Biological Conservation 170 (2014) 64–73

Contents lists available at ScienceDirect

Biological Conservation

journal homepage: www.elsevier .com/locate /b iocon

Demand for beef is unrelated to pasture expansion in northwesternAmazonia

0006-3207/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.biocon.2013.12.018

⇑ Corresponding author. Address: Department of Ecology and Evolution andConsortium for Inter-Disciplinary Environmental Research, SUNY Stony Brook, 650Life Sciences Building, Stony Brook, NY 11794-5245, USA. Tel.: +1 631 632 1554;fax: +1 631 632 7626.

E-mail addresses: liliana.davalos-alvarez@stonybrook.edu (L.M. Dávalos),jholmes@utdallas.edu (J.S. Holmes), nrodriguez@unal.edu.co (N. Rodríguez),darmenterasp@unal.edu.co (D. Armenteras).

Liliana M. Dávalos a,⇑, Jennifer S. Holmes b, Nelly Rodríguez c, Dolors Armenteras c

a Department of Ecology and Evolution and Consortium for Inter-Disciplinary Environmental Research, SUNY Stony Brook, NY, USAb School of Economic, Political and Policy Sciences, University of Texas at Dallas, Richardson, TX, USAc Landscape Ecology and Ecosystem Modelling Laboratory, Department of Biology, Sciences Faculty, National University of Colombia, Bogotá, Colombia

a r t i c l e i n f o

Article history:Received 15 September 2013Received in revised form 13 December 2013Accepted 15 December 2013

Keywords:AmazonHamburger connectionLand as wealthLivestockCattle ranching

a b s t r a c t

For decades, pastures have replaced Amazon forests, threatening biodiversity. Two dominant hypothesesexplain this process. First, according to the hamburger connection beef markets promote conversion offorests into cattle ranches resulting in deforestation. Second, the land-as-wealth hypothesis explainsdeforestation as the result of land appreciation in forest frontiers. We tested these hypotheses using landuse and socioeconomic data from northwestern Amazonia from 2000 to 2009. We also investigated twokey control variables: coca cultivation and eradication, both of which have been proposed as importantdrivers of forest loss. We found high rate of conversion from forest to pasture, and unimpeded forest frag-mentation. As predicted by the hamburger connection, pastures expanded as the cattle inventory grew.Beef demand did not drive pasture expansion, as ranching revenues declined and beef prices wereroughly stable. Coca cultivation did not respond to the eradication campaign, and declined as the popu-lation became increasingly urbanized. We propose that investment on roads has enhanced the antici-pated value of land. Cattle bolster land claims and can be used as financial collateral, so they haveinherent value even without market exchange. Cattle can then increase despite stable beef prices anddeclining ranching revenues. To protect biodiversity, policy interventions that enhance the value ofstanding forests are needed to slow deforestation and fragmentation in these rapidly changinglandscapes.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

Amazon forests encompass the largest continuous blocks oflow-latitude forest ecosystems in the world, harbor 10% of allknown species, and hold an estimated 15-years worth of globalCO2 emissions (Houghton et al., 2001). Conversion of forests to pas-ture, however, has accelerated over the last four decades (Nepstadet al., 2006). Rapid degradation has eroded ecosystem functions,and habitat loss and fragmentation have increased vulnerabilityto edge effects and biodiversity loss (Barlow et al., 2012; Fearnside,2005; Laurance et al., 2006).

Two contrasting hypotheses have emerged to explainconversion from forest to pasture in Amazonia. The hamburgerconnection proposed low-cost beef production for fast food drovethe decline of Central American forests (Myers, 1981). Demand

for beef promotes forest loss by making cattle ranching profitable.Relatively low start-up costs and institutional support for cattleranching make cattle ranching attractive (Gomes et al., 2012;Murphy et al., 1997; Van Ausdal, 2009), and the perennial demandfor beef promotes cattle ranching and ultimately causes deforesta-tion (Kaimowitz et al., 2004; McAlpine et al., 2009).

The land-as-wealth hypothesis highlights legal and fiscalpolicies that enhance the value of cattle without growing demandfor beef, and proposes land speculation is the ultimate incentivefor clearing. Clearing the forest to establish a herd of cattlestrengthens land claims in Amazonia, even if clearing andranching are not economically productive (Hecht, 1985, 1993).Legal requirements for granting titles, improving means of access,and nearly non-existent rural property taxes encourage landspeculation (Bowman et al., 2012; Parry et al., 2010). Landspeculation can be further exacerbated by local urbanizationand the closing of the forest frontier, as land quickly accrues va-lue in regions newly interconnected to larger markets (Parryet al., 2010).

We analyze the dynamics of land use change and forest frag-mentation in the northwestern Amazon (Guaviare, Colombia) be-tween 2000 and 2009 to investigate the contemporary forces

L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73 65

driving forest fragmentation and conversion to pasture. We fo-cused on total area change as well as the number, size, and isola-tion of fragments. In particular, we analyze demand for beef andland appreciation as potential drivers of land use change and forestfragmentation in the region.

2. Theory

2.1. The hamburger connection

Global demand for beef has been identified as an ultimate dri-ver of neotropical forest loss since the 1980s (Myers, 1981). In Bra-zilian Amazonia, for example, deforestation rates increased as thedecline of hoof-and-mouth disease and devaluation of the reaismade beef exports feasible and globally competitive (Kaimowitzet al., 2004). Local demand for beef from a growing, or increasinglyaffluent population can also drive deforestation, as it has in Austra-lia since the 1980s (McAlpine et al., 2009). In both examples ranch-ing revenues increased, as higher prices and expanding herdsresulted in a growing ranching sector.

In Colombia, the domestic market is more important than ex-ports because there are important obstacles to exports. Two sani-tary barriers have confined beef to the domestic market: hoof-and-mouth disease and lack of refrigeration. Before 2009 exportsto global markets were impossible because vaccination againsthoof-and-mouth disease had not reached the targeted 99.9% ofthe total cattle inventory (Anonymous, 2010). More than 72% ofbeef production during the study period was unrefrigerated andhence could not be commercialized internationally (Anonymous,2010). For these reasons, domestic beef prices were used to testthe hamburger connection.

SAN JOSE DEL GUAVIARE

CALAMAR

EL RETORNO

72˚30’W

72˚30’W

73˚0’W

73˚0’W

2˚30

’N

2˚30

’N

2˚0’

N

2˚0’

N

5 010km

Guaviare

Ariari

LEGENDTowns

MunicipalityBorders

Study Area

.Rivers

To Villavicencio

Colombia

BrazilPeru

Venezuela

Ecuador

Bolivia

Paraguay

Guyan

a

Surina

m

French

Guiana

N

Fig. 1. Location of samples in Guaviare, Colombia northwestern Amazonia. Insetshows study area in South America. Samples were selected at random and cover thenorth–south axis from San José to Calamar.

