This article was downloaded by: [UZH Hauptbibliothek / Zentralbibliothek Zürich]
On: 18 March 2015, At: 01:08
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House,
37-41 Mortimer Street, London W1T 3JH, UK
Click for updates
Cartography and Geographic Information Science
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/tcag20
Illegal drug cultivation in Mexico: an examination of
the environmental and human factors
Monica Medel
a
& Yongmei Lu
a
a
Department of Geography, Texas State University, San Marcos, TX 78666, USA
Published online: 26 Nov 2014.
To cite this article: Monica Medel & Yongmei Lu (2015) Illegal drug cultivation in Mexico: an examination of
the environmental and human factors, Cartography and Geographic Information Science, 42:2, 190-204, DOI:
10.1080/15230406.2014.985716
To link to this article: http://dx.doi.org/10.1080/15230406.2014.985716
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the
Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and
are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and
should be independently verified with primary sources of information. Taylor and Francis shall not be liable for
any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of
the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions
Illegal drug cultivation in Mexico: an examination of the environmental and human factors
Monica Medel and Yongmei Lu*
Department of Geography, Texas State University, San Marcos, TX 78666, USA
(Received 15 April 2014; accepted 16 October 2014)
Patterns of illicit narcotics cultivation are among the understudied topics. Some studies estimate the prevalence of illegal
crops using imagery and remote sensing data. These studies rely heavily on the availability and quality of the related
images, which is often an issue for many countries known as major drug producers. Using official drug crop eradication
data, this study examines the patterns of illegal drug cultivation in Mexico at the municipality level. Species distribution
models of ecology were used to guide the selection of environmental variables. A number of sociodemographic variables
were incorporated into the model to describe human factors. Global and local models were compared to discern the
determinants of marijuana and opium cultivation. Geographically weighted regression was proved overall more effective
than global ordinary least square regression despite the spatial variation of its explanation power. The models explained the
spatial patterns of opium poppy cultivation are better than those of marijuana cultivation, suggesting the possible presence
of more complicated local factors for growing illicit marijuana crops. A number of human factors such as law enforcement,
gang activities, and transportation accessibility were found significant for illicit cultivation.
Keywords: drug cultivation; geographically weighted regression; species distribution models; environmental factors;
human factors
Introduction
The patterns of illegal narcotics cultivation in Mexico are
among the understudied topics about drug activities. There
are many studies on Mexican antinarcotics policies and law
(e.g., Reuter 1988; Toro 1995; Dominguez and Fernandez
de Castro 2001), the history of drug smuggling (e.g.,
Astorga 2005; Youngers and Rosin 2005), the relationship
between drugs and economy (e.g., Resa Nestares 2003;
Thoumi 2003), and the socioeconomic impact of drug
cultivation on communities where different types of drug
crops are grown (e.g., De la Herrán 1980; Marín 2002;
Lizarraga and Lizarraga 2006). Studies can also be found
on illicit cultivation activities in other countries, including
Colombia and Peru, the chief producers of cocaine in South
America, and Afghanistan and Myanmar, the largest producers
of opium and heroin in Asia. Many of these studies
were led by the United Nations Office on Drugs and Crime
(UNODC), which conducts periodic surveys on illegal
crops, especially the cultivation of opium, coca, and marijuana
(UNODC 2000–2012). These studies incorporated
remote sensing data and techniques with fieldwork and
surveys to determine drug production. Preharvest images
with high spatial resolution, such as GeoEye, WorldView2,
Quickbird, and Ikonos data of 1.65- to 4-meter resolution,
were often used (UNODC 2011). They cover a number of
countries including Afghanistan, Bolivia, Colombia,
Ecuador, Peru, Morocco, Myanmar, and Laos, but not
Mexico, which is a main opium and marijuana producer
in the Americas (INCSR 2008, 2009).
Marijuana (Cannabis sativa) is endemic to the state of
Sinaloa, located in the northwest region of the country
(Figure 1). Its presence was noted in books as early as the
end of the nineteenth century (Ortega and Lopez Manon
1987). But its cultivation did not begin to boom until the
advent of the twentieth century when narcotics became
prohibited in the United States and the rest of the world
(Astorga 2005). By 1920, Mexico too had banned the
cultivation and sale of marijuana. As early as a decade
later, the first Mexican drug traffickers began being mentioned
in the national press as operating in the border
region with the United States and producing their crops
in state of Sinaloa and the neighboring states of Durango
and Chihuahua – an area that has been dubbed “the
Golden Triangle” of drug production in Mexico (Astorga
2005). In the subsequent decades, the cultivation of marijuana
spread to other, more distant regions, following
valleys in the Sierra Madre Mountains. Crops began
being detected far from Sinaloa, in areas ranging from
the southernmost Chiapas state to the Gulf Coast
Veracruz state and the deserts of the Baja California
peninsula (Astorga 2005). The vast quantities of marijuana
crops in so many varied areas across Mexico make it a
good drug specimen to study in order to understand the
effects of both environment factors and human actions.
*Corresponding author. Email: YL10@txstate.edu
Cartography and Geographic Information Science, 2015
Vol. 42, No. 2, 190–204, http://dx.doi.org/10.1080/15230406.2014.985716
© 2014 Cartography and Geographic Information Society
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
Opium poppy (Papaver somniferum) is not native to
Mexico but is believed to have been introduced in the late
1800s by Chinese immigrants who arrived in the country
to work in mines and on railroad construction (Astorga
2005). The result was the spread of a plant that the locals
dubbed Adormidera blanca but was none other than
opium. It began to spring up in Sinaloa shortly after the
arrival of the Chinese. Opium crops then spread rapidly to
neighboring states, including Sonora and Chihuahua as
mining developed in the area (Toro 1995). By the 1960s,
it was being cultivated even in the southern states of
Oaxaca and Chiapas and in the Pacific coast states of
Nayarit and Guerrero (Astorga 2005).
Understanding the factors that are favored by the illicit
crop cultivation could help authorities better pinpoint
where such illicit production is likely to occur, and thus
act to eradicate it. This perspective has guided a number of
studies on marijuana in the United States and Canada
(Walthall et al. 2003; Thomas et al. 2004; Partelow
2008; Hurley, West, and Ehleringer 2010; Bouchard,
Beauregard, and Kalacska 2013), but none in Mexico.
This study investigates the factors impacting the cultivation
of marijuana and opium poppy in Mexico. The goal is
to examine the relationship between the illicit cultivation
and the environmental and human conditions tied to the
cultivation activities. By adapting the species distribution
models (SDMs) from ecology, both global and local
models are employed to investigate the illicit cultivation
of marijuana and opium poppy in Mexico.
Species distribution model: a framework for illicit
cultivation study
SDM relates observation of species presence or their
abundance with environmental factors (Elith and
Leathwick 2009). It is frequently used as a framework
to study spatial patterns of species. Environmental factors
impacting species distribution can be classified into
indirect, direct, and resource variables (Guisan and
Zimmerman 2000). Resource variables describe matter
and energy consumed by the plant (like nutrients or
water). Direct variables are linked to environmental conditions
that affect plants but cannot be consumed by them
(e.g., temperature or PH). Indirect variables do not have a
physiological impact on plants but are well correlated
with species distribution (such as slope, elevation, geology,
and aspect).
