Combined CO₂ measurement record indicates Amazon forest carbon uptake is offset by savanna carbon release

In tropical South America there has been substantial progress in atmospheric monitoring capacity, but the region still has a limited number of continental atmospheric stations relative to its large area, hindering net carbon flux estimates using atmospheric inversions. In this study, we use dry-air CO2 mole fractions measured at the Amazon Tall Tower Observatory (ATTO) and airborne vertical CO2 profiles in an atmospheric inversion system to estimate net carbon exchange in tropical South America from 2010 to 2018. Given previous knowledge of a bias due to undried samples in the airborne vertical profiles, we calculate the effect of this systematic uncertainty in our inverse estimates and propose a water-vapor correction to the airborne CO2 profiles. We focus our analysis on the biogeographic Amazon and its neighboring "Cerrado and Caatinga"biomes. Including the water-vapor correction changes the posterior ensemble median from -0.33 to -0.04 PgC yr^-1 with a posterior uncertainty of 0.33 PgC yr^-1 for the Amazon and for the Cerrado and Caatinga from 0.31 to 0.50 PgC yr^-1, with an uncertainty of 0.24 PgC yr^-1. Our estimates of carbon exchange include the contributions from both net vegetation exchange and release from fires. Assuming that the correction brings the observational data closer to the truth implies that the Amazon is a weaker sink of carbon and that the Cerrado and Caatinga is a larger source. We do not find a strong spatial shift of fluxes within the biogeographic Amazon due to the correction, nor do we find a strong impact on the interannual variations. Finally, to further reduce the uncertainty in regional carbon balance estimates in tropical South America, we call for an expansion of the atmospheric monitoring network on the continent, mainly in the Amazon-Andes foothills.