Deforestation and Coca Cultivation Rooted in Twentieth-Century Development Projects

Most of the world's coca—the source of cocaine—is grown in the Amazonian forests of Colombia, Peru, and Bolivia. As cultivation continues despite eradication, a shift to giving farmers more incentives to abandon coca is currently proposed. Assuming coca cultivation is an important cause of migration and deforestation, new alternative development projects also aim to conserve forests. We show coca cultivation strongly increases near never-completed 1960s–1970s state-sponsored projects to settle the Amazon. Improved roads and colonization projects opened the western Amazon frontier to migration, generating deforestation and, once support dwindled, setting the stage for coca cultivation. New studies also show coca cultivation generates negligible direct or indirect forest loss and fails to explain migration, whereas expanding legal agriculture, roads, displacement, and eradication increase deforestation. These findings highlight the urgent need to both commit development investment for the long term and set explicit conservation goals in western Amazonia.

Last April, the Special Session on the World Drug Problem of the United Nations General Assembly met for the first time since 1998 to renew global commitments to the War on Drugs. As delegates revisited the 1961 Convention on Narcotic Drugs, a coalition of Latin American nations aimed to boost budgets for alternative development. For almost four decades, the primary antinarcotics strategy in the Andean region has involved drug-crop eradication and relatively meager investment in alternative development or else programs making legal economic alternatives more attractive to growers (Zech 2016). This is about to change with escalating regional and international support to curb coca production and to simultaneously reduce deforestation in the Amazon.

The main reason for integrating alternative development projects with forest conservation in Amazonia is the spatial association between illegal crops and deforestation hotspots (Young and León 1999, Killeen et al. 2008, Dávalos et al. 2011). These spatial analyses imply antinarcotics programs could effectively contribute to conserving forests and biodiversity by preventing additional forest loss (UNODC 2015). The new and environmentally friendly boost for alternative development could represent a win–win scenario for farmers. Economic growth from alternative development could lead farmers to both abandon coca and forego agricultural expansion into rainforests (figure 1).

Figure 1.

Clearing and coca in Puerto Asís, Putumayo, Colombia (2015). Cultivation at the frontier between settlements and Amazonian forests in Colombia, Peru, and Bolivia supplies most of the global demand for cocaine (Photo by Maria Gualdron and Leonardo Correa, SIMC, UNODC Colombia.).

Crucially, these new alternative development projects assume that coca cultivation is a key cause of deforestation—or at least an important one (UNODC 2015). They also assume the lack of economic options pushes farmers toward coca cultivation and deforestation (UNODC and MINAM 2011). This roughly corresponds to an immiserization model of deforestation, in which forest clearing is the result of nonexistent economic options for impoverished farmers (Rudel and Roper 1997). Under the immiserization model, the high economic returns and relatively stable market for coca make it the ultimate cash crop (Bradley and Millington 2008), attracting migration to lowland Amazon forests and promoting deforestation there (Killeen et al. 2008).

The alternative to the immiserization model of deforestation is the frontier model. In frontier deforestation, the public and private sectors build infrastructure into a region, enabling farmer access, leading to migration and, in turn, causing deforestation (Rudel and Roper 1997). Although farmers are direct agents of deforestation in both models, infrastructure and development are crucial only to the frontier model. If deforestation from coca cultivation is associated with frontier dynamics, alternative development projects aiming to curb deforestation face a dilemma. Projects often seek to improve local socioeconomic conditions and improve access to legal markets (UNODC 2015). As coca cultivation concentrates in relatively isolated areas with poor infrastructure (Dávalos et al. 2011, Davalos 2016), fulfilling this goal often entails road construction and improvement. If the frontier model of deforestation is the most relevant to coca cultivation, alternative development projects aiming to reduce deforestation must then balance the competing goals of improving access and conserving forests (Dávalos et al. 2014).

Research objectives

The goal of this article is to evaluate two fundamental premises of alternative development projects to conserve Amazonian forests: the important role of coca cultivation in land-use change and a deforestation model based on the immiserization of growers (as opposed to frontier dynamics). In particular, we aim to answer four questions: (1) Which model of tropical deforestation—immiserization or frontier—best explains coca cultivation in the Amazon? (2) Is coca cultivation an important cause of deforestation? (3) Does coca cultivation influence deforestation rates? (4) Does coca eradication influence deforestation rates? Because the first question pertains to the historical origin of land use in western Amazonia, we begin by reviewing the regional environmental history and its relationship to coca cultivation.

