Welfare Outcomes and the Advance of the Deforestation Frontier in the Brazilian Amazon

  • Published on

  • View

  • Download


  • nce of the Deforestation Frontier

    lian Amazon




    descre an, ewheforestr inma

    Key words Brazil, Amazon, tropical deforestation, welfare indicators, frontier development

    Brazical fordevelopmentalthe AmusespSantos,

    Sales, 2007; Perz et al., 2008; Verssimo, Cochrane, Souza, &

    (INPE, 2010), making forest loss the largest source of carbondioxide emissions in Brazil (MCT, 2004). 1

    ion isGoalsoverty1990

    il as anly atssimo,Souzarigues

    time frame than typically considered in the literature on the

    Valle, Daniel Santos, Carlos Souza Jr., Marco Lentini, and Anthony A-

    nderson for their collaboration and input to the analyses and concepts

    during the preparation of this paper. Comments from four anonymous

    World Development Vol. 40, No. 4, pp. 850864, 2012 2011 Elsevier Ltd. All rights reserved

    0305-750X/$ - see front matter

    rlddev.2011.09.002Despite the timber resources and new agricultural land thatSalomao, 2002). The government also supports frontier expan-sion by investing in infrastructure, facilitating credit for agri-culture, and recognizing and supporting new settlements ofsmall farmers in forested areas (Barreto, Pinto, Brito, &Hayashi, 2008; Fearnside, 2005; Pacheco, 2009). Since 1990,deforestation has averaged 17.5 thousand km2 per year

    *This research was supported by the Gordon and Betty Moore Founda-

    tion, the International Programs Oce of the USDA Forest Service, and

    the Fundacion Avina. We are grateful to Eugenio Arima, Paulo Barreto,

    Subhrendu Pattanayak, Guillaume Rousseau, Rodney Salomao, Denisestation frontier was galvanized in the 1960s and 1970s by gov-ernment road construction, colonization projects, andagricultural subsidies (Mahar, 1989). Today, deforestation islargely driven by private investors seeking to maximize prots,especially by supplying global and national markets fortimber, soybeans, and beef (Arima, Barreto, & Brito, 2005;Barona, Ramankutty, Hyman, & Coomes, 2010; DeFries,Rudel, Uriarte, & Hansen, 2010; Lentini, Pereira, Celentano,& Pereira, 2005; Nepstad, Stickler, & Almeida, 2006). The log-ging industry plays a critical role in opening new areas bybuilding unocial new roads (Brandao, Souza, Ribeiro, &

    et al. (2009) suggest that the current frontier development par-adigm in the Amazon perpetuates a boom-bust economic pat-tern, with prots from logging rst generating income andjobs, followed by a severe collapse as forest resources and soilfertility are depleted.This begs the question of why regional and local govern-

    ment leaders would perpetuate such a model of development.One explanation is that the short-term appeal of the boomoutweighs the long-term costs of the bust. The opposite expla-nation is also possible: development of the frontier may even-tually lead to improved socio-economic conditions, in a longerhave bpoverty1. INTRODUCTION

    il has the worlds largest remaining area of dense trop-est, representing both a classic frontier for economicment and a vitally important source of global environ-services. More than 760 thousand km2, almost 19% ofazon forest in Brazil, has been converted to other landrimarily farming and ranching (INPE, 2010; Pereira,Vedoveto, Guimaraes, & Verssimo, 2010). The defor-

    vere in the Amazon than in most of Brazil, and the regfar from achieving the Millennium Development(Celentano & Verssimo, 2007a). For example, the prate in the Amazon remained constant at 45% betweenand 2005, even while it fell from 42% to 31% in Brazwhole. Extreme poverty declined in the Amazon, but ohalf the rate as in the rest of Brazil (Celentano & Ver2007a). Schneider, Arima, Verssimo, Barreto, and(2002), Celentano and Verssimo (2007b) and RodWelfare Outcomes and the Adva

    in the Brazi

    DANIELLEIMAZON (Amazon Institute of Peop

    ERINNorth Carolina State Unive

    IMAZON, Bel


    Summary. Frontier expansion in the Brazilian Amazon is oftenacterization by mapping and estimating statistical models of welfaling for potential confounding variables and spatial autocorrelatioby exploitation of natural resources, followed by a bust duringwelfare. However, average per capita welfare increases again with dwelfare, along with the strong bivariate correlation between defoequate deforestation with development. This conrms the need fobiodiversity and carbon sequestration, that are provided by the A 2011 Elsevier Ltd. All rights reserved.

    www.elsevier.com/locate/worlddevdoi:10.1016/j.woeen tapped through this frontier development process,and other socio-economic problems remain more se-

    850LENTANOand the Environment), Belem, Brazil

    ILLSty (NCSU), Raleigh, USA PA, Brazil


    PA, Brazil

    ribed as boom-bust development. We critically assess this char-s a function of deforestation at the municipal level. After control-stimation results are consistent with a frontier boom generatedich forests continue to fall but there is no compensating gain inrestation at very high levels. This second turning point in averageation and municipal GDP/km2, may encourage local leaders toternational incentive payments for global public goods, such aszon forest.reviewers also helped us signicantly improve the manuscript. Final rev-ision accepted: August 12, 2011.

