Historical land use change and associated carbon emissions in Brazil from 1940 to 1995


Corresponding Author: C. C. Leite, Department of Agricultural Engineering, Universidade Federal de Viçosa, Viçosa 36570-000, Brazil. (ccleite@gmail.com)


[1] The evaluation of impacts of land use change is in general limited by the knowledge of past land use conditions. Most publications on the field present only a vague description of the earlier patterns of land use, which is usually insufficient for more comprehensive studies. Here we present the first spatially explicit reconstruction of historical land use patterns in Brazil, including both croplands and pasturelands, for the period between 1940 and 1995. This reconstruction was obtained by merging satellite imagery with census data, and provides a 5′ × 5′ yearly data set of land use for three different categories (cropland, natural pastureland and planted pastureland) for Brazil. The results show that important land use changes occurred in Brazil. Natural pasture dominated in the 1950s and 1960s, but since the beginning of 1970s it has been gradually replaced by planted pasture, especially in southeast and center west of Brazil. The croplands began its expansion in the 1960s reaching extensive areas in almost all states in 1980. Carbon emissions from historical land use changes were calculated by superimposing a composite biomass map on grids of a weighted average of the fractions of the vegetation types and the replacement land uses. Net emissions from land use changes between 1940 and 1995 totaled 17.2 ± 9.0 Pg-C (90% confidence range), averaging 0.31 ± 0.16 Pg-C yr−1, but reaching up to 0.47 ± 0.25 Pg-C yr−1 during the 1960s and through 1986–1995. Despite international concerns about Amazon deforestation emissions, 72% of Brazil's carbon emissions during the period actually came from deforestation in the Atlantic Forest and Cerrado biomes. Brazil's carbon emissions from land use change are about 11 times larger than its emissions from fossil fuel burning, although only about 18.1% of the native biomass has been lost due to agricultural expansion, which is similar to the global mean (17.7%).

1. Introduction

[2] Large-scale changes in land cover have an important impact on atmospheric composition, climate change, regional water cycle, carbon cycle, environmental quality, species diversity and the productivity of terrestrial ecosystems [International Geosphere-Biosphere Programme, 2005]. The impact of the land system on the earth system is complex and nonlinear, so that the simulation of environmental effects, especially the climate effect, of land-cover change remains very rough.

[3] Land use legacies can persist in forest ecosystems for tens to thousands of years. At the local scale, historical logging and agriculture can affect understory plant diversity [Dupouey et al., 2002; Gerhardt and Foster, 2002], carbon and nitrogen cycling [Compton and Boone, 2000; Fraterrigo et al., 2005], abundance of coarse woody debris [Currie and Nadelhoffer, 2002], and, ultimately, ecosystem functioning [Foster et al., 2003; Rhemtulla et al., 2007]. Land use history is very important to assess current landscape conditions as land use legacies may influence many ecological and climate processes occurring today. Forest structure and composition, for example, can be influenced by land use history [Thompson et al., 2002]. In addition, past land use practices have important effects on forest regrowth after the abandonment of agricultural and pasturelands [Moran et al., 1996; Tucker et al., 1998]. Moreover, in highly managed landscapes such as agricultural or pasturelands, historical land use management help explain current conditions of soils and water.

[4] It is now widely acknowledged that intensive agriculture, although essential for producing much of the world's fiber, food, and biofuels, is also a major source of degradation of many essential ecosystem services [Foley et al., 2005; Mooney et al., 2005; Vandermeer and Perfecto, 2007]. In addition, only recently, the importance of croplands for the determination of local climate has been recognized by the global climate modeling community. For example, McPherson et al. [2004] examined meteorological observations over the corn belt of the central United States and found that the difference in vegetation properties between the managed cropland and the natural pasture gave rise to local anomalies in near-surface temperature and humidity.

[5] Estimates of global carbon emissions due to land use change vary enormously, with estimates for emissions in the 1990s ranging from +0.5 to +3.0 Pg-C yr−1 [Houghton, 2003], although recent estimates point to around 1.5 Pg-C yr−1 [Canadell et al., 2007; Le Quéré et al., 2009]. Within this context, Brazil sits squarely under the scrutiny of the international community due to its carbon emissions associated with land use change and because of the country's large stock of biomass carbon stored in its vast vegetation [e.g., Fearnside, 2000a, 2000b]. Recent carbon emissions from land use change and deforestation in Brazil is well documented. During the period 1990 to 1994, there was a 2.4% increase of the carbon emissions due the land use change and forests – from 758.3 to 776.3 million tons of carbon dioxide, MtCO2 (0.206 to 0.212 Pg-C.yr−1 [Ministério de Ciência, Tecnologia e Inovação (MCTI), 2004]. However, estimates of historical emissions from land use changes are not available for Brazil, because of the lack of historical land use maps. In this work, first we reconstruct the historical land use distribution in the entire Brazilian territory between 1940 and 1995, and then we calculate the historical carbon emissions associated with land use change during the same period.