Background: The ghosts of development past

Plans to integrate the Amazon with global markets began with European colonization and expanded greatly during the nineteenth-century rubber “boom” (Young and León 1999). But the scale of landscape transformation remained small until the twentieth century, when hitherto disjointed and transitory efforts to populate the lowlands coalesced into a vast and ambitious initiative to develop western Amazonia (Snyder 1967). The geography of these efforts explains the location of deforestation hotspots and inroads into the forest, and their history explains both coca cultivation and its spatial association with deforestation.

Andean nations began coordinating Amazonian road construction and improvement in 1963 with the ultimate—and still unachieved—goal of completing the Marginal de la Selva road (Snyder 1967). More than just a road connecting Venezuela to Bolivia through 5500 kilometers (km) of forest, the Marginal offered a focus for large plans to develop the Amazon by populating the lowlands, apportioning land, and expanding agriculture (Snyder 1967). Unprecedented in scope and ambition, the Marginal road and associated development projects attracted domestic resources and, more importantly, international credit and development aid (Snyder 1967).

The roads were only one element of this transformation. Settling the Amazon and developing agriculture in the lowlands were as important as building and improving the roads. Expanding the agricultural land base was urgent. Andean land was highly concentrated in the hands of few and powerful landowners (Crist and Nissly 1973), and agricultural mechanization and technification were accelerating. These forces made hundreds of thousands of tenant and small farmers redundant, just as population growth was gaining momentum from twentieth-century improvements in sanitation and health. Between 1962 and 1972, for example, there were about 2.5 million people (out of a census population of about 25 million in 1973) seeking land in Colombia alone (Brücher 1977). By developing their vast Amazonian territories, national governments intended to relieve pressure on rapidly growing cities and scarce Andean land, securing food and commodities for each nation (Schuurman 1979).

State-sponsored colonization and settlement projects linked to the roads emerged through the 1960s and 1970s (table 1, figure 2). Although a few projects directed colonists to particular locations, most legitimated land claims and provided technical assistance and agricultural credit to spontaneous settlers migrating to the new frontier. Colonization projects opened at the edge of the Amazon forest, along an arc of sites that later became coca-cultivation hotspots (table 1, figure 2): Ariari-Güéjar, Guaviare, Caquetá, and Putumayo in Colombia (Brücher 1968, Wesche 1968); Alto Marañón, along the Apurímac river, Pichis-Palcazú, and the Huallaga River valley of Peru (Maass 1969, Jülich 1975); and in Alto Beni, Chapare and east of Santa Cruz in Bolivia (Schoop 1970, Brücher 1977, Schuurman 1979).

Figure 2.

Maps of (a) Amazonian colonization projects and coca cultivation in 1990 (Meta-Guaviare of Colombia) or 1992 (other sites), (b) forest cover and 2014 coca cultivation east of the Andes, and (c) modeled coca cultivation.

Table 1.

Government-sponsored colonization projects in western Amazonia, 1960s–1970s.

Project	Country	Location	Name Brücher (1977)	Name Schuurman (1979)
Ariari-Güéjar	Colombia	Ariari, Meta	–	Meta
Guaviare	Colombia	San José, Guaviare	S. José	El Retorno
Caquetá	Colombia	Florencia, Caquetá	Caquetá	Caquetá
Putumayo	Colombia	Puerto Asís, Putumayo	Puerto Leguízamo	–
Alto Marañón	Peru	Alto Marañón, Marañón	Alto Marañón	Alto Marañon
Jenaro Herrera	Peru	Jenaro Herrera, Loreto	Genaro Herrera	Jenaro Herrera
Middle Huallaga	Peru	South of Tingo María, Leoncio Prado	Mittl. Huallaga	Tingo María-Tocache
Pichis-Palcazú	Peru	Puerto Bermudez, Oxapampa	–	–
Apurímac river	Peru	San Francisco, Ayacucho	Apurímac	Apurímac
Alto Beni	Bolivia	Yungas de la Paz	Alto Beni	Alto Beni
Chapare	Bolivia	Chapare, Cochabamba	Chapare	Chimoré
Santa Cruz	Bolivia	West of Santa Cruz de la Sierra	Sta. Cruz	Yapacaní

Years before illegal coca cultivation became widespread (Young and León 1999, Etter et al. 2006), the roads and colonization projects promoted deforestation. Despite the high costs of importing food from developed regions, subsistence agriculture was discouraged, with credit and assistance conditioned on growing cash crops. Converting a high proportion of the forest to agriculture was required to claim land titles, and growers most efficiently achieved this by burning the forest and clearing for pasture (Wesche 1968, Schuurman 1979). Galvanized by new investment and reflecting the landscape footprint of tens of thousands of new settlers, Colombia's access roads, Peru's Marginal de la Selva, and Bolivia's Cochabamba–Santa Cruz road transformed the landscape, forming growing wedges into the Amazon that persist today (Young and León 1999, Etter et al. 2006, Killeen et al. 2008, Harris et al. 2012).

