The text that follows is a PREPRINT.
Please cite as:
Laurance, W.F., A.K.M. Albernaz, G. Schroth, P.M. Fearnside, S. Bergen, E.M. Ventincinque & C. da Costa. 2002. Predictors of Deforestation in the Brazilian Amazon. Journal of Biogeography 29: 737-748.
Copyright: Blackwell Science Ltd.
Send proofs to:
William F. Laurance
Smithsonian Tropical Research Institute
Fax: 507-212-8148 (Panamá)
Predictors of Deforestation in the Brazilian Amazon
William F. Laurance1,2, Ana K. M. Albernaz2, Götz Schroth2, Philip M. Fearnside3,
Scott Bergen4, Eduardo M. Ventincinque2, and Carlos Da Costa2
1Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Republic of Panamá; email: firstname.lastname@example.org
Dynamics of Forest Fragments Project, National Institute for Amazonian Research
(INPA), C.P. 478, Manaus, AM 69011-970,
of Ecology, National Institute for Amazonian Research (INPA), C.P. 478, Manaus,
for Environmental Literacy, Mt. Holyoke College,
We assessed the effects of biophysical and anthropogenic predictors on deforestation in Brazilian Amazonia, the region with the world’s highest absolute rate of forest destruction. Using a GIS, spatial data coverages were developed for deforestation and for three types of potential predictors: (1) human-demographic factors (rural-population density, urban-population size); (2) factors that affect physical accessibility to forests (linear distances to the nearest paved highway, unpaved road, and navigable river), and (3) factors that may affect land-use suitability for human occupation and agriculture (annual rainfall, dry-season severity, soil fertility, soil waterlogging, soil depth). To reduce the effects of spatial autocorrelation among variables, the basin was subdivided into >1900 quadrats of 50 X 50 km, and a random subset of 120 quadrats was selected that was stratified on deforestation intensity. A robust ordination analysis (nonmetric multidimensional scaling) was then used to identify key orthogonal gradients among the ten original predictor variables.
The ordination revealed two major environmental gradients in the study area. Axis 1 discriminated among areas with relatively dense human populations and highways, and areas with sparse populations and no highways; whereas axis 2 described a gradient between wet sites having low dry-season severity, many navigable rivers, and few roads, and those with opposite values. A multiple regression analysis revealed that both factors were highly significant predictors, collectively explaining nearly 60% the total variation in deforestation intensity (F2,117=85.46, P<0.0001). Simple correlations of the original variables were highly concordant with the multiple regression model and suggested that highway density and rural-population size were the most important correlates of deforestation.
These trends suggest that deforestation in the Brazilian Amazon is being largely determined by three proximate factors: human population density, highways, and dry-season severity, all of which increase deforestation. At least at the spatial scale of this analysis, soil fertility and waterlogging had little influence on deforestation activity, and soil depth was only marginally significant. Our findings suggest that current policy initiatives designed to increase immigration and dramatically expand highway and infrastructure networks in the Brazilian Amazon are likely to have important impacts on deforestation activity. Deforestation will be greatest in relatively seasonal, southeasterly areas of the basin, which are most accessible to major population centers and where large-scale cattle ranching and slash-and-burn farming are most easily implemented.
The Amazon basin sustains about 60% of the world’s remaining tropical rainforest and plays vital roles in maintaining biodiversity, regional hydrology and climate, and terrestrial carbon storage (Salati & Vose, 1984; Fearnside, 1999; Houghton et al., 2000). This region also has the world’s highest absolute rate of deforestation. In Brazilian Amazonia, which encompasses about 70% of the basin, deforestation rates since 1995 have averaged nearly 2 million ha yr-1 (INPE, 2000; Laurance et al., in press). In addition to outright deforestation, large expanses of forest are being degraded by habitat fragmentation, edge effects, selective logging, surface fires, illegal gold mining, overhunting, and other activities (Fearnside, 1990; Skole & Tucker, 1993; Laurance, 1998; Cochrane et al., 1999; Nepstad et al., 1999a).
Given the global significance of the Amazon, increasing attention has been focused on documenting the pattern and pace of deforestation in the region, and on identifying the proximate and ultimate drivers of forest loss. These efforts fall into four rough categories. The first are remote-sensing studies that quantify the rate and spatial extent of forest loss and fragmentation. For example, Brazil’s National Institute for Space Research (INPE) has used Landsat Thematic Mapper data to map deforestation annually in Brazilian Amazonia since 1988 (e.g. INPE, 2000). Other studies based on remote-sensing have estimated the area of Amazonian forest that is fragmented or influenced by edge effects (Skole & Tucker, 1993), quantified deforestation patterns in specific regions (e.g. Alves et al., 1999; Steininger et al., 2001a, 2001b), and estimated carbon flux from Amazonian deforestation (Houghton et al., 2000).
Studies in the second category evaluate the effects of government policies and development activities on deforestation. For example, Fearnside (1987) discusses the relationship between internationally funded development projects, highway building, immigration, land speculation, and deforestation in Brazilian Amazonia. Other studies have assessed the role of government policy and land-tenure conflicts in promoting environmental degradation (e.g. Moran, 1981; Fearnside, 2001). This category also includes local case studies that explore the interaction of social, economic, and cultural factors in determining land settlement patterns and agricultural practices (e.g. Schmink & Wood, 1992; Walker et al., 1993).
The third category involves conservation gap analyses that assess the impacts of deforestation on different vegetation types. Kangas (1990) assessed the effects of simulated deforestation scenarios on 17 major vegetation types in Brazilian Amazonia, whereas Fearnside and Ferraz (1995) evaluated the representation of vegetation types within nature reserves in each of the nine states of Brazilian Amazonia. Ferreira (2001) conducted a similar analysis but focused on the distribution of vegetation types within eco-regions defined by major river systems in Brazilian Amazonia.