  • tial autocorrelation through a spatial autoregressive model

    Amazon or the Amazonian biome (Figure 1), includes 3.7million km2 in 408 municipalities (74% of the legally dened

    WELFARE OUTCOMES AND THE ADVANCE OF THE DEFORESTATION FRONTIER 851boom-bust scenario. Yet a third explanation is that local lead-ers judge welfare dierently, for example, focusing on munici-pal GDP (which translates directly into tax revenues), ratherthan income per capita or the poverty rate.Another possibility is that the boom-bust pattern is an al-

    most inevitable feature of frontier development (Barbier,2005). This is consistent with the concept of the resourcecurse (Bulte, Damania, & Deacon, 2005; Ross, 2001). Coun-tries rich in natural resourcesespecially mineral resourcesare said to suer from this curse due to the volatility of pricesand hence income from natural resources, decline in the com-petitiveness of other economic activities, corruption, and weakinstitutions engendered by dependence on natural resources,and reduced incentives to invest in human capital (Gylfason,2001). When the resources run out, countries are not able totransition smoothly to other economic activities, and thus aneconomic bust follows the natural resource boom.In the case of renewable resources such as the Amazon forest,

    economic exhaustion is not inevitable, because the resource canbe managed (Boltz, Carter, Holmes, & Pereira, 2001; Pearce,Putz, & Vanclay, 2003; Rice, Sugal, Ratay, & Fonseca, 1998;Uhl et al., 1997; Verssimo, Barreto, Mattos, Tarifa, & Uhl,1992; Vincent, 1992). However, there has been limited uptakeof best management practices for harvesting timberlet alonelong-term sustainable forest managementdue to a variety offactors including lack of credit, uncertain land tenure, and com-petition from illegal logging (Applegate, Putz, & Snook, 2004;Bacha, 2003; Barbier, 1995; Putz, Dykstra, & Heinrich, 2000;Verssimo et al. , 2002). In practice, the Amazon frontier hasbeen characterized by highly protable mining of timber re-sources (Barreto, Amaral, Vidal, & Uhl, 1998; Holmes et al.,2002), followed by conversion of forest into crops and pasture.This still leaves open the possibility of sustained long-term eco-nomic growth and improved standards of living based on theagricultural 2, industrial, and service sectors. On the other hand,poor governance, corruption (as recent working papers by Brol-lo, Nannicini, Perotti, and Tabellini (2010) and Caselli andMichaels (2009) have found to be associated with windfall in-come in Brazilian municipalities), and failure to invest timberprots locally and productively also leave open the possibilitythat the frontier boom will be followed by a post-deforestationbust.In this paper, we rst describe the history and status of for-

    est resources and the deforestation frontier in the BrazilianAmazon. To assess evidence of a boom-bust pattern, we cate-gorize and map Amazonian municipalities into three defores-tation frontier zones based on their history of deforestation(cumulative deforestation through 2000), and we verify thatthese zones reect stages in the frontier development processby comparing their current deforestation rates and demo-graphics. We also categorize and map municipalities in threewelfare zones, dened separately for GDP per square kilome-ter, the human development index (HDI), income per capita,

    and the poverty rate. 3 This allows us to test for evidence ofboom-bust patterns through cross-tabulation of welfare anddeforestation categories. Next, we estimate and graph regres-sion models relating the welfare measures to historical cumu-lative deforestation, again testing for a boom-bust pattern(i.e., an inverted-U quadratic relationship) versus steadilyincreasing welfare with deforestation (i.e., a linear relation-ship) or a boom-bust and recovery (i.e., an N-shaped cubicrelationship, with two turning points). We compare the cubic,quadratic, and linear specications based on both overallmodel t (as represented by adjusted R2) and statistical signif-

    icance of the higher order terms. 4Brazilian Amazon). These municipalities vary in size from103 km2 to 160 thousand km2, with a mean of 9.3 thou-sand km2 (median = 3.7 thousand km2, SD = 17.1 thou-sand km2) (IBGE, 2000a). In our maps, the existence ofvery large municipalities can give the appearance of homog-enous areas of welfare that in fact simply reect municipalboundaries. However, the municipality is the smallest unitat which data on welfare (in 2000) and its covariates (in1991) are available for the entire region.