Immediately, farmers faced more rainfall and agricultural pests as well as poorer drainage and soil fertility than in the Andes. For similar reasons, the construction, improvement, and maintenance of roads proved more expensive and difficult than first anticipated. The infrastructure to support farmers remained incomplete for decades (Bradley and Millington 2008, Salisbury and Fagan 2011). Even today, basic education and health services remain scarce at the Amazon frontier of the Andean region (Davalos 2016, UNODC 2014). This combination of ecological and socioeconomic challenges limited productivity and hampered nascent local economies. By the 1970s and 1980s, with debt crises looming throughout Latin America, less institutional and financial support was available to farmers than ever before (Legrand 1989, Morales 1990, Valdivia 2012). At the same time, international demand for cocaine exploded (Boucher 1991, Caulkins et al. 2004). With external support declining and weak demand for legal commodities, farmers took to coca cultivation as a prime agricultural alternative, first in Peru and Bolivia—where development projects sometimes encouraged legal coca production (Jülich 1975, Maass 1969, Schoop 1970)—and later in Colombia.

Data sources

Our quantitative analyses estimate the association between coca cultivation and historical development projects, testing the links between coca today and frontier deforestation linked to twentieth-century development projects. The target area comprised the Amazonian biome of northwestern South America, including Colombia, Brazil, Ecuador, Peru and Bolivia (as defined by UNEP et al. 2009).

To compare historical and current coca extent, sites with a high density of coca cultivation in 1990 for Meta and Guaviare (Colombia) were obtained from UNODC (2010) and from Lee and Clawson (1993) for all other regions for 1992. Data on 2014 coca cultivation were obtained from the national monitoring surveys of the United Nations Office on Drug and Crime (UNODC) in Colombia, Peru, and Bolivia (UNODC 2015).

The main source for colonization projects was Brücher (1977), supplemented with the Ariari-Güéjar project in Meta, Colombia (Schuurman 1979), and the Pichis-Palcazú project started in 1978 (UNODC and MINAM 2011). Brücher (1977) included no projects from Ecuador, because this country focused most of its development and colonization west of the Andes and along the Cordilleras (Brücher 1977). When development projects in the Ecuadorian Amazon were implemented, these were associated with oil exploration and exploitation, with support for spontaneous migration during the 1970s confined to the road connecting the oil-exploitation sites at Lago Agrio to Quito (Bromley 1972, Schuurman 1978). The names and locations of the projects analyzed are summarized in table 1.

Statistical approach

The municipalities were obtained from the GADM database of Global Administrative Areas v. 2.8 (www.gadm.org). The distances to the development and colonization projects were calculated for every pixel of side approximately 4 km or 0.036 degrees. All layers were resampled to an approximately 12-km grid, with any coca cultivation mapped as presence for a given pixel, and the distance to the nearest colonization project was calculated in kilometers and log₁₀-transformed. A total of 3,525,806 km² of the Amazon biome were modeled in 24,898 pixels.

We adopted a spatially explicit landscape modeling approach to measure the influence of colonization projects on the current distribution of coca cultivation. Spatial autocorrelation was included by calculating the neighbor matrix within approximately 221 km of the centroid of each municipality in the sample (for 984 municipalities). The presence of coca was then modeled as a binomial response using hierarchical regressions in which the spatial effect corresponds to a municipality-specific (or random) vector defined by:

\begin{equation}{x_i}|{x_j},i \ne j,\tau \, {\sim} \, N\left( {\frac{1}{{{n_i}}}\sum\limits_{i \,{\sim} \, j} {{x_j},} \frac{1}{{{n_i}\tau }}}\right) \end{equation}