The final category involves modeling studies that attempt to identify the proximate causes of Amazonian deforestation. Such studies underlay efforts to predict the future condition of Amazon forests and evaluate proposed development schemes in the region (e.g. Carvalho et al., 2001; Laurance et al., 2001). Fearnside (1993), for instance, used regression techniques to estimate the proportion of deforestation in Brazilian Amazonia caused by small landowners versus large-scale cattle ranchers, whereas Imbernon (2000) assessed the effects of rapid population growth on forests in Rondônia. In a preliminary analysis, Wood and Skole (1998) explored the effects of demographic variables (e.g. population density, ranch and farm densities, rural immigration) on deforestation in Brazilian Amazonia. More generally, other investigators have examined the influence of micro- and macroeconomic variables on tropical deforestation (e.g. Reis & Margulis, 1991; Lambin, 1994; Kaimowitz & Angelsen, 1998).
Although these various studies have provided important insights, much remains unknown regarding the ultimate and proximate causes of Amazonian deforestation. For example, government policies in Brazil promote rapid immigration into the Amazon (Imbernon, 2000; Laurance et al., in press), but the relative effects of rural versus urban populations on deforestation is poorly understood. Likewise, new highways and roads sharply increase Amazonian deforestation (Fearnside, 1987; Imbernon, 2000; Carvalho et al., 2001; Laurance et al., 2001; Steininger et al., 2001a; Nepstad et al., in press), but their impacts on a basin-wide scale have not been rigorously quantified. Finally, biophysical factors such as annual rainfall, dry-season severity, and soil fertility may influence deforestation (Moran, 1981; Fearnside, 1984; Steininger et al., 2001a) but have been poorly studied in this context (but see Schneider et al., 2000).
In this study we assess the influence of ten biophysical and anthropogenic predictors on deforestation for the entire Brazilian Amazon. These include human-demographic variables, factors that affect physical accessibility to forests, and parameters that influence land-use suitability for human occupation and agriculture. Our analysis is quantitatively rigorous, incorporates the most recent available information on Amazonian deforestation, and includes potentially important anthropogenic and biophysical variables whose effects have not been evaluated on a basin-wide scale.
Most data processing was done at the GIS laboratory of the Biological Dynamics of Forest Fragments Project in Manaus, Brazil, using ArcView 3.2 (ESRI, 1996a), Spatial Analyst 2.0 (ESRI, 1996b), and IDRISI 32 software (Eastman, 1999) on a Windows PC. Analyses were conducted at two spatial scales, using quadrats of 2500 km2 (50 X 50 km) and 400 km2 (20 X 20 km). Because the patterns were similar, however, only results for the 2500-km2 quadrats are presented here.
There were 1927 quadrats of 2500-km2 in the Brazilian Legal Amazon (Fig. 1). This number does not include quadrats along the margins of the Legal Amazon that had <80% of their area within the region’s boundaries, which were excluded from analyses. For each quadrat, deforestation and predictor variables were extracted into data tables to permit statistical analysis.
Data on forest cover in the Brazilian Amazon were derived from 1999 imagery produced by the National Oceanographic and Atmospheric Administration using 1992 Pathfinder and 1998 AVHRR satellite data (www.ngdc.noaa.gov/dmsp/fires/brazil/brazil_main.html). Separate images for the individual states in the Brazilian Amazon (e.g. Rondônia, Acre) were image-mosaiced using Imagine 8.4 software, and the composite file was georeferenced with a second-order polynomial transformation to digital maps provided by the Brazilian Socio-Environmental Institute (ISA, 1999).
The final imagery included four categories of coverage: forest, water, areas of persistent light at night (cities), and non-forested areas. Non-forested areas included both deforested lands and open vegetation (principally cerrado savanna and open woodland). To discriminate deforested areas from open vegetation, the latter was converted to raster format and subtracted from the original image using digital vegetation maps (ISA, 1999). In addition, cities and areas flooded by hydroelectric reservoirs (e.g. Balbina & Tucuruí reservoirs) were classified as being deforested. The final result was an image, based on 1-km2 pixels, in which forested areas, deforested lands, and natural bodies of water were discriminated. Percent deforestation data were calculated by determining the proportion of deforested pixels within each 2500-km2 quadrat.
Data on existing highways (defined as being paved) were derived from digitized and georeferenced maps from ISA (1999), augmented with extensive personal knowledge of the region. Using IDRISI, the distance to the nearest highway was determined for each 1-km2 pixel within the Brazilian Amazon, and the mean distance was then calculated for all 2500 pixels in each quadrat. This mean provided an effective index of highway density that also included the potentially important effects of highways in adjoining quadrats.
Data on roads (defined as being unpaved) were quantified separately from paved highways, because roads provide less-efficient transportation than highways and may not be usable during the wet season. Data for roads were generated in the same way as those for paved highways, using data from ISA (1999).
The distribution of major navigable rivers in the Brazilian Amazon was estimated using georeferenced data from ISA (1999). We excluded from the analysis any rivers of <1 km in width and any river stretches that were isolated from the main stem of the Amazon or Rio Negro Rivers by cascades or waterfalls. For each quadrat, the distance to the nearest river was calculated for each 1-km2 pixel, and the mean value for all pixels within each quadrat was used to provide an overall index of river accessibility.
Data on the population sizes of all legally incorporated Amazonian cities were collected in the 2000 national census by the Brazilian Institute for Geography and Statistics (IBGE 2000). The data were downloaded as tables (www.ibge.gov.br) and combined with a map of city locations produced by ISA (1999). This coverage was used to determined the size of the urban population within each quadrat.
