You are on page 1of 9

ARTICLE

pubs.acs.org/est

Forests and Drugs: Coca-Driven Deforestation in Tropical


Biodiversity Hotspots
Liliana M. Davalos*
Department of Ecology and Evolution and Consortium for Inter-Disciplinary Environmental Research, SUNY Stony Brook,
650 Life Sciences Building, Stony Brook, New York 11794-5245, United States

Adriana C. Bejarano
Department of Environmental Health Sciences, Public Health Research Center 401, University of South Carolina, 921 Assembly Street,
Columbia, South Carolina 29208, United States

Mark A. Hall
Department of Ecology and Evolution, SUNY Stony Brook, 650 Life Sciences Building, Stony Brook, New York 11794-5245,
United States

H. Leonardo Correa
Sistema Integrado de Monitoreo de Cultivos Ilícitos, United Nations Office on Drugs and Crime, Calle 102 no. 17A-61,
Bogota, Colombia

Angelique Corthals
Department of Sciences, John Jay College of Criminal Justice (CUNY), 899 Tenth Avenue, New York, New York 10019,
United States

Oscar J. Espejo
Sistema Integrado de Monitoreo de Cultivos Ilícitos, United Nations Office on Drugs and Crime, Calle 102 no. 17A-61,
Bogota, Colombia

bS Supporting Information
ABSTRACT: Identifying drivers of deforestation in tropical biodiversity hotspots is critical to assess threats to particular ecosystems
and species and proactively plan for conservation. We analyzed land cover change between 2002 and 2007 in the northern Andes,
Choco, and Amazon forests of Colombia, the largest producer of coca leaf for the global cocaine market, to quantify the impact of
this illicit crop on forest dynamics, evaluate the effectiveness of protected areas in this context, and determine the effects of
eradication on deforestation. Landscape-level analyses of forest conversion revealed that proximity to new coca plots and a greater
proportion of an area planted with coca increased the probability of forest loss in southern Colombia, even after accounting for other
covariates and spatial autocorrelation. We also showed that protected areas successfully reduced forest conversion in coca-growing
regions. Neither eradication nor coca cultivation predicted deforestation rates across municipalities. Instead, the presence of new
coca cultivation was an indicator of municipalities, where increasing population led to higher deforestation rates. We hypothesize
that poor rural development underlies the relationship between population density and deforestation in coca-growing areas.
Conservation in Colombia’s vast forest frontier, which overlaps with its coca frontier, requires a mix of protected areas and strategic
rural development to succeed.

’ INTRODUCTION global extinctions. Quantifying forest loss in the hotspots is


A substantial portion of the world’s biodiversity is located in therefore critical to plan for conservation at all spatial scales.3
hotspots, regions harboring a disproportionate number of ende-
mic species.1 These hotspots predominantly encompass extremely Received: July 13, 2010
biodiverse and increasingly threatened tropical forest ecosystems.2 Accepted: December 20, 2010
The high concentration of unique species in tropical forest hot- Revised: December 15, 2010
spots increases the odds that local disturbances translate into Published: January 11, 2011

r 2011 American Chemical Society 1219 dx.doi.org/10.1021/es102373d | Environ. Sci. Technol. 2011, 45, 1219–1227
Environmental Science & Technology ARTICLE

