First, we use a property-level boundaries dataset, which includes the amounts of Legal Reserves (LR), Permanent Preservation Areas (PPAs), and the surplus of LRs (i.e., that can be legally cleared) 16. Second, we use an aboveground carbon density map at 50m pixel resolution for Brazil 25. Third, we use projections of unprotected native vegetation in private lands, considering rural establishments that can conduct legal clearings 16. Fourth, we employ a modelling approach that estimates the probability of the existence of native vegetation until 2025 26. This model considers topography, soil properties, climate data, distance to transportation corridors, and legally protected areas, including indigenous lands. This unprecedented combination of datasets allows us to estimate the potential future legal deforestation with different levels of risk.
We took a conservative approach, excluding from this analysis the areas that Freitas et al. (2018)16 projected as potential private properties to fill spatial gaps in the Brazilian territory. We only considered the rural establishments officially registered and identified at Brazilian land databases. Therefore, while Freitas et al. (2018)16 estimated ~101 Mha of unprotected native vegetation within private lands, we only considered ~69.2 Mha. We used Fendrich et al.'s (2020)26 land cover model to classify these unprotected areas according to their risk of conversion. Ours is a conservative estimate because we only considered properties with accurate land registries in the land database constructed by 16 and updated by 14, i.e., we excluded properties modelled to fill spatial registry gaps.
The land tenure database presented in Freitas et al. (2018) 16 and updated by Freitas et al. (2018) 14 was used as starting point for the analysis presented here. This database comprises public and private properties in Brazil such as Indigenous Lands, Conservation Units, Quilombola Territories, and private rural properties from the Rural Environmental Registry (Portuguese acronym CAR) and the georeferenced properties of the National Agrarian Reform Institute (Portuguese acronym SIGEF). The database also includes information on the compliance of these rural properties with the Brazilian Forest Code. It has information related to the area of native vegetation and the amount of aboveground carbon stock within each rural property 16,25. With the existent variables, it is also possible to identify the part of the native vegetation (or its carbon stocks) that is both legally protected or has the potential to be legally deforested (the latter called hereafter as unprotected native vegetation or unprotected carbon stocks). It is essential to highlight that our land tenure map differs from 14,16 because we only considered properties with accurate land registries in the land database constructed by 16 and updated by 14, i.e., we excluded properties that were modelled to fill spatial registry gaps.
Fendrich et al. (2020) 26 presented maps of the probability of the existence of native vegetation in a given pixel for the years 2017 and 2025. The probabilities are calculated based on a spatially explicit regression model that explores the relation of land cover maps of the Mapbiomas project (2019)24 and spatial variables (drivers of land cover change), such as topography, soil properties, climate data, distance to transportation corridors and legally protected areas, including indigenous lands. The land cover maps were reclassified to four classes in the model, namely, native vegetation, pasture, agriculture, and other uses. The existence of these four land cover classes was analyzed for every pixel in the entire period (from 1985 to 2017). The model captured the relation of the land cover classes and the spatial variables. Based on alternative future climate and policy scenarios (aggressive, business as usual, and conservative scenarios), Fendrich et al. 26 estimated the probability of the existence of native vegetation, agriculture, and pasture in Brazil for the year 2025.
Here we used the business as usual scenario to generate a map of the variation of native vegetation probability (VNVP map) between 2017 and 2025, where positive values represent pixels with an increase in the probability of native vegetation existence until 2025. In contrast, negative values represent pixels with a decrease in the probability of native vegetation existence until 2025. Further, we have combined the rural properties with unprotected native vegetation and the VNVP map to extract the average probability for each of these properties. Fig. S1 expresses the distribution of the average probabilities within rural properties.
The distribution of the VNVP shows that 99.6% of the properties with unprotected native vegetation have VNVP between -5% and +5%, with the majority of the properties (79,2 % of the total) presenting negative values (Fig. 2). Based on the distribution of the VNVP, we defined four classes of risk of unprotected native vegetation to be deforested:
- Properties with VNVP lower than -3% = High risk
- Properties with VNVP between -3% and -1%= Medium risk
- Properties with VNVP between -1% and 0% = Low risk
- Properties with VNVP higher or equal than 0% = No risk