Data on rural-population density were also determined using data from the 2000 national census (IBGE, 2000). Information on rural-population sizes were collected for 780 municipalities (which vary considerably in size) within the Brazilian Amazon. These data were converted to population densities for each municipality, at a spatial scale of 1-km2 pixels. For each quadrat, the rural-population density was the mean value for all pixels within the quadrat.
Data on mean annual rainfall were based on a figure derived from an analysis of 800 simple pluviometric sites in the Brazilian Amazon (Sombroek, in press). The rainfall data were used to derive rainfall isoclines at 200-400 mm intervals for the region (Sombroek, in press), which were digitized using ArcView and georeferenced using IDRISI with 20 control points. To avoid abrupt boundaries between isoclines, rainfall data were interpolated using the TIN (Triangulated Networks and Surface Generation) mode of IDRISI. This yielded a continuous rainfall surface at a spatial scale of 1-km2 pixels.
Data on average duration of the dry season were also produced from the digitized map derived from Sombroek (in press), who generated isoclines of the number of months with <100 mm rainfall. Again, to avoid abrupt boundaries between isoclines, data were interpolated using the TIN mode of IDRISI, yielding a continuous surface of dry-season severity at a scale of 1-km2 pixels.
Data on soils were based on a 1:3,000,000-scale digital soil map of Brazilian Amazonia that was produced in the 1970s by the Soils Division of the Brazilian Institute for Agricultural Research (EMBRAPA, Rio de Janeiro), which is regarded as the best available soils map for the Amazon (W. G. Sombroek, pers. comm.). The map contains 17 major soil types that are further subdivided into over 100 subtypes, using the Brazilian soil taxonomy (cf. Beinroth, 1975). The map was used to generate data layers for three indices of soil suitability for agriculture: (1) a general index of soil fertility (see below), (2) waterlogging and hydromorphy, and (3) soil shallowness/stoniness. Information for classifying the different soil subtypes was derived from published sources (especially Sombroek, 1984, 2000; Oliveira et al., 1992).
This parameter is a composite index of soil suitability for agriculture that ranges from 1 (poorest soils) to 10 (best soils), and that incorporates data on soil chemistry, texture, depth, waterlogging, stoniness, and other features. Soil-fertility classes 8-10 have the highest agricultural potential. These include alluvial soils in várzea forests (seasonally inundated by whitewater rivers that carry nutrient-rich sediments from the Andes Mountains), terra roxa soils (nutrient rich, well-structured upland soils that have formed on base-rich rock and are in high demand for cocoa and other nutrient-demanding crops), eutrophic Cambisols (young, relatively unweathered soils with high-activity clay and high nutrient status), and Vertisols (clay soils with high-activity clay minerals and high nutrient contents). All of these soil types have very limited distributions, collectively encompassing just 1.8% of the Brazilian Amazon, according to the EMBRAPA map.
Soil classes 5-7 have some agricultural potential but also important limitations, such as high acidity, low nutrient availability, shallowness, waterlogging, and concretionary soils. These soil types are very extensive, comprising 53.4% of the Brazilian Amazon according to the EMBRAPA map.
Soil classes 2-4 have restricted potential for certain low-demand uses, such as cattle pasture or undemanding tree crops. These include the intensively weathered Xanthic Ferralsols of central Amazonia, very stony and shallow soils, nutrient-poor waterlogged soils, and Plinthosols (soils that form into hardened laterite when exposed to wetting and drying cycles). According to the EMBRAPA map, these soils encompass 34.8% of the Brazilian Amazon.
Soil class 1 has no potential for agriculture. These include very sandy soils (podzols and quartz sands, some of which are waterlogged) and a small area of salt-affected soils along the ocean shore. This class encompasses 7.8% of the Brazilian Amazon, according to the EMBRAPA map.
This index quantifies waterlogging, poor drainage, and flooding risk, and has four classes. A value of 0 indicates soils with no waterlogging or flooding (77.9% of the Brazilian Amazon). Soils with a value of 1 are at risk of seasonal flooding but are not hydromorphic (anoxic), such as várzea soils (0.7% of the Brazilian Amazon). A value of 2 indicates soils that are hydromorphic at greater depth or periodically waterlogged (4.6% of the Brazilian Amazon), whereas 3 indicates soils that are hydromorphic near the soil surface and often permanently waterlogged (such as gley soils; 14.6% of the Brazilian Amazon).
Soil shallowness and stoniness
This index has three classes: 0 (not shallow or stony; 91.5% of the Brazilian Amazon), 1 (somewhat shallow or stony, including relatively young soils in mountainous regions and concetionary soils; 1.9% of the Brazilian Amazon), and 2 (very shallow or stony soils, including very young soils in mountain regions and a small area of Planosols that have a compact subsoil; 4.4% of the Brazilian Amazon). This index is important because certain young soils are chemically rich but too shallow for agricultural uses and have low water-holding capacity.
Predictors describing rural and urban populations, and distances to paved highways, roads, and navigable rivers were log-transformed to improve normality and the linear fit to deforestation data, whereas percent-deforestation data were arcsine-squareroot transformed to improve normality. To minimize spatial autocorrelation among the 1927 quadrats, a random subset of 120 quadrats was selected for statistical analysis. To ensure that a wide range of deforestation values were included, chosen quadrats were stratified on deforestation intensity by randomly selecting 40 plots within each of three deforestation categories (0-33.33, 33.33-66.67, and >66.67% deforestation).
Because many of the predictor variables were strongly intercorrelated, a robust ordination method, nonmetric multidimensional scaling (NMDS), was used to identify key orthogonal (statistically independent) gradients in the dataset (McCune & Mefford, 1997). The number of ordination axes was determined by examining the relationship between stress (unexplained variance) and number of axes, and by using randomization tests to confirm that each axis in the final analysis explained significantly more variation than expected by chance. All variables were standardized before analysis with the relativization by maximum method (Noy-Meir et al., 1975), and Sorensen’s distance metric was used to generate dissimilarity matrices.