2. Methodology

2.1. Land Use Reconstruction

[6] In a previous study, Leite et al. [2011] merged satellite-derived land cover data set with census data for the Amazon region. The same methodology is used here to develop land use maps for the entire Brazilian territory (Figure 1), henceforth referred to as HLUC-BR (Historical Land Use Change-Brazil). The methodology is briefly described below.

Figure 1.

Location of the study area, with identification of Brazilian states.

[7] Land use patterns in Brazil from 1940 to 1995 were derived by merging two data sets. The first data set is a land use classification for 2000 [Ramankutty et al., 2008] that provides the geographical location for land use types in Brazil. These data are part of a global land cover database with an average spatial resolution of 1 km2 derived from two land cover products entitled BU-MODIS [Friedl et al., 2002] and GLC2000 [Bartholomé and Belward, 2005]. Ramankutty et al. [2008] combined the 16 categories of land cover identified by the BU-MODIS product with the 22 land cover classifications from the product GLC2000, along with an extensive compilation of agricultural survey data collected at municipal, state and country levels. This was cross-referenced with the satellite data to develop global maps of the extent croplands and pasture. In the case of Brazil, the survey data came from the agricultural census of 1995 carried out by the Instituto Brasileiro de Geografia e Estatística (IBGE – Brazilian Institute for Geography and Statistics), adjusted for the year 2000 by using country totals. The resulting maps provide an estimate of the global distribution of agricultural lands for the year 2000, and describe the fraction of croplands and pastures inside grid cells with spatial resolution of 5′ (∼9 km).

[8] The second data source used was the agricultural census data provided by IBGE and compiled by the Instituto de Pesquisa Econômica Aplicada (IPEA – Institute of Applied Economic Research) for the period 1940–1995. The census data classify land use into three major categories: cultivated lands (permanent and temporary crops), natural pasture and planted pasture. Here, permanent crops refer to long-term cultures such as coffee, orange, cocoa, banana, etc., which do not need to be replanted after harvesting. In contrast, temporary crops such as rice, maize and soybeans grow for less than one year and need to be replanted after each harvest. Natural pasture is defined as non-planted areas used for pasture and animal grazing, while the planted pasture consists of planted grass species for animal grazing. Usually, natural pasture is a wood-harvested cerrado (regionally called ‘campo limpo’ or ‘campo sujo’), while the planted pastures are usually sown after tilling, liming and fertilizing the soil. The survey data were collected in 27 states at the municipality (political administrative units) level in intervals of 5 and 10 years.

[9] The present spatially explicit reconstruction of historical land use patterns in Brazil was conducted in three stages. First, a digital mesh was constructed for Brazil municipalities for each year in which agricultural census data was available. The construction of the mesh allowed the accurate quantification of the fraction of cultivated area, natural pasture and planted pasture in each municipality for each year of the census through the simple process of dividing the cultivated area in a census year by the total area of the municipality. The same techniques were used to calculate the relative proportions of natural pasture and planted pasture in each municipality. Second, the change in proportions of each kind of land use between year of the census and 1995 were calculated. Finally, historical land use patterns were reconstructed for Brazil, by combining the maps resulting from the second stage with a land use classification for 2000 provided by Ramankutty et al. [2008]. During this stage, the total pasture in the work by Ramankutty et al. [2008] was split in natural and planted pasture categories according to the 1995 proportion in the census data.

[10] A simple algorithm was applied to estimate the historical land cover changes. First, we calculated the proportions of land use categories of each year in relation to the 1995 census for each Minimum Comparable Area - MCA (more details given by Leite et al. [2011]) and after maps of land use proportion were created for each census year using rasterized values of the proportions in each MCA. Land use transitions followed the transition in the agricultural censuses, and are uniformly distributed in the pixels contained within each MCA. Finally, the map of land use for 1995 of Ramankutty et al. [2008] was used to adjust the land use proportions of each census year to create the historical maps of land use rasterized in a grid of 5′. The 5′ resolution represents an intermediate scale between the MODIS data and the average-sized political unit, and is well suited for most ongoing large-scale modeling studies. Our intention here is to adequately ‘capture’ large-scale patterns of historical changes in land use. It should be noted that this methodology does not allow for individual identification of small farms, unless in lumped form. However, despite this limitation, total land use is proportional to the agrarian survey and should be broadly representative of all farms, small and large, in particular for aggregated pixels in the data set.

