Study Area
The study area comprised the municipality (county) of Boca do Acre in the state of Amazonas (located in the southwestern portion of Brazilian Amazonia). A 3-km buffer around the municipality was also included in order to capture the maximum effect of the BR-317 (Rio Branco—Boca do Acre) Highway and avoid interference from edges in modeling. The study area covered a total of 24,133 km2, including the buffer that encompassed small portions of the neighboring municipalities of Lábrea and Pauini in the state of Amazonas and Manoel Urbano, Sena Madureira, Bujari, Porto Acre, and Senador Guiomard in the state of Acre (Fig. 1).
Boca do Acre comprises an area 21,951 km² and has annual average rainfall of 2000 to 2400 mm (Sombroek 2001); its economic base is cattle, which, with 84,954 head, is the fifth largest herd among the 62 municipalities in the state of Amazonas (Brazil, IBGE 2011). The boundary of the municipality to the east follows Highway BR-317, which is an important factor increasing its attractiveness for deforestation (Piontekowski et al. 2011).
Deforestation through 2012 in Boca do Acre totaled 2076 km² (9%), which was the second highest deforestation extent in the state of Amazonas. The annual increase of 54.9 km2 (0.24% of the municipality) in 2012 was the largest in the state (Brazil, INPE 2016). Amazonas is Brazil’s largest state with 1.57 million km2, an area approximately the size of the US state of Alaska and more than double that of the state of Texas.
Fieldwork
Fieldwork was carried out between 11 and 19 August 2012. Both small and large properties were visited to better understand the dynamics of land-use and land-cover change in the municipality. The routes of roads that were not included in official maps were collected using a GPS, and the widths of the main rivers at their maximum water levels were observed to supply information needed by the models.
Acquisition and Processing of Images
Imagery from the thematic mapper sensor on the Landsat-5 satellite was used (30-m spatial resolution) for the years 2005, 2008, and 2010. The images were obtained from the National Institute for Space Research (INPE) (http://www.dgi.inpe.br/CDSR/) for the relevant rows and points (3/66, 2/66, 1/66, 1/67, and 2/67). For 2012 we used images from the LISS3 sensor on the ResourceSat-1 satellite (23.5-m spatial resolution). These images were then resampled to 30 m. All images were georeferenced on the basis of the GeoCover 2000 mosaic of the US National Aeronautics and Space Administration (https://zulu.ssc.nasa.gov/mrsid/). The cartographic projection applied was UTM Zone 19 South and Datum WGS1984.
The portion representing the study area was cut out of a mosaic created from georeferenced images. The clipped images were classified into forest, non-forest, watercourses, secondary vegetation, and deforestation according to the methodology proposed by Graça and Yanai (2008) using as a classifier the maximum similarity determined in ENVI software. The result of this process for 2012 is shown in Fig. 2.
Delimitation of Hydrography and APPs
Watercourses and their associated APPs were divided into:
-
Watercourses < 30 m in width;
-
Watercourses ≥ 30 m in width computing the APP from the “regular” channel;
-
Watercourses ≥ 30 m in width computing the APP from the maximum water level.
For rivers with width less than 30 m, SRTM (Shuttle Radar Topography Mission) images were used (available at http://www.relevobr.cnpm.embrapa.br/download/) with a spatial resolution of 90 m resampled to 30 m. From this procedure a digital elevation model was developed for the study area using “Arc Hydro Tools” in ArcGIS software, where the watercourses (hydrography) were bounded as demonstrated by Alves Sobrinho et al. (2010).
Watercourses ≥30 m in width, were extracted from classified images (Fig. 2) and the width of each river was measured using the “measure” tool in ArcGIS for determining the APPs, as demonstrated by Reich and Francelino (2012). To measure the width of the rivers at the maximum water level a mask of the flooded areas of the Amazon region was used with 100-m spatial resolution produced from the synthetic-aperture radar sensor on the Japanese Earth Resource Satellite 1 (Hess et al. 2003, 2012). This mask was resampled to 30 m and, later, the floodplain of each river was measured using the “measure” tool to determine the APPs. The images were then clipped based on the study area, the rivers were grouped into width classes and the buffers for the APPs were created. Since the maximum resolution of the images was 30 m, we used modeled 30-m buffers to represent the APPs for all rivers with <30 m width (Table 2).
Table 2 Buffers built for APPs based on Law 12,651/2012
Maps of APPs generated for the rivers with width ≥30 m were added separately to the map of narrow watercourses, producing two separate maps, one with the APPs computed from the “regular” channel (Law 12,651/2012) and the other with the APPs computed from the maximum water level (Law 4771/1965).
