• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

A Disappearing Biome? Reconsidering Land-Cover Change in the Brazilian Savanna

Posted on: Saturday, 23 July 2005, 03:00 CDT

The Cerrado, the tropical savanna covering 22% of Brazil's territory, or approximately 1.783 million km^sup 2^, has suffered significant human impacts during the past three decades. This paper re-examines estimates of Cerrado vegetation change dynamics using high-resolution satellite remote sensing data from an area of interest extracted from eastern Mato Grosso State. This region has undergone a high degree of typical agricultural development since the early 1970s. Results indicate significant loss of original vegetation as well as high levels of regeneration, suggesting Cerrado vegetation may be more resilient to human impacts than catastrophic estimations suggest. The paper concludes with a critical review of Cerrado land-cover change studies and the implications of evidence for vegetation regeneration, land-cover dynamism and land-use intensification, paying particular attention to spatial scale and research methods. The discussion concludes that Cerrado land-cover change studied at a higher resolution and larger scales (smaller area) is required to represent more effectively the complexity of land conversion for better assessment of human impacts and environmental policy.

KEY WORDS: Brazil, land-cover change, tropical savanna, remote sensing, Cerrado, environmental policy

Introduction

The Cerrado, a tropical savanna covering 22% of Brazil's territory or approximately 1.783 million km^sup 2^, has suffered significant human impacts during the past three decades (Figure 1). Today's globally significant breadbasket hardly resembles the Cerrado surveyed by the Royal Society/ Royal Geographical Society in the late 1960s (Brown et al. 1970; Warnken 1999). Many environmental activists and ecologists argue that the expansion of cattle ranching and soybean cultivation in the Cerrado is indicative of economic globalization's voracious appetite for natural resources (Klink et al. 1993; Alho and Souza Martins 1995; Ratter et al. 1997; Branford and Freris 2000; WWF-Brasil 2000; Fearnside 2001; Klink and Moreira 2002). Although the view that the Cerrado is rapidly disappearing has gained much currency in international environmental policy circles, high-resolution spatial analysis has neither quantified the dynamics of Cerrado landcover change nor specified the spatial processes of vegetation fragmentation. The only estimates focus on aggregate vegetation loss for the entire ecoregion. One estimate suggests that Cerrado landcover loss is outpacing the 13% or 400 000 km^sup 2^ of aggregate Brazilian rainforest loss (Ratter et al. 1997). Another argues that agro-pastoral land uses have converted 40% of the Cerrado (JoIy et al. 1999). Other studies use agricultural census data to claim that approximately 50% of the Cerrado has been eliminated (Alho and Souza Martins 1995, 34). A further estimate, based upon small cartographic scale (large area) satellite remote sensing, is that only 35% of the Cerrado is in a 'relatively natural state' (Mantovani and Pereira 1998). A fifth estimate indicates that humans have modified more than 80% of the Cerrado, leaving only 20% of Original, native vegetation' (WWF- Brasil 2000, 9; Mittermeier ef al. 2000).

Notwithstanding their quantitative disparities, estimates of the 'disappearing' Cerrado create consensus around an environmental discourse of biodiversity loss and attending agenda' that promotes protected areas for conservation. Key elements on the agenda include consolidating current conservation units (only 5.2% of the Cerrado is protected by some conservation regime), developing large ecological corridors, and enhancing legislation to protect undisturbed areas (Ratter et al. 1997; Silva and Bates 2002; Calvalcanti and Joly 2002, 354-7). With estimates indicating massive and rapid loss of territorial biota to agro-pastoral activities, it seems logical that the best path for biodiversity protection is the swift demarcation of lands for preservation. In practice, little else beyond the advocacy for new or expanded protected areas of undisturbed Cerrado is being pursued by environmental policymakers (Cavalcanti and JoIy 2002).

Figure 1 Brazilian Cerrado and study region location

This paper re-examines estimates of Cerrado vegetation change dynamics. It tests the accuracy of previously stated assessments of Cerrado transformation using high-resolution satellite remote sensing data. This approach allows for a comparison of overall estimates on Cerrado land conversion with an area of interest extracted from eastern Mato Grosso State, a region in the Brazilian Cerrado that has undergone a high degree of agricultural development since the early 1970s (Figure 1). The paper begins with an overview of Cerrado biogeography and the study region in eastern Mato Grosso. Next, the paper discusses the processing and classification of the satellite remote sensing data, followed by a review of the land- cover change detection analysis and results. The paper employs change detection methods to describe the region's complex landscape mosaics and mixtures of land cover between 1986 and 1999. This type of high-resolution analysis reveals how different land covers follow time sequences and how these land covers are often reversible for the same geographical unit. The ensuing discussion section draws from the case of Cerrado land-cover change in the study region to discuss the importance of vegetation regeneration and land-cover change dynamics, remote sensing methodology, and environmental policy. The paper concludes that the analysis offers a contrasting view of Cerrado land conversion when compared with current estimates. This disparity not only holds implications for a better understanding of the dynamic and process of Cerrado landscape transformation, but also for scientific study of biological diversity, climate change, and environmental policies.

The Brazilian Cerrado and Eastern Mato Grosso

The Brazilian Cerrado is characterized by an undulating topography with wide interfluves interrupted by tributaries of the Amazon, ParanaParaguay, and the So Francisco Rivers. The Cerrado vegetation is a mix of grasses, woody plants, fireresistant twisted trees with thick, corky bark, sclerophyllous leaves, and vibrant flowers. The Cerrado originally covered a region equivalent to the combined areas of California, Nevada, Utah, Colorado, Arizona and New Mexico, spanning a geographic area from 24 S to 4 S latitudes. The ecoregion varies in elevation (300-1800m) with annual precipitation between 900 and 1800mm and an average monthly temperature of 21C (Ab'Saber 1971; Goedert 1983, 407; EMBRAPACerrados 1998). The climate is punctuated by a severe dry season that ranges between three and five months from May to September. The Brazilian Cerrado is a repository of biological diversity, boasting an estimated 160000 species of plants, mammals, fungi and notably diverse flora, showing high specificity of native angiosperms (Alho 1981; Silberbauer Gottsberger and Gottsberger 1984; Mares et al. 1986; Redford and Fonseca 1986; Ratter et al. 1997, 226; Myers et al. 2000; Silva and Bates 2002).

