Estimating air–sea CO<sub>2</sub> exchange in coastal and semi-enclosed seas is challenging due to limited spatial and temporal observational coverage. Atmospheric inverse modelling offers a complementary approach, although its ability to resolve regional marine carbon fluxes is still uncertain. In the Baltic Sea, intensive shipping further complicates the regional carbon budget through substantial anthropogenic CO<sub>2</sub> emissions. This study presents one of the first assessments combining air–sea CO<sub>2</sub> exchange and ship-derived CO<sub>2</sub> emissions in the Baltic Sea using both observation-based and atmospheric inversion approaches.</p> <p>Air–sea CO<sub>2</sub> fluxes for 2019 to 2021 were estimated using two independent methods. The first method combined a bulk flux calculation, which includes sea–air pCO<sub>2</sub> differences, wind-speed-dependent gas transfer velocity, solubility, and sea-ice cover, with shipping emissions simulated by the Ship Traffic Emission Assessment Model (STEAM). The second method used atmospheric CO<sub>2</sub> observations together with atmospheric transport modelling in the CIF-FLEXPART inversion framework to estimate surface CO<sub>2</sub> fluxes, providing an alternative observation-based approach. The results were analyzed for the entire Baltic Sea and for 13 HELCOM sub-basins.</p> <p>The STEAM-corrected bulk method indicated that the Baltic Sea acted as a net CO<sub>2</sub> sink during 2019 to 2021, with a mean flux of -1.79 mmol m<sup>−2</sup> d<sup>−1</sup>. Without shipping emissions, the estimated flux was -3.76 mmol m<sup>−2</sup> d<sup>−1</sup>, showing that ship traffic increased the net flux by approximately 1.97 mmol m<sup>−2</sup> d<sup>−1</sup>. In contrast, the CIF-FLEXPART inversion estimated a net source of 3.92 mmol m<sup>−2</sup> d<sup>−1</sup> for the Baltic Sea. The climatological bulk-flux estimate for 2003 to 2021, excluding shipping emissions, gave a near-neutral mean flux of 0.25 mmol m<sup>−2</sup> d<sup>−1</sup>. The basin-scale monthly fluxes showed moderate overall agreement between the two methods (mean Pearson correlation <em>r</em> = 0.49), with stronger correlations (<em>r</em> = 0.62–0.79) in several southern basins where observational coverage was greatest.</p> <p>The contrasting estimates of the Baltic Sea carbon balance show that considerable uncertainties remain in both bulk method and inversion method approaches and that shipping emissions need to be taken into account. More continuous and reliable observations, improved representation of coastal carbon processes, and further development of atmospheric inversion systems are needed to better constrain regional marine carbon budgets.