    (b) Data

    The variables used in this paper as well as denitions,sources, and descriptive statistics are presented in Table 1.Municipalities as dened in 2000 are our mapping unit as wellas the unit of statistical analysis. The regression analyses use399 observations, due to lack of data for four municipalitiesand the exclusion of four state capitals: Manaus (1.4 millioninhabitants in 2000), Belem in including Ananindeua (1.7 mil-lion), Porto Velho (335 thousand), and Rio Branco (253 thou-sand). 5 Between the years 1991 and 2000, 143 newmunicipalities were created in the study area by municipaldivision. Thus, for 1991, we use data from UNDP (2003),which recalculated the 1991 census data in terms of the 2000municipal boundaries using census sectors (the smallest unit(incorporating spatial lag and error), and again compare mod-el specications in linear, quadratic, and cubic forms.Last, we explore the possibility that the welfare measures re-

    ect characteristics of a mobile frontier population, ratherthan just conditions in municipalities at dierent stages offrontier development. We, therefore, provide a nal set ofmaps and statistics on migration patterns across the region,and discuss whether welfare is being imported (and exported)as the frontier develops.

    2. METHODS

    (a) Study area

    We consider all municipalities in the Brazilian Amazonwhose original vegetation cover was at least 50% Amazonrainforest, according to IBGEs maps of vegetation (IBGE,1997) and municipal boundaries (IBGE, 2001). Our studyarea, which has a dierent shape than either the legalWhile the maps and statistical models provide useful sum-maries of the observed patterns in current welfare and histor-ical deforestation, they cannot be taken as evidence ofcausality because they do not control for potential confound-ing factors that may be driving both deforestation and welfare.Thus, next, we add controls for biophysical, demographic, andaccess factors that have been identied as key constraints anddrivers of Amazonian development and deforestation(Chomitz & Thomas, 2001; Chomitz & Thomas, 2003;Laurance et al., 2001, 2002; Nepstad et al., 2001; Pfa, 1999;Schneider et al. , 2002; Walker, 2004). We only include pre-determined variables that represent conditions prior to 2000(e.g., population in 1991), are time invariant (e.g., agriculturalpotential), or that change relatively slowly over time (e.g.,mining centers recorded in 2002). We further control for spa-for which census data are geographically identied).

  • nall

    VE(c) Analyses

    (i) Spatial analysesWe rst map deforestation and welfare across Amazonian

    municipalities. Adapting the methodology of Celentano andVerssimo (2007b), we sort municipalities into three categoriesbased on cumulative deforestation and each of the welfare cat-

    Figure 1. Study area in the Brazilian Amazon (municipalities origi

    852 WORLD DEegories. We then test whether the apparent frontier boom seenin the maps is statistically signicant, checking if welfare levelsare highest among municipalities in the active frontier, and ifdeforestation is most rapid among municipalities in the highestwelfare zones.To calculate cumulative deforestation through 2000, we use

    ArcGIS 9.0 to process land cover data generated from LAND-SAT images by Inpe (the Brazilian Space Agency). We correctan error in Inpes land cover map, which makes some munic-ipalities obscured by cloud cover in 2000 appear forested eventhough they were among the rst colonized in the Amazon andare well known to have been deforested before 2000. 6 Thus,we include some information from 2001 and 2002 images clas-sied by Inpe in our maps. We exclude protected areas createdprior to 2000 (ISA, 2005) from the calculation of percent ofthe municipality deforested, since deforestation is legally pro-hibited in these areas and, therefore, tends to occur at reducedrates as compared to nonprotected areas (Adeney, Christen-sen, & Pimm, 2009; Soares-Filho et al., 2010).To categorize municipalities according to their stage of

    deforestation, we apply the natural breaks classication meth-od of statistical mapping, using Jenks optimization algorithm(Jenks, 1967 as cited in Mitchel, 1999). This algorithm calcu-lates groupings of data values (e.g., cumulative deforestationin each municipality) to minimize total error, equal to thesum of absolute deviations around each of the class medians.We partition municipalities into three categories, as suggestedby previous cluster analysis of Amazonian municipalities(Celentano & Verssimo, 2007b) and as the minimum neces-sary to explore whether there is a boom-bust pattern. To verifythat this method sorts municipalities according to their stagein frontier development, we compare annual deforestationand demographic characteristics across categories. Finally,we apply ANOVA and the Tukey test to welfare measures (de-scribed next) across the three deforestation frontier zonesidentied.To assess welfare, we use three common indicators of the

    populations welfare: the Human Development Index (HDI),

    y more than 50% forested). Sources of data: IBGE (1997, 2001).

    LOPMENTincome per capita, and the poverty rate (percent of populationin extreme poverty). We also consider the total income in themunicipality normalized by area: GDP...


View more >