Remote Sensing based Sea Surface partial pressure of CO2 (pCO2) and air-sea CO2 flux (FCO2) in the South China Sea (2003-2019)

The South China Sea (SCS) is one of the largest marginal seas worldwide. It includes a river-dominated, highly productive marginal sea on the north shelf and a wide, oligotrophic ocean-dominated basin with various dynamic sub-regions. Based on an in situ seawater partial pressure of CO₂ (pCO₂) datasets of 44 cruises/legs collected for the last two decades in the SCS, we proposed a seawater pCO₂ retrieval algorithm by combining the semi-mechanistic and machine learning (ML) methods (MeSAA-ML). The parameter selection strategy was based on the mechanistic analysis of pCO₂ variation, separating impacts of thermodynamics, biological activities, water mixing, and the atmospheric CO₂ forcing. We set a few semi-analytical parameters: pCO_{2_therm}, which was a proxy for the combined effect of thermodynamics and the atmospheric CO₂ forcing on seawater pCO₂; an upwelling index (UI_SST) and mixing layer depth (MLD) to characterize the multiple mixing processes; chlorophyll-a concentration (Chl-a) with remote sensing reflectance at 443 and 555 nm (Rrs(443) and Rrs(555)), which were the inputs to proxy the biological effect and other characteristics for distinguishing shelf, basin, and sub-regions. As the seawater pCO₂ and atmospheric pCO₂ ( pCO₂^air) have similar data values and characteristics in the vast SCS oligotrophic basin, it will cause instability of the model if one is input and the other is output; thus the difference between them (ΔpCO₂^sea-air) was set as the output, and the seawater pCO₂ was obtained finally by summing pCO₂^air and ΔpCO₂^sea-air. We compared several ML models, and the XGBoost model was confirmed as the best model. Completely independent cruise-based and observed datasets from Southeastern Asia Time-series Study (SEATS) were used to validate the satellite products, with low root mean square error (RMSE = 11.69 μatm) and mean absolute percentage deviation (APD = 1.59%). The increasing trend of satellite-derived pCO₂ (2.44 ± 0.24 μatm/yr) at the location of SEATS was found to be consistent with observed data. We presented that the SCS as a whole is a source of atmospheric CO₂, releasing an average of 11.00 ± 2.45 Tg C/yr from a total area of 3.32 × 10⁶ km^2, and the northern shelf is a sink (1.69 ± 0.53 Tg C/yr). The area-integrated CO₂ efflux over the entire SCS may decrease with a rate of 0.34 Tg C/yr during 2003–2019. This high-accuracy dataset with 1 km resolution provides a refined understanding of the air-sea CO₂ exchange dynamics in the SCS during 2003–2019.