Вы находитесь на странице: 1из 12

Applied Geography 35 (2012) 152e163

Contents lists available at SciVerse ScienceDirect

Applied Geography
journal homepage: www.elsevier.com/locate/apgeog

Forest cover and deforestation patterns in the Northern Andes (Lake Maracaibo
Basin): A synoptic assessment using MODIS and Landsat imagery
C.A. Portillo-Quintero a, *, A.M. Sanchez b, C.A. Valbuena b, Y.Y. Gonzalez b, J.T. Larreal a
a
b

Centro de Estudios Botnicos y Agroforestales, Instituto Venezolano de Investigaciones Cientcas, Apdo. 21827, Maracaibo, Venezuela
Departamento de Biologa, Facultad Experimental de Ciencias, La Universidad del Zulia, Maracaibo, Venezuela

a b s t r a c t
Keywords:
Tropical forests
Lake Maracaibo Basin
MODIS
Deforestation rates
Nested sampling
Landsat

This work synthesizes results from the application of land cover classication techniques and probability
sampling of satellite imagery for estimating forest extent and deforestation in Lake Maracaibo Basin
(Venezuela and Colombia). A forest map was produced using a semi-automated supervised classication
routine on MODIS 8-day 500-m imagery acquired in January 2010. Results show that forests occupy
29,710 km2 which represents 38% of the basins total terrestrial landmass. From this extent, 61% belongs
to Venezuela and 39% falls within the Colombian region. Findings indicate a drastic decrease in forest
cover as a result of anthropogenic agricultural and urban expansion, especially when compared to its
potential extent within the Maracaibo dry forests and the Venezuelan Andean montane forests ecoregions. Using time series of Landsat imagery, deforestation rates for the 1985e2010 time period were
calculated. The analysis was performed on 24 samples blocks of 10  10 km2 randomly allocated within
previously dened change probability strata. The general spatial distribution of deforestation rates was
predicted by a simple regression model between sample blocks and prior change probabilities at the
basin scale. Our results indicate that deforestation was low (<0.5%/y) in 85% of the basin, with highly
focalized deforestation fronts (intermediate-to-high rates, <2.5%/y) in three regions: a) the Motatn river
sub-basin in the Eastern Cordillera, b) the lower slopes of the Catatumbo river sub-basin and c) the
submontane regions of the Apn and Santa Ana river sub-basins. The results of this paper lead the way
for understanding current patterns in socioeconomic drivers of forest clearing in Lake Maracaibo Basin.
The study also demonstrates the feasibility of using alternatives methods to the time-consuming and
nancially unsustainable methods traditionally used at national and sub-national scale in Venezuela and
other Latin American countries.
2012 Elsevier Ltd. All rights reserved.

Introduction
Tropical deforestation has several deleterious effects including
species loss, soil erosion, siltation of streams and desestabilization
of watersheds as well as socioeconomic and cultural consequences
(Laurance & Bierregaard, 1997; Peres, 2011). It also leads to higher
emissions of CO2 and higher loss of carbon stock and is thought to
be a major contributor of greenhouse gas emissions in the world
(Achard et al., 2010). Its rapid pace is a major concern for human
societies around the world.
Data from the Forest and Agricultural Organization (FAO)
reports an approximate global annual change in forest area
of 52,110 km2/y for the period 2000e2010, ranging from 4100
to 39,970 km2/yr across regions of the world (FAO, 2011). This

* Corresponding author.
E-mail address: cportill@ivic.gob.ve (C.A. Portillo-Quintero).
0143-6228/$ e see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.apgeog.2012.06.015

information comes from per country annual deforestation rates


provided by national environmental agencies. These deforestation
rates have been tremendously valuable in providing information on
global, regional and intraregional trends in forest loss.
Comparatively to other parts of world, Latin American countries
have experienced the highest tropical deforestation rates during
the last decade. According to FAO (2011), forest cover change during
2000e2010 in Central America and South American (SA) countries
was estimated at 1.2% and 0.5% annual rates respectively,
compared to 0.49% annual rate in Africa, 0.19% in Asia and the
Pacic region, 0.03% in North America, and 0.09% in Europe.
Tropical countries (Colombia, Venezuela, Ecuador, Brazil, Bolivia
and Peru) account for 90% of total deforestation in SA. The highest
deforestation rates are found in Ecuador and Venezuela (1.8 and
0.6% annual rates, respectively), followed by Bolivia (0.5%), Brazil
(0.5%), Colombia (0.2%) and Peru (0.2%). In Central America,
the highest deforestation rates are found in Honduras (2.1%),
Nicaragua (2.0%), Guatemala (1.4%) and El Salvador (1.4%).

