Seediscussions,stats,andauthorprofilesforthispublicationat:http://www.researchgate.net/publication/282648241

Effectsofroadinfrastructureonforestvalueacrossatri-nationalAmazonianfrontier

ARTICLEinBIOLOGICALCONSERVATION·JANUARY2015

ImpactFactor:3.76·DOI:10.1016/j.biocon.2015.08.024

READ

1

25AUTHORS,INCLUDING:

WendesonCastro

UniversidadeFederaldoAcre

9PUBLICATIONS2CITATIONS

SEEPROFILE

AlexanderFrederickShenkin

UniversityofOxford

14PUBLICATIONS13CITATIONS

SEEPROFILE

MarcosSilveira

UniversidadeFederaldoAcre

65PUBLICATIONS1,275CITATIONS

SEEPROFILE

JaneSouthworth

UniversityofFlorida

84PUBLICATIONS1,731CITATIONS

SEEPROFILE

Availablefrom:CaraARockwell

Retrievedon:09October2015

Biological Conservation 191 (2015) 674–681

Contents lists available at ScienceDirect

Biological Conservation

j ourna l homepage: www.e lsev ie r .com/ locate /b ioc

Effects of road infrastructure on forest value across a tri-nationalAmazonian frontier

Christopher Baraloto a,b,⁎, Paula Alverga c, Sufer Baéz Quispe d, Grenville Barnes g, Nino Bejar Chura d,Izaias Brasil da Silva c, Wendeson Castro c, Harrison da Souza e, Iracema Elisabeth de Souza Moll c,Jim Del Alcazar Chilo d, Hugo Dueñas Linares d, Jorge Gárate Quispe d, Dean Kenji e, Matthew Marsik j,Herison Medeiros c, Skya Murphy b, Cara Rockwell f,m, Galia Selaya l, Alexander Shenkin g,h,Marcos Silveira c, Jane Southworth i,j, Guido H. Vasquez Colomo e, Stephen Perz k

a International Center for Tropical Botany, Department of Biological Sciences, Florida International University, Miami, FL 33199, USAb INRA, UMR “Ecologie des Forêts de Guyane”, 97387 Kourou Cedex, French Guianac Programa de Pós-Graduação Ecologia e Manejo de Recursos Naturais, Universidade Federal do Acre, Rio Branco AC 69910-900, Brazild Herbario Alwyn Gentry, Universidad Nacional Amazónica de Madre de Dios, Puerto Maldonado, Madre de Dios, Perue Universidad Amazónica de Pando, Cobija, Pando, Boliviaf School of Forest Resources and Conservation, University of Florida, Gainesville, FL 32611, USAg School of Natural Resources and Conservation, University of Florida, Gainesville, FL 32611, USAh Environmental Change Institute, School of Geography and the Environment, University of Oxford, Oxford OX1 3QY, UKi Department of Geography, University of Florida, Gainesville, FL 32611, USAj Land Use and Environmental Change Institute, University of Florida, Gainesville, FL 32611, USAk Department of Sociology and Criminology & Law, University of Florida, Gainesville, FL 32611-7330, USAl Department of Agricultural and Biological Engineering, University of Florida, Gainesville, 32611-0570m International Center for Tropical Botany, Department of Earth and Environment, Florida International University, Miami, FL 33199, USA

⁎ Corresponding author at: International Center for TBiological Sciences, Florida International University, Miam

E-mail address: cbaraloto@fiu.edu (C. Baraloto).

http://dx.doi.org/10.1016/j.biocon.2015.08.0240006-3207/© 2015 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 24 September 2014Received in revised form 4 August 2015Accepted 10 August 2015Available online xxxx

Keywords:REDDBiodiversityCarbon stocksTropical rainforestLivelihoodConnectivityRoad impactNTFPTimberBrazil nutRubberAçaiInfrastructure

Road construction demonstrably accelerates deforestation rates in tropical forests, but its consequences for forestdegradation remain less clear.Weestimated a series of forest valuemetrics including components of biodiversity,carbon stocks, and timber and non-timber forest product resources, along the recently paved Inter-OceanicHigh-way (IOH) integrating Brazil and Peru along the Bolivian border. We installed 69 vegetation plots in intactterra firme forests representative of local community holdings near and far from the IOH, and we characterized15 components of forest value for each plot.We observed strong geographic gradients in forest value components across the region, with increases fromwestto east in aboveground biomass and in the abundance of timber and non-timber forest product trees and regen-eration. Plots in communities in Pando, Bolivia, where the IOH remains in part unpaved, had the highest above-ground biomass, standing timber volumes and Brazil nut tree density. In contrast, communities inMadre deDios,Peru, where settlements and unpaved portions of the IOH have existed for decades, and in Acre, Brazil, wherepaving of the IOH has been underway for more than a decade, were more degraded. Seven of the fifteen forestvalue components we measured increased with increasing distance from the IOH, although the magnitude ofthese effectswasweak. Landscape scale remote sensing analyses showedmuch stronger effects of roadproximityon deforestation. We suggest that remote sensing techniques including canopy spectral signatures might becalibrated to characterize multiple components of forest value, so that we can estimate landscape scale impactsof infrastructure developments on both deforestation and forest degradation in tropical regions.

© 2015 Elsevier B.V. All rights reserved.

1. Introduction

The construction and paving of roads represent key factors respon-sible for tropical deforestation (Fearnside, 2005; Perz et al., 2007;

ropical Botany, Department ofi, FL 33199, USA.

Laurance et al., 2009; Ahmed et al., 2013). Beyond their physical foot-print via conversion of forest to agricultural lands, pasture and settle-ments, roads also precipitate the degradation of remnant forests, inat least two ways. First, roads increase forest fragmentation (Formanet al., 2003; Fearnside, 2005), thereby creating edges with altered spe-cies composition due to novel environmental constraints and/or dis-persal limitation (Benitez-Lopez et al., 2010). Second, roads provideaccess to areas that were previously isolated, which can lead to

675C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

overexploitation of natural resources (Forman et al., 2003; Walker et al.,2013). Critically, these two types of effects can be synergistic, with edgeeffects beingmore pronounced in areaswhere the surrounding forestma-trix has been modified by effects associated with roads such as fire or re-source over-exploitation (Nepstad et al., 2001; Laurance et al., 2007). As aresult, it remains imperative to assess the extent towhich forest degrada-tion due to roads can be mitigated by social–ecological factors such asgovernance by local communities (Perz et al., 2008; Pfaff et al., 2013) ornational legislation (Barber et al., 2014; Laurance et al., 2014).