In the case of crops, however, their distribution does not
depend solely on environmental factors. Because seeding,
cultivation, and harvest do not occur naturally, but rely
heavily on human factors, certain variables can be controlled
to create the optimal environmental conditions for
the plants to grow. For example, water can be dispensed
using top-notch irrigation systems that allow crops to thrive
Figure 1. Mexico – the study area.
Cartography and Geographic Information Science 191
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
in natural conditions that otherwise could be adverse
(Fortes, Platonov, and Pereira 2005; Liu 2009). For outdoor
cultivation of illegal drugs, human factors can play a significant
role. Aside from certain environment requirements
for temperature, slope, aspect, and soil characteristics,
proximity to roads and highways is important to make
drug smuggling easier, while hiding away from urban centers
can help minimize police surveillance (Partelow 2008).
Moreover, availability of rainfall can be supplemented
using irrigation technologies. Therefore, growing drug
crops is not fully restricted by environmental factors.
The distribution of environmental variables may introduce
spatial autocorrelation (Václavík and Meentemeyer
2009; Václavík, Kupfer, and Meentemeyer 2012) and nonstationarity
(Osborne and Suárez-Seoane 2002; Austin
2007) to species distribution. Dispersal process, or the
behaviors that lead to the propagation and aggregation of
illicit crop cultivation, may also cause nonstationarity. But
general SDMs do not account for these properties because
either data to measure these processes are unavailable or the
factors that lead to the formation of certain specific spatial
patterns in the species distribution are unknown (Václavík,
Kupfer, and Meentemeyer 2012).
Spatial autocorrelation and nonstationarity must be
accounted for when modeling the relationship between
illicit crops and their environment. Otherwise, models predicting
species distribution may be misspecified, and the
coefficient estimation may be incorrect (Lennon 2000; Keitt
et al. 2002). The inclusion of environmental variables for a
habitat modeling is not enough as spatial autocorrelation
and nonstationarity may be introduced to species distribution
by nonresource or nonenvironmental conditions (Bahn,
O’Connor, and Krohn 2006; Diniz-Filho, Bini, and
Hawkins 2003). For illicit cultivation, human factors such
as crop growers’ measures to minimize the possibility of
being captured play an important role in the dispersal
process of the crops, and so does the effectiveness of law
enforcement’s antidrug actions. Adopting a wider definition
of dispersal conditions as Lidicker et al. (1975), this study
includes a number of human factors that are important for
understanding the distribution of marijuana and opium cultivation
in Mexico.
Data and methods
Data on illicit cultivation in Mexico
Traditionally, areas along the Sierra Madre Mountains have
been most coveted by drug gangs looking to establish their
territories and corridors for moving narcotics into the
United States. These regions have low population density,
are usually well connected by road network, and are clustered
with low-income population with limited access to
basic services (CONEVAL 2010). The states with major
drug production sites are mostly marketed by their beach
resorts along the Pacific Ocean, while the drug fields
remain in the mountains and valleys in the rural areas.
Data on Mexican marijuana and opium cultivation are
scarce. Records of eradication of marijuana and opium
poppy plants for 2009 and 2010 were obtained from the
Mexican Secretary of Defense (SEDENA) through a
Freedom of Information Act Request. The data were
reported at municipality level, the administrative unit
below state. There were 2456 municipalities in Mexico
in 2010, among which 552 municipalities (22.5%) had
marijuana crops eradicated and 274 municipalities
(11.5%) had opium poppy plantations eradicated according
to SEDENA (Figure 2).
The data on eradication were used as a proxy for the
presence and quantity of marijuana and opium crops cultivation.
The total amount of hectares eradicated was used
in the study as a quantitative description of the illegal
cultivation in a municipality. The municipalities with no
eradication in 2010 were excluded from further analyses
to avoid model bias. As Bahn, O’Connor, and Krohn
(2006) suggest, a model including all municipalities,
including those with no eradications, would have modeled
the presence and absence rather than abundance. Using
abundance information makes it possible to model the
magnitude of illicit cultivation rather than only the presence/absence
as commonly done by SDMs (Iverson et al.
2011). Moreover, most statistical models developed to
deal with data sets with excess zeroes, such as zeroinflated
models, assume that part of the zero group has
no probability of having a count greater than zero (Barry
and Welsh 2002; Lee et al. 2006). This is not the case for
this study in which eradication data are used to approximate
the scope of illicit drug cultivation. Treating the 80–
90% of Mexico municipalities with zero eradications during
the study period as having no illicit cultivation is
likely to be very risky; using a zero-inflated model will
likely introduce excessive errors. This study chose to
focus on the municipalities with eradications to investigate
the quantitative relationships between the independent and
the dependent variables (Bahn, O’Connor, and Krohn
2006; Dormann et al. 2008; Iverson et al. 2011).
Data on marijuana and opium eradication in hectares
were standardized by municipality areal size. Furthermore,
because the standardized eradication data for both the
crops were highly positively skewed (with very few municipalities
having relatively large amount of eradication per
square kilometer), the variables were transformed by using
logistic transformation.
Independent variables
Environmental variables
Following SDMs, two environmental variables (temperature
and precipitation) and two topographical variables
192 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
(elevation and slope) are considered. Temperature – a
direct variable – and precipitation – a resource variable –
are fundamental for the development of plants. An appropriate
temperature range for day and night allows plants
to perform its metabolic processes and grow, while a
proper level of humidity and soil pH guarantee its ability
to acquire all the necessary nutrients (Hough et al. 2003).
The optimal day temperature range for marijuana is 14–
27°C (57.2–86°F). Even though they can resist light
frosts (−5°C) (Small, Pocock, and Cavers 2003),
extended cold weather may stop crop maturation. In
comparison, day/night temperature for opium poppy is
steady at 28°C/15°C, but leaf development increases at
19.5°C (Acock, Pausch, and Acock 1997). This study
uses precipitation level as an indicator for humidity condition.
Tetrahydrocannabinol (THC) strains, the primary
psychoactive ingredient in marijuana, grow best in humid
conditions, but it also grows well in areas with a dry
atmosphere during the plant maturation period (Clarke
1981). Opium poppies grow best in temperate and warm
climates with low humidity and not too much rainfall
during early growth period (Booth 1999).
Elevation and slope are indirect variables. Elevation is
closely related to temperature, while slope can be linked to
soil moisture or texture. Slope may also help hiding illegal
crops from the public eye by increasing access difficulty.
A report on illicit crop management by the Canadian
Police Research Centre (CPRC) identified those locations
under 1219 meters of elevation and south-facing as being
suitable for marijuana cultivation (Howell 2002).
Both the plants have specific needs for sunshine.
Marijuana cultivation requires at least 10 hours of sunlight
and between 11 and 12 hours of continuous darkness to
flower and get close to its optimal THC production
(Clarke 1981). Opium poppy needs direct sunlight for at
least 12 hours daily while maturing. Unfortunately, this
Figure 2. Eradications of illicit crops of (a) marijuana and (b) opium poppy.
Cartography and Geographic Information Science 193
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
study does not account for sunshine as a factor for crop
growing as the eradication data do not have information
about specific growing sites.
Temperature, precipitation, and elevation data were
obtained from the Institute of Statistics and Geography
(INEGI 2005) in Mexico. The data for temperature and
precipitation were polyline shapefiles and were converted
to raster format using ArcGIS. Digital elevation model
(DEM) data have a resolution of 1 arc second (about
31 meters) and were used to extract slope. The GIS files
were projected into North America Lambert Conformal
Conic Projection, and the datum was ITRF 1992. The
three raster layers were resampled and registered for
further analyses.
