Particulate Organic Carbon (POC) is a key component of the marine carbon cycle, and also serves as a critical link connecting biological productivity, lateral carbon transport and carbon sink processes within the Land-Ocean-Aquatic Continuum (LOAC). Global POC observation has long been constrained by scarce in-situ measurements, cloud contamination, sensor discrepancies, and the uncertainties inherent in traditional empirical remote sensing algorithms. In this study, we integrated remote sensing data from three satellite sensors including SeaWiFS, MODIS and VIIRS, together with oceanographic and environmental auxiliary parameters, to establish a multi-sensor machine learning framework for retrieving global surface ocean POC.
Model validation demonstrated satisfactory retrieval performance: the coefficient of determination on the logarithmic scale reached 0.859.Using the proposed model, we generated daily 4-km resolution global surface POC products.The products can well capture the spatial patterns of POC distribution, with high POC concentrations clearly identified in coastal waters, high-latitude productive regions and western boundary current systems. The long-term global surface ocean POC dataset developed in this study provides solid support for assessing lateral carbon transport, coastal carbon budgets and carbon fluxes across the Land-Ocean-Aquatic Continuum under the impacts of extreme marine events and anthropogenic activities.
Keywords: Particulate Organic Carbon; Land-Ocean-Aquatic Continuum; Ocean Color Remote Sensing; Machine Learning; Multi-sensor Fusion; Marine Carbon Cycle