Study Area
The Brazilian state of Amazonas is located in the central region of the Amazon basin (Fig. 1). With an area of 155,916,788.9 ha it is the largest state in Brazil and represents 31% of the Brazilian Legal Amazon (IBGE, 2020). We examined three legal categories of human occupied areas: SP, IL and CU. SP are naturally vegetated areas set aside for human colonization by people from outside the local region and often from outside the state (INCRA, 2004). These colonists are meant to seek sustainable livelihoods as small-scale farmers (INCRA, 2004). IL are areas occupied by indigenous peoples and recognized by the federal government (Brazil, 1988). They are permanently inhabited, used for productive activities, and essential for the preservation of natural resources necessary for the physical and cultural well-being of indigenous people. Indigenous people have exclusive use rights of their areas, except for the subsoil (Brazil, 1988). We assessed National and State Forests, categories of Sustainable Use CU, where authorized activities —primarily sustainable use of forest resources and scientific research— are considered less damaging to forests than those allowed in SP (Alencar et al., 2016; MapBiomas, 2021). The forest use in this case is usually the sustainable logging of native forests (Brazil, 2000). IL and CU (National and State Government Forests) were analyzed as Control Areas due to show little forest loss (MapBiomas, 2021) and are dispensed by NVPL (Brazil, 2012). In Amazonas there are 35 SP with a total area of 1,415,057.2 ha; 148 IL, with an area of 56,342,236.02 ha; 12 National Forests, with 9,983,819.96 ha; and eight State Forests that cover 2,584,929.62 ha (see appendix for more details). For convenience, we include areas that have at least some portion of their territory within the state of Amazonas.
Forest loss
We analyzed cover and land use maps from the MapBiomas collection 4.1 (MapBiomas, 2020) to quantify the amount forest loss in SP, IL and CU. The MapBiomas maps and datasets are freely available. The project classifies and geo-references information using LANDSAT data for all Brazilian biomes at a 30-m resolution. These data are generated annually with 95.9% accuracy level using an automatic classification routine applied to satellite images for the whole Amazon biome (https://mapbiomas.org/analise-de-acuracia). The analytical routine separates a number of cover and land use classes. However, our preliminary analysis identified that the routine overestimated the pasture class due to the miss-classification of agriculture or urban infrastructure classes to pasture. Thus, we limited our analysis to the native forest class and grouped all land under active human use into one category, an “Anthropogenic cover class.” Changes in the amount of this latter class amounted to forest loss or increase. Areas covered by water were omitted from analyses. Landsat data were analyzed for the years 2009, 2012, 2015 and 2018. Cover data for the years 2009 and 2012 were grouped and categorized as pre-implementation of NVPL (Pre-NVPL, while 2015 and 2018 data are affected by NVPL (Post-NVPL). For more details on the differences between the older Brazilian Forest Code and NVPL see Covre et al., 2015, Magano et al., 2021 and Pertille et al., 2017. The year 2018 was the most recent date with available land cover data in MapBiomas. We exclude from analysis any locations in settlements, indigenous lands, and conservation units that were created after the NVPL. SP shapefiles were provided by the National Institute for Colonization and Agrarian Reform (Instituto Nacional de Colonização e Reforma Agrária—INCRA) (http://certificacao.incra.gov.br/csv_shp/export_shp.py). We calculated land cover for each SP using its total area (perimeter). Eighty percent of the area of each SP must be preserved as a native forest Legal Reserves in the Amazon biome (Brazil, 2012). Legal Reserves for some SP are collective, rather than assigned to each individual family lot, while in others they may be assigned to each family lot. For many adjacent family lots, legal Reserves overlapped in area, potentially leading to an overestimation of these areas. Such overlap made it difficult to accurately assess cover of the Legal Reserves for each family lot. For simplicity, we therefore assumed that the total area of each SP equaled a single rural property. In addition to the requirement of forested Legal Reserves, additional natural areas, such as riparian corridors, steep slopes and other sensitive ecosystems, must also be maintained under forest cover as Permanent Protection Areas (Brazil, 2012).Shapefiles of IL were obtained from the National Indian Foundation (Fundação Nacional do Índio – FUNAI) (http://www.funai.gov.br/index.php/shape) and those for CU from the Ministry of Environment (Ministério do Meio Ambiente – MMA) (http://mapas.mma.gov.br/i3geo/datadownload.htm). All federally designated areas with at least part of their surface area within the state of Amazonas were included in the analysis.
We used QGIS v3.10 software (QGIS Development Team, 2020) to extract and analyze land cover and land use classes from the Google Earth Engine platform of MapBiomas (GEE; Gorelick et al., 2017). When shapefiles were not available on GEE (for example, all the SP and some IL and CU), we obtained data on the areas of human occupation from INCRA, FUNAI or MMA as a mask onto the Amazon raster of MapBiomas. We used the attribute table for each area under human use in each year to obtain cover and land use classes. The values obtained were extracted first as the total number of pixels and then converted into ha (each pixel ≈ 0.09 ha). We use the percentage of cover and land use classes for each area under human occupation pre- and post-NVPL, to avoid overestimation for large areas. All images were systematized in UTM projection Sirgas 2000 datum.