2.2. Land-as-wealth

The land-as-wealth hypothesis was proposed as an alternativeto the hamburger connection, which it critiques on three points(Hecht, 1985, 1993). First, pastures must be connected to marketsfor beef demand to drive conversion to pasture, and manyAmazonian forest frontiers were too remote to reach global oreven urban markets. Second, local population growth cannot ex-plain conversion to pasture, as people in the frontier often relyon food imports from more developed regions (Hecht, 1993). Fi-nally, the focus on beef production is misguided if land tenure,taxation, and ownership laws and policies reward unproductiveaccumulation of cleared land (Fearnside, 2005; Hecht, 1985).The latter is the most important point of the land-as-wealthhypothesis, as there are strong incentives for converting to pas-ture to acquire land and eventually obtain legal titles to it (Hecht,1985, 1993).

For land-as-wealth to explain conversion to pasture, thereshould be policies in place that facilitate legal claims for cleared(but not forested) lands, and land value should increase overtime. Throughout Amazonia, clearing the forest for pasture resultsin a stronger claim to a land title than other land uses (Fearnside,2005). Rural property taxes are too low and too often unenforcedto deter land grabbing, and all that is needed to promote specu-lation in the forest frontier are increasing land values (Bowmanet al., 2012; Parry et al., 2010). Urban growth (Lambin et al.,2001), road construction and improvement (Simmons, 2004),and the expectation of future titles (Hecht, 1993), all increasethe value of the land. Since cattle enhance claims to the land(Hecht, 1993), the land-as-wealth hypothesis is also consistentwith expansion of the cattle herd, as long as it is the land that in-creases in market value.

2.3. The role of coca

Since the 1980s, illegal coca cultivation for the global cocainemarket has attracted smallholders to lowland Amazonian forests(Hecht, 1993; Kaimowitz, 1997; Young, 1996). As a high-valueglobalized crop, coca was initially thought to leave a small foot-print on the forest (Dávalos et al., 2009). Since then, large-scaleland cover analyses revealed signatures of deforestation associatedwith coca cultivation, a large indirect impact that dwarfs the directfootprint of coca (Dávalos et al., 2011). Consistent with those large-scale analyses, detailed time series from Guaviare found that cocawas the first land use to fragment the forest, losing importanceover time as pastures replaced coca (Armenteras et al., 2013).

During the study period, Guaviare was at the center of cocaforced eradication policy (UNODC and Gobierno de Colombia,2012). It is possible that the local decline in coca cultivation wasa response to the eradication campaign, but this is unknown(Rincón-Ruiz and Kallis, 2013; Rincón-Ruiz et al., 2013). In con-trast, Dávalos et al. (2011) argued that coca cultivation was an indi-cator of poor access to markets and underdevelopment. If this werethe case, coca cultivation would decline in response to connectionsto markets and (legal) economic development, instead of eradica-tion (Dávalos et al., 2009).

3. Materials and methods

3.1. Study area

The region corresponded to an area of �4500 km2 in the SanJosé-Calamar axis in Guaviare, Colombia, northwestern Amazonia(Fig. 1). To reduce spatial autocorrelation and capture variationalong the axis, 44 windows of 3 � 3 km were randomly selected,representing a subsample of 396-km2. Aside from 1 window

66 L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73

overlapping with the municipality of Puerto Concordia, Meta, thesample comprised land in the municipalities of San José, Calamar,and El Retorno.

Active colonization of the San José-Calamar axis began as earlyas the 1930s, with the establishment of a small number rubberplantations (Arcila and Salazar, 2011). Government-assistedmigration in the 1960s renewed the colonization front, but didnot improve local infrastructure (Arcila et al., 1999). As a result,patterns of human occupation have corresponded to sites ofnatural resource extraction initially accessible by river, in whichpasture and annual crops were later established (Ariza et al.,1998). Until the 1990s, the road connecting San José to Villavicen-cio was not well maintained and was often impassable for weekson end (Arcila and Salazar, 2007). Since then, improvements tothe existing road and the construction of permanent bridges overthe Ariari and Guaviare rivers have linked San José to Andeancities and their markets (Arcila and Salazar, 2011). This changeis thought to have contributed to urbanization (Fig. 1) (Arcilaand Salazar, 2007, 2011). The trans-Andean Villavicencio–Bogotároad, in turn, took 4 h to traverse until the construction of a seriesof tunnels in the early 2000s cut travel time to 2 h. This recentregional history contrasts with patterns prevalent in the BrazilianAmazon, where road construction in the 1980s facilitatedcolonization and land use change at that time (Michalski et al.,2010).

Cultivation of coca (Erythroxylum sp.) for cocaine processing hasbeen an important driver of economic and land use transformationin Guaviare since the 1990s. Mechanized and/or industrial agricul-ture (e.g., soybeans or oil palm) was rare during the study period(H. Restrepo, pers. comm.).

By law, the forests of the San José-Calamar axis are part of theColombian Amazon reserve (Congreso de Colombia, 1958), and landtitling requires an act of congress to legalize the conversion fromstate to private property. In 1997, a 496,000-ha smallholder reservewas established along the axis with the interrelated goals of reduc-ing deforestation and discouraging land speculation (InstitutoColombiano de le Reforma Agraria, 1996). The non-tradable agricul-tural land units of the smallholder reserve were expected to stabi-lize the forest frontier (Ministerio de Agricultura y Desarrollo Ruraland Banco Mundial, 2005).

3.2. Land cover

We used existing land cover maps to analyze land cover andfragmentation dynamics. Details regarding the assignment andaccuracy of land cover classes can be found in Armenteras et al.(2013). Briefly, five land use classes were analyzed: (i) forests;(ii) pastures, (iii) burned areas; (iv) secondary forests, includingregenerating vegetation; and (v) coca cultivation as reported bythe United Nations Office on Drug and Crime (UNODC, Gobiernode Colombia, 2010). A sixth category comprising clouds, urbanareas, and roads was not included in analyses.

3.3. Fragmentation metrics

Given the potential impact of forest fragmentation on biodiver-sity (Barlow et al., 2012; Ewers and Didham, 2006), we investi-gated fragmentation dynamics in addition to land use change.Small, highly isolated Amazonian upland forest fragments areincompatible with the maintenance of biodiversity and ecosystemfunctions (Laurance et al., 2002, 2006). If the goal of stabilizing theforest frontier had been achieved by establishing a smallholder re-serve (Ministerio de Agricultura y Desarrollo Rural and Banco Mun-dial, 2005), then pasture expansion and forest fragmentationwould slow down over time.

We tracked the spatiotemporal trajectory of three land uses:forest, pasture, and coca cultivation. Three types of metrics werecalculated: (1) the percentage of landscape covered by class(PLAND)—which measures habitat loss and not fragmentation—;(2) measures of patch characteristics using patch number (NP)and the area-weighted average size of the fragments (PA); and 3)indices of isolation using the Euclidean nearest neighbor distance(ENN) and connectivity among patches, or cohesion. All metricswere calculated using Fragstats v. 3.3 (McGarigal et al., 2002). Adeclining land cover type will decrease in PLAND, and cohesion,exponentially increase its ENN, and at first increase and then de-crease in NP, as fragments first multiply and then vanish fromthe landscape. We modeled fragmentation statistics as a functionof PLAND by fitting equations derived from habitat fragmentationstudies encompassing a range of ecosystems including tropicallowland forests (Fahrig, 2003).