[11] The resulting land use maps are available at http://www.biosfera.dea.ufv.br/eng.

2.2. Comparison of Land Use Pattern Reconstruction

[12] Since the data set developed in this work comprises periods in which there is no remote sensing data, the data set produced cannot be fully validated. In this case, a comparison was performed with the maps of Ramankutty and Foley [1998] for 1992 and with the historical croplands data set developed by Ramankutty and Foley [1999] for 1950, 1970 and 1990. They used the same methodology that was used to create databases of historical land use of this work, but the databases developed in this study used agricultural census data at the municipal level, while the land use maps created by Ramankutty and Foley [1998, 1999] used census data at the state level. These comparison data sources are hereafter termed RF1998 and RF1999, respectively.

[13] RF1998 provides an analysis of geographic distribution and spatial extent of permanent croplands during the early 1990s for each grid cell on global 5 min (∼9 km) spatial resolution grid. The maps combine the IGBP DISCover data set [Loveland and Belward, 1997], which has a nominal resolution of 1 km and provides ample information about the location of grid cells containing croplands, with a variety of national and subnational (state or province) agricultural inventory data, but for Brazil used data at state level, to aggregate the grid cells to a 5 min spatial resolution.

[14] The RF1999 data set describes geographically explicit historical changes in global croplands from 1700 to 1992. This consists of an extensive database of historical croplands inventory data at the national and subnational level combined with croplands data from 1992, which was used to back-cast global 5 min resolution data on cropland area from 1992 to 1700. Grid cells were then aggregated to a spatial resolution of 0.5 degrees. The spatial distribution of croplands from this analysis provides a quantitative depiction of global agricultural geography.

2.3. Biomass Map and Emissions From Land Use Change

[15] We produced a map of live biomass below and above ground (BGB and AGB) for the original extent of the major vegetation physiognomies of Brazil (Figure 2) based on two vegetation maps: RadamBrasil (Levantamento de Recursos Naturais, volumes 1–31, 1973–1983) for the Amazon biome and northwestern Cerrado and Instituto Brasileiro de Geografia e Estatística [2004] for the remainder of Cerrado as well as for the other Brazilian biomes, i.e., Mata Atlântica, Pantanal, Caatinga and Pampas. To the vegetation physiognomies of these maps, we assigned total live biomass values using a series of data sets. To vegetation of the Amazon biome and forest physiognomies that extend from the Amazon into the Cerrado Biome, we assigned biomass values from Nogueira et al. [2008]. To savanna-type vegetation of the Cerrado biome, we assigned values from Fearnside et al. [2009]. For the Atlantic Forest, we found only 26 biomass plots of which we used 23 to assign biomass values to the various forest types of the Atlantic forest biome (we discarded 3 plots of secondary forests) (Table 1). To the remainder of vegetation types of Atlantic Forest and other vegetation physiognomies occurring in the Pantanal, Caatinga and Pampa biomes, we assigned biomass values from the biomass map made available by MCTI [2010]. For vegetation formations of biomes other than the Amazon and Cerrado that only AGB values were available, we converted these values into total live biomass using the ratios of 18% for physiognomies of Atlantic Forest and Pampas, 27% for the ones of Caatinga and 20% for the ones of Pantanal according to Food and Agriculture Organization [2009]. Finally we used a ratio of 0.485 to transform biomass into carbon [Nogueira et al., 2008].

Figure 2.

Brazil's biomass map (AM Amazon, CE Cerrado, MA Mata Atlântica, CA Caatinga, PT Pantanal, PM Pampas). In the inset, light gray is above ground biomass, black is below ground biomass.

Table 1. Biomass Data From Plots in the Atlantic Forest
PlotsReferencePhysiognomyElevation (m)Average AGB of Plots (Mg/ha)
1–4Alves et al. [2010]Montane Semideciduous Seasonal Forest1023–1096277.46
5–8Alves et al. [2010]Submontane Semideciduous Seasonal Forest176–395204.75
9–12Alves et al. [2010]Lowland Semideciduous Seasonal Forest33–89245.7
13Alves et al. [2010]Restinga (pioneer marine formation)10158.02
14–15Bais [2008]Dense Rain Forest500–900373.38
16–18Metzker et al. [2011]Submontane Semideciduous Seasonal Forest245–275182.93
19Ribeiro et al. [2009]Montane Semideciduous Seasonal Forest670166
20Rolim et al. [2005]Lowland Semideciduous Seasonal Forest28–65334.58
21Socher et al. [2008]Mixed Rain Forest900191.16
22Tiepolo et al. [2002]Lowland Semideciduous Seasonal Forest113178.01
23Tiepolo et al. [2002]Submontane Semideciduous Seasonal Forest153226.48