Following the ordination, best-subsets and multiple linear regressions were used to test the efficacy of the ordination axes as predictors of deforestation. Performance of the final regression model was assessed by comparing the standardized residuals to both the fitted values and to each significant predictor (Crawley, 1993). Simple product-moment correlations among the transformed variables, using Bonferroni-corrected P values to minimize Type II statistical errors, were also used to assess the environmental correlates of deforestation.
Among the 120 randomly selected quadrats (Fig. 1), the ordination revealed two major axes of variation (Table 1). Both axes explained significantly more variation than expected by chance (P=0.02 in both cases, randomization tests). Axis 1, which captured 62.0% of the total variation, mainly discriminated among areas with dense rural and urban populations and many highways, and those with sparse populations and no highways. Axis 2 captured 26.4% of the total variation, and described a gradient between sites with low dry-season severity, many navigable rivers, and few unpaved roads, and those with opposite values.
Best-subsets and multiple regression analysis revealed that both axes were highly significant predictors of deforestation (F2,117=85.46, P<0.0001), collectively explaining 59.4% of the total variation in forest destruction. Both axes were negatively associated with deforestation, indicating that forest loss increased near highways and roads, in areas with larger rural and urban populations, and in drier areas with stronger dry seasons. Of the two ordination axes, axis 1 was more important, individually explaining 53.6% of the variation in deforestation (F1,118=136.38, P<0.0001), whereas axis 2 individually explained only 9.6% of the variation (F1,118=12.57, P=0.0006). This result suggests that highways and human population density were the main determinants of local deforestation, with rainfall and unpaved roads influencing deforestation to a lesser degree.
The efficacy of the ten predictors was further assessed using product-moment correlations with a Bonferroni-corrected alpha value (P=0.005). The two strongest correlates of deforestation were distance to paved highways (r= -0.759) and rural-population density (r=0.731), followed by urban-population size (r=0.656), dry-season severity (r=0.443), and distance to unpaved roads (r= -0.355)(P<0.001 in all cases). Soil stoniness/shallowness also had a weaker but significantly negative effect on deforestation (r= -0.254, P=0.005). These analyses were concordant with the regression models and suggest that highways and human population density are the strongest predictors of local deforestation in the Brazilian Amazon. Deforestation is also associated with increasing dry-season severity, more roads, and deeper and less-stony soils.
The key predictors often had non-linear relationships with deforestation, especially when the predictors varied over several orders of magnitude. These relationships were examined in detail by plotting each variable against deforestation using all 1927 quadrats within the Brazilian Amazon. On average, deforestation rose mostly sharply within 50-100 km of paved highways (Fig. 2) and within 25-50 km of unpaved roads (Fig. 3). The relationship between deforestation and rural-population density was also nonlinear, tending to plateau beyond 100 persons/km2 (Fig. 4). Dry-season severity, however, tended to have a roughly linear relationship with deforestation (Fig. 5).
It is important to emphasize that our findings do not appear strongly influenced by the spatial scale of our analysis. When we performed a finer-scale analysis (by dividing the Brazilian Amazon into over 8000 quadrats of 20 X 20 km, and then randomly selecting 300 quadrats, stratified on deforestation intensity, for statistical analysis) the results were broadly similar to those presented here. The only notable difference is that in the fine-scale analysis, highways and roads emerged as the two strongest correlates of deforestation, followed in importance by climatic factors (annual rainfall and dry-season intensity) and rural-population density. As before, soil factors seemingly had little influence on deforestation activity.
The validity of our conclusions obviously depends upon the quality of available spatial data for Brazilian Amazonia. Data on demographic variables, human infrastructure, rivers, and deforestation were recently collected and probably quite accurate, but spatial data on soils (see below) and climatic variables were based on large-scale interpolations by investigators who relied on a relatively modest number of sampling points, given the vastness of the study area. In all cases, however, we used the most reliable information available to us, and we believe that the broad-scale trends we identified are generally reliable.
Another limitation of this study is that spatial autocorrelation among selected quadrats was not eliminated entirely, despite being substantially reduced. This occurred because deforestation in Brazilian Amazonia is concentrated in certain areas (Skole & Tucker, 1993; Alves et al., 1999), so that quadrats with the highest deforestation (>66.7% forest loss) tend to be clustered together (Fig. 1). An alternative strategy is to select quadrats purely at random, rather than stratifying on deforestation intensity, but this would result in a skewed distribution of deforestation values (with most quadrats having little or no deforestation, and a limited number having much higher deforestation). In this study, we tolerated some spatial autocorrelation to ensure that sampled sites included a wide range of deforestation values.
Finally, many of the predictors we assessed were both spatially autocorrelated (e.g. population density, highways, and roads) and functionally related (e.g. highways promote forest colonization, leading to local population increases), and thus it is difficult to assess the impacts of such factors independently. Moreover, the causal relationships between the predictors and deforestation are sometimes complex. For example, roads clearly promote deforestation, but deforestation can also promote roads by providing political justification for road-building, thereby forming a positive feedback loop (see Fearnside, 1987). Further studies that stratify sampling on each predictor, focus on subsets of the basin to reduce correlations among variables, and incorporate time-lags and path analyses will be needed to reveal more fully the functional relationships among these predictors and deforestation.
Few would be surprised to learn that highways and roads sharply increase in Amazonian deforestation (cf. Fearnside, 1987; Imbernon, 2000; Carvalho et al., 2001; Laurance et al., 2001; Steininger et al., 2001a; Nepstad et al., in press). Our analysis, however, helps to quantify the impacts of roads and highways on a basin-wide scale, and to evaluate their importance relative to other potential predictors. In this study, highway proximity emerged as the single most important predictor of deforestation. Because they promote efficient, year-round access to forests, highways tend to have considerably larger-scale impacts than roads (compare Figs. 2 and 3). It must be noted, however, that major highways tend to spawn secondary road networks, as has occurred extensively in the southern and eastern Amazon (for example, around the Belém-Brasília, Transamazon, BR-364 highways). Thus, the effects of highways and roads can not be assessed completely independently.