3.4. Variables explaining pasture expansion

To evaluate predictions from each of the hypotheses that ex-plain the growth of pastures, we compiled a database of demo-graphic and economic indicators for Guaviare and itsmunicipalities. The data comprised annual or monthly metricssummarized in Table 1.

3.4.1. Aggregate dataWe compiled annual data on Guaviare gross domestic product

(GDP) by sector (Vega Acevedo et al., 2011). To match predictionsregarding drivers of pasture expansion, we focused on three sec-tors: ranching, construction, and financial services. The ranchingvariable represented GDP for ranching productivity. The infrastruc-ture construction GDP was an indicator of investment in develop-ment through road construction, upgrade, and maintenance. Thefinancial services GDP encompassed financing, credit, and insur-ance. Financial services are expected to expand in response to eco-nomic development and can in turn accelerate economic growth(Levine et al., 2000). We interpreted the financial services GDP asa proxy for economic transactions between private parties. We ob-tained monthly data on beef prices at the closest documented mar-ket in Bogotá (Botiva León et al., 2012). To adjust the prices forinflation, we used the monthly general consumer price index(CPI) available from FAOstat (http://faostat3.fao.org/home/index.html).

3.4.2. Municipal annual dataTo examine the role of human population on land use, we ob-

tained rural and urban population data based on the 1993 and2005 censuses (Guerrero Lozano et al., 2009). The data were com-bined to estimate the proportion of urban population in eachmunicipality. Two economic variables were collected at the muni-cipal level: the cattle inventory based on site visits and ranchersurveys (Fedegan, 2011), and property taxes paid as an indicatorof real state values (Departamento Nacional de Planeación, 2012).Forced eradication of coca cultivation may explain changes in thisland use over time, so we collected data on aerial fumigation ofcoca plots from the Colombian National Police AntinarcoticsDirectorate.

3.5. Data analyses

Trends in aggregate annual and monthly data were examinedby plotting time series. All plotting and statistical analyses wereconducted in the R v. 2.15.2 statistical language (R DevelopmentCore Team, 2012). The relationships between fragmentationmetrics and land use were modeled using the lm command when

Table 1Description, coverage, and sources of proposed covariates of land use change in Guaviare. Unless otherwise noted, data were obtained for the 2000–2010 period.

Variable Description Units Source

Human population Based on 1993 and 2005 censuses Annual by municipality Guerrero Lozano et al. (2009)Cattle Standing cattle from cattle rancher census. Annual by municipality, 2001–2010 Fedegan (2011)Beef price Annual mean price of beef in closest

documented market (Bogotá)2000-constant value pesos/kg,monthly

Botiva León et al. (2012)

Gross domestic product(GDP) by sector

Economic activity in Guaviare by sector andsubsector

109 2005-constant values pesos Vega Acevedo et al. (2011)

Property taxes Taxes on property collected by localgovernment

Annual by municipality in constant2008 pesos

Departamento Nacional de Planeación(2012)

Coca cultivation Hectares of coca cultivation Hectares by year by municipalityfrom 2001 to 2011

UNODC and Gobierno de Colombia (2012)

Eradication Hectares of coca cultivation sprayed withherbicide

Hectares by year by municipalityfrom 1999 to 2011

Colombian National Police AntinarcoticsDirectorate (DIRAN)

L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73 67

fitting linear functions, and the nls command for non-linearfunctions.

We modeled pasture PLAND in northwestern Amazonia as afunction of demographic and economic variables to test the twohypotheses explaining pasture expansion. Coca cultivation hasbeen proposed as a major driver of Amazonian deforestation insouthern Colombia (Dávalos et al., 2011). To test the driver of cocadecline, we modeled coca cultivation as a function of aerial fumiga-tion the previous year.

3.5.1. Model selection for analyses of municipal dataMunicipal data were available for all three municipalities in

Guaviare (Table 1), providing finer spatial resolution for demo-graphic and economic variables. The main disadvantage of munici-pal analyses is the potential bias from non-modeled variationamong municipalities or among years within municipalities. Mod-eling the variation between clusters of observations by fitting clus-ter-specific parameters provides a statistically robust alternative toclassical regressions (Gelman and Hill, 2007).

A range of options may apply to these data (e.g., municipality-specific parameters with or without temporal correlation struc-ture), so we first conducted maximum likelihood analyses to iden-tify optimal combinations of explanatory variables and hierarchicalstructures in land use models. Variables were rescaled so that esti-mates of coefficients ranged from �100 to 100. All percentage vari-ables were expressed as 0–100 (instead of fractions), and heads ofcattle and hectares of coca or eradication were divided by 1000.Following Gelman (2005), we denote parameters estimated acrossall observations as constant across groups (equivalent to parame-ter estimates in classical regressions), and group-specific parame-ters as varying between groups. Three types of models werefitted: (1) generalized least squares (GLS) without group-specificparameters or correlation structures; (2) multilevel models withconstant coefficients for predictors and varying intercepts for eachmunicipality; and (3) as in 2, with a 1-year time-lag correlationstructure between annual observations. GLS models were fittedwith the gls function and multilevel models with the lme functionin the nlme v. 3.1-108 (Pinheiro et al., 2012) R package.

Models were compared using the Akaike Information Criterion(AIC): a model selection statistic that measures the fit of a modelto the observations. The best model fit has the lowest AIC. Fit sta-tistics were calculated using the anova.lme function in nlme.Parameters were then estimated using Bayesian inference, whichdirectly estimates the uncertainty of group-specific parametersand allows straightforward estimation of explained variance(Gelman and Pardoe, 2006).

3.5.2. Bayesian models of land useParameters of optimal models were estimated using the Bayes-

ian sampler Jags v. 3.1 (Plummer, 2003) and ran from within R

using R2jags v. 0.03-08 (Su and Yajima, 2012). Dependent variableswere modeled as linear combinations of explanatory variablesusing all observations. If the optimal model included municipal-ity-specific parameters, these were modeled as well. Non-informa-tive priors on all parameters are described in Section 4.3. Startingparameter estimates were random numbers, and each analysisran four independent Markov chains for 5000–10000 iterations.Model convergence was assessed using the potential scale reduc-tion factor (PRSF), which approaches 1.00 at parameter conver-gence, and is acceptable if <1.10 (Gelman and Rubin, 1992). Theproportion of the variance explained by the model was estimatedacross observations using the formulae described by Gelman andPardoe (2006).

4. Results

4.1. Land use and landscape trajectories

Pastures replaced forest and coca during the study period(Fig. 2A–C). Deforestation was high in the study area with forestloss of 38.3% (7121.84 ha). The mean annual forest loss rate was4.85% (SD = 6.5) between 2000 and 2009. The area of pastures in-creased from 10,800 ha in 2000 to 15,964 ha in 2009. We expectedgrowing area, decreasing isolation, and increased connectivity inpastures; and stable fragment sizes, isolation and connectivity forcoca in response to aggressive eradication. Coca cultivation in thesurvey plots declined from 689 to 182 ha.

To decouple change in land use from fragmentation, we mod-eled fragmentation statistics as functions of PLAND for forest, pas-ture, and coca cultivation (Table 2). Large forest patches wereprogressively fragmented as the forest PLAND declined. Pasturepatches grew and connectivity increased. Conversely, coca cultiva-tion had low coverage, and patches and connectivity decreased ascultivation declined. Fragmentation trends, most of which werestatistically significant for forest and coca land uses, are summa-rized in Fig. 2.