[16] Mean biomass values for replacement land uses were obtained from Gouvello et al. [2010]. Planted pasture is 8 Mg-C ha−1, natural pasture consists of 12 Mg-C ha−1, and crop biomass is 5.31 Mg-C ha−1, an average of a basket of six major Brazilian crops (sugarcane, soy, corn, cotton, rice, and beans). All these values represent half of the biomass of full-grown crops and pastures. Composite biomass of a grid cell xy (16′ spatial resolution) at a time t is a weighted average of the fractions of the vegetation types and the replacement land uses as follows:

display math

Where Np, Pp, and C are the percentages of natural pastures, planted pastures and crops, Bnp, Bpp, BC, their mean biomass values, and Bnv the mean biomass for the natural vegetation types present at a grid cell with spatial coordinates x,y. We calculated net carbon emissions from land use change from 1940 to 1995 subtracting the composite carbon biomass of each grid cell at time t from the one at time t + Δt [Intergovernmental Panel on Climate Change, 2003]. This carbon bookkeeping model was implemented on Dinamica EGO freeware (B. S. Soares-Filho, H. O. Rodrigues, and W. L. Costa, Modeling environmental dynamics with DINAMICA EGO, 2010, available at http://www.csr.ufmg.br/dinamica). Carbon emissions from soil organic matter and fluxes from logging were not considered in this study.

2.4. Uncertainties in Carbon Emissions

[17] The uncertainties in the carbon emissions are quantified as the sum of three different types of uncertainties, related to: (a) land use spatial distribution; (b) biomass estimates; and (c) raster grid discretization. Other sources of uncertainties, like the frequency of censuses periods, or concentration of subdivision of administrative units in specific periods, were not considered.

[18] To estimate the uncertainty in land use estimates, we use the bootstrap procedure [Ramankutty et al., 2008]. This technique was applied at this stage for only the combined land cover data set. We perform 100 bootstrap runs, where the entire census data was sampled with replacement each time, and re-estimate our regression model using the combined satellite-based land cover data set (Sampling with replacement used the standard statistical procedure wherein the census data formed the population (of 1578 values, the total number of MCA polygons), and for each of the 100 sample sets we randomly selected values from the population, with the sampled value being replaced back into the population. Each sampling outcome is therefore independent of the previous outcome; that is, every census data value has equal probability of being chosen during every sampling event, no matter whether it has been chosen before. We found that the 90% confidence range is about 20% of the mean. The sources of uncertainties in this kind of study are discussed in section 4.

[19] With respect to biomass estimates, we establish a range of 20% for the Amazon biome, which is the upper uncertainty bound for biomass field measurements [Brown and Lugo, 1992; Brown et al., 1995; Keller et al., 2001; Chambers et al., 2001; Chave et al., 2004] and expanded this range for the other biomes to 30% to account for the paucity of biomass plot data.

[20] To assess uncertainties regarding inaccuracy of the location of land use changes with respect to vegetation biomass values at the model spatial resolution of 16′, we ran a Monte Carlo simulation. In the Monte Carlo simulation, we rasterized the biomass map at an increased spatial resolution of 0.009°. At this increased resolution, each land use map cell of 16′ was now composed of 400 cells of 0.009°. For each Monte Carlo simulation run, the model then assigns randomly a percent of land occupied by the three uses (planted pasture, crops, native pastures), so that the average values for each set of 400 cells coincide with the ones of the respective coarse resolution cell. For each time series map, we ran 100 runs and compared at the end of the period the mean value for total emissions and associated standard deviation. Standard deviation was only 0.3%, hence representing the uncertainties regarding emission estimates by superimposing land use maps on a biomass map at 16′ spatial resolution. We neglect this source of uncertainty in the final uncertainty estimate.

[21] Our overall uncertainty range of biomass is 20% for the Amazon rain forest biome and 30% for the others. On the top of these ranges, we added 5%, which is the mean error obtained by comparing a set of 20 samples from our original biomass vector map with the same areas of a discrete raster version of this map at a spatial resolution of ≈20 km (the land use map spatial resolution), plus 20% coming from our land use change uncertainty analysis. The final range of uncertainty per biome is thus 45% for the Amazon rain forest biomes and 55% for the other biomes.