But documenting deforestation patterns is not enough: identify- and measuring the effect of current conservation policies in this
ing local and regional drivers of forest loss is indispensable to context.
address the causes of biodiversity loss, and plan for conservation
in a proactive manner.4 ’ EXPERIMENTAL SECTION
Although it is hard to synthesize global deforestation patterns
and uncover what drives these processes,5 agricultural markets, Remote Sensing and Land Cover. We used land cover maps
resource booms, and roads have been closely linked to increases generated to detect coca cultivation in 2002 and 2007 to quantify
in deforestation.6 These markets and the roads built to supply forest cover dynamics in Colombia. The illicit crop monitoring
them are, in turn, linked to changes in policy regimes and system of Colombia (SIMCI), used Landsat 7 Enhanced Thematic
demand for agricultural products, fossil fuels, or other natural Mapper Plus (ETMþ), supplemented with Aster, SPOT, and
resources.7,8 At the same time, a critical question for implement- IRS-LISS III images to assemble the 2002 and 2007 land cover
ing biodiversity conservation in the tropics is whether and how maps. The images were preprocessed over SIMCI’s planimetric
conservation policy affects deforestation.9 Because the effects of georeferencing layers for Colombia, and modified to remove
markets, as well as development and conservation policies are topographic and terrain distortions.29 To standardize the digital
specific to the environmental and socioeconomic context of each and visual interpretation, the visual images were corrected for
country, region, and location,10,11 analyses of their impact on topography using ERS radar images at 20-m resolution. Land
biodiversity hotspots can provide the level of detail necessary to cover data were extracted from the images through visual inter-
take action and reduce immediate drivers of deforestation. pretation and supervised multispectral classification using PCI
In this paper, we analyze forest cover over the past decade to Geomatics software.
elucidate local and regional drivers of forest loss in Colombia. We To validate the classification of these land cover maps,
focus on Colombia because it encompasses two of the most particularly as it applied to coca cultivation, SIMCI conducted
distinctive global biodiversity hotspots, the tropical Andes and helicopter or small aircraft reconnaissance flights approximately
the Choco,1 and ∼12% of the Amazon forest.12 Over the last 20 every 10 km in 4 types of areas: (1) ∼75% of the areas where coca
years the pace of deforestation in Colombia has accelerated,8,13,14 has been grown historically, (2) sites where authorities have
particularly in lowland forests,15,16 even as demographic pres- reported new coca cultivation, (3) areas where eradication by
sures have eased and the proportion of the population dependent aerial fumigation has been reported, and (4) areas with dense
on agriculture has declined.17 Because this period largely overlaps cloud cover in remote sensing images. Georeferenced photo-
with the explosion of coca cultivation for the cocaine market, graphs collected from these surveillance flights were then used
when Colombia went from growing 10% of the global coca to calibrate and validate land cover classification, and in particular
production in 1987 (220 km2) to 74% in 2000 (1633 km2),18 to verify the presence of coca in a given area (as opposed to other
several analyses have suggested coca cultivation directly and shrubby crops). Coca grown in the shade cannot be detected
indirectly drives deforestation in Colombia’s forested using these surveys.
frontier.15,19,20 Eradication by aerial spraying of herbicide is the On-the-ground truthing of remote sensing data was also
main and most widespread institutional response to the expan- conducted, but was more limited in scope, targeting only areas
sion of coca cultivation.21 The eradication program itself could of new cultivation or specific survey sites of particular social30 or
contribute to deforestation by facilitating the degradation of ecosystem importance.31 The aerial and on-the ground observa-
forest remnants,22 pushing growers out of targeted areas to new tions were then used to improve classification of remotely sensed
lands making colonization and deforestation more dynamic.18,23 data. Clouds, shadows and gaps in the remote sensing images
The ultimate driver of coca cultivation and the government were classified as missing data and excluded from all analyses for
programs that attempt to suppress it is the global demand for both years.
cocaine.18 However, reliable data on temporal variation of global Land cover maps generated this way divided the country
demand are lacking24 and even order of magnitude estimates of into three regions: North, Central, and South, encompassing
the cash flow in this market are unreliable.25 Therefore, our >450,000 km2, and the entire range of natural forests in the
analyses focus on the supply for this global commodity market, a northern Andes and Choco biodiversity hotspots 1 and 34 pro-
likely proximate driver of deforestation. tected areas (Figure 2a). On the basis of these maps, we com-
Even as the pace of forest loss in Colombia has accelerated, the pared land cover in this time step using IDRISI Andes software.32
country has consolidated its biodiversity conservation policy Complete forest regeneration is not expected to occur during the
around a large network of protected areas.26 The effectiveness 5-year span of our analysis, can only occur from fallows to
of Colombian protected areas against deforestation has been secondary forest. To test the accuracy of the land cover classifica-
established,27 particularly in lowland Amazonian forests,19,28 but tion, we checked the transitions for regeneration from cleared
their ability to prevent or mitigate deforestation spurred by illicit areas to primary forest (<0.03% of pixels in each region), and
crops remains to be demonstrated. We analyzed forest cover these were excluded from subsequent analyses. No transitions
data for the 2002-2007 period in Colombia with three from water to forest were recorded, corroborating the accuracy of
objectives: (1) to determine the indirect effects of coca cultiva- most of the classification.
tion on deforestation, (2) to evaluate the effectiveness of Modeling Forest Cover Change and Deforestation Rates.
protected areas in avoiding deforestation in coca-growing To measure the effect of coca cultivation and conservation policy
regions, and (3) to examine the role of coca eradication in on forest cover, we modeled change in forest cover as a function
deforestation. We used landscape analyses to investigate the of a series of environmental and policy variables, including
first two objectives, and analyzed municipality deforestation distance to the closest coca field and the area of coca cultiva-
rates to examine the third objective. Our ultimate goal was to tion in a 1-km2 cell, protection status of the landscape cell, relief,
help guide biodiversity policy and management by uncovering accessibility, climate, and remaining forest cover (Table 1,
the impact of illicit crops and their eradication on deforestation Figure 1). To summarize 19 biologically important climate
1220 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227
Environmental Science & Technology ARTICLE

Table 1. Geoferenced Data, Spatial Resolution of Variables, and Data Sources Used in Modeling Land Cover Change and
Deforestation Rates
variable type variable description spatial resolution data source

illicit crop and eradication Euclidean distance to new coca fields found 90-m grid SIMCI
between 2002 and 2006
coca cultivation (ha/km2) downscaled to 1-km grid from SIMCI
30-m grid of coca cultivation
area eradicated by aerial spraying during time step (ha) municipality SIMCI, based on reports filed
by anti narcotics agency
conservation policy IUCN-ranked protected areas (categorical 0/1) 90-m grid ref 65
population population growth (inhabitants/km2) municipality DANE66
accessibility Euclidean distance to nearest primary or secondary road 90-m grid SIMCI
(including unpaved roads) in km
Euclidean distance to nearest navigable river in km 90-m grid SIMCI
climate principal component 1 of pca of 19 climate variables 1-km grid Worldclim 33
(precipitation-derived)
principal component 2 of pca of 19 climate variables 1-km grid Worldclim
(temperature-derived)
principal component 3 of pca of 19 climate variables 1-km grid Worldclim
(remaining orthogonal climate variation)
topography elevation 90-m grid ref 67
aspect 90-m grid processed from ref 67
slope 90-m grid processed from ref 67
remaining natural habitat percent remaining forest cover in 2002 1-km grid current study