Considering the nonlinear effects of environment parameters
on species abundance (Austin 2002; Dormann
et al. 2008), mean annual precipitation was included in
the model in quadratic form. Temperature was included in
the model as a dummy variable where 1 indicated that the
average annual temperature of a municipality falls in the
proper range for the growth of the crops and 0 otherwise.
Elevation and slope were used to describe the variation in
topography, knowing that marijuana in Mexico is preferably
cultivated in flatter surfaces, while opium is grown in
the mountains. The environmental variables and their
expected relationship with illicit cultivation are explained
in Table 1.
Human variables
Human actions can greatly contribute to fragmenting the
distribution landscape of species (Wiens et al. 2009).
Further, they may work together with environmental
conditions and push the species to isolated patches of
terrain. Fung and Welch (1994) identified proximity to
transportation network and water sources, as well as forest
cover and distance from population center as human factors
that impact marijuana cultivation. Howell’s study in
Canada found that forest clear-cuts with irrigation nearby
and wetlands, mainly in park lands, are suitable for cultivation
of marijuana (Howell 2002). Thomas et al. (2004)
examined certain topographic characteristics that could
facilitate cultivation or transportation of illicit crops; they
further suggested unemployment rates as potentially being
related to illicit cultivation. More recently, in a study of
marijuana cultivation in California National Parks, Partelow
(2008) mentioned variables such as vegetation and canopy
coverage, and the effectiveness of law enforcement, for
identifying potential places for marijuana cultivation. Still
other studies suggested that, for drug cultivation, the impact
of human factors may be measured by considering distance
from urban centers, population density, and police presence
(Partelow 2008; Bouchard, Beauregard, and Kalacska
2013). Bouchard, Beauregard, and Kalacska (2013) proposed
that an index of corruption or propensity for illegal
behavior in an area can help determine the variation in
illicit activity dispersal. In a most recent study, Dube,
Garcia-Ponce, and Thon (2014) found that, in addition to
the impacts of drug cartel operations and drug killings, the
illicit cultivation of both marijuana and opium in Mexico is
negatively related to rural wages. Finally, opium poppy
cultivation and marijuana cultivation are known to show
spatial and temporal continuity – they tend to be close to
each other spatially and tend to be grown 1 year after
another if without interruption.
The data on roads, police forces, and land cover were
obtained from INEGI (2013). Road data were polyline
shapefiles, while the number of police officers for each
municipality came as statistical data. Land cover data were
polygon shapefile and were resampled to municipality
level to reflect the dominant type of land cover for each
municipality. Those municipalities where forest was the
majority were coded as ForestLandCover = 1, while the
rest were assigned a zero. Population data for each municipality
in 2010 were also obtained from INEGI, while
drug-related killing data on municipality level in 2010
were obtained from Mexico’s Presidency. Lastly, illicit
cultivation at a location tends to show spatial and temporal
continuity, meaning that it is likely to be related to the
illicit cultivation in the past as well as the illicit cultivation
of other crops nearby. Therefore, eradication data of the
same type of crop in 2009 and that of a different crop in
2010 were included in the models. All variables were
standardized by the areal size of municipalities. Table 2
explains the human variables. The distributions of both
environmental and human factors were heavily positively
skewed and therefore were log-transformed before being
included in the model.
Table 1. Environmental and topographic variables.
Variable name Explanation
Expected
relationship
TemperatureRange Dummy variable for
temperature in
suitable range for
cultivation (1 = in
range, 0 = out of
range)
+
MinimumPrecipitation Minimum
precipitation
−
MaximumPrecipitation Maximum
precipitation
−
MinimumPrecipitationSq Minimum
precipitation
squared
−
SlopeVariation Slope variation range + (opium)
− (marijuana)
ElevationVariation Elevation variation
range
+ (opium)
− (marijuana)
Notes: The symbols + and − indicate positive and negative, respectively.
194 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
Road data and police forces data were not available for
some municipalities. This missing data problem resulted in
a small number of municipalities with illicit drug eradications
being excluded from modeling. These include 18
municipalities with marijuana eradications and six municipalities
with opium eradications in 2010, leaving a sample
size of 534 municipalities for marijuana cultivation prediction
and 268 municipalities for opium illicit cultivation
prediction. It is important to point out that these dropped
observations are not among the municipalities with the
largest eradications. Rather, most of them are in the state
of Oaxaca (southern Mexico), which has a large number of
small and isolated communities. Moreover, despite that
some studies suggested a connection between economy
and drug activities (e.g., Thomas et al. 2004; Dube,
Garcia-Ponce, and Thon 2014), the analyses reported in
this article were not able to examine this dimension due to
the data unavailability at the time of the study.
Models
Ordinary least square (OLS) regression and geographically
weighted regression (GWR) were employed to
examine illicit cultivation at municipality level. Separate
models were created to analyze how the cultivations of
marijuana and opium poppy may be related to the environmental
and human variables. Moreover, the cultivation
of the same crop in the previous year and the cultivation of
the other crop in the same year are included in the models
as they may impact the illicit cultivation of a certain crop.
The models can be described in general as follows:
MarijuanaEradication10 ¼ f ðenvironment variables;
human variables;
OpiumEradication10;
MarijuanaEradication09Þ
þ error
(1)
OpiumEradication10 ¼ f ðenvironment variables;
human variables;
MarijuanEradication10;
OpiumEradication09Þ þ error
(2)
OLS is a global linear model that does not allow modeling
parameters to vary in space (Austin 2007). Global
models assume that the relationships being studied are spatially
stationary and that the parameters derived from the
regression are valid for the whole data (Foody 2004).
Global models cannot account for spatial nonstationarity
between the dependent variable and the predictors. The presence
of spatial autocorrelation among the residuals of an
OSL model commonly indicates the ineffectiveness of the
model. When the dependent variable responds differently to
the predictors across the study area, a local model should be a
better choice (Fotheringham, Brunsdon, and Charlton 2002;
Osborne and Suárez-Seoane 2002; Foody 2004).
GWR is an extension of the regression baseline
(Brunsdon, Fotheringham, and Charlton 1996; Kupfer
and Farris 2007); it allows the parameters to vary across
the study area. GWR was chosen over spatial autoregression
(SAR) for this study because it can reveal the spatial
variation of the relationship between the dependent and
independent variables (Brunsdon, Fotheringham, and
Charlton 1998; Fotheringham, Brunsdon, and Charlton
2002; Foody 2004). SAR models this relationship at the
global level; it accounts for the effect of spatial autocorrelation
but fails to reveal the nonstationarity of the relationship
among variables (Anselin 2005). For an ecosystem,
the spatial heterogeneity can result from systematical environmental,
physical, and biological processes (Legendre
and Legendre 1998). Because both the environmental factors
and the human factors may relate to the dependent
variables (i.e., illicit cultivation in 2010) differently across
the municipalities in Mexico, this study opted to compare
OLS global model with GWR local model.
Table 2. “Human” variables.