3. Results

3.1. Overall Agricultural Activity

[22] The agricultural activities in Brazil in 1940 (Figure 3a) were concentrated in the southeast and south region, especially in the states of Minas Gerais, São Paulo, Goiás and Rio Grande do Sul. In the northeast region, the states of Alagoas and Sergipe presented quite intense agricultural activities and in the other states those activities occurred at a lesser extent and are more distributed. In 1950 (Figure 3b) the distribution of the agricultural activities was similar, with increased agriculture intensity (fraction of planted area in each pixel), and showing an expansion in the states of Mato Grosso and Mato Grosso do Sul. In 1960 (Figure 3c), the agricultural activities kept intensifying, mainly in the northeast region and in the state of Tocantins. In 1970 (Figure 3d), the level of the agricultural activities continued to increase, as well as the distribution, expanding to practically all states, except in the northern region. From 1975 to 1995 (Figures 3e3h) the agricultural activities intensity continued to increase with the same geographic distribution, however expanded at higher rates in the states of Mato Grosso, Pará and Rondônia, mainly along the federal highways BR364 and BR158 in the states of Mato Grosso and Pará, BR364 in the states of Mato Grosso and Rondônia and BR230 in the state of Pará.

Figure 3.

The distribution and intensity of overall agricultural land use in Brazil from 1940 to 1995: (a) 1940, (b) 1950, (c) 1960, (d) 1970, (e) 1975, (f) 1980, (g) 1985, and (h) 1995.

3.2. Natural Pasture

[23] Figures 4 and 5 depict the distribution of natural and planted pasture in Brazil from 1940 to 1995, respectively. Natural pasture represents the largest fraction of the overall agricultural activities in Figure 3. In 1940 (Figure 4a) total pastureland was concentrated in the southeast, south and center-west regions, especially in the state of Goiás, in the west of the state of Mato Grosso do Sul, west of Minas Gerais and south of Rio Grande do Sul. In the northeast region pastures were also present (the 1940 census data did not provide a separation between natural and planted pastures). A comparison of the natural pastures and planted pastures for 1950 (Figures 4b and 5b) indicates a clear domination of natural pastures in 1950, so we assume it also dominated in 1940. There was an expansion of the natural pasture in 1950 (Figure 4b) in the states of Mato Grosso and Tocantins and an increase in intensity in the areas where it already existed. Intensity of natural pastures increased until 1980, in particular in central Brazil. After 1980, natural pastures began to decrease in most of the country, being replaced by the more productive planted pastures. The main exceptions were western Mato Grosso do Sul (Pantanal) and southern Rio Grande do Sul (Pampas).

Figure 4.

The distribution and intensity of natural pasture in Brazil from 1940 to 1995: (a) 1940, (b) 1950, (c) 1960, (d) 1970, (e) 1975, (f) 1980, (g) 1985, and (h) 1995. The 1940 map portrays overall pastureland use (natural pasture + planted pasture).

Figure 5.

The distribution and intensity of planted pasture in Brazil from 1950 to 1995: (a) 1950, (b) 1960, (c) 1970, (d) 1975, (e) 1980, (f) 1985, and (g) 1995. The 1940 planted pasture data is presented in Figure 4a, in combination with natural pasture data.

3.3. Planted Pasture

[24] Planted pastures were nearly absent in Brazil until 1970, except in the State of São Paulo. From 1970 to 1980, they increased mainly in western São Paulo and adjacent regions, and also in Minas Gerais, Bahia, Maranhão and along the BR-010 highway (Belém-Brasília). After 1980, planted pastures began to replace natural pastures in most regions of the country.

3.4. Cropland

[25] Figure 6 shows the distribution of the croplands in Brazil for the period 1940–1995. In 1940 (Figure 6a) the croplands were most concentrated in the states of São Paulo, Rio de Janeiro and Espírito Santo in the southeast, and in the states of Sergipe, Alagoas, Pernambuco and Maranhão in the northeast. However, croplands also occurred in eastern Bahia and in Minas Gerais. In 1950 (Figure 6b) the situation is similar. Nevertheless, croplands began to expand in northern Paraná, while they became more intense in the state of São Paulo. In the decade that followed (Figure 6c) croplands expanded in the states of Rio Grande do Sul, Santa Catarina and Paraná, as well as in southern Minas Gerais and Triângulo Mineiro, and intensified in the state of São Paulo. In the Brazilian northeast the croplands appeared well distributed in all the states, although with more intensity in the states of Alagoas, Pernambuco and Paraíba. In 1970 (Figure 6d) croplands occurred intensely in the states of São Paulo, Paraná, Santa Catarina and Rio Grande do Sul. In these areas, the existent crops in the previous decade virtually disappeared, giving place to natural pastures, as shown in Figure 4d. Possibly, crops were abandoned in these areas allowing regrowth of the natural vegetation (grasses). In this time, the croplands began to expand to the north region of Brazil, first in the state of Tocantins, following the paving of highways BR153 and BR010 in the central region of the state. After five years (Figure 6e) the distribution of the croplands followed the same pattern, though more intense and also reaching southern Mato Grosso do Sul. By 1980, (Figure 6f) a wider distribution of croplands occurred in every state, and after another five years (Figure 6g) the croplands decreased in some extent, both in distribution and intensity. Croplands in Mato Grosso and Mato Grosso do Sul also increased intensity in 1995 (Figure 6h).