variables comprising temperature and precipitation means, arising from spatial autocorrelation. Fitting separate intercepts or
extremes, and seasonality, 33 we obtained principal components slopes for different groups of municipalities would have a similar
(PC) by eigenvalue decomposition of the climate rasters in effect by accounting for unobserved variation through a random
ArcGIS v. 9.2.34 We used these PCs rather than the climate data effect of group assignments. Models were compared using the
as predictors in subsequent analyses to minimize collinearity AIC.35
among variables. PC1 comprised mostly precipitation and its All modeling steps were conducted in the R statistical language 41
seasonality, PC2 reflected mostly temperature and its varia- using the MASS library,42 and the ncf,43 geepack,44 ROCR,45 and
tion, and PC3 comprised the remainder of the variation in the nlme 46 packages. Details on the Experimental Section are presented
variables. in the Supporting Information.
We used the Akaike Information Criterion (AIC) to select
the best combination of predictors of forest cover change for ’ RESULTS
each region.35 Logistic regressions of change in forest cover as a
function of the predictors were fitted using both a generalized Land Cover Change between 2002 and 2007. The propor-
linear model (GLM) approach assuming independence in errors, tion of land cover classes and relative change are summarized in
and a generalized estimating equation approach (GEE), which Table 2. Natural vegetation, forest and scrub, was the most
included an autocorrelation structure for the errors of observa- prevalent land cover class, and most of it remained unchanged
tions within a 5  5 km box.36 The last approach was used to through 2007. Despite the general stability in land cover classes,
account for spatial autocorrelation in model fitting so that the Central region experienced dramatic losses in forest cover
hypothesis testing was not biased toward rejection of the null from 56% to 46% of its surface, and the South went from 82 to
hypothesis by virtue of the similarity of errors among observa- 78% forested. Substantial forest regeneration was only observed
tions close to one another. in the South, with 13% of crops and 10% of anthropogenic cover
All models were calibrated using a small subset of the observed reverting to forest.
changes, and validated using all the data 37 (Table S1). The The highest annual deforestation rate was recorded in the
accuracy of the models was assessed using the area under the North (4.70%) followed by the Central region (3.79%), and a
receiver operator characteristic curve or AUC, which ranges from much lower rate of forest loss in the South (0.81%). These annual
0 to 1, with 0.5 indicating a completely random model, and 1 rates mask large forest losses over the 5-year period: 14,322 km2
indicating a perfect model.38 lost in the South, a similar area of mostly Andean forests lost in
To measure the effect of eradication by aerial spraying the Central region (13,630 km2, Figure 2a), and 1,160 km2 lost in
(Figure 1F) we calculated annual municipal deforestation rates the North. The deforestation rate in the southern Choco was
using eq 7 from ref 39. Deforestation rates were modeled as a 0.98%, or 291 km2 of forest lost out of 6100 km2.
function of changes in the density of human population, coca Modeling Land Cover Change. The samples taken to cali-
cultivation, and eradication using multilevel linear models with brate models were very sparse relative to the data available, <1%
distance-based autocorrelation structures.40 Again, autocorrela- of the data in every case (Supporting Information, Table S1),
tion structures were included to remove bias in hypothesis testing resulting in some variation in the predictors that could be eliminated,
1221 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227
Environmental Science & Technology ARTICLE

Figure 1. Independent variables included as predictors in analyses of land cover change and deforestation rates. (A) New coca cultivation detected
between 2002 and 2007. (B) Relief map and permanently navigable rivers. The rivers layer was used to generate a distance surface and thus estimate
accessibility by river. (C) Primary and secondary road network, including nonseasonal unpaved roads. This network was used to generate a distance
surface and estimate accessibility by road. (D) First principal component (PC) derived from principal component analysis of 19 ecologically important
climate variables.33 (E) Proportion of remaining forest in 2002 at 1-km2 resolution. (F) Amount of aerial spraying conducted in each municipality
calculated as the mean annual number of ha sprayed per km2.

as well as the coefficients estimated across series of models the number of significantly autocorrelated residuals, and was not
(Supporting Information, Tables S2-S8). There was no signifi- pursued further. Figure 2 and Table 3 show the results from the
cant improvement in model fit from including the coca-related model in a series that minimized the proportion of significantly
variables in the Central region, or from including the protected autocorrelated residuals and was most appropriate for hypothesis
area variable in the North (Supporting Information, Table S2). testing.
Distance to roads and remaining forest cover were important Despite variability across samples some predictors were signifi-
predictors in every model, while elevation and aspect were elimi- cant in every model of a series (Table 3). The probability of
nated from most models (Supporting Information, Table S2). transition from forest to nonforest increased significantly with
There was significant spatial autocorrelation in the residuals of shorter distance to new coca plots, as well as with the amount of
logistic regressions assuming independent observations, indicat- new coca per km2 in the South region. Protected areas signifi-
ing that this approach was inadequate for determining the cantly decreased the probability of forest loss in the Central and
significance of predictors. Analyses using GEE reduced, but did South regions, but this effect was not consistent for all models in
not completely eliminate autocorrelation in the residuals (Figure S3). the series (Table 3 and Supporting Information Tables S4 and S5).
Increasing the clustering distance up to 100 km did not minimize Many of the predictors of forest loss were also predictors of forest
1222 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227
Environmental Science & Technology ARTICLE

Figure 2. Change in forest cover from 2002 to 2007. (A) Observed change, outline of IUCN-protected areas (national parks, sanctuaries, and biosphere
reserves), and forested areas discussed in the text: (1) Serranía del Perija, (2) Serranía de San Lucas, and (3) Pacific versant of Cordillera Occidental
(Choco forests). Observed changes were derived from direct comparison of land cover maps. (B) Modeled probability of deforestation based on region-
specific landscape models of forest change. (C) Modeled probability of reforestation, note overprediction for this change in the South. The probabilities
of change as a function of the observations in 2002 were obtained by applying the functions summarized in Table 3 to the predictors in 2002 (Table 1).