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

These estimates are often used to evaluate differences and similarities among socioeconomic drivers of forest cover change
(Forner, Blaser, Jotzo, & Robledo, 2006; Laurance, 1999). For international agencies and NGOs, this information is used to encourage
specic national environmental agencies to reinforce operational
strategies dedicated to stabilize or reduce elevated deforestation
rates in their countries.
However, in the case of Venezuela e the country with second
highest deforestation rate in South America according to FAO
(2011) e national deforestation estimates rely on reports that
have been questioned by independent research teams. Global
Forest Watch (2002, p. 134), Pacheco-Angulo, Aguado Suarez, and
Mollicone (2011) and FAO/JRC (2011) explain that these estimates
are derived from forecasts based on outdated forest cover data from
the 70s and early 80s. The Global Forest Watch (2002, p. 134) report
on the state of Venezuelas forests emphasizes the lack of clarity
and transparency regarding methodologies and source data used.
National and sub-national estimates from Venezuelan national
databases (at the province or ecoregional level) have the same
inconsistencies. Consequently, using this data for understanding
trends in deforestation rates within Venezuela might lead decision
makers to inaccurate conclusions on the future of its forests and
therefore, should be interpreted with caution.
Accurate and consistent forest cover data and estimates of GHG
emissions from tropical deforestation have been urged by the
United Nations Framework Convention on Climate Change
(UNFCCC) and the UNCBD Strategic Plan for Biodiversity
2011e2020. One of the technical issues emphasized at UNFCC on
the issue of reducing emissions from deforestation and degradation
(REDD), is that all estimates should be transparent, consistent and
as accurate as possible, as well as open to independent review
(Achard et al., 2010; Eckert, Ratsimba, Rakotondrasoa, Rajoelison, &
Ehrensperger, 2011).
Satellite remote sensing has played an important role in
generating more accurate and reliable information about forest
cover, vegetation type and land use changes. Tucker and
Townshend (2000) state that remote sensing procedures adopted
by scientic teams across the globe, have achieved high accuracies
through a combination of the use of high-resolution satellite data,
integration of automated methods and human interpreters. Such
procedures and technical capabilities have advanced since the early
1990s and standardized operational forest monitoring systems at
the national and sub-national level are now a feasible goal for most
developing countries (DeFries et al., 2007).
The application of automated land cover classication techniques using high resolution and moderate resolution imagery have
greatly improved our knowledge of forest extent at global, regional
and local scales since the 1970s (Eva et al., 2004; Fuller, 2006;
Portillo-Quintero & Snchez-Azofeifa, 2010). Techniques include
unsupervised and supervised classication using clustering algorithms (e.g. isodata, maximum likelihood classier) and nonparametrical methods such as machine learning, supported by
ground-truth information from high-resolution imagery, aerial
photography or eld data collection (Jensen, 2007).
Regarding deforestation mapping, one of the most practical and
robust methodologies applied today is the probability sampling of
satellite imagery (Achard et al., 2010; DeFries et al., 2007; Stehman,
Sohl, & Loveland, 2003). Probability sampling relies on the
assumption that deforestation events are not randomly distributed
in space, and therefore, particular attention is needed to ensure that
the statistical design is adequately sampled within areas of
potential deforestation (Achard et al., 2007). Comparatively to wallto-wall or complete coverage approaches, probability sampling of
satellite imagery is preferred because is less expensive, it has
a smaller classication error bias and allows for timelier reporting