Although the distinction between deforestation and forest degrada-tion has been recognized for some time, there remains debate regardingdefinitions of forest degradation (Sasaki and Putz, 2009; Asner et al.,2010). The most oft-employed approaches are based on forest coverestimates derived from remote sensing techniques (e.g., Hosonumaet al., 2012). However, forest cover may not be a valid proxy of forestvalue, or the tangible (e.g., timber) and intangible (e.g., water conserva-tion) benefits conferred by particular forested areas (Ninan and Inoue,2013; Baraloto et al., 2014). Forests with the same cover metric maydiffer substantially in structure and composition based on site factorssuch as climate and soil (Quesada et al., 2012), as well as the impactsof human disturbances such as logging and fire (Nepstad et al., 2001).Empirical field data therefore remain a necessary tool to provide preciseestimates of both tangible and intangible components of forest value(Putz and Redford, 2008). Typically, studies examining road ecologyhave focused on biodiversity of specific taxa (Cousins, 2006;Benitez-Lopez et al., 2010; Hoskin and Goosem, 2010), even thoughmitigation of road impacts and forest degradation will require an inte-grative assessment of forest value combining multiple ecological attri-butes and ecosystem services (Freudenberger et al., 2013).

Debates about indicators of forest degradation in turn raise ques-tions about whether roads affect different aspects of forests in thesame way. Very clear general patterns have emerged regarding the ef-fects of road proximity on forest cover, with for example 95% of defores-tation events in the Brazilian Amazon found within 5 km of roads or1 km of streams (Barber et al., 2014). However, contrasting predictionsmay emerge for the effects of road infrastructure on forest value. Onthe one hand, roads may accelerate forest degradation via increased re-source harvesting (Nepstad et al., 2001; Perz et al., 2007). On the otherhand, increased access to markets and alternative sources of livelihoodmay lessen the pressure on extant forest resources in some forest settle-ments (Perz et al., 2013a; Laurance et al., 2014).

Here we evaluate the impacts of a road infrastructure project onmul-tiplemetrics of forest value in forest communities of the tri-national fron-tier in the southwestern Amazon. The southwestern Amazon provides auseful study region inwhich to evaluate the impacts of road infrastructureon metrics of forest value. This region has significant forest diversityrepresenting important biogeographic gradients from the Andes moun-tains to the lowland Amazonian forests (Phillips et al., 2006), includingseveral economically important timber aswell as non-timber forest prod-uct (NTFP) species. The region also contains a major road corridor, theInter-Oceanic Highway (IOH), which was recently paved across the tri-national frontier of Bolivia, Brazil and Peru. Under the Initiative for theIntegration of Regional Infrastructure in South America (IIRSA), the lastunpaved segments of the IOH were paved in the southwestern Amazonfrom 2006 to 2010 to permit trans-boundary commerce between Braziland Peru along the Bolivian border. The objective was to integrate thesouthwestern Amazon and other portions of the interior of SouthAmerica into the global economy by providing export outlets via Atlanticports in southern Brazil as well as Pacific ports in Peru (Perz et al., 2013a).Paving of the IOH has resulted in marked reductions in travel times tomarkets and urban centers for some of the remote communities (Perzet al., 2013a) and has been linked with reductions in forest cover ofabout 10% in the region, most of which has occurred within 20 km ofthe IOH (Southworth et al., 2011).

Communities in the region practicemultiple use forestmanagement,which includes harvesting of both timber, particularly cumaru (Dipteryx

spp.), cumaru cetim (Apuleia leiocarpa (Vogel) J. F. Macbr.) and cedro(Cedrela spp.); and NTFPs including Brazil nuts (Bertholletia excelsaKunth.), rubber (Hevea brasiliensis (Willd. ex A. Juss.) Müll. Arg.), sapresins and oils (Copaifera spp., Myroxylon balsamum (L.) Harms;Hymenaea courbaril) and palm fruits for food (Euterpe precatoriaMart.), in addition to subsistence hunting and materials consumption(Duchelle et al., 2011; Baraloto et al., 2014). There are incipient programsin place for environmental service payments related to carbon in the re-gion, such as Acre's System of Incentives for Environmental Services,which was passed into state law in 2010 (Duchelle et al., 2013), alongwith multiple pilot REDD+ projects.

In this paperwe evaluate the impacts of road infrastructure on forestvalue derived from empiricalmeasures in forest vegetation plots. In par-ticular, we respond to the following questions:

1) How do forest value components vary across the region with in-creasing distance from the foothills of the Andes?

2) How does proximity to roads, particularly the paved IOH, and tourban hubs of colonization affect forest value components acrossthis gradient?

3) Are patterns of road effects consistent between the forest value esti-mates of this study and published deforestationmetrics in the region(Southworth et al., 2011)?

2. Material and methods

2.1. Study sites

Siteswere chosen to represent communities along the Inter-OceanicHighway in Acre, Pando and Madre de Dios, Peru (MDD), as well asalong the Cobija access road in Pando, Bolivia (Fig. 1). In each of thethree border countries, 12–15 communities were selected for study torepresent geographic variability and themajor land tenure classes iden-tified in the region, including agroextractive projects in Acre (Perz et al.,2013a). Broad variation in soils and climate in the region follows a gen-eral direction of the IOH along which we sampled, with the most fertilesoils in the southwest and decreasing as one travels northeast into Acre(Phillips et al., 2003; Quesada et al., 2012). The three countries also con-trast in (1) the paving status of the highway at the time of fieldwork,with Pando unpaved, MDD being paved, and Acre already paved; and2) governance effort, with Acre having high governance of forests andMDD and Pando with weaker forest governance.

In each community, we worked with community leaders to identifyforest sites that were representative of the area circumscribed by thecommunity. Our objective was to focus on the average human impactson the more common terra firme forests, so we avoided seasonally-inundated forests and areas that were not representative of the locallandholdings because of extreme human impacts (or lack thereof).In most communities we chose one plot to be near (b2 km), and oneplot to be farther (N5 km) from the IOH, with at least 5 km distance be-tween any two plots. Someof the communities and sites thatwere iden-tified for the study could not be included because inhabitants deniedentry to inventory forest resources on their lands. Overall, we sampled69 sites. A sensitivity analysis of statistical power indicated that thissampling was sufficient to detect hypothesized effects of the IOH onforest value components (Appendix S1).