An unexpected result is that distance to navigable rivers was not a significant predictor of deforestation. It is important to emphasize, however, that our analysis focuses on identifying general, basin-wide predictors. Significant deforestation has obviously occurred along some major rivers (especially whitewater rivers such as the Solimões and Amazon, which carry relatively fertile sediments), but this is minor compared to the vast areas of deforestation associated with highways and roads.
Another key conclusion of our study is that deforestation was most concentrated in drier, more seasonal areas of the Brazilian Amazon. In Bolivian Amazonia, Steininger et al. (2001a) also found that drier, more deciduous forests were most vulnerable to deforestation. The most obvious reason for this pattern is that drier forests are easier to burn, reducing the effort needed to clear forests and maintain pastures and croplands. There might also be some tendency for drier forests to overlay more fertile soils, because heavy rainfall can leach soil nutrients. On a basin-wide scale, our soil-fertility index was weakly but significantly related to annual rainfall (R2=1.2%, F1,1925=22.87, P<0.0001, slope negative; linear regression analysis).
It must be emphasized, however, that the effects of rainfall are complicated by other factors. In Brazilian Amazonia, drier forests are most extensive along the southeastern margin of the basin, which is also most accessible to immigrants and markets from southern population centers. However, deforestation has also been very high in areas such as Roraima state, which contains extensive seasonal forests but is located in northern Amazonia far from major population centers. In a recent study, Schneider et al. (2000) asserted that rainfall, rather than proximity to population centers, is the more important determinant of deforestation in the Brazilian Amazon.
Because deforestation is greatest in seasonal areas, the vulnerability of different Amazonian habitat types is not uniform. Deciduous and semi-deciduous forests, woody oligotrophic vegetation such as campina and campinarana, regenerating forests, and ecotonal forests at the rainforest-cerrado woodland boundary are likely to be most vulnerable (Kangas, 1990; Fearnside & Ferraz, 1995). These forest types have suffered the heaviest deforestation and are poorly represented in parks and reserves (Fearnside & Ferraz, 1995; Ferreira 2001).
The destruction of Amazonian forests is a major source of atmospheric carbon emissions (Fearnside, 1997; Houghton et al., 2000). The high vulnerability of drier forests could influence carbon emissions, because seasonal forests have lower aboveground biomass than wetter forests (Houghton et al., 2000). However, drier forests probably contain more belowground biomass, because they require extensive root systems to access deep soil water (Nepstad et al., 1994). As a result, the destruction of seasonal forests in Amazonia could ultimately produce nearly as much atmospheric carbon emissions per hectare as the loss of dense rainforest.
Among our most surprising results was the seemingly weak effect of soils on deforestation; soil fertility and waterlogging had nonsignificant effects, and soil depth was of only limited importance. In part, this might reflect limited knowledge about the spatial distribution of Amazonian soils. The soil map we used is regarded as the best available, but is coarse-scale and undoubtedly incomplete. Amazonian soils can be patchy at a spatial scale of hundreds of meters (Moran, 1981; Fearnside, 1984; Chauvel et al., 1987; Laurance et al., 1999), limiting the effectiveness of coarse-scale soil maps. It is difficult, moreover, to create a general index of soil fertility because different crops have varying soil requirements (e.g. stoniness is no problem for pastures and many tree crops but it is for annual crops, whereas waterlogging precludes most crops but is suitable for rice). Finally, soil-texture data were not consistently available to us, and certain soil types (e.g. Ferralsols, Acrisols) can include a wide range of different texture classes that substantially affect their agricultural potential.
Despite these limitations, it is quite possible that soils have played a relatively insignificant role in determining deforestation activity on a basin-wide scale. In Brazilian Amazonia, most large-scale colonization and development projects have been implemented with little or no information about soil characteristics (Moran, 1981; Fearnside, 1984). Even on a local scale, soil quality was found to be much less important than distance from roads in determining deforestation by colonists (Fearnside, 1986). Rather than determining where deforestation occurs, soil features may influence how quickly land is abandoned after being deforested, and how rapidly forest regenerates after abandonment (Saldarriaga et al., 1986; Laurance et al., 1999). It is also not inconceivable that farmers and ranchers on low-fertility soils are forced to clear larger areas of forest than those on better soils, in order to remain viable; this would complicate a simple relationship between soil fertility and deforestation.
This study clearly implicates local population density as a key correlate of deforestation (see also Wood & Skole, 1998; Imbernon, 2000; Steininger et al., 2001a). Our analyses suggest that rural-population density is a somewhat better predictor of local deforestation than urban-population size, which accords with some earlier analyses (Wood & Skole, 1998; Laurance et al., in press). Nevertheless, rural and urban populations in Amazonia interact in complex ways, with frequent movements of people and materials between the two (Browder & Godrey, 1997; Imberon, 2000). Although rural populations have the most direct impact on forests, urban populations influence demand for forest resources, market availability, human migration, and other factors (Browder & Godfrey, 1997).
In Brazilian Amazonia, not all segments of the rural population are equally important drivers of deforestation. Large-scale cattle ranchers, rather than small-scale farmers, appear to be responsible for roughly 70% of all forest loss (Fearnside, 1993; Wood & Skole, 1998; Nepstad et al., 1999b). However, small-scale farmers and government-sponsored forest-colonization projects have caused extensive deforestation in some areas (Schmink & Wood, 1992; Alves et al., 1999; Imbernon, 2000), and forest exploitation on small farms is often relatively intensive (Fearnside, 1993). Mounting social pressures for agrarian reform, which involves redistributing lands from large to small landowners, could lead to further deforestation at the hands of small-scale farmers (Fearnside, 2001).