Anthropogenic and landscape factors
We evaluated the impact of area accessibility and landscape-scale land use as factors that potentially contribute to forest loss (Table 1). All variables and their potential impact are listed in Table 1. To assess accessibility, we measured the distances to nearest cities, roads and waterways. Distance to the variable of interest was measured as the minimum distance from the centroid of the SP polygon to the nearest city, road or waterway. Shapefiles of cities were obtained from the Brazilian Institute of Geography and Statistics (Instituto Brasileiro de Geografia e Estatística – IBGE) https://www.ibge.gov.br/geociencias/downloads-geociencias.html); of roads from the Instituto do Homem e Meio Ambiente da Amazônia (Imazon; https://imazongeo.org.br/#/); and of waterways from IBGE (http://www.metadados.geo.ibge.gov.br/) and the National Water Agency (Agência Nacional de Águas – ANA; https://metadados.ana.gov.br/geonetwork/srv/pt/main.home). For landscape-scale land use characteristics areas we used: 1) the population size of the county where the study area is embedded, 2) area size, 3) date established (creation time), 3) geographical location (latitude and longitude), and 4) external deforestation pressure calculated as the amount of forest loss in a 5 km buffer surrounding the borders of the areas under study. These variables have been previously identified as driving forest conversion to agricultural land use (see Table 1). County seats provide the most accurate population estimates for a geopolitical unit for the state of Amazonas. Population size was collected from IBGE (https://www.ibge.gov.br/estatisticas/sociais/populacao/9103-estimativas-de-populacao.html). Area size, creation time and geographical location data were collected from INCRA, FUNAI and MMA. Data for calculating external deforestation pressure were collected from the GEE platform in MapBiomas, as it provides a raster of polygons in the 5 km buffer area around study areas with land use categories identified. If the data were unavailable in the GEE platform, we generated these buffers using QGIS software and shapefiles for the state of Amazonas, as a base layer. All data on the anthropogenic factors were measured pre- and post-NVPL period.
Table 1
The anthropogenic and landscape factors examined, their impact and source of information
Factors | Impact | Sources |
Anthropogenic (accessibility) | | |
Distances: to cities | Proximity to cities tends to increase forest loss. For example: markets for agricultural and other products increase deforestation; roads permit access for tractors and heavy machinery, ease gasoline transportation and, movement by people. | Berenguer et al., 2021 Pfaff, 1999 Pfaff et al., 2015 |
to roads | Road proximity increases forest loss. Roads connect forests to cities and increase the flow of goods and people. Roads also increase the opening of illegal roads. | Barber et al., 2014 Brandão Jr & Souza Jr, 2006 Laurance et al., 2002 Jusys, 2018 |
to waterways | Function is similar to roads but without cars, because of these waterways are somewhat less harmful. Historically, this is one of the main means of travel in the Amazon | Barber et al., 2014 Laurance et al., 2002 |
Landscape-scale land use characteristics | | |
Human population size of nearest county | Areas with bigger population should show more forest loss. This because more people mean more demand for natural resource, greater flow of forest products, and more people to invade forests. | Laurance, 1999 Tritsch & Le Tourneau, 2016 |
Area size | Smaller areas should show greater forest loss. This because these areas are more exposed to surrounding property owned by non-indigenous people. | Cabral et al., 2018 de Almeida-Rocha & Peres, 2021 |
Establishment time | Amount of area deforested is linked to the age of areas. | Alencar et al., 2016 |
Geographical location (Latitude and Longitude) | Proxy for locating where deforestation happens most. For example, areas located to the south and west of the Brazilian Amazon have a greater record of forest loss. | Fearnside & De Alencastro Graça, 2006 A. M. dos Santos et al., 2021 |
External pressure (buffer 5km around) | Amount of area deforested outside the border of area. | Cabral et al., 2018 de Almeida-Rocha & Peres, 2021 Ferreira et al., 2005 |
Data analyses
We compare forest loss pre- and post-NVPL for SP, IL and CU using total area (ha) and percentage of area deforested. To assess the effectiveness of the NVPL for forest preservation, we compared the mean percentage of forest loss between pre- and post-NVPL periods by sampling unit. We test the significance of differences using a paired t-test. Percent of forest cover was transformed into log + 1 to achieve normality in data distribution. We analyzed forest loss (response variable) by anthropogenic factors (predictor variables) using the Negative Binomial GLM (Zuur et al., 2009), using the ‘MASS’ statistical package (Venables & Ripley, 2002). We tested for multicollinearity between predictor variables for each time frame and area class of human occupation, using the Spearman correlation test. None of the predictor variables were correlated (r < 0.75). GLM models were used to analyze the relationship between the response and predictor variables. We estimated the independent contributions of each predictor variable of the GLM models using hierarchical partitioning of the ‘hier.part’ package (Mac Nally & Walsh, 2004). All statistical analyses were performed with R version 3.5.1 (R Core Team, 2018).