(1)

where n_i is the number of neighbors of municipality i, i∼j are the two neighboring municipalities i and j, and the precision parameter τ captures the variance in influence between different sets of neighbors (Besag and Kooperberg 1995). The influence of the distance to the colonization projects on the presence of coca was modeled as a sample-wide (or fixed) coefficient. A series of models were fitted using approximate Bayesian inference for latent Gaussian models through integrated nested Laplace approximation (Rue et al. 2009). Models with and without municipality-specific (random) intercepts were also fitted to estimate the effect of factors contributing to the variation between municipalities but excluded from these analyses (e.g., roads, population density). All models were fitted using the INLA package v.0.0-1455098891 (Rue et al. 2016) implemented in R v.3.2.4 (R Development Core Team 2016), data and scripts are available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.1hb1f.

To measure the empirical evidence for particular models, we calculated Bayes factors and compared all models against the spatial process model (equation 1) lacking any additional data (e.g., distance to colonization projects, or municipality-specific intercepts). The Bayes factor is the summary of the evidence provided by the data in favor of one statistical model compared with another (Kass and Raftery 1995), given by

\begin{equation}logBF = log(mL(mode{l_1})) - log(mL(mode{l_2})) \end{equation}

(2)

where mL is the marginal likelihood of a model, and model₂ in this case corresponds to the purely spatial model. To interpret Bayes factors, we follow guidelines by Kass and Raftery (1995): BF < –2 suggests preference for model₂, BF > 2 for model₁, 6 < BF < 10 strong evidence for model₁, and BF > 10 represents decisive evidence for model₁. A summary of models fitted, log marginal likelihoods and Bayes factors is presented in table 2. The best-supported model was then used to estimate the coca presence. Posterior estimates of parameters for this best-supported model are shown in table 3.

Table 2.

Spatial models fitted, log marginal likelihoods, and log Bayes factors (BF). Bayes factors compared the evidence in favor of the statistical model against the purely spatial (first) model. BF > 10 represents decisive evidence for the model given (Kass and Raftery 1995). The best-supported model is shown in bold.

Sample-wide (fixed) term	Municipality-specific (random) term	Log marginal likelihood	Log BF
None	Spatial	–6340.77	–
None	Spatial, municipality intercept	–5879.25	461.52
Distance to nearest project	Spatial	–6061.23	279.54
Distance to nearest project	Spatial, country	–5955.12	385.65
Distance to nearest project	Spatial, nearest project	–5963.22	377.55
Distance to nearest project	Spatial, province political unit	–5894.68	446.09
Distance to nearest project	Spatial, municipality intercept	–5121.02	1219.75

Results

Coca cultivation remained spatially clustered along the edges of lowland forests in Western Amazonia for more than 20 years, with closely matching clusters for the 1990s and 2014 (figure 2). The best-supported model of coca cultivation included the distance to the nearest development project as a predictor of the probability of finding coca in a given pixel in 2014 (table 2). Increasing distance to the nearest project corresponded to decreasing probability of coca cultivation (figure 3). In addition, individual municipalities experienced differences in the probability of observing coca summarized by the variation in the municipality-specific (random) term (table 3). A review of data on direct coca deforestation uncovered four independent studies from Colombia and Peru, three of them from Amazonia (Dávalos et al. 2011, UNODC and MINAM 2011, Armenteras et al. 2013b, Chadid et al. 2015). All found deforestation rates from coca cultivation to be at least one order of magnitude smaller than deforestation for legal uses (figure 4). In four analyses, all from Colombia, researchers examined coca cultivation as a covariate of deforestation rates while controlling for other factors (Dávalos et al. 2011, Armenteras et al. 2013a, Sánchez-Cuervo and Aide 2013, Fergusson et al. 2014). None found coca cultivation to be a factor strongly influencing deforestation rates when other covariates (such as roads or armed conflict) were analyzed. In two analyses from Colombia, coca eradication was related to migration and deforestation, revealing that eradication moved coca cultivation to new forests and was associated with forced displacement (Rincón-Ruiz and Kallis 2013, Rincón-Ruiz et al. 2013).

Figure 3.

The modeled probability of coca presence as a function of distance from the nearest development project. Parameter means are shown on the dark gray line; 95% high-probability density is shown in light gray (table 3). Individual municipalities (not shown) may have lower or higher intercept values. Observed pixel values are shown in 
black dots, jittered for visualization.

Figure 4.