153

of results (Stehman et al., 2003). Achard et al. (2002) developed


a statistical sampling strategy using satellite imagery to provide
a reliable measurement of change in tropical forests in a uniform,
independent and repeatable manner. In this approach, also known
as a nested sampling approach, wall-to-wall coarse resolution data
is used to produce a deforestation risk map (indicating spatial
trends in forest loss probabilities), which is later categorized in risk
strata. Sample sites among strata are then randomly selected using
a stratied sampling method. Finally, deforestation is calculated for
each sample site using time-series of high resolution data
(20e30 m spatial resolution). This approach has been applied in
several studies at the global (Hansen, Stehman et al., 2008),
regional (Broich, Stehman, Hansen, Potapov, & Shimabukuro, 2009;
Hansen, Shimabukuro, Potapov, & Pittman, 2008; Potapov, Hansen,
Stehman, Loveland, & Pittman, 2008; Stehman, Hansen, Broich, &
Potapov, 2011), national (Hansen, Shimabukuro et al., 2008;
Hansen et al., 2009) and sub-national level (Broich et al., 2011). For
its global assessments, the Global Forest Resource Assessment led
by FAO, has begun to estimate forest change for the periods
1990e2000e2005 based on samples of high-resolution satellite
imagery (e.g. Landsat TM, ETM, ASTER) in order to complement
national reports (FAO/JRC, 2011).
In South America, systematic monitoring systems based on
a similar nested sampling approach have only been developed by
the Brazilian National Space Agency (INPE) for the Legal Amazon
through a comprehensive annual national monitoring program
called PRODES. The Brazilian PRODES monitoring system for the
Brazilian Amazon uses a hot-spot approach to identify critical
areas based on the previous years monitoring (INPE, 2006). These
critical areas are priorities for analysis in the following year to be
used as samples for high resolution deforestation mapping using
Landsat imagery.
For Venezuela and other developing countries in South America,
the implementation of a standardized national deforestation
monitoring system in a cost-effective and efcient manner is at
hand, but the feasibility of a nested sampling design has only been
addressed within the Brazilian Legal Amazon (BLA) area. Operational strategies and technical issues outside the BLA might have to
take into account differences in the availability of ancillary information and the uniqueness of local geomorphological and
biophysical characteristics of the Andean and Llanos bioregions.
The purpose of this study was to quantify forest extent and
deforestation rates using a combination of remote sensing and GIS
techniques applied at the sub-national level in Venezuela and to
identify challenges and needs in the process of designing and
implementing a forest monitoring system in Venezuela. The study
focuses within the limits of the Lake Maracaibo Basin (LMB), one of
the countrys most degraded regions in terms of the historical
anthropogenic pressure posed on local dry and humid forests
(Bisbal, 1988; Llamozas et al., 2003, p. 555; Olson & Dinerstein,
2002; Rodrguez, Rojas-Surez, & Hernndez, 2010, p. 324).
Methods
Study area
Lake Maracaibo Basin, Venezuela
The Lake Maracaibo Basin (LMB) is a large coastal depression
located between the 9 and 12 N and the 70 and 72 W in northwestern Venezuela and north-eastern Colombia (Fig. 1). The region
is anked by the bifurcation of the Andean Cordillera in northern
South America and extends for approximately 121.823 km2
(Medina & Barboza, 2006; Rivas et al., 2005). Topographic relief is
variable across the basin, and it reaches altitudes of 3000 to 5000 m
in the western and eastern anks of the Andean cordilleras. Fig. 1

154

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

Fig. 1. Study area in western Venezuela: Lake Maracaibo Basin (LMB). The LMB is formed by fteen river sub-basins which drain into the lake. Each sub-basin is labeled with its
name. The dotted line indicates international boundaries between Colombia and Venezuela within the LMB. Shades of grey show altitude ranges within the LMB (0 m above sea
level black; >5000 m above sea level white).

shows the relative location of the study area and the geographical
distribution of its topographical relief. The cordilleras act as natural
barrier to winds coming from the northeeast, creating different
precipitation and evapotranspiration conditions that range from

arid, to semiarid, to humid environments in a north-south direction


(Medina & Barboza, 2006). The region is known for being one of the
largest crude oil reservoirs in Venezuela, which has been intensively exploited since the beginning of the twentieth century. The