Variable name Explanation
Expected
relationship
MarijuanaEradication09 Marijuana crops eradication in hectares in 2009 (weighted by municipality area) +
MarijuanaEradication10 Marijuana crops eradication in hectares in 2010 (weighted by municipality area) +
OpiumEradication09 Opium crops eradication in hectares in 2009 (weighted by municipality area) +
OpiumEradication10 Opium crops eradication in hectares in 2010 (weighted by municipality area) +
DrugKillings Drug-related killings in 2010 (weighted by population in each municipality) −
HighwayDensity Length of federal administration or at least four-lane roads by municipality (weighted by
the municipality area)
+
PopulationDensity Population density by municipality −
PolicePresence Police force in 2010 (weighted by population in each municipality) −
ForestLandCover Dummy variable for majority land cover type (1 = forest; 0 = everything else) +
Notes: The symbols + and – indicate positive and negative, respectively.
Cartography and Geographic Information Science 195
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
MarijuanaEradication10i ¼ f ðenvironment variables at i
municipality; human variables
at i municipality;
OpiumEradication10i;
MarijuanaEradication09iÞ
þ errori
(3)
OpiumEradication10i ¼ f ðEnvironment Variables at i
municipality; HumanVariables
at i municipality;
MarijuanEradication10i;
OpiumEradication09iÞ þ errori
(4)
Results
A preliminary correlation analysis among the independent
variables revealed no significant collinearity issue. For
OLS, the exploratory regression tool in ArcGIS10.1 was
used to evaluate all possible combinations of the explanatory
variables to predict the dependent variables. Various
models were tested with between three and nine explanatory
variables, an acceptable adjusted R2
of 0.5 and a
maximum p-value of 0.05 for explanatory variables.
ArcGIS exploratory regression tool takes these parameter
settings and the given dependent and independent variables
to examine all the possible regression models; the
return from this tool is a report on the performance of the
best models, a summary of the significant explanatory
variables, and statistics of the relationship between each
explanatory variable and the dependent variable.
Examining opium cultivation
A total of 268 municipalities with eradications in 2010 were
included in the analyses. Among the many models that
passed the cut-off criterion of an adjusted R2
of 0.50
(Table 3), a five-variable model revealed the greatest explanatory
power with an adjusted R2
of 0.692 and an Akaike
information criterion (AIC) of −846.408 (Model (5)).
OpiumEradication10 ¼ β0 þ β1Ã
MarijuanaEradication10
þ β2Ã
OpiumEradication09
À β3Ã
DrugKillings
þ β4Ã
PopulationDensity
þ β5Ã
HighwayDensity þ error
(5)
Three variables in this model were significant
(OpiumEradication09, DrugKillings, and Population
Density). The joint F-statistic and the joint Wald statistic
were 121.08 and 124.19 at a significant level of 0.001,
showing a good fit of the model. However, the Koenker’s
studentized Breusch-Pagan (BP) statistic was significant
(p < 0.001), suggesting a nonstationarity problem. The
Jarque–Bera statistic revealed that the residuals are not
from a normal distribution (p < 0.001), a warning for
possible spatial autocorrelation (Figure 3). Moran’s I
value for the residuals is 0.028 with a Z-score of 2.061,
suggesting that the residuals are spatially clustered at 95%
confidence level.
Being a local model to account for nonstationarity,
GWR was applied to examine the variation of the relationship
between the explanatory variables and the opium
eradication. The best five-variable OLS model was
adapted for the local model.
OpiumEradication10i ¼ β0i þ β1iMarijuanaEradication10i
þ β2iOpiumEradication09i
þ β3iDrugKillingsi
þ β4iPopulationDensityi
þ β5iHighwayDensityi þ errori
(6)
where “i” denotes a measurement for municipality i. The
local variable coefficients in GWR are a function of
the spatial kernel surrounding municipality i (Foody 2004;
Kupfer and Farris 2007). That means that close observations
have a greater influence than distant ones on the resulting
coefficients. Because the data for this study are embedded in
Table 3. Summary exploratory regression for opium poppy.
Variable
Percentage
of the
tested
models
where the
variable
was
significant
Percentage
of the
models
where the
coefficient
was
positive
Percentage
of the
models
where the
coefficient
was
negative
OpiumEradication09 100 100 0
PopulationDensity 89.44 100 0
SlopeVariation 69.73 7.67 92.33
DrugKillings 59.83 50.89 49.11
ElevationVariation 52.72 82.53 17.47
ForestLandCover 50.89 98.12 1.88
MaximumPrecipitation 36.72 98.83 1.17
MarijuanaEradication10 20.21 100 0
HighwaysDensity 9.55 100 0
MinimumPrecipitation 9.34 40.17 59.83
TemperatureRange 1.47 0.51 99.49
PolicePresence 1.42 51.60 48.40
196 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
irregular, asymmetrical polygons, adaptive kernel was used
to make local estimation for the parameters.
The GWR model showed improvement compared with
the OLS model when the residuals and AIC are concerned
(Figures 3 and 4 and Table 4). The condition number,
which evaluates local collinearity, remained much lower
than the threshold of 30, suggesting that the model has
reliable, stable results (Table 4). The values of local R2
(Figure 5) ranged between 0.42 and 0.88, and their spatial
distribution suggested that the GWR model explains
opium cultivation better for central Pacific coast and the
north regions of Mexico than the southern region.
Examining marijuana cultivation
A total of 534 municipalities were included in the modeling
of marijuana cultivation. But the overall performance
of the OLS models for marijuana cultivation was not as
good as that for opium. The best OLS model has an
adjusted R2
value of 0.16, accounting for only 16% of
the variance in marijuana eradication in 2010.
MarijuanaEradication09 was revealed by multiple top models
as significant predicting variable and with the largest
coefficient; the variable is significant in more than half of
all the possible models, and it is always positively related to
the dependent variable (Table 5). Other common variables
across the different models are MaximumPrecipitation and
PolicePresence, being significant for around 25% of all
possible models. However, MaximumPrecipitation is
always negatively related to marijuana eradication, while
PolicePresence is always positive.
The OLS model with the highest adjusted R2
and the
lowest AIC to predict marijuana eradication is a ninevariable
model with explanatory variables including
ForestLandCover, MarijuanaEradication09, Opium
Eradication10, DrugKillings, PolicePresence, Elevation
Figure 3. Residuals of the OLS regression model for (a) marijuana and (b) opium poppy.
Cartography and Geographic Information Science 197
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
Variation, MinimumPrecipitation, MaximumPrecipitation,
and SlopeVariation. The last two variables are negatively
related to the dependent variable, while other variables are
positively related, and only MarijuanaEradication09 and
MaximumPrecipitation are significant. However, the
Jarque–Bera statistic, which tests for normality of residuals,
and the Koenker (BP) statistic both pointed to
autocorrelation among the residuals, indicating possible
nonstationarity in the relationship between the dependent
and independent variables (Table 6).
Since GWR does not allow for dummy variables, an
OLS model without the ForestLandCover variable must be
selected in order to compare with GWR model.
Furthermore, GWR model requests that the independent
variables are free from global and local multicollinearity.
To prevent global multicollinearity, variance inflation factor
(VIF) was forced to remain below 7.5. To preclude
local collinearity, the independent variables were analyzed
via thematic mapping and those with low spatial variability,
particularly SlopeVariation, ElevationVariation, and
MaximumPrecipitation, were excluded from the model.
The best OLS model that is free from collinearity problem
was used for comparison with GWR model. As a result, a
four-variable model was selected through ArcGIS OLS
exploratory regression analysis:
MarijuanaEradication10 ¼ β0 þ β1Ã
MarijuanaEradication09
þ β2Ã
DrugKillings
þ β3Ã
PolicePresence
þ β4Ã
HighwayDensity þ error
(7)
This four-variable model, measured by F-statistic and
joint Wald statistic, was significant at 0.01 level.