Table 2. Transition of Land Cover Classes for the 2002-2007 Perioda


region 2002V/2007f forest crops other natural vegetation anthropogenic % of total area

North forest 0.68 0.23 0.08 0.00 30


crops 0.20 0.68 0.11 0.01 10
other natural vegetation 0.02 0.03 0.94 0.00 58
anthropogenic 0.00 0.07 0.02 0.91 1
% of total area 24 16 59 1
Central forest 0.67 0.26 0.08 0.00 56
crops 0.24 0.68 0.08 0.00 34
other natural vegetation 0.08 0.16 0.75 0.01 10
anthropogenic 0.04 0.17 0.04 0.75 <1
% of total area 46 39 14 <1
South forest 0.93 0.05 0.02 0.00 82
crops 0.13 0.75 0.11 0.00 15
other natural vegetation 0.16 0.13 0.71 0.00 3
anthropogenic 0.10 0.36 0.06 0.49 <1
% of total area 78 16 6 <1
a
The anthropogenic category included buildings, roads (paved and unpaved), airstrips, (paved and unpaved), and any other anthropogenic land cover
classes excluding cultivation and pastures.

Table 3. Coefficients from GEE regression and performance of models with the fewest instances of significant spatial
autocorrelation in residuals.a

distance to climate
remaining coca protected
region coca road river forest (%) (ha/km2) PC1 PC2 PC3 aspect elevation slope area (binary) AUC

deforestation
North 0.05 -4.45* -0.27* -0.40*** -0.17 -1.09* -0.45* -0.08 -0.03*** 0.83
Central -3.03*** -0.19** -0.24*** -0.43*** 0.54*** -0.08 -0.02*** -0.80** 0.76
South -0.82*** -2.32*** -0.34** -0.60*** 0.16** -0.67*** 0.32** -0.90*** -0.16* -0.26* -0.03*** -0.78** 0.93
reforestation
North -1.98 -1.11*** -0.27 0.92
Central -2.52** -0.62*** -0.23*** -0.11 -0.25*** -0.15** -0.63 0.85
South -0.32 -2.54*** -0.33** -0.71*** 0.13 -0.09 -0.36*** -0.43*** -0.33** -17.36*** 0.96
a
* Significant at p < 0.05, ** significant at p < 0.01, *** significant at p < 0.001. Coefficients in bold were significant in every model of the series of 10 for
each data set.

1223 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227


Environmental Science & Technology ARTICLE

gain, reflecting more dynamic landscapes. The performance of


the models, summarized in Table 3 and Figure 2 ranged from fair
(>0.7 AUC) in the Central region to excellent (>0.9 AUC) in the
South.
Modeling Deforestation Rates. Rates of forest loss/gain
varied widely, from -1.13;or 113% annual deforestation in a
municipality in the Central region where ∼0.3% of the original
forest remained after the 5-year period, to >48% annual regen-
eration;where the forest grew to 11 times its original size;in a
municipality on the Amazonian versant of the Andes in the South
(Supporting Information, Figure S3a). The median municipal
deforestation rate was -0.025 (2.5% annual loss; std. dev. =
16.29%). Supporting Information, Figure S3 summarizes defor-
estation/reforestation rates by municipality and by 1-km2 pixel. Figure 3. Deforestation rate as a function of change in population
Eradication and coca cultivation were not significant predictors density in the South, fitted line shows relationship (P = 0.0001).
of deforestation/reforestation rates in any analyses (Supporting
Information, Table S9). Change in population density was a but the demographic and eradication data can help assess the role
significant covariate of deforestation rates when assuming an of immigration and aerial fumigation in deforestation.
independent sampling structure, but this effect was lost in the Coca cultivation had no effect on municipal deforestation rates
best models, which included a distance-based error correlation (Supporting Information, Table S9), despite the landscape-level
structure (Supporting Information, Table S9). These nonsigni- effects of coca on the probability of conversion (Table 3). It
ficant relationships between changes in population density and could be that the relationship between coca cultivation and
deforestation rates are summarized in Supporting Information, probability of deforestation is captured across the landscape,
Figure S4. Assuming independent sampling in municipal analyses whereas on the aggregate, migration and many forms of exploita-
resulted in highly autocorrelated residuals, and neither applying tion (not measured here) mediate the relationship between coca
distance-based correlation structures nor fitting intercepts for cultivation and deforestation rates (e.g., ref 48). This explanation
each region reduced residual autocorrelation (Supporting Informa- implies that even if coca cultivation attracts immigration, defor-
tion, Figure S5). estation rates arise from many other activities and are therefore
Separate single-level models with a Gaussian autocorrelation only weakly associated with coca cultivation.
structure for each region found no significant effect of either If coca attracts new growers who then convert the forest,
eradication or coca cultivation in any region (P > 0.2793), and population change should be positively linked to coca cultivation,
a significant effect of change in population density in the South and deforestation rates should be positively related to population
(P = 0.0001), but not in the North or Central region (P > 0.13). change in coca-growing areas. We modeled population change as
The South regional model had minimal residual autocorrelation a function of new coca cultivation across municipalities and
relative to both whole-country models and other regional models found a nonsignificant relationship (P = 0.3222 with all data, P =
(Supporting Information, Figure S5). Deforestation rates for the 0.2330 with data from municipalities that recorded new coca).
South based on the regional model are shown in Supporting There was some evidence of the link between deforestation rates
Information, Figure S5b. and population gains in coca-growing areas, since the only
significant relationship was found in the South where most coca
is grown (P = 0.0001; Figure 3). To evaluate this relationship
’ DISCUSSION
more broadly we reanalyzed data from municipalities where new
In this study we have shown that (1) coca cultivation increases coca was detected (not restricted to the South, N = 227), and
the probability of forest conversion in the northern Andes and found that deforestation rates increased with growing population
Choco of southern Colombia, (2) changes in population density density (P = 0.0005). In contrast, municipalities where no new
predict deforestation rates in this same region, and (3) protected coca was found (N = 267) showed no significant relation-
areas decrease the probability of forest loss, even after controlling ship between change in population density and deforestation
for accessibility by factoring distances to roads and rivers. rate (P = 0.3689). Although we found no evidence that coca
How Does Coca Cultivation Promote Deforestation? The cultivation attracts immigrants, our analyses indicate that popu-
forces driving deforestation result from complex interactions lation density and deforestation rates are linked in coca-growing
between policy decisions and socioeconomic processes as they municipalities. This relationship is not explained by coca cultivation
unfold in an environmental space more or less propitious to because coca was not associated with population density.
agriculture and other human activities. Considering the illegality As with coca cultivation, eradication in coca-growing munici-
of coca and local socioeconomic conditions, four nonexclusive palities could translate into higher deforestation rates because
mechanisms have been proposed to explain coca as a driver of eradication and law enforcement may result in relocation of coca
deforestation: (1) armed conflict associated with coca produc- growers and new clearings.18,20,22,49 As law enforcement pressure
tion and trafficking may drive growers away from existing crops increases, eradication would displace people,21 thus explaining
promoting further deforestation,20 (2) higher income from coca deforestation rates better than coca cultivation.49 However,
cultivation attracts new growers and drives existing growers to eradication had no effect on population density in coca-growing
expand their production,47,48 (3) eradication and law enforcement municipalities (P = 0.7004), or on deforestation rates (Supporting
force growers to relocate promoting further deforestation,20,49 Information, Table S9).
and (4) eradication may drive deforestation directly.22 Analyses Neither the amount of coca cultivation nor eradication is a
of the role of armed conflict are beyond the scope of this study, satisfactory explanation for the relationship between population
1224 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227
Environmental Science & Technology ARTICLE