Canopy cover, type of plant cover and species composition distinguish the Cerrado's diverse vegetation structure. Cerrado vegetation varies in height and density. The vegetation ranges from an open herbaceous land cover to orchard-like treescrub savanna or almost a closed canopy of 12-15 m. In addition to the savanna vegetation and forest formations, seasonal savannas are present to a lesser extent in the Cerrado region. Land covers studied in the paper are based upon types found in the study region. They include the following: (1) Cerrado (sensu lato), (2) seasonally inundated savannas, and (3) forests (Ribeiro and Walter 1998; Oliveira Filho and Ratter 2002). Three general land covers can be further classified into subtypes. Cerrado sensu IaW is most common in the ecoregion, and it can be subdivided into three types: campo cerrados (grasses); cerrado sensu stricto (savanna); and cerrado (woodland). Seasonal savannas can be further subdivided into parque do cerrado (parkland) and vereda. Forest formations, including riparian and mesophytic, represent the third broad land cover in the Cerrado and study region. These landscape vegetation differences within the biome are determined by fire, plant-available moisture and nutrients in soils, and topography (Coutinho 1977 1982; Goodland and Pollard 1973; Oliveira-Filho et al. 1989; Furley 1994 1996 1999; Kaufman et al. 1994; Bilbao et al. 1996; Ratter et al. 1997, 225; Mistry 1998, 430-2; 2000).

Concern for biological diversity loss due to rapid land conversion drives the Cerrado environmental conservation agenda. The Cerrado ranks twelfth on a list of global 'hot spot' areas, ecoregions that contain high levels of plant endemism and are 'under threat' (Mittermeier et al. 1998, 519). The 'hot spot' designation shapes the actions of various environmental organizations, in particular Conservation International (Cl), which develops and supports conservation priority setting workshops (Government of Brasil and FUNATURA 1999; Cavalcanti and Joly 2002; Cl 2003). National and global partnerships for Cerrado conservation also have focused on improving national and state parks or encouragedlocal, sustainable uses of the region's biodiversity. For example, the Brazilian environmental organization FUNATURA along with the Global Environmental Facility (GEF) are expanding protected areas near major Cerrado parks through the establishment of private national heritage reserves (Reservas Particulares do Patromnio Nacional, RPPNs) (Lake 1999; GEF 2003; FUNATURA 2003). Moreover, FUNATURA and The Nature Conservancy jointly manage the 84 000 ha Grande Serto Veredas National Park, a conservation unit established under Brazil's only debt-for-nature swap.

The case study is located in eastern Mato Grosso state, a region located on the low-relief Serra do Roncador Plateau between the Xingu River, a southern Amazon tributary, and the Araguaia River. The region's vegetation cover is typical of the Cerrado. It ranges from campo limpo to cerrado, as described in a set of reports from a geographical expedition (Askew et al. 1970; Brown et al. 1970, 371- 2). The study region's northern reach forms the transition belt between the savanna and the dry tropical forest.

Private agricultural colonization has played an integral role in the conversion of the Cerrado to agriculture (Jepson 2003). Eastern Mato Grosso is a relatively older agro-pastoral region in which local farmers practiced intensive land uses common in other Cerrado areas. Beginning in the early 1970s, more than 800 families from southern Brazil joined a colonization cooperative or purchased land from a private colonization firm to settle in eastern Mato Grosso's Cerrado. These colonization projects initiated almost a decade of migration from southern Brazilian states to municipalities of Canarana, gua Boa, and Nova Xavantina (Figure 2). By 1990, private colonization initiatives had sold more than 460 040 ha in 25 projects in the region's Cerrado. Major agricultural activities at present include cattle ranching and annual cultivation of upland rice, maize and soybeans.

Figure 2 Eastern Mato Grosso and area of interest

Table 1 Landsat data

Remote sensing methods

Data processing

Three Landsat high-spatial resolution satellite images were acquired for the years 1986, 1992 and 1999. An area of interest (3896 km2) was extracted from the images and examined in a change detection analysis (Figure 2). The area of interest (AOI) covers major areas of settlement and colonization. Two Landsat images and one Landsat +ETM image, all of which have a spatial resolution of 30 m x 30 m, were obtained at a six- and seven-year interval, respectively. Near-anniversary date images, which cover one Landsat scene, minimize the effects of seasonal phenological differences that may cause spurious change to be detected in the analysis. To reduce scene-to-scene variability, all ground data were collected during the dry season, between August and September (Table 1).

Geometric and radiometric corrections followed standard processing methods (Jensen 1996). Landsat data (bands 1, 2, 3, 4, 5, 7) were geometrically rectified to the UTM (22-S) coordinate system to create an objective spatial basis for comparing the three scenes. The 1999 scene was resampled to a common coordinate system using over 100 ground control points distributed throughout the image and obtained by a global positioning system, a Carmin GPS base station system with an accuracy of between 5 and 10m. All subsequent scenes were rectified to the UTM coordinates by resampling each to the corresponding 1999 image. The AOI was extracted from the data in the resampling process. Radiometric corrections took into account the sensor parameters for 1999 +ETM to transform the original digital numbers into surface reflectance measurements. The 1999 image, which was corrected for the sensor gain and bias, was used as a reference image to reduce the effects of sensor difference when comparing the 1986 and 1992 images. The output images and the reference image were classified and subsequently used in change detection process.

Unsupervised classification

Landsat bands 3 (visible red), 4 (near infrared) and 5 (mid- infrared) were selected from the original data set to reduce the processing time, data volume, and band data correlation. Preliminary data exploration was conducted using bands 2, 3 and 4. The conclusion was that the chosen bands better distinguished land covers of interest. Bands 3, 4, and 5 combined to make a composite 8- bit image that was the input for the self-iterative module (ISOCLUST) for cluster seeding. The iterative process follows the maximum likelihood procedure to assign each pixel to 19 clusters. Once the clusters were established, land covers were identified, assigned, and aggregated into three basic classes.

1 Cerrado. This class includes cerrado sensu latu (campos cerrados, cerrado sensu strito, and cerrado) and seasonal savannas.

