Inadequate temporal and spatial coverage of observational data can obscure short-term or small-scale variations in air-sea carbon dioxide(CO₂) flux, potentially leading to inaccurate evaluations of oceanic CO₂ uptake or release. Existing research has largely struggled to meet the gap between underway observations that are asynchronous and buoy observations that cover only small areas. This has made it challenging to capture synchronous short-term fluctuations in air-sea CO₂ flux. To overcome these limitations, this study integrates data-driven machine learning techniques with process-based analysis to develop a remote sensing model for sea surface partial pressure of carbon dioxide (pCO₂) using Geostationary Ocean Color Imager(GOCI). This model enables the observation of daily air-sea CO₂ exchange processes in the East China Sea(ECS) during summer. Model validation shows good performance, with a root mean square error (RMSE) of approximately 27 μatm and an R² of around 0.96 on the test dataset. Though seawater pCO₂ did not exhibit a significant upward trend, the carbon sink effect in the East China Sea has strengthened, likely driven by increasing atmospheric pCO₂. However, estimates of the ECS’s CO₂ uptake for June 2011-2020, derived from GOCI, are about half of those obtained from polar-orbiting satellite products at a monthly scale. This discrepancy may be due to substantial data gaps in nearshore areas in the polar-orbiting satellite datasets. These findings, based on high-frequency observations, offer new insights into marine carbon sink dynamics and highlight the need for further research in this area.

Air-sea CO2 flux,sea surface pCO2,Remote sensing (RS),east china sea,east china sea,Semi-analytical algorithm (MeSAA),Machine learning,Geostationary Ocean Color Imager(GOCI)