2.2. Sampling methods

We sampled woody vegetation at each site between 2008 and 2010using a modification of the Phillips et al. (2006) modified Gentry plotmethod (Baraloto et al., 2011), which has been demonstrated to be ef-fective in measuring both aboveground biomass and floristic composi-tion in Neotropical forests (Baraloto et al., 2013). Each woody plantrooted within the transect area and with a diameter at breast height(DBH at 1.3 m) of ≥2.5 cm was included and measured for both DBH

Fig. 1. A map illustrating the study area in southwestern Amazonia, with the Inter-Oceanic Highway (in red) and other roads (in black and gray), the 69 sample plots (green circles),and major cities (black circles with names). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

676 C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

and height. Where a plant had multiple stems, we recorded it as oneindividual, but took separate diameter and height measures for eachstem to be used for calculations of aboveground biomass. We modifiedthe original Gentry plot protocol to provide more accurate measures offorest biomass and timber and non-timber resources (Baraloto et al.,2013). Each 2 × 50 m subplot was extended to a 10 m × 50 m subplotin which all woody stems with diameter at breast height of ≥20 cmwere sampled (Baraloto et al., 2011). Where necessary, diameterswere measured above buttresses and other stem irregularities.

We collected voucher specimens for each putative morphospeciesand whenever there was any uncertainty to identity. A full duplicateset of collections from each country is deposited in each of their threelocal herbaria (the National Amazonian University of Madre de Dios(UNAMAD) in Puerto Maldonado, Peru; the Center for Research onAmazon Protection of the Amazonian University of Pando (CIPA-UAP)in Cobija, Bolivia; and the Zoobotanical Park of the Federal Universityof Acre (PZ-UFAC) in Rio Branco, Brazil). Samples have been cross-referenced among countries by the lead author, but since referencecollections at these herbaria are under construction, we could identifymany of our sterile specimens only to the genus level with confidence(e.g., Barlow et al., 2012). Overall, 93.5% of stems were assigned genuslevel identificationswith confidence, and 99.5% were assigned to familylevel.

2.3. Forest value components

Weestimated 15 variables describing forest value (forest value com-ponents; Table 1) in this region. First, we calculated indices describingwoody plant diversity in each plot. We used the Simpson index to

describe genus-level diversity, and the complementary Rao quadraticentropy to describe functional diversity (Baraloto et al., 2012a). Weapplied a functional trait database describing leaf and wood tissuecharacteristics of more than 1200 species across the Amazon region(Fortunel et al., 2012). Trait values were applied to each individual inthe sample by taxonomic correspondence. For species or genera in theplot data that were not present in the functional trait database, wesubstituted the mean values for all individuals pertaining to the nexthigher taxonomic level, as these traits showweak but significant phylo-genetic signal (Baraloto et al., 2012b). Of the 397 genera present in thedataset, 177 were assigned genus-level traits and 166 were assignedmean family level trait values.

Second, we estimated aboveground live biomass using allometricequations for different size classes as described in Baraloto et al.(2011) (see Appendix S2 for details). We also estimated three otherindices that relate forest value to local livelihoods, including timbervolume of commercial species harvested in the region and the abun-dance of principal NTFP species. We compiled a list of 28 commerciallyimportant genera/species complexes in the region based on interviewswith community members (Perz et al., 2013a). We calculated standingtimber volume for stems of all commercial species with DBH N45 cmin each plot using the Nogueira et al. (2008) allometry for the southernAmazon region. We also calculated future crop tree density of timberspecies (FCT; Rockwell et al., 2007a) as the density of all trees of thesame species group with DBH between 20 and 45 cm, and the regener-ation density as all stems with DBH between 2.5 and 20 cm.

We identified four commercially important species as non-timberforest products in the region based on interviews with communitymembers (Perz et al., 2013a), including Brazil nut (B. excelsa,

Table 1Forest value components defined in this study, their units and description, and their mean values (with standard deviation) in each of the three regional states across the tri-nationalfrontier. Significant differences among countries are indicated by different letters, following Tukey HSD tests and univariate ANOVA.

Attribute Units Description Pando,Bolivia

Acre, Brazil Madre de Dios,Peru

Tree genus diversity Simpson index, i.e. probability of drawing a pair of individuals from two different genera 0.94 (0.03)a 0.95 (0.04)ab 0.96 (0.01)b

Tree functionaldiversity

Rao's quadratic entropy (Pavoine and Bonsall, 2011), integrating differences among generain leaf and wood functional trait values with relative abundances of genera in each site

0.92 (0.04)a 0.92 (0.04)a 0.92 (0.04)a

Abovegroundbiomass

Mg ha−1 Biomass of stems with diameter greater than 10 cm, calculated using the allometry ofChave et al. (2005)

227 (92)a 162 (88)b 197 (130)ab

Small stem biomass Mg ha−1 Biomass of stems with diameter less than 10 cm (and greater than 2.5 cm), calculatedusing the allometry of Hughes et al. (1999)

7.6 (1.7)ab 8.0 (1.6)a 6.8 (2.0)b

Timber volume m3 ha−1 Total commercial volume of 28 genera (see Table S1), calculated using the allometryof Nogueira et al. (2008)

126 (110)a 22 (31)b 52 (52)b

Future crop trees Stems ha−1 Total density of future timber crop trees, defined as having DBH between 20 and 45 cm,of 28 commercial timber genera (see Table S1)

80 (16)a 59 (52)a 57 (42)a

Timber regeneration Stems ha−1 Total density of regenerating juveniles, defined as having DBH between 2.5 and 20 cm,of 28 commercial timber genera (see Table S1)

263 (130)a 208 (82)a 112 (62)b

Brazil nut trees Stems ha−1 Density of adult (DBH N 50 cm) Brazil nut (B. excelsa) trees 5.6 (4.4)a 2.2 (2.7)b 0.7 (1.4)b

Brazil nutregeneration

Stems ha−1 Density of regenerating juvenile (DBH between 2.5 and 50 cm) Brazil nut (B. excelsa) trees 16.7 (23)a 3.2 (5.5)b 0.7 (2.7)b

Rubber trees Stems ha−1 Density of adult (DBH N 30 cm) rubber (H. brasiliensis) trees 1.7 (3.5)a 1.1 (2.3)a 3.3 (10.3)a