Figure 6.

The distribution and intensity of croplands in Brazil from 1940 to 1995: (a) 1940, (b) 1950, (c) 1960, (d) 1970, (e) 1975, (f) 1980, (g) 1985, and (h) 1995.

3.5. Comparison With Previous Studies

[26] In a previous study, we have validated the estimates of changes in land use for Amazonia against independent estimates of changes in land cover using LandSat imagery [Leite et al., 2011]. However, these validations are limited to the satellite era, and can only be done from the 1970s. Before this period, we are limited to comparisons to other land use reconstruction studies, like the maps produced by Ramankutty and Foley [1998, 1999], which are similar in terms of data and methodologies used, and are therefore not totally independent from each other. The main difference between their study and ours is that they used state level census information (average state size in Brazil is 315 000 km2, while we use MCA level census information (average size 5400 km2). Our estimates are thus much more disaggregated.

[27] Figure 7 shows a comparison between the cropland maps of developed in this work (HLUC-BR) with the one by Ramankutty and Foley [1998] (RF1998) for 1992, where the former has a spatial resolution of 5-min and the latter has a spatial resolution of 0.5 degree. We observe that in the map RF1998 (Figure 7a) the croplands appear well distributed, mainly in the southeast regions, center-west and north, while in the HLUC-BR-1992 map, the croplands are more concentrated in the south regions, southeast and northeast, especially in the states of São Paulo, Santa Catarina, Rio Grande do Sul, Sergipe, Alagoas, Pernambuco and Paraíba. The states of Mato Grosso, Mato Grosso do Sul and Goiás also present some areas with concentrated croplands. These considerable differences between the two databases is probably due to the fact that they were developed using different sources from remote sensing data, as well as from agricultural census at different spatial levels.

Figure 7.

Comparison between the croplands maps for Brazil in 1992 at a spatial resolution of 5 min: (a) HLUC-BR-1992 map developed in this work, (b) croplands map for 1992 developed by RF1998, and (c) the difference between the two.

[28] The comparison between the historical croplands maps for the years 1950, 1970 and 1990 developed in this work (HLUC-BR) at a spatial resolution of 5-min and the maps produced for the same periods by Ramankutty and Foley [1999] (RF1999) at a spatial resolution of 0.5 degree is shown in Figure 8. In 1950 (Figures 8a and 8d) the croplands occur mainly in the states of São Paulo, Rio de Janeiro and Espírito Santo, though more concentrated in RF1999 map and more spread in HLUC-BR-1950 map. The croplands in northeast region, in the RF1999–1950 map, are concentrated in the state of Piauí, while in the HLUC-BR-1950 map they are more dispersed. By 1970 (Figures 8b and 8e) there was an expansion of croplands, especially in the south, southeast and central west regions, in both HLUC-BR-1970 and RF1999–1970 maps, however in the RF1999–1970 map the croplands appear very intense in the states of Mato Grosso do Sul, Goiás, Tocantins e Amazonas, while in the HLUC-BR-1970 map it does not occur. Croplands on the RF1999–1990 map had a great intensification, mainly in the southeast regions and central west, especially in the states of Minas Gerais and Mato Grosso. In the northeast region the state of Bahia also had a major expansion. In the HLUC-BR-1990 map the croplands had great intensification in the states of Mato Grosso and Mato Grosso do Sul and begin to occur in the state of Rondônia and in the northeast region of Pará (Figures 8c and 8f).

Figure 8.

Comparison between the croplands maps for Brazil in 1950, 1970 and 1990. (a, d, and g) HLUC-BR maps developed in this work at a spatial resolution of 5 min, (b, e, and h) croplands maps developed by RF1999 at a spatial resolution of 0.5 degree, and (c, f, and i) the difference between the two.