4.2. Covariates of land use change

By the end of the decade each of the municipalities had morepercent land in pasture and its population was more urban(Fig. 3). The expansion of pastures mirrored the growth of the cat-tle herd in all municipalities, with a full doubling of the cattleinventory in El Retorno (Fig. 4A). Inflation-adjusted beef prices inBogotá fluctuated considerably during the study period, and theirnet increase was �8% between 2000 and 2009 (Fig. 4B). Ranchingrevenues in Guaviare declined by �70% in the decade between2000 and 2010 (Fig. 4C).

Per-capita property taxes, reflecting land value, increased andwere consistent with the land-as wealth hypothesis (Supplementary

1400

1600

1800

2000

2200

05 54 04 53 03

Forest

NP

1500

2000

2500

28 30 32 34 36 38 40

Pasture

150200250300350400

5.2 0.2 5.1 0.1 5.0

Coca

500

550

600

650

700

05 54 04 53 03

PA (h

a)

150

200

250

300

350

28 30 32 34 36 38 40

2468

101214

5.2 0.2 5.1 0.1 5.0

90

100

110

05 54 04 53 03

PLAND

ENN

(m)

95100105110115120

28 30 32 34 36 38 40

PLAND

300

400

500

600

5.2 0.2 5.1 0.1 5.0

PLAND

BA

G

C

E FD

H I

253035404550

2001 2004 2007 201025

30

35

40

2001 2004 2007 2010

1

2

3

2001 2004 2007 2010

PLAN

D

K L M

Fig. 2. Percentage of land area (PLAND) and fragmentation statistics in relation to PLAND for each land use. PLAND for (A) forest; (B) pasture; and (C) coca cultivation.Relationships between fragmentation statistics and PLAND were fitted based on the equations of Fahrig (2003). Curve-fitting parameters are shown in Table 2. The x-axes ofthe forest and coca PLAND were reversed, and linear and exponential coefficients multiplied by �1, to reflect the decrease in both uses through time (see panels: (A and C)number of patches (NP) as a function of PLAND for (D) forest; (E) pasture; and (F) coca cultivation. Mean patch area (PA) as a function of PLAND for (G) forest; (H) pasture; and(I) coca cultivation. Mean Euclidean nearest-neighbor distance (ENN) as a function of PLAND for (K) forest; (L) pasture; and (M) coca cultivation.

68 L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73

Fig. 1). Investment in infrastructure, a potential driver of land values,also increased in Guaviare and mirrored variation in property tax re-ceipts from San José and Calamar (Supp. Fig. 1A and B). Land titlingdata were not available for Guaviare, however, transactions betweenprivate parties measured through the financial sector also increased(Supp. Fig. 1C).

4.3. Modeling results

The optimal model explaining the percentage land area in pas-ture only included cattle as a predictor (i.e., other variables did notincrease model fit enough to merit inclusion in the final model,Supplementary Table 1). This model did not include municipal-ity-specific parameters or a time correlation structure, and priorsfor both the linear intercept and coefficient were centered on 0with a variance of 104. The prior on the standard deviation of theobservations was set to a uniform distribution between 0 and100, appropriate for the variation expected in a percentage re-sponse variable. Chains ran for 10,000 samples, and PRSFs for allparameters were 61.001 after discarding the first half of the chainsare burn-in. The majority of the expansion of PLAND pasture wasexplained by increases in cattle (measured in thousand of headsin Table 3), as shown in Fig. 5A.

The best model to explain coca cultivation included eradicationthe previous year and percentage urban population, as well asmunicipality-specific intercept terms (Supplementary Table 2). Pri-ors for the overall intercept and eradication coefficients, as well asthe municipality-specific coefficients, were centered on 0 with avariance of 104. Precision on the prior on the percentage urbanpopulation was reduced, centering on 0 with variance of 105. Priorson the standard deviation of observations and the municipality-specific intercepts were uniform from 0 to 100. Chains ran for5,000 samples, and PRSFs for all parameters were 61.006 after dis-carding the first half of the chains are burn-in. Most of the decreasein coca cultivation (in thousands of hectares) was explained by in-creases in percentage urban population (Table 3). The coefficienton eradication the year before included 0, i.e., eradication did notinfluence coca cultivation in this time series. San José almost dou-bled the coca cultivation of the other two municipalities at thebeginning of the study period as shown in Fig. 5B, and estimatedin thousands of hectares in Table 3.

5. Discussion

The main findings of our analyses are: (1) landscape metricsconsistent with unimpeded fragmentation; (2) increases in the

Table 2Model parameters (standard error of the estimate), and significance for relationships between selected fragmentation statistics and the percentage land area under use cover(PLAND) shown in Fig. 2. The Euclidean nearest-neighbor distance (ENN) was modeled as: ENN ¼ aebPLAND, where a = intercept, and b = coefficient. NP = number of patches,PA = mean patch area.

Response variable Parameter Estimate (standard error) P-value

NPforest Intercept �5993 (3293) 0.128Coefficient PLANDforest 403.3 (170.3) 0.064

Coefficient PLAND2forest

�4.993 (2.148) 0.068

NPpasture Intercept �19,696 (13,364) 0.201Coefficient PLANDpasture 1320 (798.0) 0.159

Coefficient PLAND2pasture

�19.60 (11.73) 0.159

NPcoca Intercept 84.25 (101.8) 0.446Coefficient PLANDcoca 214.4 (144.2) 0.197

Coefficient PLAND2coca

�42.81 (42.33) 0.358

PAforest Intercept 102.8 (71.23) 0.199Coefficient PLANDforest 11.64 (1.686) 0.000

PApasture Intercept �389.0 (96.39) 0.007Coefficient PLANDpasture 18.48 (2.916) 0.001

PAcoca Intercept �2.170 (1.555) 0.212Coefficient PLANDcoca 4.851 (1.018) 0.003

ENNforest Intercept 173.3 (18.02) 0.000Coefficient PLANDforest �0.014 (0.002) 0.001

ENNpasture Intercept 128.9 (37.25) 0.013Coefficient PLANDpasture �0.072 (0.009) 0.445

ENNcoca Intercept 644.2 (57.00) 0.000Coefficient PLANDcoca �0.326 (0.074) 0.005

20253035

Calamar

30.032.535.037.540.042.5

Retorno

30

35

40

45

San Jose

4,000

4,250

4,500

4,750

5,000

4,000

6,000

8,000

30,000

32,500

35,000

37,500

40,000

5,000

6,000

7,000

8,000

2001 2004 2007 2010

11,200

11,400

11,600

2001 2004 2007 2010

19,000

19,500

20,000

20,500

2001 2004 2007 2010

BA

G

C

E FD

H I

Urb

anPo

pula

tion

Rur

alPo

pula

tion

PLAN

DPa

stur

e

Fig. 3. Percentage area (PLAND) covered in pasture and population by municipality. Population data between 2000 and 2005 are estimated based on the 1993 and 2005censuses, and projections after 2005 (Guerrero Lozano et al., 2009). PLAND for pasture in (A) Calamar; (B) Retorno; and (C) San José. Urban core population for (D) Calamar; (E)Retorno; and (F) San José. Rural population for (G) Calamar; (H) Retorno; and (I) San José.