Rubber regeneration Stems ha−1 Density of regenerating juvenile (DBH between 2.5 and 30 cm) rubber (H. brasiliensis) trees 12.0 (12.6)a 6.0 (7.0)a 6.3 (12.8)a

Copaiba trees Stems ha−1 Density of adult (DBH N 25 cm) copaiba (Copaiba spp.) trees 0.1 (0.5)a 0.2 (0.9)a 0 (0)a

Copaibaregeneration

Stems ha−1 Density of regenerating juvenile (DBH between 2.5 and 25 cm) copaiba (Copaiba spp.) trees 0.7 (2.6)a 0.4 (2.0)a 0 (0)a

Açai trees Stems ha−1 Density of adult (DBH N 10 cm) açai (E. precatoria) trees 13.9 (15.0)ab 8.6 (11.5)a 20.4 (15.0)b

Açai regeneration Stems ha−1 Density of regenerating juvenile (DBH between 2.5 and 10 cm) açai (E. precatoria) trees 40.0 (31.8)a 8.4 (9.9)b 23.3 (24.4)a

677C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

Lecythidaceae), rubber (H. brasiliensis; Euphorbiaceae), açai (here thesingle-stemmed, upland E. precatoria, Arecaceae) and copaiba oil(Copaifera spp., Fabaceae). For each of these taxa,we estimated the den-sity of productive adults, defined here for each species as the size atwhich the product (fruit, latex, oil) is harvested, and the density of re-generation as all stems between 2.5 cm DBH and the size of productiveadults (Table 1). While we do not consider these four species to consti-tute an exhaustive list of NTFP species in the study region, we considerour NTFP index to be a reasonable indicator of value as perceived bylocal community members.

2.4. Data analysis

A geographic information system (ArcGIS 10.1, ESRI, Redlands,California) was developed for the area, based on the geo-referenced2005 Landsat image (Southworth et al., 2011) and digitized communitymaps indicating land tenure, land-use history, extractive zones, trails,protected areas, and precise locations of our samples (Perz et al.,2013a). We also digitized all maps we could obtain describing pavedand unpaved roads in all three countries of the region. We used theGIS to calculate four classes of features representing biogeographicgradients, urban centers of colonization, and highway infrastructure.

First, we calculated the distance to the foothills of the Andes (the500 m a.s.l. contour line) as a proxy for biogeographic variation in theregion, using the 30 meter digital elevation model derived from ASTERstereo image data (Hirano et al., 2003).

Second, we calculated the distance to the nearest access point alonga hierarchical series of roads, including (i) the paved IOH, (ii) secondaryroads, defined as any road that originates from the paved sections of theIOH and paved primary roads, and (iii) tertiary roads as any road thatoriginates from a secondary road.

Third, we estimated the date of paving of the section of the IOHnearest each sampling plot, based on (i) proximity to the IOH and(ii) the approximate distance along the IOH between starting and end-ing dates of paving according to paving dates provided by local commu-nity leaders. We then calculated the time since paving as the differencebetween the date of sampling plot fieldwork at the site and the estimat-ed paving date. Sampling plots in Bolivia along unpaved sections of theIOH were therefore assigned negative values.

Fourth, we calculated the distance to the center of the nearest urbancenter along the highway (see Fig. 1), defined as having a recent popula-tion estimate ofmore than 20,000. These included Rio Branco (269,505),Puerto Maldonado (44,381), Cobija (32,124), Epitaciolandia–Brasileia(21,470), and Sena (21,356).

We then used ArcGIS Network Analyst to calculate minimum traveltimes with speeds and network connectivity defined by Perz et al.(2013b) to calculate the minimum travel time from each samplingplot to reach (i) the closest segment of the IOH and (ii) the closesturban center.

We first simplified the GIS-derived variables using correlation anal-ysis and principal components analysis. We then used multiple regres-sion models to examine the relationships between each forest valuecomponent and independent variables chosen to represent variationin biogeographic gradients, and geographic gradients related to urbancenters of colonization and highway infrastructure.

We contrasted results from our plot-based analyses of forest degra-dation with landscape scale analyses of deforestation for communitiesin the region conducted on Landsat TM image data that had been proc-essed, mosaicked and classified (Southworth et al., 2011). We focusedon forest cover change at the scale of entire communities withinwhich many of which our plots were embedded, using a dataset offorest cover changes in 93 communities with total surface area of atleast 200 ha (Perz et al., 2013a). Given the triangular nature of relation-ships between roadproximity and deforestation in this dataset,we usedquantile regressions with permutation tests for significance to evaluatethe effect of road proximity on non-forest land cover in the 2010 imageset, which corresponds most closely to the timing of our sampling.

All analyses were performed in the R statistical platform v 3.0(R Development Core Team), using packages, ade4 (Dray and Dufour,2007), glmulti (Calcagno, 2013), quantreg (Koenker, 2013), and vegan(Oksanen et al., 2013).

3. Results

3.1. Gradients in forest value components across the tri-national frontier

Across the 69 plots, we sampled 15,314 stems representing 87families, 397 genera, and at least 1009 species. We found very strikingdifferences in forest value across the international borders in the tri-

Table 2Correlations among biogeographic and geographic variables across the 69 sampling sites. Shown are Pearson correlation coefficients with significant correlations (alpha= 0.05) in bold.

Distanceto IOH

Distance to2ry road

Distance to3ry road

Distance tourban center

Time sinceIOH paving

Travel time tourban center

Travel timeto IOH

Distance to IOHDistance to 2ry road −0.29Distance to 3ry road 0.03 −0.08Distance to urban center −0.36 0.37 −0.09Time since IOH paving −0.48 −0.24 −0.02 −0.18Travel time to urban center −0.14 0.42 −0.09 0.85 −0.21Travel time to IOH 0.99 −0.27 −0.00 −0.37 −0.47 −0.13Distance to Andes 0.33 −0.44 0.03 −0.34 0.55 −0.09 0.31

678 C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

national frontier. Plots in the roadside communities in Pando, Boliviahad on average 50% more aboveground biomass than those in Acre,Brazil and nearly twice the standing timber volume (Table 1). Brazilnut tree density was also much higher in plots in communities inPando than those in Acre or Madre de Dios, Peru. In contrast, forestplots in communities inMadre deDios had higherwoody plant commu-nity diversity and more abundant E. precatoria adults (Table 1).