2 Forests. This class includes seasonally dry tropical and riparian forests.

3 Agro-pastoral. This class includes all pastures and agricultural fields.

Based upon knowledge of the region's environment, ground data and general reflectances, land covers were assigned to clusters. Problems arose in the classification process whereby burnt Cerrado and water were difficult to distinguish in the cluster process. Two categories were aggregated into the Cerrado class, as the overall area of water was very limited and would not significantly alter the final analysis of change. This is evident in the final accuracy assessment table. Areas of burnt Cerrado were concluded not to be agricultural fields, and thus could be aggregated logically into the Cerrado land cover. Spectral reflectances of areas, satellite data before and after the date of burning, and field knowledge of the region indicate that this land area was not cultivated for that time period; thus, the burning scars represented wildfire in the Cerrado, not agricultural fire used for cropland or pasture renewal.

Ground data were used to assess the accuracy of the unsupervised classification. Between July and August 2000, detailed ground observations of vegetation cover were collected to assess the accuracy of unsupervised land-cover classification on the 1999 TM image. Over 160 land-cover polygons provided at least 30 pixels for 16 land classes. Land classes included bare soil, millet fields, no- till agriculture, various types of pasture, relative densities of Cerrado vegetation and forest formations. No attempt to carry out statistically based sampling was made. Rather, all ground observations were made based on access to land and roads and permission to enter private property. All field observations were geo-referenced with a global positioning system, a Carmin GPS hand- held and base station system with accuracy between 5 and 10m. Polygons were mapped and converted to a raster file. This file was compared with the unsupervised classification for the 1999 image to assess classification accuracy.

Table 2 Contingency table and accuracy indicators for classified image (1999)

A contingency table, which summarizes the accuracy of the initial classification to ground data, established overall accuracy and Kappa Index of Agreement for the 1999 image (Table 2). Observations were recorded as the number of pixels selected for three classes (Cerrado, Agro-pastoral and Water). Forest was not identified during the ground data exercise due to lack of access to these land covers during fieldwork. Riparian areas were densely vegetated and difficult to access; forest tracts also were inaccessible by paths or roads. However, the spectral reflectances of forest and riparian vegetation are clearly distinguished from Cerrado and agro-pastoral areas. Thus, this lack of ground truth information was not detrimental to the overall classification process.

The overall accuracy for the 1999 unsupervised classification exceeded minimum targets of an overall accuracy of 85% with no class less than 70% accurate (Congalton 1991; Foody 2002). The Kappa Index of Agreement, which takes into consideration chance agreement, also was above the minimum 85% threshold. The important error to note is of commission. Forested areas were not identified in the ground data, but some forest areas were classified for Cerrado. This can be explained by the confusion of very dense cerrado with forest. Accuracy of the unsupervised classifications for 1986 and 1992 could not be assessed as the necessary ground data, such as aerial photos, either do not exist or were unavailable at the time of analysis. Although no ground data were available for those dates, standard image processing protocol was followed for radiometric rectification to guarantee that the change detected in the image was not due to simple radiometric noise or differences in the sensor's gains or biases.

Table 3 Land-cover change matrices, 1986-92 and 1992-9

Land-cover change detection

Analysis

Change detection for this study was developed in two stages. First, land-cover change analysis used Boolean or binary land cover maps (i.e. CerradoNon-Cerrado) to compare a set of two successive classified images, 1988-92 and 1992-9. Postclassification change detection produced a change detection matrix (Table 3). The change detection matrix identifies, pixel by pixel, detailed information about the land-cover status, and it indicates landcover changes (i.e. from Cerrado to agro-pastoral) for the same geographic area at the highest resolution allowed for this database. One major advantage of this method is that this 'from-to' is directional information, but the analyst always needs to consider that the process is dependent upon the accuracy of land-cover classification (Jensen 1996, 269). In this case, the accuracy was greater than minimum standards for classification (Table 2).

Table 4 Cerrado land-cover sequences between 1986 and 1999 (area and proportion of total affected by each trajectory)

Table 5 Estimated area of Cerrado, total (gross) converted area, and conversion rate in are\a of interest, 1986-99

The second step established a multiple sequence of land-cover change that revealed pixel-by-pixel shifts in land cover over the entire 13-year period (1986-1992-1999). This approach discloses shifts between different land covers for the same geographic area during the two observation years (Mertens and Lambin 2000). To achieve this result, the binary maps (Cerrado-Non-Cerrado) for each set of observation years were cross classified. These results were again cross classified to create all eight possible land-cover sequences. Total land area and proportion of land area for each landcover trajectory were calculated (Table 4). Eight possible land- cover sequences are represented on the maps as separate coverages. Two classes of the final table represent stable land covers (stable forest or agro-pastoral = 1 ; stable Cerrado = 8). The other classes define a series of possible conversion scenarios during the 13-year period. It is also important to note that the ecological condition of Cerrado regrowth after agro-pastoral land uses is unknown as there are no studies on this topic in the broader biogeographical or botanical literature. The eight land-cover sequences are

1 Stable forest or agropastoral land cover. No change detected.

2 Older Cerrado clearing. This sequence represents a shift from Cerrado in 1986 to non-Cerrado by 1992.

3 Clearing in Cerrado regrowth. Previously cleared Cerrado experienced regrowth between 1986 and 1992; the area was again cleared between 1992 and 1999.

4 Recent Cerrado clearing. Cerrado cleared only between 1992 and 1999.

5 Recent Cerrado regrowth. Pre-1986 cleared areas were allowed to regenerate between 1992 and 1999.

6 Cerrado regrowth in recently cleared area. Cerrado that experienced clearing between 1986 and 1992; the area was allowed to regenerate between 1992 and 1999.

7 Cerrado regrowth in pre-1986 cleared area. Cerrado regeneration on areas cleared before 1986.

8 Stable Cerrado land cover. No change detected in Cerrado change during the time series.

Results

Change detection matrices for the successive years 1986-92 and 1992-9 reveal an accumulated or gross reduction of 1986 Cerrado by 1338.49 km^sup 2^, but only a net reduction of 384.85 km^sup 2^. This indicates relatively high levels of vegetation regeneration over the 13-year period. Table 5 summarizes the total or accumulated Cerrado land-cover change between 1986 and 1999. Total Cerrado loss measured for this area of interest between 1986 and 1999 is 70.10%. The rate of total or gross annual Cerrado loss, that is, the rate of Cerrado converted without consideration of regeneration, is 102.96 km^sup 2^ per year or 5.39%. Over the observation years, there is a slight decrease in area converted per year, from 105.78 to 100.54 km2 per year. Between 1986 and 1992, the annual conversion rate is 5.54%, while between 1992 and 1999 the annual conversion rate declined to 5.26%.