Our results highlight the critical role of highways, roads, and population growth in determining local forest destruction. Clearly, the rapid expansion of the Amazonian population, which rose from about 2.5 million in 1960 to over 20 million today (IBGE, 2000), is increasing pressures on forests. Such dramatic growth has largely resulted from government policies designed to accelerate immigration and economic development in the region, including large-scale colonization schemes, credit and tax incentives to attract private capital, and major transportation projects such as the TransAmazon and Manaus-Boa Vista Highways (Moran, 1981; Smith, 1982; Fearnside, 1987). As a result, the Amazon has the highest rate of immigration of any region in Brazil, and has often been characterized as an “escape valve” for reducing overcrowding, social tensions and displacement of agricultural workers in other parts of the country (Anon., 2001). In addition to rapid immigration, existing populations in the region are growing quickly because most residents begin bearing children at a relatively young age and the population is strongly skewed toward young individuals currently in or entering their reproductive years (Brown & Pearce, 1994; Wood & Perz, 1996).
As part of its policies to promote immigration and rapid economic development, the Brazilian federal government is planning to expand sharply the existing network of paved highways and infrastructure within Brazilian Amazonia. Under the auspices of its Avança Brasil (Advance Brazil) program, the government intends to invest over US$40 billion in highways, railroads, gas lines, power lines, hydroelectric reservoirs, and river-channelization projects that will criss-cross large expanses the basin, greatly increasing accessibility to remote frontier areas. About 7500 km of highways will be paved (Laurance et al., in press) and road networks will also be expanded markedly. Modeling studies that incorporate spatial data on planned Amazonian highways and infrastructure projects suggest that the Avança Brasil program will lead to dramatic increases in forest loss and fragmentation (Carvalho et al., 2001; Laurance et al., 2001). The large-scale fragmentation of Amazonian forests would have diverse ecological impacts (Lovejoy et al., 1986; Skole & Tucker, 1993; Laurance et al., 1997, 1998, 2000), and the surviving forest remnants would be far more vulnerable than intact forests to predatory logging (Nepstad et al. 1999), wildfires (Cochrane & Laurance, in press), and other degrading activities.
It is vital to emphasize, however, that the potentially dire losses of Amazonian forests projected by recent studies (Carvalho et al., 2001; Laurance et al., 2001) are not yet a faite accompli. Pressures from the international community and from foreign investors (which provide significant financial support for Avança Brasil and other Amazonian development initiatives) can strongly influence development policy, planning, and environmental assessment in Brazil. Cooperative resource-management programs supported by wealthy nations (e.g. Anon. 1999a) and nongovernmental organizations (e.g. Anon 1999b) can also have major environmental benefits. Such efforts are crucially needed to ensure that Amazonian forests are not irreversibly degraded in the coming decades.
We thank William Magnusson, Jeff Chambers, and Mark Cochrane for commenting on a draft of the manuscript. Support was provided by the NASA-LBA program, A. W. Mellon Foundation, Conservation International, National Institute for Amazonian Research (INPA), and Smithsonian Tropical Research Institute. This is publication number --- in the BDFFP technical series.
Alves, D. S., Pereira, J., de Sousa, C., Soares, J. & Yamaguchi, F. (1999) Characterizing landscape changes in central Rondônia using Landsat TM imagery. International Journal of Remote Sensing. 20, 2877-2882.
Anon. (1999a) Pilot Program to Conserve the Brazilian Rain Forest: annual report on progress. World Bank, Brasília, Brazil.
Anon. (1999b) Programa Fogo: emergência crônica. Friends of the Earth: Brazilian Amazonia, São Paulo, Brazil.
Anon. (2001) Amazônia cede as terras e o governo se esquece das verbas. O Liberal,
Belém, Brazil, 4 April.
Beinroth, F. H. (1975) Relationships between U.S. soil taxonomy, the Brazilian system,
and FAO/UNESCO units. Soil management in tropical America (ed. by E.
Bornemisza and A. Alvarado), pp. 97-108. North Carolina State University,
Raleigh, North Carolina.
Browder, J. O. & Godfrey, B. J. (1997) Rainforest cities: urbanization, development, and globalization of the Brazilian Amazon. Columbia University Press, New York.
Brown, K. & Pearce, D. W., editors.
(1994) The causes of tropical
deforestation: the economic and statistical analysis of factors giving rise to
the loss of tropical forests.
University College London Press,
Carvalho, G., Barros, A. C., Moutinho, P. & Nepstad, D. C. (2001) Sensitive development
could protect the Amazon instead of destroying it. Nature. 409, 131.
Chauvel, A., Lucas, Y. & Boulet, R. (1987) On the genesis of the soil mantle of the region
of Manaus, Central Amazonia, Brazil. Experientia. 43, 234‑240.
Cochrane, M. A., Alencar, A., Schulze, M., Souza, C., Nepstad, D., Lefebvre, P., &
Davidson, E. (1999) Positive feedbacks in the fire dynamics of closed canopy
tropical forests. Science. 284, 1832-1835.
Cochrane, M. A. & Laurance, W. F. (In press) Fire as a large-scale edge effect in
Amazonian forests. Journal of Tropical Ecology.
Crawley, M. J. (1993) GLIM for ecologists. Blackwell Scientific, London, U.K.
Eastman, J. R. (1999) ISRISI 32: guide to GIS and image processing. Clark Labs,
Fearnside, P. M. (1984) Initial soil quality conditions on the TransAmazon Highway of
Brazil and their simulation in models for estimating human carrying capacity.
Tropical Ecology. 25, 1‑21.