density and deforestation rates in municipalities where coca is to the national economy through road construction, has been
expanding. Previous analyses to establish the characteristics of proposed as a way to curb coca cultivation (see Foreign Opera-
regions where coca expands have revealed that (1) the potential tions Congressional Budget requests at ref 55). Our landscape-
for expansion based on climate alone coca encompasses virtually level analyses, along with several recent studies,8,14,19,28 have
all lowland forests and some of the subtropical forest remnants in shown that access, particularly by road, is a strong covariate of
the country and (2) the prevalence of abject poverty and low deforestation. Development of coca-growing areas that improves
accessibility sets apart coca growing districts from other agricul- access to the frontier without alleviating the conditions that make
tural areas.50 Both household surveys and analyses above the level of forest conversion the main source of income has the potential to
municipalities have shown that socioeconomic conditions are poor greatly accelerate forest loss in the Central and South regions.
and living standards low for most coca growers 18,21,30 The patterns of forest conversion along the Colombia-Ecuador
In light of previous research, we hypothesize that what sets border suggest that orderly, development-driven colonization
coca-growing municipalities apart is poor rural development. can lead to the stabilization of the frontier with a much greater
Gains in rural population density relate to higher deforesta- proportion of remnant forest,14 and that the spontaneous
tion rates because most or all economic activities that absorb waves of colonization experienced in southern Colombia
immigrants, or used to occupy emigrants, require forest clearing. stabilize at a much lower proportion of forest.15,56 An alter-
Municipalities without new coca would have a diverse suite of native development scenario would be a more orderly coloni-
economic activities to accommodate population growth, so that zation process, with roads catalyzing deforestation along their
the relationship between population and deforestation breaks immediate vicinity, but larger tracts of forest remaining as
down. In our data, new coca cultivation and the existing road economic opportunities that do not convert the forest become
network are complementary, suggesting that coca does not available.57
expand in areas with substantial development (cf., Figure 1a Prospects for Biodiversity Conservation. Our results show
and c). The fact that coca, a crop that cannot be grown where that protected areas reduce the probability of forest conver-
government and law enforcement are active, is expanding in sion,19,27,28 not just by virtue of being in remote areas, and even in
these municipalities further points to their lack of socio-political coca-growing regions. An additional, even larger, effect would
and economic development. The expansion of coca itself is an follow from conserving a larger proportion of forest (Table 2).
indication that these municipalities constitute the agricultural Larger forest remnants would, in turn, enhance the prospects for
frontier, where settled land ends and new inroads begin. If so, conserving more endemic species than a mosaic where forest
these municipalities should have a greater proportion of their remnants are too small and isolated to support many species.58
surface in forest because socio-political integration and economic Expanding the protected area network to conserve large tracts of
development have produced massive forest loss in Colombian the most biodiverse forest remnants in the three study regions is
history.16 We investigated this prediction by modeling both warranted considering the high rates of deforestation observed
the presence of new coca and its quantity as functions of the over the past decade in the North and Central region, and
proportion of the municipality that remained in forest in 2002. southern Choco, and the historical trajectory toward stabilization
Both models were highly significant (N = 594, P < 2  10-16 for at very low forest cover in midelevations.59,60
the logistic regression, and P = 0.0011 for the linear model), Four of eight protected areas slated for official government
confirming the expectation that these municipalities are the protection in the 2009/2010 period encompass biodiversity
hitherto undeveloped forested frontier. hotspots in the Serranía del Perija, Serranía de San Lucas
Our municipal-level analyses suggest that the relationship (Figure 2a), dry forests in the easternmost extreme of the mid-
between coca cultivation and deforestation is more complex, Central region, and the Serranía del Pinche in the Choco (see
and the policy context in which landscape-level processes unfold http://www.parquesnacionales.gov.co/PNN/portel/libreria/php/
has more influence on eventual outcomes 48,49 than previously decide.php?patron=01.1103). On the basis of these immediate
proposed.18,20,22 We propose poor rural economic development expansion plans, lowland southern Choco forests would con-
signaled by the expansion of coca as the context underlying the tinue to be underrepresented in the protected area network.
relationship between population density and deforestation rates, Unlike some of the vast parks in Amazonian forests, these new
even in the North and Central region, where coca cultivation was protected areas would have to be responsive to already-established
not a covariate of the probability of deforestation at the landscape local populations and their current economy.61,62 Considering
level. Analyses of the economic covariates of coca have shown that structural underdevelopment would underlie the relation-
that proportionally larger rural populations and poverty predomi- ship between migration and deforestation rates in Colombia,
nate in coca-growing regions,51 bolstering this interpretation. support for a diverse smallholder economy is crucial to break the
Coca is expanding in these municipalities because they are cycle of environmental degradation and poverty that traps small
underdeveloped, rather than the converse. Coca is therefore a farmers throughout the agricultural frontier53,63 and simultaneously
symptom rather than the ultimate cause of deforestation, and conserve biodiversity beyond strictly protected areas.64 To this end,
structural features such as socioeconomic inequality, failed agri- one of the primary goals of conservation should be to articulate its
cultural development policies, and armed conflict are the large- priorities within a larger framework seeking to both promote and
scale drivers of deforestation.51-54 More data at levels ranging regulate rural development throughout the forest frontier.
from the household to the municipality are needed to test the
hypothesis that the drivers of deforestation in coca growing areas ’ ASSOCIATED CONTENT
are socioeconomic and linked to underdevelopment and possibly
conflict. bS Supporting Information. Detailed experimental section,
Aside from eradication, development, broadly construed to nine tables and five figures of results, and supporting references.
encompass projects to replace coca with alternative crops, This information is available free of charge via the Internet at
strengthen local institutions, and integrate coca-growing regions http://pubs.acs.org/.
1225 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227
Environmental Science & Technology ARTICLE