L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73 69

number of cattle explain the expansion of pastures, but neitherprices nor revenues reflect growing demand for beef; and (3) erad-ication is not a covariate of decline in coca cultivation, a growingproportion of urban population is. We examine each of these find-ings in detail.

5.1. Land use and fragmentation trajectories

The declines of forests and coca cultivation and increase in pas-tures were consistent with the regional trends beyond the SanJosé-Calamar axis in all northwestern Amazonia (Armenteras

et al., 2013). Close to 5%, the local annual rate of forest loss is muchhigher than the regional mean of 61% (Armenteras et al., 2013),and high even when compared with other dynamic deforestationfronts (Armenteras et al., 2006; Dávalos et al., 2011). Forest lossis compounded by fragmentation: remaining forest patches wereon average 40% smaller and more isolated than in 2000 (Fig. 2).Forest fragmentation metrics revealed unimpeded fragmentationas forest was lost (at a = 0.1, Table 2, Fig. 2). The axis started outas a transition zone, in which modal land use changed from forestto non-forest, and became a settlement zone with the highestdeforestation rates of any spatial arrangement (Rodríguez et al.,

A

B

C

Calamar

El Retorno

San Jose

30,000

60,000

90,000

10

20

30

Year

Cat

tlePr

ice

of b

eef (

peso

s/Kg

) (1

09 pes

os)

2000 2002 2004 2006 2008 2010

1,600

1,800

2,000

2,200

Ran

chin

g G

DP

Fig. 4. Cattle ranching indicators for Guaviare. (A) Municipal bovine inventory instanding heads of cattle. (B) Monthly inflation-adjusted beef prices for the Bogotámarket in 2000 constant pesos. (C) Ranching gross domestic product (GDP) forGuaviare in 2005 constant 109 pesos.

Table 3Bayesian parameter estimates for optimal models relating land use to demographic and ecestimates correspond to the median of the posterior distribution. HPD = high probability d

Response variable Parameter

PLAND pasture InterceptCoefficient 103 heads of cattle

Coca cultivation (103 ha) InterceptCoefficient eradication (103 ha)Coefficient percentage population urbCalamar interceptRetorno interceptSan José intercept

A B

20

30

40

30,000 60,000 90,000

Cattle

Perc

enta

ge la

nd p

astu

re

2,0

4,0

6,0

Coc

a cu

ltiva

tion

(ha)

Fig. 5. Modeled relationships between percentage land area (PLAND) in pasture andcorrespond to optimal models described in Table 3. (A) PLAND pasture as a function of camaximum likelihood fits were significant (Supplementary Table 3).

70 L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73

2012). If left unchecked, the potential for biodiversity conservationin the remaining patches will deteriorate from increasing isolation,edge effects, and greater ease of further deforestation (Barlowet al., 2012; Laurance et al., 2006).

Coca cultivation declined and had little direct impact on forestcover. Álvarez (2003) proposed that the end state of areas initiallycolonized as smallholder coca plots was consolidated large pas-tures in the hands of a few, often absentee landholders. Land coverchanges were consistent with this model, but no data on land own-ership were available to evaluate the hypothesis of land concentra-tion (Ministerio de Agricultura y Desarrollo Rural and BancoMundial, 2005). Field data are needed to test this hypothesis, ascoca smallholders may over time become pasture holders them-selves. A 2005 smallholder survey revealed that 47% of the surfaceof coca farms in Guaviare and neighboring Meta was covered inpasture (UNODC, 2006).

5.2. Demand for beef is unrelated to pasture expansion

Connecting beef demand to pastures requires two links: frompastures to cattle, and from cattle to demand for beef (McAlpineet al., 2009). The first link is demonstrated with our data, headsof cattle strongly predict pasture PLAND across municipalities(Supplementary Tables 1 and 3, and Fig. 5A). Demand for beefdid not rise to a comparable extent: the cattle herd doubled from2001 to 2009, but prices fluctuated and their increase was smallat best (Fig. 4B). Even as the cattle herd grew, cattle ranching plum-meted as a contributor to Guaviare’s economy.

onomic variables in Guaviare municipalities. Model R2 is shown. Bayesian parameterensity estimate from the posterior distribution.

Bayesian estimate (95% HPD) R2

23.52 (18.804, 28.278) 0.550.206 (0.124, 0.285)

13.78 (2.19, 26.70) 0.64�0.100 (�0.229, 0.029)

an �0.219 (�0.296, �0.142)11.68 (8.18, 15.05)11.93 (8.83, 15.03)17.94 (12.85, 22.93)

MunicipalityCalamarEl RetornoSan Jose

00

00

00

30 40 50 60

Percentage population urban

cattle, and coca cultivation and percentage population in urban core. Fitted linesttle. (B) Coca cultivation as a function of percentage of population in urban core. All

L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73 71

There are three possible interpretations for the decline in ranch-ing revenues. Cattle ranchers may have traded their cattle inunderground markets, underreporting their contribution to GDP.For this unobserved market to explain ranching expansion, under-ground prices for standing beef would have to be higher than pricesrecorded in the Bogotá market. This would require that, relative toBogotá, there was greater demand or lower supply within Guavi-are. In 2010 Bogotá had a population of �7 million inhabitants,while San José had �60,000. Relative demand for beef is unlikelyto have been greater in San José than in Bogotá so that rancherscould sell their cattle for a better price in Guaviare. Conversely,per capita beef supply could not be larger in Bogotá than in SanJosé, particularly considering how much the herd expanded inGuaviare (Fig. 4A). I.e., an underground market for standing beefcannot explain the expansion of the cattle herd in Guaviare.

Another explanation for the loss of ranching revenue is that thevolume of cattle trading dropped: landholders could hold ontotheir cattle to strengthen their claims to the land or awaiting high-er prices for their herd. It was not just ranching that contracted inthe 2000s. The economic value of the general agricultural (includesranching) sector in Guaviare dropped by half from 2000 to 2009. Bycomparison, construction more than doubled, and transportationand communication more than tripled. Thus, the proportional con-tribution of ranching to GDP plummeted, while herds expanded.This also rejects the hamburger connection, since beef prices can-not drive the conversion to pasture without trading to produce rev-enue. Finally, it could be that both the exchange value of cattle andthe volume of trading declined. This is consistent with the GDP andbeef price data, and also rejects the hamburger connection becausebeef prices are not driving the growth of the cattle stock. Instead,cattle trading would have decreased as an adjustment to price fluc-tuations. In short, reasons other than the demand for beef mustdrive ranchers to transform forest to pastures and expand theircattle herds.

5.2.1. Use and exchange: the logic of livestock reduxThe question of why pastures expand and the cattle herd

grows as ranching revenues drop and/or beef prices fluctuate de-mands closer investigation. Although often considered inextrica-bly linked, pastures and cattle need not tightly covary ifpastures are a means to the end of obtaining legal titles. Afterall, if ‘‘improvements’’ to the land are all that is needed to claimtitle (Mendelsohn, 1994), why the additional investment in cat-tle? And yet, in Amazonia, even pastures directly associated withspeculative land grabs have cattle (Bowman et al., 2012). The lo-gic of livestock offers an explanation to why cattle and land spec-ulation are so tightly linked.