The annual deforestation rates from coca (dark gray, negative values) or legal agriculture (light gray, positive values). Where no forest type is indicated, the values correspond to old-growth and secondary forests combined. Sources: Dávalos et al. (2011), UNODC and MINAM (2011), Armenteras et al. (2013b), and Chadid et al. (2015).

Table 3.

The posterior parameter estimates for the best-supported model. See table 2 for Bayes factors in favor of this model.

Parameter	Type	2.5 percentile	Mean	Median	97.5 percentile
Intercept	Sample-wide (fixed)	–1.2671	–1.1540	–1.1540	–1.0410
Distance to nearest project	Sample-wide (fixed)	–0.6096	–0.5613	–0.5613	–0.5129
Precision municipality intercept	Municipality-specific (random)	1.116	1.4582	1.4381	1.910
Precision spatial	Municipality-specific (random)	0.243	0.5211	0.4716	1.079

Discussion

The review of Andean regional history and deforestation analyses as well as spatially explicit analyses of coca cultivation illuminates the origin of this crop and its effects on Amazonian deforestation. First, coca cultivation in the Amazon is embedded within the larger forest frontier along a series of wedges spatially associated with twentieth-century development projects. Second, coca cultivation is not a dominant cause of direct deforestation. Third, there is little evidence that coca cultivation increases deforestation rates, independent of the dynamics already prevalent at the western Amazon frontier. The first two results support a frontier model of deforestation, in which the process of deforestation is primarily determined by infrastructure to open the region, and in particular road construction (Barber et al. 2014), whereas the absence of a relationship between migration and coca cultivation rejects the immiserization model (Dávalos et al. 2011). Finally, there is quantitative evidence that coca eradication, perhaps mediated by forced displacement, displaces cultivation to new sites, increasing deforestation.

Expanding legal agriculture, not coca, explains deforestation

Our results imply coca cultivation is not an idiosyncratic driver of deforestation, but only one element of the agricultural frontier linked to twentieth-century development projects. The importance of this history is clear: The spatial overlap between beachheads of settlement, deforestation, and coca persists to this day (table 3, figure 2), despite large investments to fight this crop. The sites targeted for road construction, settlement, and development in the 1960s and 1970s became resource frontiers and deforestation hotspots, emitting more carbon today than the rest of western Amazonia (Harris et al. 2012). The spatial overlap between coca and expanding agriculture, as well as a keen focus on heavily monitored coca cultivation in Amazonia (Dávalos et al. 2011, Armenteras et al. 2013b), have obscured the ultimate origin of frontier dynamics in decades-old infrastructure and development projects.

With twentieth-century colonization projects long in the past, coca cultivation might prove a powerful attractor to the western Amazon forest frontier. A growing number of studies throughout the Andean region, however, reveals coca cultivation has little influence on deforestation rates. In the Colombian Amazon frontier, migration strongly correlates with deforestation, but coca cultivation explains neither deforestation rates nor migration (Dávalos et al. 2011). In Peru, deforestation rates at frontier sites without coca tend to be higher than at sites with coca in the Huallaga valley (UNODC 2014). Although areas with coca crops in the Chapare of Bolivia experienced high deforestation rates in the 2000s, so did sites where legal farming and harvesting forest products were the predominant land uses (Killeen et al. 2008), indicating that expanding agriculture and not coca drives deforestation.

The most detailed analyses of deforestation rates have been conducted in Colombia. When roads and migration are included, coca cultivation fails to explain Amazonian deforestation rates (Armenteras et al. 2013a). Analyses of an independent, annual data on deforestation rates from the 2000s revealed no relationship with coca cultivation for any region of Colombia at any time (Sánchez-Cuervo and Aide 2013). Instead, armed conflict and forced displacement explained higher deforestation rates (Sánchez-Cuervo and Aide 2013). There is only one study in which coca cultivation correlated with forest loss (Fergusson et al. 2014). In that case, deforestation rates rose together with coca cultivation as a consequence of the violent takeover by armed groups, not through the spontaneous expansion of illegal agriculture (Fergusson et al. 2014).

Another line of evidence for the relationship between coca and deforestation is the trajectory of landscapes where cultivation has declined (Dávalos et al. 2014). Far from leading to forest regrowth, steep declines in coca cultivation between 2000 and 2010 in the Colombian Guaviare were associated with the closing of the forest frontier, when minimal or no forest remains in a matrix of agricultural uses. Taken together, these analyses show that coca is not a predictor of deforestation rates but rather that roads and migration are, further supporting a frontier model of deforestation. Furthermore, coca by itself fails to cause migration, and its disappearance fails to reduce deforestation.