3.2. Effects of the IOH versus urban proximity versus biogeographicgradients

Many of theGIS-derived variableswere strongly correlated (Table 2),especially strong positive correlations between distance metrics andthe derived travel time indices. A first principal component reflecteda strong positive correlation between distance from urban centersand the distance from secondary roads, whereas a second componentreflected a negative correlation between distance from the IOH andtime since paving, with many of the communities sampled in theearliest-paved regions found within close proximity of the IOH. Thetertiary road network proved largely independent of the IOH and itssecondary roads. We chose four variables to summarize these gradientsin subsequent analyses: distance from the IOH, distance from urbancenters, distance from tertiary roads, and distance from the Andes.

The broader country differences were only partly reflected in thebiogeographic gradients summarized by the distance from Andes,when the other geographic variables were considered in the stepwise

Table 3Variation in forest value components across biogeographic and geographic distances —from the Andean foothills, from the Inter-Oceanic Highway (IOH), from the nearest tertia-ry road, and from the nearest urban center with population greater than 20,000. Shownare coefficients for log-transformed distance values from themost parsimonious stepwisemultiple regression models for each forest value component, as determined by AIC. Thevariance explained by each model is indicated in the last column.

Component Andes IOH 3ry road Urbancenter

Modelr2

Tree genus diversity −0.0067⁎⁎ −0.0086⁎⁎ 0.105Tree functional diversity −0.020⁎⁎ −0.022⁎⁎ 0.180Aboveground biomass 31.6 14.7 0.086Small stem biomass 0.648⁎ 0.065Timber volume 21.4⁎⁎ 41.6⁎⁎ 0.177Timber regeneration 40.9⁎⁎ 24.9⁎⁎ 0.355Future crop trees 6.40 0.043Brazil nut trees 1.07⁎⁎ 0.136Brazil nut regeneration 2.96⁎⁎ −5.82⁎⁎ 0.344Rubber trees −3.17⁎ −2.28 0.111Rubber regeneration −6.72⁎⁎ 2.91⁎⁎ 0.153Copaiba trees −0.113 0.069Copaiba regeneration NAAçai trees 4.27⁎ 0.095Açai regeneration 4.41⁎⁎ 3.98⁎⁎ 0.160

NA best model retained intercept only.⁎ P b 0.05.⁎⁎ P b 0.01.⁎⁎⁎ P b 0.001.

models. In particular, rubber trees were less frequent in forests in east-ern Pando and Acre, but overall smaller stems and timber tree regener-ation was higher with increasing distance from the Andes (Table 3).

After accounting for the biogeographic gradients associated withdistance from the Andes, we did find some evidence for forest value de-creasingwith increasing proximity to the IOH (Table 3). Notably, timbervolume and Brazil nut tree density increased in the sites farthest fromthe IOH, most of which occur in Pando, Bolivia; when examined withineach country, these trends were not significant (Fig. 2).

We found little evidence for an effect of proximity to urban centerson forest value in the region, beyond that explained by proximity tothe IOH (Table 2). More remote sites tended to have higher standingtimber volume, even when adjacent to the IOH (Table 2). We foundlittle effect of tertiary road networks on forest value, other than a posi-tive effect of distance of nearest road on Euterpe palm densities.

3.3. Contrasts with regional deforestation data

The proportion of non-forest cover estimated from remote sensingin the 93 communities of the region showed a strong triangular rela-tionship with the proximity of communities to the nearest paved road(Fig. 3); that is, communities near the road had highly variable forestcover estimates, whereas communities far from the road all had veryhigh forest cover (i.e., low non-forest proportions). Quantile regression(tau= 90%) between 2010 non-forest cover and distance to road werehighly significant (P b 0.001). Across the region, forest cover increasedby about 2.4% for each 1 km of distance from the road (Fig. 3). This pat-tern was largely driven by communities in Brazil, which also had thehighest overall proportion of non-forest cover. In Peru, this relationshipdissolved, with high variance around the mean of ca. 15% non-forestedarea (Fig. 3).

4. Discussion

4.1. Differences in forest value across the tri-national border

Our broad sampling network allowed us to describe some inherentgeographic patterns in forest structure and composition and to furtherexamine distinctions among the three countries of this dynamic region.We foundmarked differences across political borders of the three coun-tries, for several components of forest value (Table 1). These differencesappear to be concordant with both regional history and differences ingovernance (such as via land tenure) among the countries. For example,forest value components related to forest structure, including timberstocks and aboveground biomass, were much higher in Pando, wherehighway paving remains incipient (Perz et al., 2013a). In Pando, thecommunities in which we sampled were mostly tenured as communallands where smallholders are invested in sustainable management offorest resources, or were newly created in areas where resources hadnot been exploited for long time periods (Duchelle et al., 2011; Perzet al., 2013a). In contrast, most of the roadside communities in whichwe sampled in Madre de Dios were privately tenured as agriculturallands, since we were granted access to few of the concesiones castañales

Fig. 2. Effects of distance to the Inter-Oceanic Highway (IOH; not log scale) on four forest value components. Points represent each of the forest sample plots, with symbols accordingto countries including Bolivia (gray circles), Brazil (black squares) and Peru (red diamonds). Coefficients from stepwise mixed models on these relationships are reported in Table 2.(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

679C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

(where resource management is restricted for Brazil nut harvest) thatoccur directly along the road. Many of the Peruvian sites we sampledhad been high-graded for commercial timber, resulting in lower timberstocks and aboveground biomass, but likely favoring regeneration and

Fig. 3. Effects of distance to the Inter-Oceanic Highway (IOH) on non-forest cover in 2010in 93 forest settlements including those that were sampled for this study. Symbols aredistinguished by country as Bolivia (gray circles), Brazil (black squares) and Peru (reddiamonds). The triangular relationship was tested using permutations following quantileregressions, with trend lines illustrated for each country with the same colors. Data arefrom Perz et al. (2013a, b). (For interpretation of the references to color in this figurelegend, the reader is referred to the web version of this article.)

overall woody plant diversity (Fig. 2). Acre comprises very diverseland tenure situations differing in concession size and regulations forforest use, including extractive reserves, agro-extractive projects, andrural and urban colonies (Perz et al., 2013a,b). Further research con-trasting land tenure types within Acre will be important to understandhow national and regional policies are mitigating road impacts in thisrapidly developing state.

