Air-sea CO2 flux in the Gulf of Mexico from observations and multiple machine-learning data products

Quantifying air-sea carbon dioxide (CO₂) flux from observations is subject to uncertainties due to missing data, uneven data distribution, and a relatively short observation period in the Gulf of Mexico (GOM). Despite the publication of multiple seawater partial pressure of CO₂ (pCO_2sw) products, their reliabilities in the GOM have been relatively understudied. We compare the Surface Ocean CO₂ Atlas (SOCAT) observation-based synthesis with eight regional and global machine-learning pCO_2sw data products in the GOM. SOCAT reveals significant spatial and seasonal variations in pCO_2sw in the GOM owing to complex local nonthermal (physical and biological) dynamics, particularly in the Louisiana Shelf (LAS) and Western Florida Shelf (WFS). The regional pCO_2sw data product outperforms the global products in capturing small-scale pCO_2sw variations. When averaging climatology across the entire northern GOM, the spatial heterogeneity of pCO_2sw and CO₂ flux resulting from local nonthermal processes tends to counterbalance across the entire GOM in all pCO_2sw data products. Consequently, the regional data product and the ensemble mean of seven global products yield pCO_2sw climatology that closely aligns with the SOCAT observations with a small difference (< ±3 µatm). During the overlapping period from 2003 to 2017 (15 years), the average flux from the eight products indicates that the entire GOM is CO₂-neutral, with an ocean uptake flux of 0.08 ± 0.12 mol C/m²/yr or 1.50 ± 2.25 TgC/yr, which is about 0.6 % of the global coastal ocean CO₂ sink. Observations show that the pCO_2sw trend also exhibits notable spatial differences, with the river plume area acting as an increasing CO₂ sink and the WFS acting as an increasing CO₂ source. Due to limited observations and large spatiotemporal variations, the true values of the decadal trend still have large uncertainties in the highly dynamic river plume area. In most other subregions, pCO_2sw increases following atmospheric CO₂. Uncertainties persist across all pCO_2sw data products in simulating the decadal trend, given that the regional product displays essentially no trend (<0.5 µatm/yr), while the ensemble average of global products exhibits a trend that follows atmospheric pCO₂ (∼+2.0 µatm/yr). Our findings demonstrate that existing pCO_2sw data products effectively simulate the climatology of pCO_2sw in the GOM, providing valuable information for CO₂ flux quantification in the GOM. Future research should emphasize the development of pCO_2sw data products designed to accurately predict small-scale variations and temporal shifts, while also delving into the underlying dynamics responsible for these changes.