Table 6 Estimated area of Cerrado, net converted area, and conversion rate in area of interest, 1986-99

Vegetation regeneration represents an important contribution to overall land cover between 1986 and 1999 (Table 6). Overall net loss of Cerrado vegetation is 20.15% (384.85 km^sup 2^), or 50% less than total Cerrado loss, as it accounts for the difference between Cerrado area converted and the area of Cerrado regeneration. That is, 50% of Cerrado converted during some of this period of time experienced regeneration. While the total annual rate of Cerrado loss decreases over time, the net annual rate of conversion experiences a rise between 1992 and 1999. During the first time period (1986-92), net Cerrado loss is 0.95% and during the second time period it rises to 1.97%.

Change detection matrices permit the description of complex landscape mosaics and mixtures of land cover (Table 4). Land classes follow multiple time sequences and are often reversible over more than two observation years. Over 45% (1771 km^sup 2^) of the total land area shifted at least once and 13.4% (522.36 km^sup 2^) of the total land area shifted land cover twice. Rates of conversion from tropical savanna to agro-pastoral land covers are generally stable over the 13-year period. Older Cerrado clearing (between 1986 and 1992) represents 10.19% (397.16 km^sup 2^) and recent Cerrado clearing (between 1992 and 1999) is 10.77% (419.59 km^sup 2^). Stable or continuously Cerrado areas only account for 21.93% (854.41 km^sup 2^) while stable forests account for 3.3% (128.25 km^sup 2^) and stable agro-pastoral accounts for 29.2% (1142.71 km^sup 2^) of total land cover between 1986 and 1999.

Discussion

Rapid Cerrado disappearance is attributed to modern agro- pastoral land uses, particularly soybean production and cattle ranching. Most explanations of environmental change broadly generalize the process as a unidirectional loss of vegetation to agricultural land uses. As reviewed in the introduction, estimates of Cerrado vegetation loss speculate that modern agro-pastoral expansion has converted up to 80% of the savanna and that the pace of change is increasing. As a first attempt to quantify anthropogenic Cerrado land-cover change using highresolution data for a sample region, this study indicates high levels of vegetation regeneration, although overall Cerrado loss is still high in the long-occupied Cerrado agro-pastoral region. The following discussion will review the implications of the evidence for vegetation regeneration, land-cover dynamism and land-use intensification, paying particular attention to spatial scale and research methods. The discussion concludes that Cerrado land-cover change studied at a higher resolution and larger scales (smaller area) is required to represent more effectively the complexity of land conversion to better inform policymakers.

Vegetation regeneration: scalar implications

The implications of vegetation regeneration for understanding Cerrado land cover vary according to geographical scale. First, at the landscape scale, data from this analysis indicate large areas of regenerated and fragmented undisturbed Cerrado, vegetation that has not been under agro-pastoral land uses. Most revealing is that over 50% of land cover identified as Cerrado was 'reversible', representing some form of secondary vegetation land cover. Currently there are few published studies on the regeneration capacity of Cerrado environments after land clearing for agro-pastoral uses. The exceptions are Durigan et al. (1998) and research on Cerrado regeneration after Eucalyptus (Eucalyptus citriodora) plantations in western So Paulo state (Durigan et al. 1997). Overall, a lack of detailed scientific information on regeneration limits the broader claims about ecological integrity of secondary stands that could be drawn from evidence of Cerrado regrowth. However, the relatively high levels of regrowth broadly suggests resiliency rather than vulnerability after agropastoral land uses have ceased. In addition, many Cerrado fragments that have not been ploughed under for agro- pastoral land uses still persist, despite continued developments of the agricultural economy.

The documented presence of remaining Cerrado fragments begs the question of their influence on biotic and abiotic processes. First, biodiversity assessments, which are of particular concern to Cerrado ecologists, would be better served by considering the ecological implications of patches, fragment connectivity, corridors and edge characteristics of secondary growth and Cerrado fragments (Dale et al. 1994; Forman 1995; Dauber et al. forthcoming). Second, by focusing on regenerated fragments in the complex cultivated Cerrado landscape, future studies may better assess their ecological integrity rather than rely on the theoretical assessment of biodiversity loss, which is based solely on global land-cover change estimates prevalent in the environmental literature (WVVFBrasil 2000; Myers et al. 2000; Mittermeier et al. 2000; Wood et al. 2000). Once secondary growth is taken into account, more precise accounts may be made of how the region's habitat fragmentation and regeneration affect biodiversity, such as displacement of native grasses (Pivello ef a/. 1999) and wildlife (Tubelis and Cavalcanti 2000; Marini 2001 ; Bates 2002). Consequently, environmental policies based on observed patterns are better suited to intervene more precisely rather than simply 'salvaging' (Colchester 1997) large territories for Cerrado preservation and leaving remaining Cerrado 'sacrificed' for agro-pastoral development.

Evidence of regeneration also raises new questions about society- nature interactions within the entire ecoregion. Previous characterizations of Cerrado conversion simply describe the capacity of modern agricultural systems to ravage the natural landscape, driven as they are by profit and dominated by large-scale soybean 'monoculture' and cattle ranches (Alho and Souza Martins 1995; Ratter et al. 1997; Branford and Freris 2000; WWFBrasil 2000; Fearnside 2001). One would expect this pattern of land use to be reflected in the satellite data. Yet, in eastern Mato Grosso, a region that has undergone considerable agricultural development since the early 1970s, land-cover dynamics reflect a different trajectory, one that appears more complex and dynamic than simply large-scale farming and ranching. Moreover, looking at rates of Cerrado loss suggests an intensification of land use; land, which was allowed to regenerate previously, is being renovated for continued cultivation or pasture. Both of these findings require an investigation of the precise causes of land-use change at the scale of individual land manager (McCracken et al. 1999; Perz and Walker 2002; Southworth and Tucker 2001). For example, studies of fire as an ecological transformation factor or a maintenance tool and its role at different moments in the landcover change trajectory, may be better understood at the landscape sc\ale (Eva and Lambin 2000, 773). This would focus consideration of land-cover change on the decision-making processes behind the patterns of regeneration, loss and fragmentation at the landscape scale, rather than on vague 'drivers' of global capitalism (Klink ef a/. 1993; WWF-Brasil 2000). Environmental policy prescriptions that follow from individual land- resource user dynamics, those that capitalize on the preservation of Cerrado fragments and regeneration, may offer alternative conservation policies to current topdown, park-oriented efforts (Brannstrom 2001). These may include 'productive conservation' initiatives (Hall 1997), watershed management approaches (Brannstrom 2001), and more effective incentives for RPPNs (Lake 1999).