’ AUTHOR INFORMATION (19) Armenteras, D.; Rodríguez, N.; Retana, J. Are conservation
strategies effective in avoiding the deforestation of the Colombian
Corresponding Author Guyana Shield? Biol. Conserv. 2009, 142 (7), 1411–1419.
*E-mail: ldavalos@life.bio.sunysb.edu; phone: þ1 631 632 1554; (20) A  lvarez, M. D. Forests in the time of violence: conservation
fax: þ1 631 632 7626. implications of the Colombian war. J. Sustainable For. 2003, 16 (3-4),
49–70.
’ ACKNOWLEDGMENT (21) Dion, M. L.; Russler, C. Eradication efforts, the state, displace-
We thank Resit Akc- akaya for helpful discussion on drivers of ment and poverty: explaining coca cultivation in Colombia during Plan
Colombia. J. Latin Am. Stud. 2008, 40, 399–421.
land cover change, Jessie Stanton, Maria Uriarte and Charles
(22) Messina, J. P.; Delamater, P. L. Defoliation and the war on drugs
Yackulic for guidance on analyses in R, Jim Rohlf for insights on in Putumayo, Colombia. Int. J. Remote Sens. 2006, 27, 121–128.
spatial autocorrelation, Elizabeth Simola for helping get the project (23) Holmes, J. S.; Gutierrez De Pi~neres, S. A.; Curtin, K. M. Drugs,
started, and Eleonora Davalos and Leonardo Zurita for discussions violence, and development in Colombia: A department-level analysis.
on economic and social drivers of coca cultivation and deforestation. Latin Am. Polit. Soc. 2006, 48 (3), 157–184.
(24) Mejía, D.; Posada, C. E. Cocaine production and trafficking:
’ REFERENCES What do we know? Res. Work. Pap. 2008, 1, 1–62.
(25) Thoumi, F. E. The numbers game: Let’s all guess the size of the
(1) Myers, N.; Mittermeier, R. A.; Mittermeier, C. G.; da Fonseca,
illegal drug industry. J. Drug Issues 2005, 35 (1), 185–200.
G. A. B.; Kent, J. Biodiversity hotspots for conservation priorities. Nature
(26) Armenteras, D.; Gast, F.; Villareal, H. Andean forest fragmenta-
2000, 403 (6772), 853–858.
tion and the representativeness of protected natural areas in the eastern
(2) Laurance, W. F. Have we overstated the tropical biodiversity
Andes, Colombia. Biol. Conserv. 2003, 113 (2), 245–256.
crisis? Trends Ecol. Evol. 2007, 22 (2), 65–70.
(27) Bruner, A. G.; Gullison, R. E.; Rice, R. E.; da Fonseca, G. A. B.
(3) Gaston, K. J.; Jackson, S. F.; Cantu-Salazar, L.; Cruz-Pi~ non, G.
Effectiveness of parks in protecting tropical biodiversity. Science 2001,
The ecological performance of protected areas. Annu. Rev. Ecol. Evol.
291 (5501), 125–128.
Syst. 2008, 39 (1), 93.
(28) Etter, A.; McAlpine, C.; Wilson, K.; Phinn, S.; Possingham, H.
(4) Brooks, T. M.; Mittermeier, R. A.; da Fonseca, G. A. B.; Gerlach,
Regional patterns of agricultural land use and deforestation in Colombia.
J.; Hoffmann, M.; Lamoreux, J. F.; Mittermeier, C. G.; Pilgrim, J. D.;
Agric., Ecosyst. Environ. 2006, 114 (2-4), 369–386.
Rodrigues, A. S. L. Global biodiversity conservation priorities. Science
(29) Bauer, T.; Schneider, W. Illicit Crop Monitoring in Colombia:
2006, 313 (5783), 58–61.
Review of Methodology; Institute of Surveying, Remote Sensing and Land
(5) Asner, G. P.; Rudel, T. K.; Aide, T. M.; Defries, R.; Emerson, R. A
Information: Vienna, Austria, 2008; pp 1-25.
contemporary assessment of change in humid tropical forests. Conserv.
(30) UNODC Características Agroculturales De Los Cultivos De Coca
Biol. 2009, 23 (6), 1386–1395.
En Colombia; Sistema Integrado de Monitoreo de Cultivos ilícitos:
(6) Rudel, T. K. Tropical Forests: Regional Paths of Destruction and Bogota, Colombia, 2006; pp 1-116.
Regeneration in the Late Twentieth Century; Columbia University Press: (31) Gobierno de Colombia; UNODC; Parques Nacionales
New York, 2005; pp 1-131. Naturales de Colombia. Analisis Multitemporal De Cultivos De Coca En
