Methods

Datasets

To calculate the FC balance, we employed the CAR database from January 2022. We only analyzed private properties, excluding thus settlement projects and collective lands, such as quilombola (maroon) territories. We disregard CARs that overlap with conservation units, except Permanent Preservation Areas – APAs, and indigenous lands, according to the MPF⁴ protocol criteria, and those canceled by the SFB. In addition, the model employs as input maps of state and municipal limits, municipal fiscal modules, the limit of Legal Amazon, vegetation distribution, drainage, land use, deforestation, and protected areas.

We used the map of municipalities from IBGE (Brazilian Institute for Geography and Statistics), so as to assign the municipality geocode to each CAR record. Each Brazilian municipality has a size for the fiscal module⁵. Through the municipality geocode, the size of fiscal module is attributed to the CAR. The FC considers as a small property those from one up to four fiscal modules, a medium property those between 4 and 15 fiscal modules, and as large properties the ones larger than 15 fiscal modules.

Public domain nature conservation units (except APAs) and demarcated indigenous lands⁶ are used to calculate the percentage of a municipality and a state occupied by these land use categories, and the resulting numbers are assigned to the CAR via the IBGE geocode.

The boundary of the Legal Amazon has been extended several times as a result of changes in the political division of the country. For our modeling exercise, the limit of the Legal Amazon⁷ was used to set the requirements of Legal Reserve (LR).

Vegetation formations from the Radam-Brasil are used to determine the percentage of LR in the Legal Amazon, i.e., 80% for forest formations, and 35% for other vegetation types. Outside of Legal Amazon, the FC establishes the percentage of 20% of the property for LR. When a property overlaps different biomes (i.e., Cerrado and Amazon), a weighted average is applied.

For calculating APP conservation and restoration requirements, we used the drainage maps, including springs and water bodies, from the National Water and Sanitation Agency (ANA).

Our Land-use map is a mosaic composed of water bodies from ANA, land-use categories identifying native vegetation remnants and agricultural areas (so-called “consolidated areas) from Mapbiomas, (collection 6), and maps of annual deforestation from PRODES-Amazon and PRODES-Cerrado^8-12.

The model

We have applied the rules and definitions of the Forest Code (FC)¹ for each rural private property from the CAR dataset obtained from SICAR—the Online National Rural Environmental Registry System. In doing so, we provide estimates of the FC level of compliance, i.e., landowners’ deficits—areas that must be reforested at the owners’ expenses, or and surpluses, areas of native vegetation that exceed the FC conservation requirements (Fig. 1).

To this end, we have developed an innovative geoprocessing set of tools that handle big data by employing PostgreSQL and PostGIS extension, and Dinamica EGO 7 freeware¹³. This system takes advantage of full parallel processing¹⁴. Dinamica EGO parallel execution system uses a variable number of execution threads (called workers) boosted by task-stealing algorithms to provide load balancing and increase the flexibility for running parallel tasks. In theory, all model components can run in parallel, including independent operators, loops, and map tiles^15,16.

Substantive improvements in our computing capacity and modeling tools enabled fine-scale reanalysis of the FC^3,17, making it feasible to estimate the FC balance; i.e., level of compliance, throughout the Brazilian territory at the property-level. These advances allowed us to frog-leap from a 60-meter spatial resolution³ to a 5-meter (the narrowest APP width for restoration) by using parallel processing and memory allocation optimization. All processing relied on the computing resources of the Center for Remote Sensing¹⁸ of the Federal University of Minas Gerais (Belo Horizonte, Brazil). All calculations can be replicated by downloading the software and opening the FC models (csr.ufmg.br/radiografia_do_car) using Dinamica EGO’s user-friendly graphical interface.

To calculate the forest balance (deficit and surpluses), the model first calculates the total area of each property where the law is applicable. Next, the model generates buffer sizes along river, spring and water bodies according to the rules of the FC (Fig. 1). To define the buffer width either for APP conservation or restoration requirements, the model uses the property size (defined in the number of fiscal modules as specified for each municipality) and river width. To calculate riparian APP buffer width to be restored, the model applies a set of rules so-called “escadinha” (little ladder), which specifies the buffer size to be restored according to the property size (defined in the number of fiscal modules as specified for each municipality) and river width.

Thereafter, the model applies the FC rules according to the property sizes to define LR requirements. In the Amazon biome, LR can be reduced by up to 50% in municipalities that have more than 50% of their territory occupied by conservation units and indigenous reserves (Art. 12, II – § 4). The FC exempts small landowners (up to 4 fiscal modules) to restore LR deficit (Art. 67). In addition, the law establishes a maximum percentage of the property for LR restoration (Art. 61-B), depending on the total extent of its riparian APPs (Art. 15). Here we consider the increase in the size of the Legal Reserve (LR) from 50% to 80% established by Provisional Measures 1,511 of 1996 and 2,166-67 of 2001. The FC also establishes that the percentage of LR for forest restoration can be reduced to 50% in the Amazon states that have the ecological-economic zoning approved.

In addition, article 68 of the of FC reviewed in 2012 states that landowners that suppressed native vegetation respecting the legislation in force at the time need not to recover LR to the percentage mandated by the current law, i.e., 80%. Therefore, it corrected conflicting past legislation to bring to legality “properties pushed into illegal status”.

The difference in LR definition is the reason that we separated deforestation before 2002 and this year onwards. Deforestation before and after the decree must be analyzed with respect to different specification of LR size. Note that the time of deforestation occurring is also evidence for article 68 of the 2012’s FC as specified in Paragraph 1, as follows:

“Owners of rural properties may prove their history of occupation by documents such as the description of historical facts of the region, commercialization records, data, agricultural activities, contracts and bank documents related to production, and by all other means of evidence permitted by law”¹.

The main sequence to obtain the FC balance is depicted in Fig.1. For each property, the model subtracts the total area required for LRs from the areas of native vegetation remnants within each private property and the areas of native vegetation within the customized APP buffer sizes to arrive at the level of compliance. We define a positive result as an environmental surplus and a negative result as an environmental deficit.

Uncertainties in the FC estimates arise from overlaps of properties and different drainage bases, as well as the accuracy of the mappings.

The SeloVerde platform

For traceability purpose the results per property are integrated with annual deforestation maps^10,11, soy cropping maps (Mapbiomas, collection 7), and GTA documents (permit to transport animals). The analyses of FC thus allow us to map potentially legal or illegal post-2008 deforestation (in APP or below a minimum of RL) — the amnesty deadline for past-deforesters³ – so as to link deforestation to cattle and soy supply from each cattle ranch or soy farm on the SeloVerde Platform.

The CAR 2.0

In turn, the CAR 2.0 uses mapping and spatially explicit modeling based on high resolution images to automatically analyze the environmental compliance of each rural property through the methods described above. Properties without overlaps and without significative LR and APP deficits are, as a result, directed to the Canal Verde (Green light channel), a simplified procedure for joining the PRA based on the landowner’s self-report, hence without the need to rectify the RL features, hydrography, land use and others features input by the landowner.

Datasets

The model

The SeloVerde platform

The CAR 2.0

References