla, CDOM, and turbidity, enabling systematic assessment of seasonal and spatial variability across the WMS. Results reveal distinct seasonal regimes, with elevated Chla during winter and summer linked to nutrient-driven productivity, and higher CDOM and turbidity during spring associated with enhanced terrestrial runoff and sediment transport. Spatial analyses highlight strong nearshore terrestrial influence and more stable, lower concentrations in marine-dominated regions. Importantly, this study identifies key spectral indices that enhance model performance by explicitly capturing parameter-specific bio-optical signals. Overall, this work provides a scalable, data-driven framework that advances estuarine water quality monitoring beyond traditional approaches, supporting long-term ecosystem assessment and informing management strategies under changing climatic, hydrological, and anthropogenic conditions.