To understand the logic of livestock, we must first contrast theuse and exchange values for cattle and land (Hecht, 1985, 1993).The hamburger connection emphasizes the exchange value of cat-tle (e.g., for beef in domestic and international markets), but cattlealso have value in use. Livestock can be used to buttress landclaims (Hecht, 1993), provide a safe investment that crops do not(Rudel and Horowitz, 1993), and in parts of Amazonia can be usedas collateral on bank loans (Murphy et al., 1997). In Colombia, thesame trade guild that conducts the cattle census (Fedegan) hasagreements in place with the BBVA—a Spanish-Argentinianconglomerate that became the most important shareholder of theBanco Ganadero (Cattle-Ranching Bank) in 1996—to providelow-interest rotating credit lines and credit cards to its affiliates(Fedegan, 2013). Finally, the value of cattle can convert from useto exchange by selling the cattle when necessary.

5.2.2. Use and exchange of land without legal titlesAt the core of the land-as-wealth hypothesis is the exchange va-

lue of the land. The land is worth holding, even if its present value in

use is low, because its exchange value will eventually be high(Hecht, 1985). Proxies of property value and transactions were ex-pected to rise with pasture expansion, and both property taxes andfinancial GDP did (Fig. 5). But property taxes are indicators of thevalue of urban, not rural properties. Taxation on rural propertiesin Colombia is so rare that the average landholder pays the equiva-lent of a single bottled drink in property taxes each year(Kalmanovitz, 2006). It is the urban properties, then, that are payingmore taxes. Financial services increased, but GDP from real estatetransactions (not shown) remained static at 5 � 109 2005 pesos/year (Vega Acevedo et al., 2011). I.e., the volume of private transac-tions increased, but this increase is not driven by recorded real es-tate transactions.

Outside of its urban cores the San José-Calamar axis is a nomi-nal forest reserve (Congreso de Colombia, 1958), so land titles can-not be obtained without officially subtracting from this reserve.Despite the lack of legal titles, real estate transactions may occurthrough informal tenure systems. A coca smallholder survey foundthat 77% of local smallholders regarded farms as their own(UNODC, 2006). Parties to transactions involving untitled land of-ten notarize agreements to secure some standing of their claimto the land (Carlos A. Salazar, pers. comm.). There may have beenchanges in the volume of informal real state transactions that werenot captured by the real estate GDP.

The expectation of future land titles may drive conversion topasture, as happened in 1980s Brazil (Fearnside, 2005; Hecht,1993). Subtraction from the forest reserve has been the primarymeans of legalizing land in Colombian Amazonia (Arcila et al.,1999; Arcila and Salazar, 2011). Smallholders know that clearingresults in informal tenure (UNODC, 2006), and presumably futuretitles. The notion that landholders anticipate future titling is sug-gestive, but it does not explain why the last decade appearedpromising enough to clear the land and add cattle even as the ruralshare of the population dwindled (Fig. 3).

5.2.3. Road improvements and cattleWe propose that improvements to both the San José-Calamar

and Bogotá-Villavicencio roads have promoted urbanization, forestconversion, and cattle ranching in Guaviare. Since cattle hold valuein use and can help establish land claims, pastures have beenstocked with cattle despite stable beef prices and a declining con-tribution to the Guaviare economy. Three lines of evidence showthat Guaviare is urbanizing as its roads improve: (1) the populationis increasingly urban, (2) it relies less on rural sources of revenuesuch as ranching and coca agriculture, and (3) the urban tax-payingbase enables financing local infrastructure development.

An important portion of Guaviare’s infrastructure GDP has beendevoted to improvement and maintenance of the road that con-nects San José to Calamar, and the central government has im-proved the roads that connect Bogotá to San José (Arcila andSalazar, 2007, 2011; Instituto Nacional de Vías, 2010). Similartrends in property taxes and construction suggest taxes are leviedto pay back the investment, invested directly, or both investmentand taxes were planned in advance to enhance local development.Those changes in infrastructure were not captured at the 100-mresolution of our land use analyses, but they do change prospectsfor landholders that currently have tenure of the land and lack le-gal titles.

Urbanization can promote deforestation by connecting land-holdings to local markets and/or through land appreciation (Parryet al., 2010). The San José-Calamar axis is more integrated to mar-kets thanks to improving roads, and this increases the value of ur-ban properties. Most study plots are close enough to each of thethree town cores to become integrated into local markets if reliablyconnected by road (Fig. 1). Since those who clear land have a muchstronger claim to the land than those who do not, land appreciation

72 L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73

driven by urbanization and road improvement can promote pas-tures (and cattle) without increasing ranching revenues or beefprices.

5.3. Urbanization and access excludes coca, eradication does not

Two key findings in analyses of coca cultivation and rural devel-opment in the 2000s were: (1) coca does not expand where there ishigh road accessibility, and (2) municipalities with high percent for-est cover tend to attract more coca cultivation than already-defor-ested ones (Dávalos et al., 2011). Both investment ininfrastructure and the local decline in forest cover signal theimpending closure of the forest frontier in the San José-Calamaraxis. Rather than reflect the success of the eradication campaign,the decline of coca cultivation in Guaviare is in line with regionaleconomic shifts away from subsistence agriculture. One of the sig-natures of this closing frontier is the decline of coca. This does notimply that deforestation and road construction drive the declineof coca cultivation. Rather, the connection of land to marketsthrough road access makes properties both more accessible to lawenforcement and remaining forest easier to clear. The relationshipbetween percent urban population and coca cultivation capturesthe decline of coca as an economic activity in these increasingly ur-ban municipalities (Fig. 5B).

6. Conclusion

By showing that pastures and cattle expand without greater de-mand for beef, our analyses clarify the forces underlying deforesta-tion, fragmentation, and biodiversity loss. These results can guidepolicy toward effective outcomes, based on the actual drivers of for-est loss. Analyses of the impact of market integration and urbangrowth on tropical deforestation have focused on economically pro-ductive land uses and far-flung markets (DeFries et al., 2010; Rudelet al., 2009). In northwestern Amazonia, the availability of creditraises the value of cattle in use, and local urbanization and develop-ment raises the exchange value of land. As a result, the footprint ofinterconnection and urbanization has been rapid conversion to pas-tures. The forest reserve has been utterly ineffective: the absence oftitles has not stopped conversion to pasture. Instead, future sustain-able economic development requires the appreciation of standingforests, for example by granting equal standing to (limited) landclaims on forested and cleared lands, or through direct payments,is necessary to conserve forests and their ecosystem functions inrapidly urbanizing Amazonian frontiers.