Finally, the presence of vegetation regeneration calls into question the aggregate estimates of Cerrado conversion or, perhaps, explains the divergent speculation reviewed in the introduction. Between 1986 and 1999, the case study area has experienced gross or aggregate vegetation loss, which approaches the worst-case estimates of Cerrado conversion. However, the observed landcover complexity illustrates significant processes of vegetation regrowth which significantly changes the net vegetation loss estimates.

The importance of quantifying the reversibility of land covers, identifying complex trajectories, and assessing the landscape fragmentation has important implications for broader global environmental change concerns of ecological services, not just biodiversity (Woodwell 2002). An accurate assessment of secondary vegetation growth better specifies the contribution of Cerrado land conversion and agro-pastoral land uses to overall carbon emissions and climate change (Henderson-Sellers and McGuffie 1995), such as carbon sequestration (Miranda and Miranda 2000; Vourlitis et al. 2001), greenhouse gas emissions (Kirchhoff and Avala 1996; Lardy et al. 2002; Pinto et al. 2002), and biomass burning (Ferek et al. 1998; Kisselle ef al. 2002). The irony is that the type of investigation outlined above has been carried out for years in the humid forests of Amazonia (Moran et al. 1994 1996; Fearnside 1996 1997; Moran and Brondizio 1998; Steininger 1996 2000; Zarin eta/. 2001; Lucas eta/. 2002); perhaps, it is time for similar approaches to be employed for that biome's southern neighbour.

Studying Cerrado land-cover change: remote sensing and census data

Studies of Cerrado land-cover change have only begun to assess the effects of post-1970 human impacts. Primary data used to explore these transformations include remote sensing and agricultural census material. Both data sets offer important insights into the process of Cerrado conversion. However, findings on vegetation regeneration support a methodological re-evaluation. In particular, the evidence for vegetation regeneration highlights analytical problems associated with studies that employ a subjective land-cover classification system, single change detection matrix, and agricultural census data, all of which contribute to an environmental 'zero-sum' discourse of Cerrado land-cover change and represent the biome as disappearing 'grain by grain'. Indeed, the reliance on 'irreversible' land-cover change estimates may overestimate Cerrado loss because they are unable to identify vegetation regeneration. This would not be the first time that local, quantitative methods have corrected or reframed (mis)readings of the tropical savanna landscapes (Fairhead and Leach 1996 1998; Bassett and Zueli 2000).

More broadly, a critique of remote sensing methods is useful to evaluate the limited number of available Cerrado land-cover change studies. As one geographer observes, the categories of analysis which a remote sensing practitioner employs to represent vegetation are not impartial tools to measure land-cover change; rather, in many instances, they 'fix' certain interpretations of the environment, interpretations that are either political or based on subjective assumptions about 'nature' (Robbins 2001). My interest is not to go as far as Robbins's post-structuralist position that all landcover classifications are political. However, I agree that the act of naming and describing land covers must be qualitatively as well as quantitatively interrogated. This, in turn, reveals how subjective views of human-society interaction shape certain interpretive practices in remote sensing analysis.

Mantovani and Pereira (1998) is the only publicly available study using satellite data to assess Cerrado conversion to agro-pastoral activities. The study compares a digitized vegetation map (1:5 000 000) to 144 Landsat images from 1988 to 1993.' in this study, the perceived impact of human activity defines land-cover categories. The authors use a classification system that includes categories such as 'not Cerrado', 'Cerrado not anthropogenically altered', 'Cerrado anthropogenically altered', and 'Cerrado strongly anthropogenically altered'. There are two problems with Mantovani and Pereira's categorization. First, the study does not provide a clear discussion of how categories are developed or how ground data are collected to test coverage accuracy. Second, their assumptions about nature underpin the categories, reinforcing the notion of land cover 'irreversibility'. That is, the level of 'naturalness' is the defining feature of each land cover rather than its physical characteristics. Not only is this definition highly subjective, but it forecloses the possibility that secondary growth is a land cover worth assessing in an overall project to quantify vegetation change. Similar arguments can be applied to how Mantovani and Pereira construct the land-cover change matrix. Simply comparing two data sets, one an ancillary small (cartographic) scale vegetation map representing a baseline2 and the other a set of Landsat data, to identify biome 'change', also precludes the study of vegetation regeneration or land-cover dynamics. Despite these serious methodological shortcomings, the categories used by Mantovani and Pereira are 'fixed' in the quantification of Cerrado land-cover change and these estimates are reproduced as Objective' in multiple publications to bolster the view of a 'disappearing biome' (Government of Brasil and FUNATURA 1999; WWF-Brasil 2000; JoIy ef a/ . 1999).

Agricultural census data have provided another means to examine how the Cerrado has been transformed over the past 30 years. Two high-profile publications use municipal level (county-level) agricultural production data to estimate the pattern and extent of land uses central to Cerrado landcover change biodiversity loss (Alho and Souza Martins 1995; WWF-Brasil 2000). In both cases, land- cover change is equated with annual cropping area and artificial pasture. Thus, agricultural production is a proxy measure for land- cover change. That is, for every hectare of production, 1 ha of Cerrado habitat is lost. The cumulative effect is interpreted as a loss in biodiversity.

Observed vegetation regrowth, however, undermines the use of agricultural census data for estimating land-cover change. First, agricultural census data do not offer any means to represent vegetation regeneration. Second, increases in annual cropping or pasture are represented as permanently replacing original vegetation even though they may not necessarily cause Cerrado loss (Angelsen and Kaimowitz 2001, 3). There are other drawbacks to using agricultural census data to estimate land-cover change. Besides possible problems in obtaining accurate agricultural census data for these relatively remote regions, the proxy data only provide 'snapshot' views of the land cover. These cannot accurately represent the changes of land uses for a specific area smaller than a county. They cannot capture the dynamic that one field, for example, undergoes multiple changes over time and different uses - from fallow to pasture to annual crops. In conclusion, agricultural census data should not substitute measures of land-cover change with easily available satellite data.