MarijuanaEradication09 was the only significant
Figure 4. Residuals of the GWR model for (a) marijuana and (b) opium poppy.
198 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
Figure 5. Local R2
of the GWR models for (a) marijuana and (b) opium poppy.
Table 4. OLS and GWR models predicting opium.
Variable Coefficient
Best OLS model
Adjusted R2
= 0.692; AIC = −846.408; maximum VIF = 1.262; JB = 0; K (BP) = 0
MarijuanaEradication10 0.1173
Killings −0.0027**
OpiumEradication09 0.8128*
PopulationDensity 0.0052**
HighwayDensity 0.0017
Comparing GWR model
Adjusted R2
= 0.734; AIC = −873.079; condition number < 13.392
MarijuanaEradication10 −0.0765 ~ 0.4407
DrugKillings −0.0065 ~ −0.0008
OpiumEradication09 0.2322 ~ 0.9284
PopulationDensity −0.0011 ~ 0.0103
HighwayDensity −0.0019 ~ 0.0073
Notes: *p < 0.001; **p < 0.05. VIF, variance inflation factor; JB, Jarque–Bera; K (BP), Koenker (BP).
Cartography and Geographic Information Science 199
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
(p < 0.05) predicting variable. However, the adjusted R2
was 0.094, indicating that only 9.4% of the variation in the
cultivation of marijuana was explained by the model. The
VIF was lower than 7.5, indicating no concerns about
multicollinearity. But a significant (p < 0.01) Jarque–
Bera statistic indicates that the residuals are not from a
normal distribution (Figure 3). Further investigation of the
residuals revealed a Moran’s I value of 0.074 with Z-score
as 6.462, suggesting spatial clustered residuals at a 99%
confidence level (Table 4).
A GWR model was run in comparison with the fourvariable
OLS model (Model (7)). The model detail is as
follows:
MarijuanaEradication10i ¼ β0i þ β1iMarijuanaEradication09i
þ β2iDrugKillingsi
þ β3iPolicePresencei
þ β4iHighwayDensityi þ errori
(8)
Overall, the GWR model showed an improved performance
(Figures 3 and 4). The AIC for the GWR model
was −1708.04, compared to AIC of −1662.05 for the fourvariable
OLS model. The GWR analyses revealed that the
relationships between marijuana cultivation and the environmental
and human variables were nonstationary. With
the exception of MarijuanaEradication09, the coefficients
for all the other variables changed the sign in the GWR
model compared with the OLS model (Table 6). The local
R2
showed that the model explains marijuana cultivation
better in the north areas of Mexico than in the south and
central regions (Figure 5). The condition number, which
evaluates local collinearity, remained below 5.1, suggesting
reliable, stable results.
Findings and discussion
GWR models performed better than OLS models for
explaining the patterns of opium and marijuana cultivation.
As a global model, OLS parameters are estimated
averages of the processes that may potentially exhibit
great variation (Fotheringham, Martin, and Brunsdon
2001). The spatial variation of the coefficient values for
GWR independent variables illustrates clearly the nonstationary
nature of the relationship between the independent
variables (i.e., the examined environmental and human
variables) and the dependent variables (i.e., illicit cultivation
of marijuana or opium). Moreover, the local R2
of
GWR models varies across municipalities. Higher explanation
power of the GWR models, as indicated by larger
local R2
in Figure 5, can be found in the areas where the
largest drug producers are and where drug cultivation has
been a tradition for decades.
The GWR model for opium poppy cultivation
appeared to perform much better compared to that for
marijuana cultivation. The GWR model for opium cultivation
explained between 59% and 88% of the variance in
Table 5. Summary on exploratory regressions for marijuana
eradications in 2010.
Variable
Percentage
of the
tested
models
where the
variable
was
significant
Percentage
of the
models
where the
coefficient
was
positive
Percentage
of the
models
where the
coefficient
was
negative
MarijuanaEradication09 50.89 100 0
MaximumPrecipitation 25.34 0 100
PolicePresence 24.89 100 0
MinimumPrecipitation 8.18 49.21 50.79
DrugKillings 5.89 96.39 3.61
OpiumEradication10 3.71 100 0
ForestLandCover 0.86 92.08 7.92
PopulationDensity 0.05 63.59 36.41
TemperatureRange 0 99.75 0.25
HighwaysDensity 0 98.32 1.68
SlopeVariation 0 0 100
ElevationVariation 0 49.11 50.89
Table 6. OLS and GWR models predicting marijuana.
Variable Coefficient
Best OLS model (nine-variable)
Adjusted R2
= 0.16; AIC = −1694.91; maximum VIF = 5.26;
JB = 0; K (BP) = 0
ForestLandCover 0.0044
MarijuanaEradication09 0.2141***
OpiumEradication10 0.0975
DrugKillings 0.0015
PolicePresence 0.0013
ElevationVariation 0.0110
MinimumPrecipitation 0.0064
MaximumPrecipitation −0.0127**
SlopeVariation −0.0559
Comparing OLS model (four-variable)
Adjusted R2
= 0.094; AIC = −1662.05; maximum VIF = 1.12;
JB = 0; K (BP) = 0.01
MarijuanaEradication09 0.2360**
DrugKillings 0.0006
PolicePresence 0.0039
HighwayDensity 0.0007
Comparing GWR model
Adjusted R2
= 0.174; AIC = −1708.04; condition
number < 5.089
MarijuanaEradication09 0.1377 ~ 0.8186
DrugKillings −0.0016 ~ 0.0004
PopulationDensity −0.0011 ~ 0.0061
HighwayDensity −0.0028 ~ 0.0027
Notes: **p < 0.05; ***p < 0.1. VIF, variance inflation factor; JB, Jarque–
Bera; K (BP), Koenker (BP).
200 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
the major drug-producing regions. The best results ranging
from 74% to 88% of local R2
were found in the
states of Guerrero, Michoacán, Jalisco, and Nayarit.
These were not the early drug cultivation regions in the
late 1800s, but they are among the top producers of
opium and heroin in Mexico nowadays, particularly
Guerrero (Bucardo et al. 2005). The explanation power
of the GWR model decreased to about 50% for the areas
that are new for opium cultivation, such as some municipalities
in Oaxaca, Veracruz, and Chiapas, in the southern
parts of the country. Most of these regions are not among
the top opium producers.
The local R2
varies from 7% to about 50% for the
marijuana GWR model. The lowest explanation power
explains between 7% and 18% of the variance in marijuana
cultivation, similar to the best OLS models. These
fall in areas that are relatively new for marijuana cultivation,
including Puebla, Quintana Roo, and Veracruz,
which are in the heart of central Mexico and the Gulf
Coast. The GWR model did not perform well for marijuana
cultivation in the municipalities in Guerrero and
Oaxaca states, which are heavy producers of opium poppies
and have produced marijuana for decades, though in
less quantity. On the contrary, the local models showed
apparent improvement in explaining marijuana cultivation
in the states where the drug has been harvested for decades,
including Sinaloa, Chihuahua, Durango, and Sonora.
The GWR model explained more than a third of the
variance for these areas.
It is interesting to note that the GWR models clearly
separate the regions that are traditional and larger drug
producers from the newer and less active areas for drug
cultivation, including both opium poppy and marijuana.