3.6. Carbon Emissions

[29] Original carbon content of native vegetation of Brazil's biomes amounted to 115.7 ± 33.8 Pg-C (uncertainties are reported as the 90% confidence range): 68.4 Pg-C for the Amazon biome, 21.3 Pg-C for the Cerrado, 17.6 Pg-C for the Mata Atlântica, 6.5 Pg-C for the Caatinga, 0.65 Pg-C for the Pantanal, and 1.3 Pg-C for the Pampas (Table 2). Net emissions from land use changes between 1940 and 1995 totaled 17.2 ± 9.0 Pg-C, averaging 0.31 ± 0.16 Pg-C yr−1 (Table 2), but reaching up to 0.47 ± 0.25 Pg-C yr−1 during the 1960s and through the 1985–1995 period (Please note that, since the carbon emissions calculation does not account for time-varying emissions, such as decay of dead wood following conversion, this may contribute to additional uncertainty in the emission estimates, in particular for the 5-year periods, and to part of the variability for each emissions period). In addition, emissions from land use change prior to 1940 totaled 3.8 ± 2.1 Pg-C, summing up 21.0 ± 11.0 Pg-C emissions by 1995 (Table 2). Historically, land use change in the Atlantic Forest biome accounted for 43% of total carbon emissions from 1940 to 1995, followed by Cerrado with 28% and the Amazon with 25% (Table 2). Nevertheless, from 1970 onwards, deforestation in the Amazon has become the main source of carbon emissions from land use changes, accounting from 36% to 65% of Brazil's emissions (Table 2).

Table 2. Carbon Emissions per Period and per Biomea
BiomeNet EmissionsEstimated Original Carbon Pool (Pg-C)Fraction of the Original Carbon Pool Emitted (%)
Before 19401941–1950 (%)1951–1960 (%)1961–1970 (%)1971–1975 (%)1976–1980 (%)1981–1985 (%)1986–1995 (%)1941–1995
  • a

    For each time period, the percentages indicate the percent contribution to total land use change emissions from each biome.

Pampas 010001110.12 ± 0.071.3 ± 0.469.2
Pantanal 33102−1110.24 ± 0.130.65 ± 0.2336.9
Caatinga 031010110.21 ± 0.126.5 ± 2.33.2
Mata Atlântica 566355172740437.4 ± 4.117.6 ± 6.242.0
Cerrado 33322835353421284.9 ± 2.721.3 ± 7.523.0
Amazonia 11−21465554036254.3 ± 1.968.4 ± 17.16.3
Average emissions per period (Pg-C/yr)   
Cumulative emissions (Pg-C)3.8 ±  115.8 ± 33.818.1

[30] Analysis per biome indicates that Mata Atlântica deforestation was the major emission source (7.4 ± 4.1 Pg-C during 1940–1995), followed by Cerrado (4.9 ± 2.7 Pg-C) and Amazonia (4.3 ± 1.9 Pg-C). This is because the Atlantic Forest is the biome that had most of its original biomass converted from changes in land use (42%). Cerrado had about 23% of its biomass converted, while the Amazon had a relatively small fraction (6%) of its large carbon pool (68 Pg-C) emitted by 1995.

4. Discussion and Conclusion

[31] The combination of satellite and census data allows analysis of the geographic patterns of agricultural land use, and the relative changes between censuses allowed us to estimate changes in agricultural activities in a way that is consistent with maps produced from satellite imagery [Cardille and Foley, 2003].

[32] To reconstruct geographically explicit changes in natural pasture, planted pasture and croplands from 1940 to 1995 for Brazil, we have employed a straightforward algorithm that uses a product generated by merging agricultural census data with maps derived from remote sensing developed by Ramankutty et al. [2008] and census data at the municipal level. The results show that the largest changes occurred in natural pasture that gradually gave way to planted pasture, especially in the southeast and central west since 1970. This substitution probably occurred because the planted pasture is more productive for cattle raising. The croplands began its expansion in the 1960s to reach almost all states in 1980, but with less intensity in the northern states. However, intensification took place during the 1980s and 1990s, which is confirmed by the recent study of Gibbs et al. [2010].

[33] Although this data set has been partially validated by a previous study [Leite et al., 2011], it should be noted that the reconstructed data set has a number of uncertainties related to the merging of two data sets utilizing different observations – satellite-based land cover classification and agricultural census data. The remote sensor can observe only the top of the vegetation, and thus provides only indirect information on land use. In addition, agricultural census data include information that is inconsistent with the remotely sensed data. For example, while the census data identifies arboreal cultures as permanent cultures, it is not clear if the land cover satellite-based classification consider arboreal cultures as forests or as deforested area. Moreover, satellite remote sensing is usually unable to discriminate between some Cerrado physiognomies and the existence of natural pasturelands. This is a serious limitation as most farmers in central Brazil use logged Cerrado as natural pasturelands, which is recorded as Cerrado by remote sensing and reported as pastureland in the census. This land use is therefore represented differently (and more accurately) in our data set.