Fearnside, P. M. (1986) Human carrying capacity of the Brazilian Amazon. Cambridge
University Press, Cambridge, U.K.
Fearnside, P. M. (1987) Deforestation and international economic development projects in
Brazilian Amazonia. Conservation Biology. 1, 214-221.
Fearnside, P. M. (1990) Environmental destruction in the Amazon. The future of
Amazonia: destruction or sustainable development? (ed. by D. Goodman and A.
Hall), pp. 179-225. MacMillan,
Fearnside, P. M. (1993) Deforestation in the Brazilian Amazon: the effect of population
and land tenure. Ambio. 8, 537-545.
Fearnside, P. M. (1997) Greenhouse gases from deforestation in Brazilian Amazonia:
net committed emissions. Climatic Change. 35, 321-360.
Fearnside, P. M. (1999) Biodiversity as an environmental service in Brazil’s Amazonian
forests: risks, value and conservation. Environmental Conservation. 26, 305-321.
Fearnside, P. M. (2001) Land-tenure issues as factors in environmental destruction in
Brazilian Amazonia: the case of southern Pará. World Development. 29, 1361-
Fearnside, P. M. & Ferraz, J. (1995) A conservation gap analysis of Brazil’s Amazonian
vegetation. Conservation Biology. 9, 1134-1147.
Ferreira, L. V. (2001) A conservation gap analysis for the Brazilian Legal Amazon. Ph.D.
Thesis, National Institute for Amazonian Research (INPA), Manaus, Brazil.
Houghton, R. A., Skole, D. L., Nobre, C. A., Hackler, J. L., Lawrence, K. T. &
Chomentowski, W. H. (2000) Annual fluxes of carbon from deforestation and
regrowth in the Brazilian Amazon. Nature. 403, 301-304.
IBGE. (2000) Censo demográphico, dados distritais: XI recenseamento geral do Brasil
2000. Brazilian Institute of Geography and Statistics (IBGE), Brasília, Brazil.
IDRISI. (1996a) ArcView GIS. Environmental Research Systems Institute, Redlands,
IDRISI. (1996b) ArcView Spatial Analyst. Environmental Research Systems Institute,
Imbernon, J. (2000) Deforestation and population pressure in the state of Rondônia,
Brazil. Bois et Foréts des Tropiques. 266, 23-33.
INPE. (1998) Deforestation estimates in the Brazilian Amazon, 1995‑1997. Unpublished
report, National Institute for Space Research (INPE), São José dos Campos, Brazil.
INPE. (2000) Deforestation estimates in the Brazilian Amazon, 1998-1999. Unpublished
report, National Institute for Space Research (INPE), São José dos Campos, Brazil.
ISA. (1999) PROBIO databases. Instituto Socioambiental (ISA), São Paulo, Brazil.
Kaimowitz, D. & Angelsen, A. (1998) Economic models of tropical deforestation: a review.
Center for International Forestry Research (CIFOR), Bogor, Indonesia.
Kangas, P. (1990) Deforestation and diversity of life zones in the Brazilian Amazon: a
map analysis. Ecological Modelling. 49, 267-275.
Lambin, E. F. (1994) Modelling deforestation processes: a review. Office of Official
Publications of the European Commission, Luxembourg, Belgium.
Laurance, W. F. (1998) A crisis in the making: responses of Amazonian forests to land
use and climate change. Trends in Ecology and Evolution. 13, 411-415.
Laurance, W. F., Albernaz, A. K. M. & Da Costa, C. (In press) Is deforestation
accelerating in the Brazilian Amazon? Environmental Conservation.
Laurance, W. F., Cochrane, M. A., Bergen, S., Fearnside, P. M., Delamonica, P., Barber,
C., D’Angelo, S. & Fernandes, T. (2001) The future of the Brazilian Amazon.
Science. 291, 438-439.
Laurance, W. F., Delamonica, P., Laurance, S. G., Vasconcelos, H. L. & Lovejoy, T. E.
(2000) Rainforest fragmentation kills big trees. Nature. 404, 836.
Laurance, W. F., Ferreira, L. V., Rankin-de Merona, J. M. & Laurance, S. G. (1998) Rain
forest fragmentation and the dynamics of Amazonian tree communities. Ecology.
Laurance, W. F., Fearnside, P. M., Laurance, S. G., Delamonica, P., Lovejoy, T., Rankin-de
Merona, J., Chambers, J. & Gascon, C. (1999) Relationship between soils and
Amazon forest biomass: a landscape-scale study. Forest Ecology and Management.
Laurance, W. F., Laurance, S. G. Ferreira, L., Rankin-de Merona, J., Gascon, C. &
Lovejoy, T. (1997) Biomass collapse in Amazonian forest fragments. Science.
Lovejoy, T. E., Bierregaard, R. O., Rylands, A., Malcolm, J., Quintela, C., Harper, L.,
Brown, K., Powell, A., Powell, G., Schubart, H. & Hays, M. (1986) Edge and
other effects of isolation on Amazon forest fragments. Conservation biology: the science of scarcity and diversity (ed. by M. E. Soulé), pp. 257-285. Sinauer, Sunderland, Massachusetts.
McCune, B. & Mefford, M. J. (1997) PC‑ORD: multivariate analysis of ecological data.
MjM Software, Glenenden Beach, Oregon.
Moran, E. F. (1982) Developing the Amazon. Bloomington: Indiana University Press.
Nepstad, D., Carvalho, G., Barros, A. C., Alencar, A., Capobianco, J., Bishop, J., Moutinho, P., Lefebvre, P. & Silva, Jr., U. (In press) Road paving, fire regime feedbacks, and the future of Amazon forests. Forest Ecology and Management.