Overall, the GWR models show stronger explanation
power (with larger local R2
values) for the traditional
and larger drug-producing areas. Chihuahua, Durango,
and especially Sinaloa all are big for drug production
and trafficking, which is intertwined with a history of
local corruption and lawlessness (Astorga 2005), a characteristic
that makes producing marijuana easier than in
other states (Bouchard, Beauregard, and Kalacska 2013).
This suggests that, different from ordinary plants, the
illicit cultivation in these areas is closely related to the
human factors that are included in the models. In other
words, the social, cultural, and policing factors are significant
for the distribution of illicit cultivation. Nevertheless,
the fact that the models perform not as well for the
relatively new drug-producing regions indicates they
failed to capture certain factors for illicit drug growth in
these areas. It could be that these newly developed drug
cultivation areas have incorporated more adaptive measures
and new technologies, making illicit cultivation
less dependent on the traditional factors.
There are clear differences in the performance of the
OLS models for explaining marijuana and opium
cultivation. For opium poppy eradication, the overestimated
regions are about the same across the OLS and
GWR models, so are the underestimated regions, suggesting
the similarity in goodness of fit for both the models.
However, the OLS model for marijuana underestimates
almost all the major and traditional drug-producing
regions, while the GWR model exhibits both overestimation
and underestimation. The systematic error by the OLS
model for marijuana cultivation seems to confirm its limitation
in accounting for the spatial variations based on the
particular idiosyncrasy of certain regions in Mexico. These
variations may not follow administrative boundaries, but
they play a significant role in defining the patterns of illicit
cultivations. One such example is the gang influence in the
“golden triangle”, laying northwest of Mexico between the
states of Sinaloa, Chihuahua, and Durango (Astorga
2005). Increased gang activities may make it harder for
illicit crops being detected, leading to model underestimation.
Moreover, the fact that the OLS model for opium
cultivation performed better than that for marijuana cultivation
may suggest that the model failed to capture the
more complicated local factors impacting marijuana cultivation.
In recent years, the Mexican Army has found
several marijuana fields with complex irrigation systems,
materials that provide protection from the sun, and even
genetically modified plants that are resistant to pesticides
and that can be cut and removed without fully pulling out
the roots (Llana 2006; Reuters 2011). On the contrary, the
conditions required for growing opium poppy are relatively
straightforward and call for little modification.
Opium poppies develop best in regions featuring warm
temperature and moderate moisture, and they do not
require special irrigation, which makes the Sierra Madre
Mountains a perfect area for the crop to grow. Opium
poppy tends to perform poorly in the other areas of
Mexico where more tropical climates prevail (Duke 1983).
Overall, both human variables and environmental factors
were revealed to be associated with the illicit cultivation
activities at municipality level. Factors that are related
to the easiness of starting illicit cultivation, the presence
and effectiveness of law enforcement, and the magnitude
of drug activities are significantly related to the illicit crop
cultivation. Some environmental variables may affect drug
cultivation through impacting human factors. Slope variation
was a consistent predictor for opium poppy cultivation
in OLS models, being significant at about 70% of the
times. In addition to ensuring necessary climate conditions
for the crops, increased slope variation may help hiding
the crops from being found by law enforcement.
Population density was significant at almost 90% of the
times for the opium cultivation OLS models, but it showed
spatial variation in the GWR models. Population may
positively contribute to illicit cultivation as it provides a
pool of labor; it may be a negative factor, however,
because more people would increase the risk of cultivation
Cartography and Geographic Information Science 201
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
being detected and cracked down by police force. The
latter may explain the negative coefficients for population
density in GWR for some northern municipalities. It is
very important to note the spatial and temporal continuity
of illicit cultivation activities. As can be seen in Tables 5
and 7, the cultivation eradication was found to be significantly
related to that in the previous year in all of the OLS
models for opium poppy cultivation and more than half of
the models for marijuana cultivation. Furthermore, more
than 20% of the OLS models revealed that opium cultivation
is positively related to the cultivation of marijuana in
the same year.
The limitations of this study warrant some discussions
here. First, the fact that the eradication data were available
only at municipality level created limits to the study. GWR
model has proved that nonstationarity is embedded in the
relationship between illicit cultivation and the independent
variables. Because the data were set at a certain spatial
structure, it is impossible to test for the local variation
across different scales. Second, the environmental data
from INEGI were the mean annual minimum and maximum
temperature in degree Celsius and the mean annual
precipitation in millimeters. These data do not reflect the
real extremes of temperature or precipitation, which can be
critical for the survival and development of plants.
Moreover, the environmental variables were aggregated
to and recorded at municipality level. This process may
introduce errors. Future research is warranted to examine
illicit crop cultivation using site-specific data. Last, by
only including those municipalities with illicit drug eradications
into the models, the spatial continuity of the study
area is interrupted at some places (although not in the
major drug production areas). Although adaptive bands
were used for the GWR models, there may still be potential
impact for the modeling results, which should be
investigated by future studies.
Summary
Cultivation of marijuana and opium poppy in Mexico is
related to the distribution of favorable climate and geographical
conditions – the physical environment required
to grow the crops. But drug production is also a narrative
of decades of social and political processes, lately seasoned
by increasing access to high-tech means to help
plants grow faster and produce more of the raw materials
that make illicit drugs. The human factors have actually
begun to play an increasing role, sometimes making the
environmental factors more or less secondary while the
diversity of drug production and the easiness of managing
and sustaining the producing process are being more
important.
In fact, human factors that account for ineffective law
enforcement, the corruption in social and management
processes, and the easiness of the setup process for crop
growing are the most influential when trying to explain the
opium and marijuana cultivation in Mexico. Traditional
environmental variables like temperature and precipitation
cannot explain drug production by themselves; rather, to a
great extent, they may be significant for explaining the
patterns of illicit drug cultivation when they are examined
as part of the easiness of the setup process.
The drug cultivation in Mexico, however, shows a
notable spatial variation, reflecting differences in not
only climate and topography but also governance styles
and drug gang activity, among other factors. Furthermore,
more marijuana plantations are starting to appear in the
middle of the Mexican desert, supported by complex
water irrigation systems, and light, net-like covers to provide
protection from the sun and hiding from police surveillance
(Reuters 2011). All these add up to spatial
variation in the illicit crop growing process. Human
actions are making the crop cultivation respond differently
to the traditional explanatory variables across different
growing locations, thereby making researching outdoor
drug cultivation an even more challenging task.
References
Acock, M. C., R. C. Pausch, and B. Acock. 1997. “Growth and
Development of Opium Poppy (Papaver somniferum L.) as a
Function of Temperature.” Biotronics 26: 47–57.
Anselin, L. 2005. Exploring Spatial Data with GeoDaTM: A
Workbook. Urbana: Center for Spatially Integrated Social
Science, Department of Geography, University of Illinois.
https://geodacenter.asu.edu/system/files/geodaworkbook.pdf
Astorga, L. 2005. El Siglo de las Drogas: el Narcotrafico, del
Porfiriato al Nuevo Milenio. Mexico: Random House
Mondadori.
Austin, M. 2002. “Spatial Prediction of Species Distribution: An
Interface between Ecological Theory and Statistical
Modelling.” Ecological Modelling 157 (2–3): 101–118.
doi:10.1016/S0304-3800(02)00205-3.
Austin, M. 2007. “Species Distribution Models and Ecological
Theory: A Critical Assessment and Some Possible New
Approaches.” Ecological Modelling 200 (1–2): 1–19.
doi:10.1016/j.ecolmodel.2006.07.005.