[34] Despite these uncertainties, this data set is the first estimate of the historical changes of the croplands, natural pasture and planted pasture in Brazil and associated carbon emissions, using municipal level agricultural census data. First, this is an advance over previous studies that have produced historical land use data sets which have been created using state level agricultural census data [Ramankutty and Foley, 1998, 1999; Cardille et al., 2002; Cardille and Foley, 2003], given the greater disaggregation of the municipal level census data. In addition, this represents a considerable advance over previous emissions studies, which either considered only regional emissions (e.g., Amazon emissions reviewed by Houghton et al. [2009] and Potter et al. [2009]), or cover only recent periods (e.g., the Brazilian communications to the UNFCCC [MCTI, 2004, 2010]). Moreover, the use of a more disaggregated data set allows, in principle, a more precise regional estimate of carbon emissions. Furthermore, the reconstruction provides a strong database for future attempts to model environment-climate interactions in Brazil.

[35] Recently, Pitman et al. [2009] found a wide range of bio-geophysical climate impacts from historical land cover change when modeled in a suite of current Global Climate Models (GCMs) in the Land-Use and Climate study. Historical land use applied in GCMs helps us understand the causes of the impacts of these changes on regional and global climate.

[36] Overall anthropogenic land use increased from 106.4 M ha by 1940 (12.5% of national territory) to 219.54 M ha by 1995 (25.8% of national territory). These estimates do not include lands that were once converted and were later abandoned. Thus, this study does not take into account the potential carbon uptake from native forests. In this regard, Gouvello et al. [2010] shows that only regrowth of forest remnants in the Atlantic forest biome, which were extensively deforested during the forties and fifties for charcoal and left for regeneration since then, is currently removing from the atmosphere 30 ± 6 Mg-C ha−1 yr−1.

[37] Net emissions from land use changes by 1940 totaled 3.8 ± 2.1 Pg-C, and between 1940 and 1995 totaled 17.2 ± 9.0 Pg-C, summed up 21.0 ± 11.0 Pg-C emissions by 1995. This indicates that current carbon stored in remaining native vegetation of Brazil is not greater than 91.8 ± 27.5 Pg-C.

[38] Our estimates of Brazilian carbon emissions for 1940–1995 are equivalent to about 62% of the Houghton [2008] estimates from land use change from South and Central America combined (27.7 Pg-C in the period) and to 23% of the global carbon emissions from land use change in the period (74.5 Pg-C).

[39] Our estimates also indicate that historical emissions from changes in land use in Brazil were about 11 times higher than the emissions from fossil fuels burning (1.56 Pg-C for 1940–1995 – CDIAC, Carbon Dioxide Information Analysis Center, http://cdiac.ornl.gov). This confirms that Brazil carbon emissions are typically from land use change, although the historical rate (11:1) is much higher than recent rates (7.2:1 during 1985–1995, calculated using CDIAC fossil fuel emissions 0.657 Pg-C).

[40] Although emissions are high, the total Brazilian carbon emissions by 1995 amount to only 18% of its original biomass (Table 2), which is similar to the 1850–1995 global numbers. Assuming 141.2 Pg-C global emissions from 1850 to 1995 [Houghton et al., 2001], and the standard 800 Pg-C of global above ground biomass [Intergovernmental Panel on Climate Change, 2007], a calculation using a comparable methodology yields that 17.7% of global biomass has been emitted to the atmosphere in the last 150 years.

[41] Moreover, if we account for carbon emissions only from Amazon deforestation between 1995 and 2010 (emissions from other biomes deforestation not included), historical emissions from land use changes in Brazil from pre-colonial times to the present amount to 24.0 ± 11.9 Pg-C. This value is equivalent to about 12 years of worldwide emissions from land use change during the last decades [Houghton, 2008].

[42] Natural ecosystems provide locally important services including serving as habitats for endangered plants and animals, preserving major elements of the global hydrological cycle, protecting massive watersheds and storing carbon in its biomass. On the other hand, expansion of agriculture and pastureland over natural vegetation inevitably entails carbon emissions to the atmosphere.

[43] In future work we will develop new databases of historical land use that disaggregates historical land use in major agricultural crops (e.g., soybeans, corn, beans, etc.) during the period 1940 to 1995 and current periods using annual agricultural census data for the period 1990 to 2008. This will allow a more accurate calculation of carbon emissions due to land use changes for the crops.


[44] C. C. Leite is supported through a CNPq fellowship. Britaldo Soares-Filho receives support from CNPq, the Betty and Gordon Moore Foundation, the Climate and Land Use Alliance, and NASA, grant NNX11AE56G.