Nepstad, D. C., Carvalho, C., Davidson, E., Jipp, P., Lefebre, P., Negreiros, P., Silva, E.,
Stone, T., Trumbore, S. & Vieira, S. (1994) The role of deep roots in the
hydrological cycles of Amazonian forests and pastures. Nature. 372, 666-669.
Nepstad, D. C., Moreira, A. G. & Alencar, A. A. (1999b) Flames in the rain forest:
origins, impacts, and alternatives to Amazonian fires. Pilot Program to Conserve
the Brazilian Rain Forest, Brasília, Brazil.
Nepstad, D. C., Verissimo, A., Alencar, A., Nobre, C., Lima, E. Lefebre, P., Schlesinger,
P., Potter, C., Moutinho, P., Mendoza, E., Cochrane, M. & Brooks, V. (1999a)
Large-scale impoverishment of Amazonian forests by logging and fire. Nature. 398, 505-508.
Noy‑Meir, I., Walker, D. & Williams, W. T. (1975) Data transformations in ecological
ordination. II. On the meaning of data standardization. Journal of Ecology.
Oliveira, J. B., Jacomine, P. K. T. & Camargo, M. N. (1992) Classes gerais de solos do
Brasil. FUNEP, Jaboticabal, Brazil.
Reis, E. J. & Margulis, S. (1991) Perspectivas econômicas do desflorestamento da
Amazônia. Textos para Discussão No. 215. Instituto de Pesquisa Econômica Aplicada
(IPEA), Brasilia, Brazil.
Saldarriaga, J. G., West, D. C. & Tharp, M. L. (1986) Forest succession in the Upper Rio
Negro of Colombia and Venezuela. Environmental Sciences Division Publication
2694, Oak Ridge National Laboratory, Oak Ridge, Tennessee.
Schmink, M. & Wood, C. H. (1992) Contested frontiers in Amazonia. Columbia
University Press, New York.
Schneider, R. R., Arima, E., Veríssimo, A., Barreto P. & Souza, Jr., C. (2000) Amazônia sustentável: limitantes e oportunidades para o desenvolvimento rural. World Bank, Brasília and IMAZON, Belém, Brazil.
Skole, D. & Tucker, C. J. (1993) Tropical deforestation and habitat fragmentation in
the Amazon: satellite data from 1978 to 1988. Science. 260, 1905-1910.
Smith, N. J. H. (1982) Rainforest corridors: the Transamazon colonization scheme.
University of California Press, Berkeley, California.
Sombroek, W. G. (1984) Soils of the Amazon region. The Amazon – limnology and
landscape ecology of a mighty tropical river and its basin (ed. by H. Sioli), pp. 521-
535. W. Junk, Dordrecht, Netherlands.
Sombroek, W. G. (2000) Amazon landforms and soils in relation to biological diversity.
Acta Amazonica. 30, 81-100.
Sombroek, W. G. (In press) Annual rainfall and dry-season strength in the Amazon region
and their environmental consequences. Ambio.
Steininger, M. K., Tucker, C. J., Townshend, J. R. G., Killeen, T. J., Desch, A., Bell, V. &
Ersts, P. (2001a) Tropical deforestation in the Bolivian Amazon. Environmental
Conservation. 28, 127-134.
Steininger, M. K., Tucker, C. J., Ersts, P., Killeen, T. J., Villegas, Z. & Hecht, S. B.
(2001b) Clearance and fragmentation of tropical deciduous forest in the Tierras
Bajas, Santa Cruz, Bolivia. Conservation Biology. 15, 856-866.
Walker, R. T., Homma, A., Scatena, F., Conto, A., Carvalho, R., Rocha, A., Ferreira, C.,
Santos, A. & Oliveira, R. (1993) Sustainable farm management in the Amazonian
piedmont. Congresso Brasileiro de Economia e Sociologia Rural. 31, 706-720.
Wood, C. H. & Perz, S. (1996) Population and land-use change in the Brazilian Amazon.
Population growth and environmental issues (ed. by S. Ramphal and S. Sindig),
pp. 95-108. Praeger, Westport, Connecticut.
Wood, C. H. & Skole, D. (1998) Linking satellite, census, and survey data to study
deforestation in the Brazilian Amazon. People and pixels: linking remote sensing
and social science, pp. 70-92. National Academy Press, Washington, D.C.
Table 1 Product-moment correlations of ten predictor variables with two ordination axes produced by nonmetric multidimensional scaling. Asterisks indicate significant correlations, using a Bonferroni-corrected alpha value (P=0.0025).
Variable Axis 1 Axis 2
Rural-population density -0.881* -0.130
Urban-population size -0.904* -0.180
Distance to nearest paved highway 0.791* 0.424*
Distance to nearest unpaved road 0.355* 0.579*
Distance to nearest navigable river 0.377* -0.740*
Annual rainfall 0.069 0.782*
Dry-season severity -0.381* -0.769*
Soil fertility 0.301* 0.284*
Soil waterlogging -0.440* 0.290*
Soil depth and stoniness 0.396* -0.129
Variation explainedA 62.0% 26.4%
ACoefficients of determination for correlations between ordination distances and distances in the original n-dimensional space.
Figure 1 The Brazilian Legal Amazon divided into 2500-km2 quadrats; shaded areas indicate the 120 quadrats used in this study.
Figure 2 Relationship between deforestation and the mean distance to paved highways in the Brazilian Legal Amazon. Data for quadrats further than 600 km from highways are not shown.
Figure 3 Relationship between deforestation and the mean distance to unpaved roads in Brazilian Amazonia. Data for quadrats further than 300 km from roads are not shown.
Figure 4 Relationship between deforestation and rural-population density in Brazilian Amazonia. Data for rural densities above 500 persons/km2 are not shown.
Figure 5 Relationship between dry-season duration (number of months with <100 mm of rainfall) and deforestation in Brazilian Amazonia.