Some of the differences among countries may also be explained bybiogeographic variation across the region, as estimated by the distancefrom the foothills of the Andes (Table 2). The biogeographic patternswe observed may be due to differences in climate or soil, both ofwhich have been found to explain general patterns of forest structureand composition across broader gradients in the Amazon (Quesadaet al., 2012). Our sampling design, which focused on terra firme forests,controlled for some of the topographically-related variation in foreststructure that occurs in the larger region (Baraloto et al., 2011).Among terra firme sites such as those we sampled, broad variation insoils in the study region follow the axis of distance from the Andes,with themost fertile soils in the southwest and decreasing as one travelsnortheast into Acre (Phillips et al., 2003; Quesada et al., 2012). It is verylikely that the geographic patterns we observed reflect in large partthese differences in soil structure and fertility. An important remainingobjective will be to follow the impacts of both land use and climatechange in the region via continued resampling of these study plots,and we plan to implement standardized soil sampling as part of thisfuture work.

4.2. Effects of distance to highway on forest value

Nevertheless, we did observe someweak effects of IOH proximity onforest value, with seven of the 15 forest value components increasing

680 C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

significantly with increasing distance from the IOH (Table 3). Notably,the forests we sampled closest to the IOH, almost all of which were inAcre and Madre de Dios, had much lower standing timber volumesand density of adult Brazil nut trees (Fig. 2, Table 2). However, above-ground biomass was highly variable across our sample and did not cor-relate with distance from roads or urban centers. Moreover, lowcomposite forest value was also frequent at sites tens of kilometersfrom the IOH (Fig. 2). In fact, one of themore striking results of our sam-pling was the frequency of sites with very low stocks of timber andNTFPs, across all three countries and irrespective of the distance fromthe IOH (Fig. 2).

At a more local scale around Puerto Maldonado, Peru, one of theurban areas studied here, the distance of forest plots to the nearest an-thropogenic edge (maximum of 9 km) was not a significant explanatoryvariable for the variation infloristic composition or abovegroundbiomass(Phillips et al., 2006). These authors interpreted their observations asconsistent with the hypothesized resilience of southwestern Amazonianforests to moderate anthropogenic disturbances, because these forestsgenerally have higher intrinsic rates of turnover than forests in centraland eastern Amazonia (Quesada et al., 2012) and may therefore bemore adapted to disturbances (Barlow et al., 2012). Our study is partiallyconsistent with these observations, with almost the full breadth of varia-tion of each forest value component found at sites within 10 km of theIOH; but in our sample the sites closest to the roads were also the mostheavily degraded (Fig. 2).

4.3. Contrasts in the effects of distance to road on deforestation and forestdegradation

Theweak relationship between road proximity and forest value con-trasts with consistent negative effects of road proximity on forest coveracross the globe (Fearnside, 2005; Perz et al., 2007; Laurance et al.,2009; Ahmed et al., 2013; Barber et al., 2014). Communities near theIOH had much lower forest cover than communities farther from theroad in an analysis of forest cover in 2010 (Fig. 3). These results are con-sistent with recent reports by Barber et al. (2014) that most Amazoniandeforestation occurs within 5 km of unpaved roads. We have found noother study contrasting deforestation and forest degradation in re-sponse to road infrastructure, but our results are consistent with workby Arroyo-Rodríguez et al. (2012, 2009), who report weak effects offorest fragmentation on forest structure and composition. Neverthelessour results contrast markedly with those of fragmentation experimentsin which edge effects of deforestation have led to accelerated degrada-tion of forest value components including plant species compositionand forest structure (Laurance et al., 2007).

We offer three alternative but not exclusive explanations for theweak effects of IOH proximity on forest value that we observed. First,forests in this region are extremely heterogeneous in the absence ofany recent anthropogenic disturbances. Indeed, the region is character-ized by forests dominated by the woody bamboo Guadua, which tendto have lower basal area and a distinct floristic composition (Barlowet al., 2012; Rockwell et al., 2014). We attempted to limit the effectof forest types in our sampling by focusing on terra firme forests thatwere representative of the larger community holdings, but some ofthese communities were situated predominantly in forest types suchas Guadua-dominated forests that had particular characteristics evenbefore recent human colonization (Barlow et al., 2012). Nevertheless,while forest variability may have introduced noise into our analysis, itwas unlikely to drive any of the observed patterns, since the spatial dis-tribution of Guadua-dominated forests is not related to road proximity(r = 0.08, P = 0.76) nor to country (F = 2.2, P = 0.14) in our sample.Disentangling these effects will require a more comprehensive andstratified sampling strategy. It would also be useful to couple this sam-pling with remote sensing approaches to sample forest structure andleaf spectral composition (Féret and Asner, 2012), of communities atvarying distances from the road.

Second, our sampling within any particular community was limitedby access decisions of local community leaders and landowners, and theability of our field teams to conduct fieldwork within a timeframe thatalso accommodated their academic schedules. Therefore, it is possiblethat the plots we sampled are not completely representative in all com-munity forests. In particular, few of our forest sites were distant fromany secondary or tertiary road, even though we did sample acrossbroad distances from the IOH (Fig. 1). It will be interesting to addressany discrepancy between our samples and local peoples' perceptionsof available forest resources by confronting our plot inventory datawith data from interviews conducted with community members (Perzet al., 2013a).

Third, the effects of infrastructure change on forest value in the re-gion are likely to be much more complicated than simple relationshipswith road proximity (Perz et al., 2013a). Indeed, in our experience com-munities distant from the IOH have experienced some of the most dra-matic changes. For example, several Agro-extractive Settlement Projects(PAEs) in Acre, Brazil have been able to integrate timber managementplans in large part because of access to markets provided by road im-provement (Rockwell et al., 2007a). These communities remain rela-tively distant from both the IOH and market and urban centers, butthe infrastructural changes accompanying the IOH paving have beencited as important to their considerations of implementing timberman-agement plans, with variable outcomes for forest value (Rockwell et al.,2007b). In some cases, roads may generate socioeconomic benefitsthat actually reduce pressure on resources in remaining forest patches(Weinhold and Reis, 2008; Laurance et al., 2014), by providing accessto markets and supplies that permit the maintenance of livelihoodswithout additional investment in unsustainable harvesting of forest re-sources (Perz et al., 2013a).

4.4. Conclusions and perspectives

Our study provides an important complement to global analyses ofroad impacts (e.g., Forman et al., 2003; Laurance et al., 2014) in twokey ways. First, we show that distinguishing between deforestationand forest degradation complicates the study of road impacts becausethe two do not always respond in the same way. Second, we provide arepeatable approach to define forest value components that can be im-plemented across sites at different spatial scales, and thatwill ultimatelypermit comparative analyses of road impacts on forest degradation.

