We show that forward simulations of global CO2 using an atmospheric transport model (ATM) at 0.5°×0.625° and 4°×5° resolutions differ significantly in vertical and meridional distribution. Comparing two observing simulation system experiments at 4°×5° resolution that assimilate pseudo observations sampled from the two forward simulations, we isolated the impact of coarse-resolution ATM error on regional flux estimates that a significant amount of annual carbon uptake from the ocean and tropics is improperly redistributed to the land and extratropics, respectively. In addition, this error leads to an underestimated seasonal amplitude in the northern extratropical land and a reversed seasonal phase in the northern extratropical ocean. The reversed seasonal phase has also been shown in a real data assimilation experiment and state-of-the-art inversions, suggesting that ocean glint retrieval error may not be as significant as previously thought and reasonable ocean flux estimates depend strongly on the accuracy of ATM.