Conclusion

Is the Cerrado disappearing? The Cerrado has undergone tremendous transformation over the last three decades, from the isolated savanna described in the expedition published in The Geographical journal to the global agricultural breadbasket of today. However, assessments that one of Brazil's largest ecoregions is losing space' misrepresent and oversimplify the land-cover change dynamics on the ground. Results using high-resolution digital satellite data of land cover over multiple time periods indicate significant loss of original vegetation as well as high levels of regeneration, suggesting that Cerrado vegetation may be more resilient to human impacts than the catastrophic estimations suggest. A shift in the focus of current studies of Cerrado change to one of the landscape and mesoscale is necessary to assess more reliably the complexity of human impacts. Transparent methods of assessment that focus on high- resolution data analysis, employ multiple-year change detection methods, and provide clear land-cover classifications are necessary to shift from aggregate perceived 'loss' to vegetation and landscape dynamics. This shift will allow for a conceptual remapping of human- environment relations in the Cerrado that includes anthropogenic landscapes and vegetation regrowth in the scenario of environmental conservation. Rather than focusing on the disappearance of the Cerrado, a more insightful question could be posed: why are areas of the cultivated Cerrado landscapes preserved or allowed to regenerate? The answer to this question opens analytical space to explore complex changes in the landscape, particularly, the importance of secondary growth. A clear understanding of these processes could provide a better scientific basis that may focus future environmental policy initiatives.

Acknowledgements

I am grateful to the National Science Founda\tion (Geography and Regional Science Program, Dissertation Improvement Grant, SBR- 99485), the Organization of American States (PRA-1328) and the Latin American Center at the University of California, Los Angeles for funding the purchase of satellite data and supporting the in- country travel and fieldwork. I would like to thank Donald Sawyer at the Institute de Sociedade Populao e Natureza (ISPN, Institute of Society, Population and Nature) in Brasilia, Hugo de Souza Dias, Otvio de Souza Dias, Christian Brannstrom, and Thomas Gillespie.

Notes

1 Other remote sensing studies of the Cerrado evaluate landcover changes in parks (Almeida-Filho and Vitorello 1997) or characterize agricultural settlements (Passes 1998).

2 See Brannstrom (2002) for a critique of the Brazilian Geography and Statistics Institute (IBGE) vegetation map as an empirical baseline.

References

Ab'Saber A N 1971 A organizao natural das paisagens e subtropicais brasileiras in Ferri M ed III Simposio sobre o Cerrado Editera Universidade de So Paulo, So Paulo 1-14

Alho C J R 1981 Small mammal populations of the Brazilian Cerrado: the dependence of abundance and diversity on habitat complexity Revista Brasileira de Biologia 41 223-30

Alho C J R and Souza Martins E de 1995 De Gro em Gro, o Cerrado Perde Espace/ (Cerrado - Impactos do Processo de Ocupao) WWF- Brasil, Brasilia, DF

Almeida-Filho R and Vitorello I 1997 Remote sensing and field data in the definition of hydrothermically altered areas in vegetated terrains, central Brazil International journal of Remote Sensing 18 1835-42

Angelsen A and Kaimowitz D 2001 Agricultural technologies and tropical deforestation Cabi Publications in association with Center for International Forestry Research, New York

Askew G P, Moffal D J, Montgomery R F and Searl P L 1970 Soil landscapes in northeastern Mato Grosso Geographical Journal 136 210- 27

Bassett T and Zueli K 2000 Environmental discourses and the Ivorian savanna Annals of the Association of American Geographers 90 67-95

Bates J M 2002 The genetic effects of forest fragmentation on five species of Amazonian birds journal of Avian Biology 33 276-95

Bilbao B, Braithwaite R, Dall'Aglio C, Moreira A, Oliveira P E, Ribeiro J F and Stott P 1996 Biodiversity, tire, and herbivory in tropical savannas in Solbrig M ed Biodiversity and savanna ecosystem processes Springer, Berlin 197-203

Branford S and Freris N 2000 One great big hill of beans The Ecologist 30 46-7

Brannstrom C 2001 Conservation-with-development models in Brazil's agro-pastoral landscapes World Development 29 1345-59

Brannstrom C 2002 Rethinking the 'Atlantic Forest' of Brazil: new evidence for land cover and land value in Western So Paulo, 1900- 1930 Journal of Historical Geography 28 42039

Brown E H, Askew G P, Thornes J B, Young A, Townshend J R G and Daultery S G 1970 Geographical research on the Royal Society/Royal Geographical Society's expedition to Northeastern Mato Grosso, Brazil: a symposium Geographical Joumal 136 36509

Cavalcanti R B and Joly C A 2002 Biodiversity and conservation priorities in the Cerrado region in Oliveira P S and Marquis R J eds The cerrados of Brazil Columbia University Press, New York 351-67

Colchester M 1997 Salvaging nature: indigenous people and protected areas in Ghimire K B and Pimbert M P eds Social change and conservation: environmental politics and impacts of national parks and protected areas Earthscan Publications, London 97-130

Congalton R G 1991 A review of assessing the accuracy of classification of remotely sensed data Remote Sensing of the Environment 37 35-47

Conservation International (Cl) 2003 Cerrado-biodiveristy hotpots (vital signs) (http://www.biodiversityhotspots.org/xp/hotspots/ cerrado/?showpage=ConservationAction/) Accessed 1 July

Coutinho L 1977 Aspetos ecologicos do fogo no cerrado. II. As queimadas e a disperso de sementes em algumas espcies anemorcoricas do estrato herbaceo-subarbustivo Boletim Botnica 5 57-64

Coutinho L 1982 Ecological effects of fire in Brazilian Cerrado in Huntley B ) and Walker B H eds Ecology of tropical savannas Springer, Berlin 273-91

Dale V H, Pearson S M, Offerman H L and O'Neill R V 1994 Relating patterns of land-use change to faunal biodiversity in the central Amazon Conservation Biology 8 1027-36

Dauber J, Hirsch M, Simmering D, Waldhardt R, Otte A and Wolters V forthcoming Landscape structure as an indicator of biodiversity: matrix effects on species richness Agriculture, Ecosystems and Environment

