This study explores an optimal inversion strategy for assimilating the Orbiting Carbon Observatory-2 (OCO-2) column-averaged atmospheric CO2 concentration (XCO2) observations to constrain air-sea CO2 fluxes. The performance of different inversion set-ups is evaluated through Observing System Simulation Experiments (OSSEs) by comparing the optimized fluxes with assumed true fluxes. The results indicate that the conventional inversion, simultaneously optimizing terrestrial biosphere and air-sea fluxes, reduces root mean square errors (RMSEs) in regional monthly air-sea fluxes by up to 22–24% and 6–10% in the low (<40°) and high (>40°) latitudes, respectively, with up to 22% error reduction in global annual air-sea fluxes. These limited adjustments are associated with an order of magnitude higher variability of terrestrial biosphere fluxes compared to the air-sea fluxes. To isolate ocean signals within XCO2 variations, we employ a sequential inversion, first optimizing terrestrial biosphere fluxes with land XCO2 data and then optimizing air-sea fluxes with ocean XCO2 data while prescribing the optimized terrestrial biosphere fluxes. This approach achieves an 11% additional error reduction in global annual air-sea fluxes and a 33% further RMSE reduction in monthly air-sea fluxes in the southern high latitudes. However, we find that potential biases (+0.2 ppm) in ocean XCO2 measurements over this region could induce a 24% RMSE increase despite the application of sequential inversion. Our results show that sequential inversion is a promising technique for improving seasonal air-sea flux estimates in the Southern Ocean but mitigation of OCO-2 measurement biases is required for practical applications.