(7) Laurance, W. F. Switch to corn promotes Amazon deforestation. Las A reas Del Sistema De Parques Nacionales Naturales: Periodo 2001-
Science 2007, 318 (5857), 1721b. 2004; UNODC: Bogota, Colombia, 2005.
(8) Armenteras, D.; Rudas, G.; Rodriguez, N.; Sua, S.; Romero, M. (32) Clark Labs. IDRISI Andes; Clark University: Worcester, MA,
Patterns and causes of deforestation in the Colombian Amazon. Ecol. 2006.
Indic. 2006, 6 (2), 353–368. (33) Hijmans, R. J.; Cameron, S. E.; Parra, J. L.; Jones, P. G.; Jarvis, A.
(9) Jha, S.; Bawa, K. S. Population growth, human development, and Very high resolution interpolated climate surfaces for global land areas.
deforestation in biodiversity hotspots. Conserv. Biol. 2006, 20 (3), 906–912. Int. J. Climatol. 2005, 25 (15), 1965–1978.
(10) Kaimowitz, D. Factors determining low deforestation: the (34) ESRI. ArcGIS Desktop, 9.2; Environmental Systems Research
Bolivian Amazon. Ambio 1997, 26 (8), 536–540. Institute, Inc.: Redlands, CA, 2006.
(11) Alvarez, N. L.; Naughton-Treves, L. Linking national agrarian (35) Burnham, K. P.; Anderson, D. R. Model Selection and Multi-
policy to deforestation in the Peruvian Amazon: A case study of model Inference; Springer-Verlag: New York, 2002.
Tambopata, 1986-1997. Ambio 2003, 32 (4), 269–274. (36) Dormann, C. F. Effects of incorporating spatial autocorrelation
(12) Sanderson, E. W.; Jaiteh, M.; Levy, M. A.; Redford, K. H.; into the analysis of species distribution data. Global Ecol. Biogeogr. 2007,
Wannebo, A. V.; Woolmer, G. The human footprint and the last of the 16 (2), 129–138.
wild. Bioscience 2002, 52 (10), 891–904. (37) Pontius, R. G.; Huffaker, D.; Denman, K. Useful techniques of
(13) Vi~na, A.; Cavelier, J. Deforestation rates (1938-1988) of validation for spatially explicit land-change models. Ecol. Model. 2004,
tropical lowland forests on the Andean foothills of Colombia. Biotropica 179 (4), 445–461.
1999, 31 (1), 31–36. (38) Pontius, R. G.; Batchu, K. Using the Relative Operating
(14) Vi~na, A.; Echavarria, F. R.; Rundquist, D. C. Satellite change Characteristic to Quantify Certainty in Prediction of Location of Land
detection analysis of deforestation rates and patterns along the Colombia- Cover Change in India. Trans. GIS 2003, 7 (4), 467–484.
Ecuador border. Ambio 2004, 33 (3), 118–125. (39) Puyravaud, J. P. Standardizing the calculation of the annual rate
(15) Etter, A.; McAlpine, C.; Phinn, S.; Pullar, D.; Possingham, H. of deforestation. For. Ecol. Manage. 2003, 177, 593–596.
Unplanned land clearing of Colombian rainforests: Spreading like (40) Bivand, R. Implementing spatial data analysis software tools in
disease? Landsc. Urban Plann. 2006, 77 (3), 240–254. R. Geogr. Anal. 2006, 38 (1), 23–40.
(16) Etter, A.; McAlpine, C.; Possingham, H. Historical patterns and (41) R Development Core Team. R: A Language and Environment for
drivers of landscape change in Colombia since 1500: A regionalized Statistical Computing; R Foundation for Statistical Computing: Vienna,
spatial approach. Ann. Assoc. Am. Geographers 2008, 98 (1), 2–23. Austria, 2010.
(17) Mejía, D.; Ramírez, M. T.; Tamayo, J. The Demographic (42) Venables, W. N.; Ripley, B. D., Modern Applied Statistics with S.,
Transition in Colombia: Theory and Evidence; Banco de la Republica de 4th ed.; Springer: New York, 2002.
Colombia: Bogota, Colombia, 2008. (43) Bjornstad, O. N. ncf: Spatial Nonparametric Covariance Functions,
(18) Davalos, L. M.; Bejarano, A. C.; Correa, H. L. Disabusing version 1.1-3; 2009 http://CRAN.R-project.org/package=ncf.
cocaine: Pervasive myths and enduring realities of a globalised com- (44) Højsgaard, S.; Halekoh, U.; Yan, J. The R package geepack for
modity. Int. J. Drug Policy 2009, 20 (5), 381–386. generalized estimating equations. J. Stat. Software 2006, 1–11.