Durigan G, Confier! W A, Franco GADC and Garrido M A O 1998 Induo do processo de regenerao da vegetao de cerrado em area de pastagem, Assis, SP Acta Botnica Brasileira 12 (Suppl) 421-9

Durigan G, Franco, GADC, Pasture J A and Aguiar O T 1997 Regenerao natural da vegetao de cerrado sob floresta de Eucalyptus citriodora Revista do lnstituto Florestal (SSo Paulo) 9 71-85

EMBRAPA-Cerrados 1998 EMBRAPA-Cerrados e a Regio dos Cerrados: Informaes Bas/cas e Dados Estatfsticos EMBRAPA-Cerrados, Planaltina, DF

Eva H and Lambin E F 2000 Fires and land-cover change in the tropics: a remote sensing analysis at the landscape scale Journal of Biogeography 27 765-76

Fairhead J and Leach M 1996 Misreading the African landscape: society and ecology in a forest-savanna mosaic Cambridge University Press, Cambridge

Fairhead J and Leach M 1998 Reframing deforestation: global analysis and local realities: Studies in West Africa Routledge, London

Fearnside P M 1996 Amazonian deforestation and global warming: carbon stock in vegetation replacing Brazil's Amazon Forest Forest Ecology and Management 80 21-34

Fearnside P M 1997 Greenhouse gases from deforestation in Brazilian Amazonia: net committed emissions Climatic Change 35 321- 60

Fearnside P M 2001 Soybean cultivation as a threat to the environment in Brazil Environmental Conservation 28 23-38

Ferek R J, Reid J S, Hobbs P V, Blake D R and Liousse C 1998 Emission factors of hydrocarbons, halocarbons, trace gases and particles from biomass burning in Brazil Journal of Geophysical Research-Atmospheres 103 32107-18

Foody G 2002 Status of land cover classification accuracy assessment Remote Sensing of Environment 80 185-201

Forman R 1995 Land mosaics: the ecology of landscapes and regions Cambridge University Press, Cambridge

FUNATURA 2003 Projecto de establecimento de rservas particulares do Patromnio natural (RPPNs) no Cerrado Brasileiro PNUD BRA/00/G35 (http://www.funatura.org.br) Accessed 1 July

Furley P A 1994 Savanna formations: ecology and environment Progress in Physical Geography 18 276-94

Furley P A 1996 The influence of slope on the nature and distribution of soils and plan communities in the central Brazilian Cerrado in Anderson M G and Brooks S eds Advances in hillslope processes Wiley, Chichester 327-46

Furley P A 1999 The nature and diversity of neotropical savanna vegetation with particular attention to the Brazilian Cerrados Global Ecology and Biogeography 8 223-41

Global Environment Facility (GEF) 2003 Brazil-establishment of private national heritage reserves in the Brazilian Cerrado (http:// www.gefonline.org/projectDetails.cfm?projlD=868) Accessed 1 )uly

Goedert W J 1983 Management of the Cerrado soils of Brazil: a review journal of Soil Science 34 405-28

Goodland R and Pollard R 1973 The Brazilian Cerrado vegetation: a fertility gradient journal of Ecology 61 219-24

Government of Brasil and FUNATURA 1999 Aes Prioritrias para a Conservao da Biodiversidade do Cerrado e Pantanal (http:// www.bdt.fat.org.br/workshop/cerrado/br/) Accessed 1 July 2003

Hall A 1997 Sustaining Amazonia: grassroots action for productive conservation Manchester University Press, Manchester

Henderson-Sellers A and McGuffie K 1995 Global climate models and 'dynamic' vegetation changes Global Change Biology 1 63-75

Jensen J 1996 Introductory digital image processing Prentice Hall, Upper Saddle River, NJ

Jepson W 2003 A political ecology of land-use and land-cover change in the central Brazilian savannas of eastern Mato Grosso since 1970 Unpublished doctoral dissertation Department of Geography, University of California Los Angeles

Joly C A, Aidar M P, Klink C A, McGrath D G, Moreira A G, Moutinho P, Nepstad D C, Oliveira A A, Pott A, Rodal M J N and Sampaio EVSB 1999 Evolution of the Brazilian phytogeography classification systems: implications for biodiversity conservation Cincia e Cultura 51 331-48

Kaufman J B, Cummings D L and Ward D E 1994 Relationship of fire, biomass and nutrient dynamics along a vegetation gradient in the Brazilian Cerrado journal of Ecology 82 519-31

Kirchhoff V and Alvala P C 1996 Overview of an aircraft expedition into the Brazilian Cerrado for the observation of atmospheric trace gases Journal of Geophysical ResearchAtmospheres 101 23973-81

Kisselle K W, Zepp R C, Burke R A, Pinto A D, Bustamante M M C, Opsahl S, Varella R F and Viana L T 2002 Seasonal soil fluxes of carbon monoxide in burned and unburned Brazilian savannas tournai of Geophysical Research-Atmospheres 107 18-29

Klink C A and Moreira A G 2002 Past and current human occupation and land use in Oliveira P S and Marquis R J eds The Cerrados of Brazil: ecology and natural history of a neotropical savanna Columbia University Press, New York 6988

Klink C A, Moreira A G and Solbrig O T 1993 Ecological impact of agricultural development in the Brazilian Cerrado in Young M D and Solbrig O T eds The world's savannas. Economic driving forces, ecological constraints and policy options for sustainable !and use Parthenon Publishing, London 259-83

Lake A 1999 Reserving judgment: a critical examination of RPPN policy in the Cerrados of Brazil Unpublished master's thesis Department of Geography, University of Edinburgh

Lardy L C, Brossard M, Lopes Assad M L and Laurent J-Y 2002 Carbon and phosphorus stocks of clayey Ferralsols in Cerrado native and agroecosystems, Brazil Agriculture, Ecosystems and Environment 92 147-58 Lucas R M, Honzak M, Amaral I \D, Curran P J and Foody G M 2002 Forest regeneration on abandoned clearances in central Amazonia International Journal of Remote Sensing 25 96588

Mantovani J E and Pereira A 1998 Estimativa da integradidade da cobertura vegetal do cerrado/pantanal atravs de dados TM/Landsat Simposio Brasiliero de Sensoriamento Remoto, 9 Institute) Nacional de Pesquisas Espaciais, So Jose dos Campos