Our study also underlines the need for integration between plot-based estimates of forest valuewith improvedmeasures of forest degra-dation from remote sensing. Changes in forest value may accelerate inthis region as a result of interactions between changes in land use andclimate with an increased incidence of drought and fire (Phillips et al.,2009). Our field data from this understudied region can be used to cal-ibrate emerging approaches in remote sensing such as spectranomics(Féret and Asner, 2012), that seek to characterize not only changes inforest structure but also changes in forest diversity and species compo-sition in this region and elsewhere.

Acknowledgments

This study was supported by the US NSF (HSD 0527511, and CNH1114924) and the French agriculture ministry (MAAP BGF GuyaSpaSEproject). We thank the institutions supporting the training coursesthat were the foundation of fieldwork in the three countries, includingCentro do Trabalhadores da Amazonia (CTA) in Rio Branco, Acre, andthe Center for Research on Amazon Protection (CIPA) in Cobija,Pando; and the numerous students and community members in thethree countries who provided assistance with fieldwork. We are alsograteful to the taxonomic specialists who are investing in capacitybuilding in the region andwho assistedwith the identification of botan-ical specimens from the project, including P. AcevedoRodriguez, C. Berg,

681C. Baraloto et al. / Biological Conservation 191 (2015) 674–681

D. Daly, R. Foster, and C. Vriesendorp. We also thank A. Duchelle and M.Menton for comments on previous versions of the manuscript.

Appendix A. Supplementary data

Detailed methods (Appendix S1 and S2) are available online. Theauthors are solely responsible for the content and functionality ofthese materials. Queries (other than absence of the material) shouldbedirected to the corresponding author. Supplementary data associatedwith this article can be found, in the online version, at http://dx.doi.org/10.1016/j.biocon.2015.08.024.

References

Ahmed, S.E., Souza Jr., C.M., Ribeiro, J., Ewers, R.M., 2013. Temporal patterns of road net-work development in the Brazilian Amazon. Reg. Environ. Chang. 1–11.

Arroyo-Rodríguez, V., Cavender-Bares, J., Escobar, F., Melo, F.P., Tabarelli, M., Santos, B.A.,2012. Maintenance of tree phylogenetic diversity in a highly fragmented rain forest.J. Ecol. 100, 702–711.

Arroyo-Rodriquez, V., Pineda, E., Escobar, F., Benitez-Malvido, J., 2009. Value of smallpatches in the conservation of plant‐species diversity in highly fragmented rainforest.Conserv. Biol. 23, 729–739.

Asner, G.P., Loarie, S.R., Heyder, U., 2010. Combined effects of climate and land‐use changeon the future of humid tropical forests. Conserv. Lett. 3, 395–403.

Baraloto, C., et al., 2011. Disentangling stand and environmental correlates of above-ground biomass in Amazonian forests. Glob. Chang. Biol. 17, 2677–2688.

Baraloto, C., et al., 2012a. Contrasting taxonomic and functional responses of a tropicaltree community to selective logging. J. Appl. Ecol. 49, 861–870.

Baraloto, C., et al., 2012b. Using functional traits and molecular phylogenetic trees toexamine the assembly of tropical tree communities. J. Ecol. 100, 690–701.

Baraloto, C., Molto, Q., Rabaud, S., Hérault, B., Valencia, R., Blanc, L., Fine, P.V., Thompson, J.,2013.Rapid simultaneous estimation of abovegroundbiomass and tree diversity acrossNeotropical forests: a comparison of field inventory methods. Biotropica 45, 288–298.

Baraloto, C., et al., 2014. Trade-offs among forest value components in community forestsof southwestern Amazonia. Ecol. Soc. 19 (4).

Barber, C.P., Cochrane, M.A., Souza Jr., C.M., Laurance, W.F., 2014. Roads, deforestation, andthe mitigating effect of protected areas in the Amazon. Biol. Conserv. 177, 203–209.

Barlow, J., et al., 2012. Wildfires in bamboo-dominated Amazonian forest: impacts onabove-ground biomass and biodiversity. PLoS One 7 (3), e33373.

Benitez-Lopez, A., Alkemade, R., Verweij, P.A., 2010. The impacts of roads and other infra-structure on mammal and bird populations: a meta-analysis. Biol. Conserv. 143,1307–1316.

Calcagno, V., 2013. Glmulti: model selection and multimodel inference made easy. RPackage Version 1.0.7.

Chave, J., Andalo, C., Brown, S., Cairns, M.A., Chambers, J.Q., Eamus, D., Folster, H., Fromard,F., Higuchi, N., Kira, T., Lescure, J.P., Nelson, B.W., Ogawa, H., Puig, H., Riera, B.,Yamakura, T., 2005. Tree allometry and improved estimation of carbon stocks andbalance in tropical forests. Oecologia 145, 87–99.

Cousins, S.A., 2006. Plant species richness in midfield islets and road verges–the effect oflandscape fragmentation. Biol. Conserv. 127, 500–509.

Dray, S., Dufour, A.B., 2007. The ade4 package: implementing the duality diagram forecologists. J. Stat. Softw. 22 (4), 1–20.

Duchelle, A.E., Cronkleton, P., Kainer, K.A., Guanacoma, G., Gezan, S., 2011. Resource theftin tropical forest communities: implications for non-timber management, liveli-hoods, and conservation. Ecol. Soc. 16.

Duchelle, A.E., et al., 2013. Linking forest tenure reform, environmental compliance, andincentives: lessons from REDD+ initiatives in the Brazilian Amazon. World Develop-ment in press.

Fearnside, P.M., 2005. Deforestation in Brazilian Amazonia: history, rates, and conse-quences. Conserv. Biol. 19, 680–688.

Féret, J.-B., Asner, G.P., 2012. Semi-supervised methods to identify individual crowns oflowland tropical canopy species using imaging spectroscopy and LiDAR. RemoteSens. 4, 2457–2476.

Forman, R.T.T., et al., 2003. Road ecology: science and solutions. Island Press, Washington,DC.

Fortunel, C., Fine, P.V., Baraloto, C., 2012. Leaf, stem and root tissue strategies across 758Neotropical tree species. Funct. Ecol. 26, 1153–1161.