1226 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227


Environmental Science & Technology ARTICLE

(45) Sing, T.; Sander, O.; Beerenwinkel, N.; Lengauer, T. ROCR: MAIN&BASE=CG2005BASICO&MAIN=WebServerMain.inl (22 April,
Visualizing classifier performance in R. Bioinformatics 2005, 21 (20), 2010).
3940–1. (67) Jarvis, A.; Guevara, E.; Reuter, H. I.; Nelson, A. SRTM 90m
(46) Pinheiro, J.; Bates, D.; DebRoy, S.; Sarkar, D. nlme: Linear and Digital Elevation Data. http://srtm.csi.cgiar.org/ (2010).
Nonlinear Mixed Effects Models, version 3.1-96; 2009 http://CRAN.
R-project.org/package=nlme.
(47) Young, K. R. Threats to biological diversity caused by coca/
cocaine deforestation. Environ. Conserv. 1996, 23 (1), 7–15.
(48) Bradley, A.; Millington, A. Agricultural land-use trajectories in a
cocaine source region: Chapare, Bolivia. In Land-Change Science in the
Tropics: Changing Agricultural Landscapes; Springer: New York, 2008;
pp 231-250.
(49) Bradley, A. V.; Millington, A. C. Coca and colonists: quantifying
and explaining forest clearance under coca and anti-narcotics policy
regimes. Ecol. Soc. 2008, 13 (1), 31.
(50) Davalos, L. M.; Correa, H. L.; Bejarano, A. C. Fighting the
Wrong War: Unmet Basic Needs and Coca Cultivation in Colombia. In
The Social Life of Forests: New Frameworks for Studying Change; Hecht,
S. B., Morisson, K., Padoch, C., Eds.; University of Chicago: Chicago, In
Review.
(51) Angrist, J. D.; Kugler, A. D. Rural windfall or a new resource
curse? Coca, income, and civil conflict in Colombia. Rev. Econ. Stat.
2008, 90 (2), 191–215.
(52) Sanchez, F.; Diaz, A. M.; Formisano, M., Conflicto, Violencia y
Actividad Criminal en Colombia: Un Analisis Espacial. Documento
CEDE 2003, 5.
(53) Heath, J.; Binswanger, H. P., Policy-induced effects of natural
resource degradation: The case of Colombia. In Agriculture and the
Environment: Perspectives on Sustainable Rural Development; Lutz, E., Ed.;
World Bank Publications: Washington, D.C., 1998; pp 22-34.
(54) Deininger, K.; Olinto, P. Rural nonfarm employment and
income diversification in Colombia. World Dev. 2001, 29 (3), 455–
465.
(55) Center for International Policy Colombia Program: U.S. gov-
ernment fact sheets and reports. http://www.ciponline.org/colombia/
aidgovt.htm (15 June, 2010).
(56) Etter, A.; McAlpine, C.; Phinn, S.; Pullar, D.; Possingham, H.
Characterizing a tropical deforestation wave: A dynamic spatial analysis
of a deforestation hotspot in the Colombian Amazon. Glob. Change Biol.
2006, 12 (8), 1409–1420.
(57) Ewers, R. M. Interaction effects between economic develop-
ment and forest cover determine deforestation rates. Glob.Environ.
Change 2006, 16 (2), 161–169.
(58) Fjeldsa,
 lvarez, M. D.; Lazcano, J. M.; Leon, B. Illicit crops
J.; A
and armed conflict as constraints on biodiversity conservation in the
Andes region. Ambio 2005, 34 (3), 205–211.
(59) Etter, A.; McAlpine, C.; Pullar, D.; Possingham, H. Modelling
the conversion of Colombian lowland ecosystems since 1940: Drivers,
patterns and rates. J. Environ. Manage. 2006, 79 (1), 74–87.
(60) Etter, A.; van Wyngaarden, W. Patterns of landscape transfor-
mation in Colombia, with emphasis in the Andean region. Ambio 2000,
29 (7), 432–439.
(61) Kaimowitz, D.; Sheil, D. Conserving what and for whom? Why
conservation should help meet basic human needs in the tropics.
Biotropica 2007, 39 (5), 567–574.
(62) Cernea, M. M.; Schmidt-Soltau, K. Poverty risks and national
parks: Policy issues in conservation and resettlement. World Dev. 2006,
34 (10), 1808–1830.
(63) Aide, T. M.; Grau, H. R. ECOLOGY: Enhanced: Globalization,
migration, and Latin American ecosystems. Science 2004, 305 (5692),
1915–1916.
(64) Perfecto, I.; Vandermeer, J. The agroecological matrix as
alternative to the land-sparing/agriculture intensification model. Proc.
Natl. Acad. Sci., U.S.A. 2010, 107 (13), 5786–5791.
(65) UNEP-WCMC, World Database of Protected Areas. In UNEP-
IUCN, 2010.
(66) DANE Sistema de Consulta Informacion Censal: Censo 2005.
http://190.25.231.242/cgibin/RpWebEngine.exe/PortalAction?&MODE=

1227 dx.doi.org/10.1021/es102373d |Environ. Sci. Technol. 2011, 45, 1219–1227

You might also like