Abstract Soil is the largest terrestrial carbon reservoir and can be a source or sink of CO 2 for the atmosphere, depending on management practices. CO 2 emissions from the soil surface (FCO 2 ) are directly related to the biological and physicochemical soil properties. Our objective was to estimate and spatialize the net ecosystem production (NEP) for the Brazilian territory, using visible (400−700 nm), near infrared (700−1100 nm), shortwave infrared (1100−2500 nm), (and mid‐infrared (2500−25,000 nm, 4000−400 cm − 1 ) reflectance spectroscopy, digital soil mapping, and machine learning. We created FCO 2 and carbon sequestration potential prediction models using soil physical, chemical, and microbiological properties as covariates, while the spatialization was based on a bare soil image, relief, climate, and soil mineralogy. A multivariate regression model with R 2 of 0.35 was fitted for FCO 2 and a spatial error model with R 2 0.76 for carbon sequestration. The accuracy of the spatialization ranged from 0.41 to 0.76, with a correlation of 0.56 in an external validation. The NEP map highlights negative balances in the Cerrado, Mata Atlântica, Caatinga, and Amazon biomes, with strong influence of mineralogy, where soils rich in iron oxides are below their carbon‐storage capacity. Our methodology can be used as an approximation of the C fixation potential in agroecosystems and contribute to climate change mitigation.
Rodríguez-Albarracín et al. (Fri,) studied this question.