Abstract
Coastal phytoplankton play a fundamental role in regulating marine primary productivity and carbon sequestration; however, their quantification in optically complex waters remains a major source of uncertainty in Earth system assessments. This study develops a hybrid observation–learning framework that integrates Sentinel-3 Ocean and Land Colour Instrument (OLCI) data with the Case-2 Regional CoastColour (C2RCC) processor and an Artificial Neural Network (ANN) to resolve phytoplankton chlorophyll-a (Chl-a) dynamics in the Sundarbans delta. A seasonally stratified dataset comprising 100 in-situ sampling stations (2021–2022) was used for model calibration and validation. The results reveal pronounced seasonal and interannual variability, with higher phytoplankton concentrations observed in 2022 relative to 2021, driven by intensified riverine nutrient fluxes and hydrodynamic forcing from the Ganges–Brahmaputra–Meghna (GBM) system. The ANN model achieved high predictive skill (R2 = 0.94–0.99), demonstrating robust performance and improved retrieval stability under monsoon-driven turbidity where conventional algorithms typically degrade. Spatial analysis identified persistent productivity hotspots within estuarine mixing zones, highlighting strong coupling between hydrological variability and coastal biogeochemical processes. By addressing a critical limitation in coastal ocean-colour retrievals, the proposed ANN framework provides a reliable and scalable approach for improving Chl-a estimation in data-scarce, high-turbidity environments. These findings advance the integration of Earth observation and machine learning for coastal ecosystem monitoring and support climate-resilient management, with direct relevance to carbon cycling, ecosystem services, and policy frameworks aligned with UN SDGs 13 and 14.