Mares M A, Ernest K A and Gettinger D D 1986 Small mammal community structure and composition in the Cerrado province of central Brazil Journal of Tropical Ecology 2 289399

Marini M A 2001 Effects of forest fragmentation on birds of the Cerrado region, Brazil Bird Conservation international 111325

McCracken S, Brondizio E, Nelson D, Moran E, Siqueira A and Rodriguez-Pedraza C 1999 Remote sensing and CIS at farm property level: demography and deforestation in the Brazilian Amazon Photogrammetric Engineering and Remote Sensing 65 1311-20

Mertens B and Lambin E F 2000 Land-cover change trajectories in southern Camaroon Annals of the Association of American Geographers 90 467-94

Miranda H S and Miranda A C 2000 Queimadas e estoques de carbono no Cerrado in Moreira A G and Schwartzman S eds As Mudanas Climaticas Globais e os Ecossistemas Brasileiros IPAM, Brasilia, DF 75-88

Mistry J 1998 Fire in the Cerrado (savannas) of Brazil: an ecological review Progress in Physical Geography 22 42548

Mistry J 2000 Savannas Progress in Physical Geography 24 273-9

Mittermeier R A, Myers N and Mittermeier C 2000 hotspots: Earth's biologically richest and most endangered ecoregions CEMEX, Mexico City

Mittermeier R A, Myers N, Thomsen J B, Fonseca GAB and Olivieri S 1998 Biodiversity hotspots and major tropical wilderness areas: approaches to setting conservation priorities Conservation Biology 12 516-20

Moran E F and Brondizio E 1998 Land-use change after deforestation in Amazonia in Liverman D, Moran E f, Rindfuss R R and Stern P C eds People and pixels: linking remote sensing and social science National Academy Press, Washington DC 94-121

Moran E F, Brondizio E, Mausel P and Wu Y 1994 Integrating Amazonia vegetation, land-use, and satellite data: attention to differential patterns and rates of secondary succession can inform future policies Bioscience 44 329-38

Moran E F, Packer A, Brondizio E and Tucker J 1996 Restoration of vegetation cover in the eastern Amazon Ecological Economics 18 41- 54

Myers N, Mittermeier R A, Mittermeier C, Fonseca GAB and Kentand J 2000 Biodiversity hotspots for conservation priorities Nature 403 853-8

Oliveira-Filho A T and Ratter J 2002 Vegetation physiognomies and woody flora of the Cerrado biome in Oliveira P S and Marquis R J eds The Cerrados of Brazil: ecology and natural history of a neotropical savanna Columbia University Press, New York 91-120

Oliveira-Filho A T, Shepherd G ), Martins F R and Stubblebine W H 1989 Environmental factors affecting physiognomic and floristic variation in an area of Cerrado in central Brazil Journal of Tropical Ecology 5 413-31

Passos M M 1998 Amazonia: teledeteo e colonisao Editora UNESP, So Paulo

Perz S G and Walker R T 2002 Household life cycles and secondary forest cover among small farm colonists in the Amazon World Development 30 1009-27

Pinto A D, Bustamante MMC, Kisselle K, Burke R, Zepp R, Viana L T, Varella R F and Molina M 2002 Soil emissions of N2O, NO, and CO, in Brazilian savannas: effects of vegetation type, seasonality, and prescribed fires Journal of Geophysical Research-Atmospheres 107 57- 65

Pivello V R, Shida C N and Meirelles S T 1999 Alien grasses in Brazilian savanna: a threat to the biodiversity Biodiversity and Conservation 8 1281-94

Ratter J A, Ribeiro J F and Bridgewater S 1997 The Brazilian Cerrado vegetation and threats to biodiversity Annals of Botany 80 223-30

Redford K H and Fonseca GAB 1986 The role of gallery forests in the zoogeography of the Cerrado's non-volant mammalian fauna Biotropic 18 126-35

Ribeiro J F and Walter BMT 1998 Fitofisionomies do bioma Cerrado in Sano S M and Almeida S P eds Cerrado: Ambiente e Flora EMBRAPA, Planaltina, DF 89-166

Robbins P 2001 Fixed categories in a portable landscape: the causes and consequences of land-cover categorization Environment and Planning A 33 161-79

Silberbauer Goltsberger I and Gottsberger G 1984 Cerrado cerrado: a comparison with respect to number of species and growth forms Phylocoenologia 12 293-303

Suva J M C and Bates j M 2002 Biogeographic patterns and conservation in the South American Cerrado: a tropical savanna hotspot BioScience 52 225-34

Southworth J and Tucker C 2001 The influence of accessibility, local institutions, and socioeconomic factors on forest cover change in the mountains of western Honduras Mountain Research and Development 21 276-83

Steininger M K 1996 Tropical secondary forest regrowth in the Amazon: age, area and change estimation with Thematic Mapper data International Journal of Remote Sensing 17 927

Steininger M K 2000 Satellite estimation of tropical secondary forest above-ground biomass: data from Brazil and Bolivia International Journal of Remote Sensing 211139-57

Tubelis D P and Cavelcanti R B 2000 A comparison of bird communities in natural and disturbed non-wetland open habitats in the Cerrado's central region, Brazil Bird Conservation International 10 331-50

Vourlitis G L, Priante N, Hayashi MMS, Nogueira J D, caseiro F T and Campelo J H 2001 Seasonal variations in the net ecosystem CO2 exchange of a mature Amazonian transitional tropical forest (cerradao) Functional Ecology 15 388-95

Warnken P F 1999 The development and growth of the soybean industry in Brazil Iowa State University Press, Ames, IA

Wood A, Stedman-Edwards P and Mang J 2000 The root causes of biodiversity loss Earthscan, Sterling, VA

Woodwell G M 2002 On purpose in science, conservation and government Ambio 31 432-6

WWF-Brasil 2000 Expansla e Perda da Biodiversidade no Cerrado WWF- Brasil, Brasilia, DF

Zarin D J, Ducey M J, Tucker Jand Salas W A 2001 Potential biomass accumulation in Amazonian regrowth forests Ecosystems 4 658- 68

WENDY JEPSON

Department of Geography, 803D Eller O&M Building, Texas A&M University, College Station, TX 77843-3147, USA

E-mail: wjepson@geog.tamu.edu

This paper was accepted for publication in February 2004

Copyright Royal Geographical Society Jun 2005

Source: Geographical Journal, The

More News in this Category