Acknowledgements

The United Nations Office on Drugs and Crime and its SistemaIntegrado de Monitoreo de Cultivos Ilícitos in Colombia made this re-search possible by hosting the first researcher network meeting inFebruary 2012, and through agreements with LMD and DA. CarolFranco helped us generate Fig. 1. Carlos Salazar from Instituto Sinchiand Hernando Restrepo (Guaviare Secretary of Agriculture) providedlocal information; Claudia Romero Schadegg and Ismael Zuñiga fromFedegan made the cattle inventory available; Leonardo Correa andMarye Saenz from SIMCI made coca cultivation and eradication dataavailable; and Heather Lynch helped with Jags. Angelique S.P. Cor-thals, Samantha DelSerra, Omar M. Warsi, and Laurel R. Yohe madehelpful comments on earlier versions of the manuscript.

Appendix A. Supplementary material

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.biocon.2013.12.018.

References

Álvarez, M.D., 2003. Forests in the time of violence: conservation implications of theColombian war. J. Sustain. Forest. 16, 49–70.

Anonymous, 2010. Diagnóstico del sector en el mundo y punto de partida ydiagnóstico del sector en Colombia: Sector: Carne bovina, AT Kearney, pp. 1–275.

Arcila, O.H., Gonzalez, G.I., Salazar, C.A., 1999. Guaviare: población y territorio.Tercer Mundo Editores, Bogotá.

Arcila, O.H., Salazar, C.A., 2007. Sur del Meta: Territorio Amazónico. InstitutoAmazónico de Investigaciones Científicas, Bogotá.

Arcila, O.H., Salazar, C.A., 2011. La Amazonia colombiana: poblada y urbanizada.Colombia Amazónica 4, 37–55.

Ariza, E., Ramirez, M.C., Vega, L., 1998. Atlas cultural de la Amazonía colombiana: laconstrucción del territorio en el siglo XX. Instituto Colombiano de Antropologíae Historia, Bogotá.

Armenteras, D., Rodriguez, N., Retana, J., 2013. Landscape dynamics in northwesternAmazonia: an assessment of pastures, fire and illicit crops as drivers of tropicaldeforestation. PLoS ONE 8, e54310.

Armenteras, D., Rudas, G., Rodriguez, N., Sua, S., Romero, M., 2006. Patterns andcauses of deforestation in the Colombian Amazon. Ecol. Ind. 6, 353–368.

Barlow, J., Parry, L., Gardner, T.A., Ferreira, J., Aragp

£o, L.E.O.C., Carmenta, R.,Berenguer, E., Vieira, I.C.G., Souza, C., Cochrane, M.A., 2012. The criticalimportance of considering fire in REDD+ programs. Biol. Conserv. 154, 1–8.

Botiva León, M.A., Arciniegas Riveros, E., Velásquez Chaverra, E., Arciniegas Muñoz,J.C., Arias Walteros, C.H., Ruiz Romero, A., Alvarado, L.F., Lozano de Manjares, M.,Pereira, C.A., Villamil, M.A., Díaz, L.E., Yucumá Castillo, Y., Hernández, N.J., DíazCovaleda, J.M., Pescador, M., Osuna Correa, F., Castillo, L., Rodríguez, E., Ordoñez,A., Romero, C., 2012. Anuario Estadístico del Sector Agropecuario 2011:Resultados Evaluaciones Municipales 2011, Ed. Grupo Sistemas deInformación, Ministerio de Agricultura y Desarrollo Rural, Bogotá, pp. 1–204.

Bowman, M.S., Soares-Filho, B.S., Merry, F.D., Nepstad, D.C., Rodrigues, H., Almeida,O.T., 2012. Persistence of cattle ranching in the Brazilian Amazon: a spatialanalysis of the rationale for beef production. Land Use Policy 29,558–568.

Congreso de Colombia, 1958. Ley 2 de 1959, ed. Congreso de Colombia, Bogotá,Colombia.

Dávalos, L.M., Bejarano, A.C., Correa, H.L., 2009. Disabusing cocaine: pervasivemyths and enduring realities of a globalised commodity. Int. J. Drug Policy 20,381–386.

Dávalos, L.M., Bejarano, A.C., Hall, M.A., Correa, H.L., Corthals, A., Espejo, O.J., 2011.Forests and drugs: coca-driven deforestation in tropical biodiversity hotspots.Environ. Sci. Technol. 45, 1219–1227.

DeFries, R.S., Rudel, T., Uriarte, M., Hansen, M., 2010. Deforestation driven by urbanpopulation growth and agricultural trade in the twenty-first century. NatureGeosci. 3, 178–181.

Departamento Nacional de Planeación, 2012. Ejecuciones Presupuestales de losMunicipios 2000–2010. Departamento Nacional de Planeación, Bogotá.

Ewers, R.M., Didham, R.K., 2006. Confounding factors in the detection of speciesresponses to habitat fragmentation. Biol. Rev. 81, 117–142.

Fahrig, L., 2003. Efects of habitat fragmentation on biodiversity. Annu. Rev. Ecol.Evol. Syst. 34, 487–515.

Fearnside, P.M., 2005. Deforestation in Brazilian Amazonia: history, rates, andconsequences. Conserv. Biol. 19, 680–688.

Fedegan, 2011. Inventario Nacional Bovino a Nivel Municipal. FederaciónColombiana de Ganaderos, Bogotá.

Fedegan, 2013. Tarjeta Agroinsumos BBVA-Fedegan.Gelman, A., 2005. Analysis of variance – why it is more important than ever. Ann.

Stat. 33, 1–31.Gelman, A., Hill, J., 2007. Data Analysis Using Regression and Multilevel/hierarchical

Models. Cambridge University Press, Cambridge.Gelman, A., Pardoe, I., 2006. Bayesian measures of explained variance and pooling in

multilevel (hierarchical) models. Technometrics 48, 241–251.Gelman, A., Rubin, D.B., 1992. Inference from iterative simulation using multiple

sequences. Stat. Sci. 7, 457–472.Gomes, C.V.A., Perz, S.G., Vadjunec, J.M., 2012. Convergence and contrasts in the

adoption of cattle ranching: comparisons of smallholder agriculturalists andforest extractivists in the amazon. J. Latin Am. Geogr. 11, 99–120.

Guerrero Lozano, B., Cifuentes Noyes, M., Sardi Perea, E., Pachón, A.E., CubillosRafael, Leonel Gutiérrez, Montenegro, L.A., Rucinque, D.N., Daza, C.E., Vargas,E.M., Cabezas, J., Rubriche, D., Ospina, M.F., Hincapié, C., Roldán, J.J., Pinzón, J.M.,Castellanos, L.E., 2009. Metodologia proyecciones de población y estudiosdemográficos, Departamento Nacional de Estadística, Bogotá, pp. 1–66.

Hecht, S.B., 1985. Environment, development and politics: capital accumulation andthe livestock sector in Eastern Amazonia. World Dev. 13, 663–684.

Hecht, S.B., 1993. The logic of livestock and deforestation in amazonia. Bioscience43, 687–695.

Houghton, R.A., Lawrence, K.T., Hackler, J.L., Brown, S., 2001. The spatial distributionof forest biomass in the Brazilian Amazon: a comparison of estimates. Glob.Change Biol. 7, 731–746.

Instituto Colombiano de le Reforma Agraria, 1996. Resolución No. 041 de 1996:Determinación de extensiones para las UAFs, ed. Instituto Colombiano de laReforma Agraria. Diario Oficial, Bogotá, Colombia.