Freudenberger, L., Hobson, P., Schluck, M., Kreft, S., Vohland, K., Sommer, H., Ibisch, P.L.,2013. Nature conservation: priority-setting needs a global change. Biodivers. Conserv.2013, 1–27.

Hirano, A., Welch, R., Lang, H., 2003. Mapping from ASTER stereo image data: DEM valida-tion and accuracy assessment. ISPRS J. Photogramm. Remote Sens. 57 (5), 356–370.

Hoskin, C.J., Goosem, M.W., 2010. Road impacts on abundance, call traits, and body size ofrainforest frogs in northeast Australia. Ecol. Soc 15, 15.

Hosonuma, N., Herold, M., De Sy, V., De Fries, R.S., Brockhaus, M., Verchot, L., Angelsen, A.,Romijn, E., 2012. An assessment of deforestation and forest degradation drivers indeveloping countries. Environ. Res. Lett. 7, 044009.

Hughes, R.F., Kauffman, J.B., Jaramillo, V.J., 1999. Biomass, carbon, and nutrient dynamicsof secondary forests in a humid tropical region of Mexico. Ecology 80, 1892–1907.

Koenker, R., 2013. Quantreg: quantile regression. R Package Version 4.98.Laurance, W.F., Nascimento, H.E.M., Laurance, S.G., Andrade, A., Ewers, R.M., Harms, K.E.,

Luizão, R.C.C., Ribeiro, J.E., 2007. Habitat fragmentation, variable edge effects, andthe landscape-divergence hypothesis. PLoS One 2, e1017.

Laurance, W.F., Goosem, M., Laurance, S.G., 2009. Impacts of roads and linear clearings ontropical forests. Trends Ecol. Evol. 24, 659–669.

Laurance, W.F., et al., 2014. A global strategy for road building. Nature in press.Nepstad, D., et al., 2001. Road paving, fire regime feedbacks, and the future of Amazon

forests. For. Ecol. Manag. 154, 395–407.Ninan, K.N., Inoue, M., 2013. Valuing forest ecosystem services: what we know and what

we don't. Ecol. Econ. 93, 137–149.Nogueira, E.M., Fearnside, P.M., Nelson, B.W., Barbosa, R.I., Keizer, E.W.H., 2008. Estimates

of forest biomass in the Brazilian Amazon: new allometric equations and adjustmentsto biomass from wood-volume inventories. For. Ecol. Manag. 256, 1853–1867.

Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., Minchin, P.R., O'Hara, B., Simpson, G.L.,Solymos, P., Stevens, M.H.H., Wagner, H., 2013. Vegan: Community Ecology Package.R Package Version 2.0-7.

Pavoine, S., Bonsall, M., 2011. Measuring biodiversity to explain community assembly: aunified approach. Biol. Rev 86, 792–812.

Perz, S.G., Caldas, M.M., Arima, E., Walker, R.J., 2007. Unofficial road building in theAmazon: socioeconomic and biophysical explanations. Dev. Chang. 38, 529–551.

Perz, S., et al., 2008. Road building, land use and climate change: prospects for environ-mental governance in the Amazon. Philos. Trans. R. Soc. B 363, 1889–1895.

Perz, S.G., Shenkin, A., Rondon, X., Qiu, Y., 2013a. Infrastructure upgrades and rural–urbanconnectivity: distance disparities in a tri-national frontier in the Amazon. Prof. Geogr.65, 103–115.

Perz, S.G., et al., 2013b. Trans-boundary infrastructure and land cover change: highwaypaving and community-level deforestation in a tri-national frontier in the Amazon.Land Use Policy 34, 27–41.

Pfaff, A., Robalino, J., Lima, E., Sandoval, C., Herrera, L.D., 2013. Governance, Location andAvoided Deforestation From Protected Areas: Greater Restrictions can Have LowerImpact, Due to Differences in Location. World Development in press.

Phillips, O.L., Vargas, P.N., Monteagudo, A.L., Cruz, A.P., Zans,M.E.C., Sánchez,W.G., Yli‐Halla,M., Rose, S., 2003. Habitat association amongAmazonian tree species: a landscape‐scaleapproach. J. Ecol. 91, 757–775.

Phillips, O.L., Rose, S., Mendoza, A.M., Vargas, P.N., 2006. Resilience of southwesternAmazon forests to anthropogenic edge effects. Conserv. Biol. 20, 1698–1710.

Phillips, O.L., et al., 2009. Drought sensitivity of the Amazon rainforest. Science 323,1344–1347.

Putz, F.E., Redford, K.H., 2008. The importance of defining ‘Forest’: tropical forest degrada-tion, deforestation, long-term phase shifts, and further transitions. Biotropica 42,10–20.

Quesada, C., et al., 2012. Basin-wide variations in Amazon forest structure and functionare mediated by both soils and climate. Biogeosciences 9, 2203–2246.

Rockwell, C., Kainer, K., Staudhammer, C., Baraloto, C., 2007a. Future crop tree damage ina certified community forest in southwestern Amazonia. For. Ecol. Manag. 242,108–118.

Rockwell, C., Kainer, K.A., Marcondes, N., Baraloto, C., 2007b. Ecological limitations of re-duced impact logging at the smallholder scale. For. Ecol. Manag. 238, 365–374.

Rockwell, C., Kainer, K., Staudhammer, C., Baraloto, C., 2014. Logging in bamboo-dominated forests in southwestern Amazonia: caveats and opportunities for small-holder forest management. For. Ecol. Manag. 315, 202–210.

Sasaki, N., Putz, F.E., 2009. Critical need for new definitions of “forest” and “forest degra-dation” in global climate change agreements. Conserv. Lett. 2, 226–232.

Southworth, J., Marsik, M., Qiu, Y., Perz, S., Cumming, G., Stevens, F., Rocha, K., Duchelle, A.,Barnes, G., 2011. Roads as drivers of change: trajectories across the tri-national fron-tier in M-A-P, the southwestern Amazon. Remote Sens. 3, 1047–1066.

Walker, R., Arima, E., Messina, J., Soares-Filho, B., Perz, S., Vergara, D., Sales, M., Pereira, R.,Castro, W., 2013. Modeling spatial decisions with graph theory: logging roads andforest fragmentation in the Brazilian Amazon. Ecol. Appl. 23, 239–254.

Weinhold, D., Reis, E., 2008. Transportation costs and the spatial distribution of land use inthe Brazilian Amazon. Glob. Environ. Chang. 18, 54–68.