The A-eco Model
The model used in the present study, denominated “A-eco,” was simplified from the AGROECO model that was created for simulating deforestation considering the influence of roads (IR) and the preservation offered by CUs (Fearnside et al. 2009). The main modifications refer to the simulated transition rate feedback and the inclusion of “regions” in the model.
In the AGROECO model transition rates were calculated in Vensim, which is a non-spatial simulation software (Ventana Systems Inc. 2007). In AGROECO, Vensim was coupled interactively with the 32-bit version of Dinamica-EGO, which performed the spatial allocation of the rates. In the A-eco model, transition rates were calculated using only the operators (“functors”) in the 64-bit version of Dinamica-EGO, thus no longer requiring the use of Vensim. This change was made because the coupling with Vensim hindered the use of maps with a large number of cells and because the 64-bit version of Dinamica-EGO is incompatible with Vensim. The 64-bit version of Dinamica-EGO allowed more-detailed raster maps to be used and has better performance in terms of processing time.
The approach consists of partitioning the “regions” into which we divided the study area so that processing is done separately for each of the “regions.” At the end of each iteration, the regions are grouped again into a single map. The iterations (repetitions of the model calculations) in this case represent years. In the present study, the regionalization of the study area allowed calculating the rates of deforestation, with projection of road construction specific to each region. This allowed capturing the particularities of deforestation for each agent or intrinsic focus of deforestation. In addition, using this approach it was possible to compute the transition rates for each region in order to construct the scenarios used for comparing forest loss and carbon emissions under different versions of the Forest Code and under different assumptions regarding enforcement.
Input Variables in the A-eco Model
The spatial resolution used in the input maps was 30 m and the cartographic projection applied was UTM Zone 19 South and Datum WGS-1984. The inputs to the model were:
-Map of static variables: Vegetation (Brazilian Institute of Geography and Statistics: IBGE); soil (IBGE); altitude (SRTM); slope (derived from the altitude map); Watercourses (extracted from the land-cover map); roads (from the Remote Sensing Center, Federal University of Minas Gerais: CSR/UFMG) updated with the roads identified in satellite images (2005 and 2012) and on-site during fieldwork in 2012; CUs (from IBGE and the Amazonian Protection System: IBGE/SIPAM); and ILs (from the National Indian Foundation: FUNAI);
-Maps of friction and attractiveness: Created in Dinamica-EGO through multi-criteria analysis by assigning values (weights) to features that have a predisposition to either attract or repel the construction of roads and, consequently, speed or slow deforestation. Factors of attraction are roads and watercourses, while repulsive factors are ILs, CUs and areas with steep slopes (Soares-Filho et al. 2009);
- Land-cover map: For the year 2012 (Fig. 1);
- Road map: (CSR/UFMG) updated with roads identified in satellite images for 2005 (calibration phase) and 2012 (simulation phase) and on-site in 2012; this is necessary for the model’s “road-builder” module and for calculation of transition-probability maps and rates of deforestation within the program;
-Map of regions: This map compartmentalizes the study area into “regions” and projects deforestation in a different way for each region (Table 3 and Fig. 3). This considers the level of protection of the area, deforestation dynamics along rivers and roads, and APPs. The study area was divided into six regions in the “Baseline Scenario” and the “1965 Scenario” and into seven regions in the “2012 Scenario” (Table 3). For each region, the transition rates, weights of evidence, and dynamics of road construction were distinct.
Table 3 Regions used in the A-Eco model
Weights of evidence represent the susceptibility of a cell to changing from one state to another. For example, cells in the forest class that are located away from deforested areas or from roads are less susceptible to changing from forest to deforestation, since they have lower weights compared with forest cells located next to these areas. The transition rates represent the overall amount of change, i.e., they determine the number of cells that will undergo the transition in each iteration. The transitions used were:
Deforestation rates were obtained according to the equation used by Yanai et al. (2012), where the rates are updated in each iteration in accord with the increment of roads in the model. Rates of cutting and regeneration of secondary vegetation were determined from the calculated transition matrix in Dinamica-EGO.
Calibration and Validation
“Calibration” refers to the “estimation and adjustment of model parameters and constants to improve the agreement between model output and a data set,” while “validation” means that a model is “acceptable for its intended use because it meets specified performance requirements” (Rykiel 1996). In the process of calibration, weights of evidence and transition rates were determined using the 2005–2010 period.
Validation was carried out by applying the weights and rates found in the same study area for the period from 2005 to 2012. As input, the simulation used the land-use map for 2005 and ended by simulating the map for 2012, which was then compared to the map of real deforestation by that year (from PRODES: Brazil, INPE 2016) in an effort to achieve the maximum possible spatial similarity.