Instituto Nacional de Vías, 2010. Departamento del Guaviare. Instituto Nacional deVías de Colombia, Ministerio de Transporte, pp. 1–23.

L.M. Dávalos et al. / Biological Conservation 170 (2014) 64–73 73

Kaimowitz, D., 1997. Factors determining low deforestation: the Bolivian Amazon.Ambio 26, 536–540.

Kaimowitz, D., Mertens, B., Wunder, S., Pacheco, P., 2004. Hamburger ConnectionFuels Amazon Destruction, CIFOR, Bogor, pp. 1–10.

Kalmanovitz, S., 2006. El Impuesto Predial. In: El Espectador. Bogotá.Lambin, E.F., Turner, B.L., Geist, H.J., Agbola, S.B., Angelsen, A., Bruce, J.W., Coomes,

O.T., Dirzo, R., Fischer, G., Folke, C., George, P.S., Homewood, K., Imbernon, J.,Leemans, R., Li, X., Moran, E.F., Mortimore, M., Ramakrishnan, P.S., Richards, J.F.,Skånes, H., Steffen, W., Stone, G.D., Svedin, U., Veldkamp, T.A., Vogel, C., Xu, J.,2001. The causes of land-use and land-cover change: moving beyond the myths.Global Environ. Change 11, 261–269.

Laurance, W.-F., Lovejoy, T.-E., Vasconcelos, H.-L., Bruna, E.-M., Didham, R.-K.,Stouffer, P.-C., Gascon, C., Bierregaard, R.-O., Laurance, S.-G., Sampaio, E., 2002.Ecosystem decay of Amazonian forest fragments: a 22-year investigation.Conserv. Biol. 16, 605–618.

Laurance, W.F., Nascimento, H.E.M., Laurance, S.G., Andrade, A., Ribeiro, J.E.L.S.,Giraldo, J.P., Lovejoy, T.E., Condit, R., Chave, J., Harms, K.E., D’Angelo, S. 2006.Rapid decay of tree-community composition in Amazonian forest fragments. In:Proceedings of the National Academy of Sciences, USA, vol. 103, pp. 19010–19014.

Levine, R., Loayza, N., Beck, T., 2000. Financial intermediation and growth: causalityand causes. J. Monetary Econ. 46, 31–77.

McAlpine, C.A., Etter, A., Fearnside, P.M., Seabrook, L., Laurance, W.F., 2009.Increasing world consumption of beef as a driver of regional and globalchange: a call for policy action based on evidence from Queensland (Australia),Colombia and Brazil. Global Environ. Change 19, 21–33.

McGarigal, K., Cushman, S.A., Neel, M.C., Ene, E., 2002. FRAGSTATS: Spatial PatternAnalysis Program for Categorical Maps. University of Massachusetts, Amherst.

Mendelsohn, R., 1994. Property Rights and Tropical Deforestation. Oxford EconomicPapers, vol. 46, pp. 750-756.

Michalski, F., Metzger, J.P., Peres, C.A., 2010. Rural property size drives patterns ofupland and riparian forest retention in a tropical deforestation frontier. GlobalEnviron. Change 20, 705–712.

Ministerio de Agricultura y Desarrollo Rural, Instituto Interamericano deCooperación para la Agricultura, Banco Mundial, 2005. Marco normativo delas Zonas de Reserva Campesina: legislación vigente. In: Proyecto Piloto. Zonasde Reserva Campesina. Bogotá, pp. 1–51.

Murphy, L., Bilsborrow, R., Pichón, F., 1997. Poverty and prosperity among migrantsettlers in the Amazon rainforest frontier of Ecuador. J. Dev. Stud. 34, 35–65.

Myers, N., 1981. The hamburger connection: how Central America’s forests becameNorth America’s hamburgers. Ambio 10, 3–8.

Nepstad, D.C., Stickler, C.M., Almeida, O.T., 2006. Globalization of the amazon soyand beef industries: opportunities for conservation. Conserv. Biol. 20, 1595–1603.

Parry, L., Peres, C.A., Day, B., Amaral, S., 2010. Rural–urban migration bringsconservation threats and opportunities to Amazonian watersheds. Conserv.Lett. 3, 251–259.

Pinheiro, J., Bates, D., DebRoy, S., Sarkar, D., The R Core Team, 2012. nlme: Linear andNonlinear Mixed Effects Models.

Plummer, M., 2003. JAGS: A program for analysis of Bayesian graphical modelsusing Gibbs sampling. In: Proceedings of the 3rd International Workshop onDistributed Statistical Computing. Technische Universität Wien, Vienna,Austria.

R Development Core Team, 2012. R: A language and environment for statisticalcomputing. R Foundation for Statistical Computing, Vienna, Austria.

Rincón-Ruiz, A., Kallis, G., 2013. Caught in the middle, Colombia’s war on drugs andits effects on forest and people. Geoforum 46, 60–78.

Rincón-Ruiz, A., Pascual, U., Flantua, S., 2013. Examining spatially varyingrelationships between coca crops and associated factors in Colombia, usinggeographically weight regression. Appl. Geogr. 37, 23–33.

Rodríguez, N., Armenteras, D., Molowny-Horas, R., Retana, J., 2012. Patterns andtrends of forest loss in the colombian guyana. Biotropica 44, 123–132.

Rudel, T.K., Defries, R., Asner, G.P., Laurance, W.F., 2009. Changing drivers ofdeforestation and new opportunities for conservation. Conserv. Biol. 23, 1396–1405.

Rudel, T.K., Horowitz, B., 1993. Tropical Deforestation: Small Farmers and LandClearing in the Ecuadorian Amazon. Columbia University Press, New York.

Simmons, C.S., 2004. The political economy of land conflict in the eastern BrazilianAmazon. Ann. Assoc. Am. Geogr. 94, 183–206.

Su, Y.-S., Yajima, M., 2012. R2jags: A Package for Running Jags from R, p. Using thisPackage to Call Jags from R.

UNODC, 2006. Características agroculturales de los cultivos de coca en Colombia.Sistema Integrado de Monitoreo de Cultivos Ilícitos, Bogotá, pp. 1–116.

UNODC, Gobierno de Colombia, 2010. Censo de cultivos de coca 2009, In Monitoreode Cultivos de Coca. UNODC, Bogotá, pp. 1–110.

UNODC, Gobierno de Colombia, 2012. Colombia: Censo de Cultivos de Coca 2011.UNODC, Bogotá, pp. 1–112.

Van Ausdal, S., 2009. Pasture, profit, and power: an environmental history of cattleranching in Colombia, 1850–1950. Geoforum 40, 707–719.

Vega Acevedo, A.V., Lozano, J., Mora Zambrano, H., Aldana Castellanos, D.G., CruzZamora, A.M., Garcia Marin, L.M., Herrera Barrera, A., Parrado Cortes, D.R.,Tarazona Camacho, M., 2011. Metodología Cuentas Departamentales Base 2005.In: Gomez Vallejo J.E. (Ed.), Departamento Administrativo Nacional deEstadistica, Dirección de Síntesis y Cuentas Nacionales DSCN, Bogota, pp. 1–64.

Young, K.R., 1996. Threats to biological diversity caused by coca/cocainedeforestation. Environ. Conserv. 23, 7–15.