Bahn, V., R. J. O’Connor, and W. Krohn. 2006. “Importance of
Spatial Autocorrelation in Modeling Bird Distributions at a
Continental Scale.” Ecography 29 (6): 835–844.
doi:10.1111/j.2006.0906-7590.04621.x.
Barry, S. C., and A. H. Welsh. 2002. “Generalized Additive
Modelling and Zero-Inflated Count Data.” Ecological
Modelling 157: 179–188. doi:10.1016/S0304-3800(02)
00194-1.
Booth, M. 1999. Opium: A History. New York: St. Martin’s
Griffin.
Bouchard, M., E. Beauregard, and M. Kalacska. 2013. “Journey
to Grow: Linking Process to Outcome in Target Site
Selection for Cannabis Cultivation.” Journal of Research in
Crime and Delinquency 50 (1): 33–52. doi:10.1177/
0022427811418001.
Brunsdon, C., S. Fotheringham, and M. Charlton. 1996.
“Geographically Weighted Regression: A Method for
Exploring Spatial Nonstationarity.” Geographical Analysis
28 (4): 281–298. doi:10.1111/j.1538-4632.1996.tb00936.x.
202 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
Brunsdon, C., S. Fotheringham, and M. Charlton. 1998.
“Geographically Weighted Regression – Modelling Spatial
Non-Stationarity.” Journal of the Royal Statistical Society:
Series D (The Statistician) 47 (3): 431–443. doi:10.1111/
1467-9884.00145.
Bucardo, J., K. C. Brouwer, C. Magis-Rodriguez, R. Ramos, M.
Fraga, S. G. Perez, T. L. Patterson, and S. A. Strathdee. 2005.
“Historical Trends in the Production and Consumption of
Illicit Drugs in Mexico: Implications for the Prevention of
Blood Borne Infections.” Drug and Alcohol Dependence 79
(3): 281–293. doi:10.1016/j.drugalcdep.2005.02.003.
Clarke, R. C. 1981. Marijuana Botany: An Advanced Study: The
Propagation and Breeding of Distinctive Cannabis, 109–
110. Berkeley, CA: RONIN.
Consejo Nacional de Evaluación de la Política de Desarrollo
Social (CONEVAL). 2010. La Pobreza Por Ingresos En
México. Mexico, DF: CONEVAL.
De la Herrán, J. 1980. “La migración campesina en Sinaloa.”
Ciencia y Universidad 13: 77–102.
Diniz-Filho, A. F., L. M. Bini, and B. A. Hawkins. 2003.
“Spatial Autocorrelation and Red Herrings in Geographical
Ecology.” Global Ecology and Biogeography 12 (1): 53–64.
doi:10.1046/j.1466-822X.2003.00322.x.
Dominguez, J., and R. Fernandez de Castro. 2001. The United
States and Mexico: Between Partnership and Conflict. New
York: Routledge.
Dormann, C. F., O. Purschke, J. R. García Márquez, S.
Lautenbach, and B. Schröder. 2008. “Components of
Uncertainty in Species Distribution Analysis: A Case Study
of the Great Grey Shrike.” Ecology 89 (12): 3371–3386.
doi:10.1890/07-1772.1.
Dube, O., O. Garcia-Ponce, and K. Thon. 2014. From Maize to
Haze: Agricultural Shocks and the Growth of the Mexican
Drug Sector. Working Paper 355. Washington, DC: Center
for Global Development. Accessed October 10. http://www.
cgdev.org/sites/default/files/maize-haize-agricultural-shocks-
growth-mexican-drug-sector_0.pdf.
Duke, J. A. 1983. “Papaver somniferum L.” Handbook of
Energy Crops, unpublished. http://www.hort.purdue.edu/
newcrop/duke_energy/papaver_somniferum.html.
Elith, J., and J. R. Leathwick. 2009. “Species Distribution
Models: Ecological Explanation and Prediction Across
Space and Time.” Annual Review of Ecology, Evolution,
and Systematics 40: 677–697. doi:10.1146/annurev.
ecolsys.110308.120159.
Foody, G. 2004. “Spatial Nonstationarity and Scale Dependency
in the Relationship between Species Richness and
Environmental Determinants for the Sub-Saharan Endemic
Avifauna.” Global Ecology and Biogeography 13 (4): 315–
320. doi:10.1111/j.1466-822X.2004.00097.x.
Fortes, P. S., A. Platonov, and L. S. Pereira. 2005. “GISAREG–A
GIS Based Irrigation Scheduling Simulation Model to
Support Improved Water Use.” Agricultural Water
Management 77 (1–3): 159–179. doi:10.1016/j.
agwat.2004.09.042.
Fotheringham, S., C. Brunsdon, and M. Charlton. 2002.
Geographically Weighted Regression: The Analysis of
Spatially Varying Relationships. Chichester: Wiley.
Fotheringham, S., C. Martin, and C. Brunsdon. 2001. “Spatial
Variations in School Performance: A Local Analysis Using
Geographically Weighted Regression.” Geographical &
Environmental Modelling 5 (1): 43–66. doi:10.1080/
13615930120032617.
Fung, D. S., and R. Welch. 1994. Modeling Canabis Cultivation in
North Georgia. Technical Paper, ASPRS/ACSM Annual
Convention & Exposition: Reno, Nevada, 217–220. Bethesda,
MD: American Society for Photogrammetry and Remote
Sensing-American Congress on Surveying and Mapping.
Guisan, A., and N. E. Zimmerman. 2000. “Predictive Habitat
Distribution Models in Ecology.” Ecological Modelling 135
(2–3): 147–186. doi:10.1016/S0304-3800(00)00354-9.
Hough, M., H. Warburton, B. Few, T. May, L. Man, J. Witton, and
P. J. Turnbull. 2003. A Growing Market: The Domestic
Cultivation of Cannabis. York: Joseph Rowntree Foundation.
Howell, T. 2002. “TR-03-01, Operation SABOT and Illicit Crop
Information Management Using Satellite Imagery.” Canadian
Police Research Centre (CPRC). Accessed March 31, 2014.
http://publications.gc.ca/collections/collection_2008/ps-sp/
PS63-2-2001-3E.pdf
Hurley, J., J. West, and J. Ehleringer. 2010. “Tracing Retail
Cannabis in the United States: Geographic Origin and
Cultivation Patterns.” International Journal of Drug
Policy 21 (3): 222–228. doi:10.1016/j.drugpo.2009.08.001.
Instituto Nacional de Estadísticas y Geografía (INEGI). 2005. “Guía
Para la Interpretación de Cartografía Climatológica.” Mexico.
Accessed October 26, 2013. http://www.inegi.org.mx/.
Instituto Nacional de Estadísticas y Geografía (INEGI). 2013.
“Catálogo de Municipios Nacionales.” Accessed October 26.
http://geoweb.inegi.org.mx/mgn2k/catalogo.jsp.
International Narcotics Control Strategy Report (INCSR). 2008,
2009. “Drug and Chemical Control.” Vol. I, Mexico Chapter.
Accessed October 24, 2013. http://www.state.gov/p/inl/rls/
nrcrpt/2009/vol1/116522.htm.
Iverson, L. R., A. Prasad, S. Matthews, and M. P. Peters. 2011.
“Lessons Learned While Integrating Habitat, Dispersal,
Disturbance, and Life-History Traits into Species Habitat
Models Under Climate Change.” Ecosystems 14 (6): 1005–
1020. doi:10.1007/s10021-011-9456-4.