The role of coca eradication

For decades, coca eradication has been the primary antinarcotics strategy in the Andean region (Zech 2016), but its success is often transitory. A few short years of declining coca production since 2010 were upended by increases in cultivation in 2015. In the 40 years since the start of the War on Drugs, cultivation remains clustered around the Amazonian colonization wedges, and it has also expanded to new sites (figure 2). Illegal coca is found farther into the Amazon today than even 15 years ago (Armenteras et al. 2013b). If coca cultivation fails to explain much deforestation, perhaps drug policy contributes little to deforestation dynamics. But the evidence suggests that forced eradication promotes migration, increasing deforestation (Salisbury and Fagan 2011, Rincón-Ruiz and Kallis 2013).

In Colombia, detailed spatial analyses have begun to measure the consequences of coca eradication, finding increased movement following fumigation campaigns. In 2001, coca cultivation concentrated in municipalities in Putumayo and Caquetá with less forest compared with that of their neighbors—that is, the decades-old Amazonian frontier (Rincón-Ruiz et al. 2013). By 2008, cultivation had moved to municipalities with more forest than their neighbors. This suggests some farmers relocated to less accessible forests as the eradication campaigns intensified. The spatial shift in coca cultivation from the Amazon colonization frontier to other forests was also associated with forced displacement (Rincón-Ruiz and Kallis 2013). As independent analyses find displacement to be a strong predictor of deforestation (Sánchez-Cuervo and Aide 2013), the relationship between eradication and displacement may possibly lead to more deforestation.

Drug policy as conservation

Recent findings that social investment in education, health, and infrastructure prevents coca cultivation more efficiently than forced eradication support the shift in antidrug policy from mostly eradication to other forms of investment, even if alternative development investment did not reduce cultivation in those analyses (Davalos 2016). Our analyses additionally show two fundamental assumptions behind the push for alternative development as a means for Amazonian forest conservation are flawed. First, coca cultivation is neither an important cause of direct deforestation nor a powerful factor determining deforestation rates. Instead, the large wedges of deforestation in which coca cultivation takes place originated at sites targeted for development and colonization projects decades ago. Second, the current frontier dynamics is not a consequence of coca cultivation by itself but rather the result of efforts to develop western Amazonia by improving access through the Marginal de la Selva and allocating land. This history of twentieth-century development projects compels consideration for the long-term commitment and investment necessary to successfully foster economic development and prevent deforestation.

Because coca cultivation is neither an important direct cause nor a covariate of deforestation rates, focusing on coca cultivation as a substantive driver of Amazonian forest loss is misguided. Instead, conservation and development policy should focus on stabilizing the agricultural frontier—which effectively means avoiding or mitigating the displacement associated with forced coca eradication (Salisbury and Fagan 2011, Rincón-Ruiz and Kallis 2013). Despite migration and conflict over territorial control, protected areas have remained effective at conserving forests (Dávalos et al. 2011, Armenteras et al. 2013a, Barber et al. 2014). Protected areas need to be strengthened along with governance and basic services, and infrastructure needs to be oriented toward stabilizing agriculture and settlement by defining and enforcing conservation targets.

Although some assume any agricultural uses other than coca cultivation will inevitably lead to more deforestation (Álvarez 2001), the reality is much more complex (Bradley and Millington 2008). Still, successful examples of long-term coca reduction often involve closing the forest frontier, with formerly forested lands transformed into pastured properties (UNODC and MINAM 2011, Dávalos et al. 2014). The lingering effects of long-abandoned projects to develop the Amazon prove current policy can shape land use decades into the future. To avoid repeating the mistakes of the past, alternative development projects must work with Amazonian farmers and conservationists to set explicit conservation goals and plan for decades and not just years of investment.

This work was partially supported by a commission from the Gesellschaft für Internationale Zusammenarbeit (GIZ) to LMD. We thank A. Corthals for German-language translations, and S. Lo for encouragment. For comments, we thank the three anonymous reviewers, A. Corthals, E. Dávalos, E. Lauterbur, M. Lim, J. Retana, and L.R. Yohe. We thank C. Bussink, L. Correa, M. Gualdrón, A. Korenblik, L. Vallejos and A. Vella from the United Nations Office on Drug and Crime (UNODC). All opinions are exclusively the authors’ and do not reflect GIZ or UNODC positions.