The weights and the rates were obtained and applied to each region. In the scenario for Law 12,651/2012, the sizes of the regions were changed and the weights and transition rates were therefore recalculated for this scenario.
For the allocation of the land-cover classes, the model was validated spatially with 51% minimum similarity for a 5 × 5 pixel window (Fig. 4). For the quantitative validation, the difference between the real map and simulated map are given in Table 4.
Table 4 Quantitative validation the A-eco model applied to the study area
The difference between the real and the simulated map in the validation step shows that the model underestimated the forest and deforestation classes and overestimated the secondary vegetation (Table 4). The overestimation of the amount of secondary vegetation can be attributed to the difference between the calibration period (2005 to 2010) and the 2012 map used for validation, since in 2012 there was 24% less secondary vegetation than in 2005 and 30 percent less than in 2010.
Simulated Scenarios
Three deforestation scenarios were simulated from 2013 to 2025, using Dinamica-EGO software:
-Baseline Scenario: The transition rates consider the deforestation trend in recent years. There is no restriction on the use of APPs, a premise that is closest to the real situation, considering that only a few of the landholders respect the legislation in Boca do Acre. In this scenario, only six regions (Table 3) were considered and the APPs were based on Law 4771/1965.
-Law 4771/1965 Scenario (1965 Scenario): This is a scenario where forest legislation regarding APPs along the banks of watercourses (calculated on the basis of the maximum water level) was fully respected beginning from the first iteration (i.e., no deforestation occurs in these areas). This scenario assumes that starting to respect the Forest Code in private properties would stimulate “leakage,” where deforestation that would otherwise occur in the APPs moves elsewhere to areas of intact vegetation that are unprotected (public and non-designated forest areas) (Sparovek et al. 2012). All of the gross rate of deforestation and of cutting secondary vegetation in the APPs was transferred and recalculated in terms of the net rate for cutting secondary vegetation in adjacent areas. This made it possible to observe and compare the effect of the two legislations in the scenarios. The APP region in the 1965 Scenario was the same as that used in the Baseline Scenario.
-Law 12,651/2012 Scenario (2012 Scenario): The APPs built for this scenario were based on the “regular” channel of each watercourse. In these areas, the Forest Code was fully respected from the first iteration, thus preventing deforestation in APPs. In addition, the “APP2008 region,” which refers to deforestation through 2008 in APPs, was added. Under Law 12,651/2012, these cleared areas are exempt from being fully recovered, and agricultural activities can be continued. Requirements for recovery of the vegetation are in accordance with the size of the property. The largest restoration is required for properties with areas greater than four tax modules (i.e., 4 × 100 ha in Boca do Acre), with the width of APP restoration being at least 20 m and the maximum requirement being 100 m. The transitions for cutting secondary vegetation were maintained in the simulation for the APP2008 region due to the spatial resolution used being 30 m and because it is assumed that the minimum required under the 2012 Forest Code will be adopted. Note that the APP2008 region was only used in the simulation of this scenario, this area being included in the APP region in all of the other analyses.
In all scenarios a mask was used to nullify values in urban areas in order to prevent regeneration in these areas, even when they were located on the banks of rivers. These sites have human occupation, impeding regeneration of the vegetation.
Estimates of Carbon Stock Loss and Annual Carbon Emissions
Biomass values were obtained based on the forest type indicated at each location by the vegetation map of the IBGE (Brazil, IBGE 1992) and the dry mass value above ground and below ground for each forest type calculated by Nogueira et al. (2008a). For areas of forest with a predominance of bamboo, which is abundant in Boca do Acre (Nelson et al. 2006), we used the methodology presented by Vasconcelos et al. (2013). This methodology uses the values for biomass of trees and palms with diameter at breast height (measured 1.3 m above the ground or above any buttresses) greater than 5 cm (Nogueira et al. 2008b), and adds the values obtained from the biomass equations developed by Nelson et al. (1999) for bamboos and by Gehring et al. (2004) for lianas, applied to the inventory carried out in Acre by de Oliveira (2000). Finally, necromass values are added (Nogueira et al. 2008a), obtaining the total biomass for the forest type with predominance of bamboo.
To determine the loss of carbon stocks, biomass values above ground and below ground were multiplied by the average proportion of carbon in dry biomass as determined by da Silva (2007). This proportion is 0.485.
The calculation of annual emissions included the secondary vegetation biomass based on the mean biomass growth rate of secondary vegetation found in abandoned cattle pastures in the municipalities of Paragominas and Altamira, Pará (Fearnside and Guimarães 1996). The average age of secondary vegetation was considered to be 5 years (Almeida 2009). Carbon was considered to represent 45% of the dry biomass of secondary forest (da Silva 2007).