Keitt, T. H., O. N. Bjørnstad, P. M. Dixon, and S. Citron-Pousty.
2002. “Accounting for Spatial Pattern When Modeling
Organism-Environment Interactions.” Ecography 25 (5):
616–625. doi:10.1034/j.1600-0587.2002.250509.x.
Kupfer, J., and C. Farris. 2007. “Incorporating Spatial Non‐
Stationarity of Regression Coefficients into Predictive
Vegetation Models.” Landscape Ecology 22 (6): 837–852.
doi:10.1007/s10980-006-9058-2.
Lee, A. H., K. Wang, J. A. Scott, K. K. W. Yau, and G. J.
Mclachlan. 2006. “Multi-Level Zero-Inflated Poisson
Regression Modelling of Correlated Count Data with
Excess Zeros.” Statistical Methods in Medical Research 15
(1): 47–61. doi:10.1191/0962280206sm429oa.
Legendre, P., and L. Legendre. 1998. Numerical Ecology.
Amsterdam: Elsevier Science.
Lennon, J. J. 2000. “Red-Shifts and Red Herrings in Geographical
Ecology.” Ecography 23 (1): 101–113. doi:10.1111/j.1600-
0587.2000.tb00265.x.
Lidicker, W. Z. 1975. “The Role of Dispersal in the Demography
of Small Mammals.” In Small Mammals: Their Productivity
and Population Dynamics, edited by F. B. Golley, et al.,
103–128. London: Cambridge University Press.
Liu, J. 2009. “A GIS-Based Tool for Modelling Large-Scale CropWater
Relations.” Environmental Modelling & Software 24
(3): 411–422. doi:10.1016/j.envsoft.2008.08.004.
Lizarraga, A., and O. Lizarraga. 2006. Narcotráfico, Violencia
y Emigración Femenina. Estudio de Caso. Mexico:
Universidad Autónoma de Sinaloa. Accessed March 31,
2014. http://estudiosdeldesarrollo.net/coloquio2006/docs2006/
11772.pdf
Llana, M S. 2006. “Mexican Army Invades Pot Plantations.”
USA Today, December 21. Accessed February 10, 2014.
Cartography and Geographic Information Science 203
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015
http://usatoday30.usatoday.com/news/world/2006-12-21-
mexico-drug-effort_x.htm
Marín, A. 2002. “Cambios en el financiamiento de las labores del
campo Mexicano desde las reformas estructurales de 1983
hasta la fecha. El caso de los municipios serranos de sinaloa”.
Revista de Antropología Experimental 2: 1–12.
Ortega, S., and E. Lopez Manon. 1987. Sinaloa, Textos de su
Historia. Vol. 2. Mexico: Instituto de Investigaciones Jose
María Luis Mora.
Osborne, P., and S. Suárez-Seoane. 2002. “Should Data Be
Partitioned Spatially before Building Large-Scale
Distribution Models?” Ecological Modelling 157 (2–3):
249–259. doi:10.1016/S0304-3800(02)00198-9.
Partelow, C. 2008. “Using GIS to Depict Resource Risk from
Probable Cannabis Cultivation Sites.” Master thesis, San
Jose State University. http://scholarworks.sjsu.edu/cgi/view
content.cgi?article=4584&context=etd_theses&sei-redir=1&
referer=http%3A%2F%2Fscholar.google.com%2Fscholar%
3Fq%3DGis%2Bcannabis%2Bcultivation%2Bpolygon%
2Bfile%26hl%3Den%26btnG%3DSearch%26as_sdt%3D1%
252C44%26as_sdtp%3Don#search=%22Gis%20cannabis%
20cultivation%20polygon%20file%22.
Resa Nestares, C. 2003. El Valor de las Exportaciones Mexicanas
de Drogas Ilegales 1961–2000. Colección Documentos,
Universidad Autónoma de Madrid.
Reuter, P. 1988. “Quantity Illusions and Paradoxes of Drug
Interdiction: Federal Intervention into Vice Policy.” Law
and Contemporary Problems 51 (1). doi:10.2307/1191721.
Reuters. 2011. “Mexico Marijuana Plantation Reportedly Biggest
Ever.” 14 July. Accessed May 1, 2012. http://www.huffing-
tonpost.com/2011/07/14/mexico-marijuana-plantation-_n_
898630.html#s309904.
Small, E., T. Pocock, and P. B. Cavers. 2003. “The Biology of
Canadian Weeds. 119. Cannabis sativa L.” Canadian
Journal of Plant Science 83: 217–237. doi:10.4141/P02-021.
Thomas, M., S. Hallman, M. Plaisent, and P. Bernard. 2004.
“Using IT and GIS to Improve Crop Assessments.”
Proceedings of the 15th Annual Conference of the IIMA
(International Information Management Association):
Technology Twenty Years After Orwell. Chicago, IL.
Accessed February 6, 2012. http://psugeo.org/Publications/
MJ%20Paper%20Rev%204%20IIMA.doc.
Thoumi, F. 2003. Illegal Drugs, Economy, and Society in the
Andes. Washington, DC: Woodrow Wilson Center Press.
Toro, M. C. 1995. Mexico’s War on Drugs. Boulder, CO: Lynne
Rienner.
UNODC. 2000–2012. “Illicit Crop Monitoring.” Accessed February
6, 2012. http://www.unodc.org/unodc/en/crop-monitoring/index.
html.
UNODC. 2011. “South-East Asia Opium Survey 2011. Lao
PDR, Myanmar.” Accessed February 8, 2012. http://www.
unodc.org/documents/crop-monitoring/sea/
SouthEastAsia_2011_web.pdf.
Václavík, T., J. A. Kupfer, and R. Meentemeyer. 2012.
“Accounting for Multi-Scale Spatial Autocorrelation
Improves Performance of Invasive Species Distribution
Modelling (ISDM).” Journal of Biogeography 39 (1): 42–
55. doi:10.1111/j.1365-2699.2011.02589.x.
Václavík, T., and R. Meentemeyer. 2009. “Invasive Species
Distribution Modeling (Isdm): Are Absence Data and
Dispersal Constraints Needed to Predict Actual
Distributions?” Ecological Modelling 220 (23): 3248–3258.
doi:10.1016/j.ecolmodel.2009.08.013.
Walthall, C., C. Daughtry, V. Vanderbilt, M. Higgins, T. Bobbe,
J. Lydon, and M. Kaul. 2003. “What Do We Know about
the Spectral Signatures of Illegal Cannabis Cultivation?”
Paper Presented at the 2003 ONDCP International
Technology Symposium, San Diego, CA, July 8–11.
Accessed March 31, 2014. http://psugeo.org/Interagency%
20MJ%20Study%20Project/AppendixB03.doc.
Wiens, J. A., D. Stralberg, D. Jongsomjit, C. Howell, and M. Snyder.
2009. “Colloquium Papers: Niches, Models, and Climate
Change: Assessing the Assumptions and Uncertainties.”
Proceedings of the National Academy of Sciences 106
(Supplement 2): 19729–19736. doi:10.1073/pnas.0901639106.
Youngers, C., and E. Rosin. 2005. Drugs and Democracy in
Latin America: The Impact of U.S. Policy. Boulder, CO:
Lynne Rienner.
204 M. Medel and Y. Lu
Downloadedby[UZHHauptbibliothek/ZentralbibliothekZürich]at01:0818March2015