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

lake itself and the lowland plains contain more than 5000 oil wells
and 20,000 km of oil pipes (Gardner, Cavaletto, Bootsma,
Lavrentyev, & Troncone, 1998).
Deciduous forest formations in the LMB, also known as the
Maracaibo Dry forests ecoregion, once dominated the colluvioalluvial lowlands surrounding the lake, but little is left after 60
years of human intervention. The area is now mostly dominated by
cattle ranching, agricultural activities, infrastructure for the oil
industry, residential areas and urbanisms (WWF, 2002). Several small
to large remnants of submontane and montane forests are distributed
along the western (Perij mountains) and south-eastern cordilleras
(Mrida cordillera) surrounded by populations of small and mediumsize-farm landholders that produce a diversity of crops (from vegetables to plantain and coffee plantations). Moist forests can be found
in the south-western region of the basin, along the ColombianVenezuelan international border, in the valley of the Catatumbo
river. Also known as the Catatumbo moist forests ecoregion, the core
area of these forests surround the lower slopes and lowlands of the
Perij and Mrida cordilleras. These are considered among the richest
in oral diversity in humid tropical forests of Venezuela and Colombia
(Viloria, 2002, p. 220; WWF, 2002).
Historical forest cover data for the LMB dates back to the 1920s
when the rst vegetation map of Venezuela was published by Henri
Pittier (Pacheco-Angulo et al., 2011; Rodrguez et al., 2010, p. 324).
The phytogeographical maps of Venezuela, as well as the Holdridge
Life zone maps contributed to have a better knowledge of the forest
distribution in the basin, but it was not until the early 80s that
forest cover was mapped by visual interpretation of satellite
imagery and aerial photography at 1:250,000 scale or less. Recently,
Rodrguez et al. (2010, p. 324) published an updated map of vegetation of Venezuela using the same techniques, and added information on the national and local trends of forest loss through the
comparison of historical maps using coarse resolution datasets.
Prior to the work presented in this paper, there are no studies on
forest cover and deforestation at the scale of the LMB. Its forest
cover has only been assessed as part of national inventories and few
independent studies at the landscape level.
Forest cover mapping
Satellite imagery
In 1998, NASA launched the rst Moderate Resolution Imaging
Spectroradiometer (MODIS) providing improved capabilities for
terrestrial satellite remote sensing aimed at meeting the needs of
global change research (Justice et al., 1998). The design of the land
imaging component combines characteristics of the AVHRR and the
Landsat Thematic Mapper sensors, adding spectral bands in the
middle and long-wave infrared (IR) and providing a spatial resolution of 250 m, 500 m and 1-km global coverage on a near daily
imaging capability. The MODIS sensor is a key instrument aboard
the Terra (EOS AM) and Aqua (EOS PM) satellites providing global
surface reectance data on a near-daily basis through the Land
Processes Distributed Active Archive Center e LPDAAC website (Giri
& Jenkins, 2005).
The imagery used for this analysis included the rst seven bands
from MODIS-Terra surface reectance 8-day 500-m L3 Global
composites data. This MODIS 8-day product provides Bands 1e7 at
500-m resolution in an 8-day gridded level-3 product in the
Sinusoidal projection. Each pixel contains the best possible observation during an 8-day period as selected on the basis of high
observation coverage, low view angle, the absence of clouds or
cloud shadow, and aerosol loading. We selected tiles h10/v07, h10/
v08, h11/v07 and h11/v08 from the rst week of January 2010, at
the beginning of the dry season, due to very low cloud presence
(Fig. 4a). An additional MODIS dataset was needed in order to

155

successfully identify seasonally dry forests. The dataset consisted


on a time-series of 240 mosaics of MODIS Vegetation Indices 16-day
L3 Global 250 m product (tiles h10/v07, h10/v08, h11/v07 and h11/
v08) from January 2001 to January 2010. Mosaicking and reprojection to UTM19/Datum WGS84 was performed using the MODIS
Reprojection Tool (available from http://lpdaac.usgs.gov/).
Landsat TM and ETM imagery (Level 1T e terrain corrected)
was acquired through the Global Land Cover Facility (http://www.
landcover.org) with very low cloud/haze cover (less than 10%) for
acquisition dates ranging from 2008 to 2011 (Circa 2010). A total of
nine Landsat satellite scenes (p006/r052; p006/r053; p006/r054;
p007/r052; p007/r053; p007/r054; p008/r052; p008/r053; p008/
r054) were necessary in order to cover the study area. This data was
used as supporting land cover information at higher resolution for
the automated classication of MODIS imagery and also, for
calculating deforestation rates in Section 2.3.
Ancillary data
Digital hydrographic information for geographic information
systems (GIS) was obtained from the USGSs HydroSHEDS database
(http://hydrosheds.cr.usgs.gov) in order to dene the geographical
limits of the LMB. Geographical information on the ecoregional
limits within the basin between tropical humid and dry forests was
obtained from the WWF terrestrial ecoregions GIS database
(available at www.worldwildlife.org). The WWF ecoregions database is a result of the exploration of existing global maps of oristic,
zoogeographic provinces and broad vegetation types, and the
consultation of regional experts (Olson et al., 2001). Additional
supporting information on forest extent at 1:250,000 was available
from a National GIS database of land use and land cover completed
in 2007 (known as the MARNOT project) and distributed by the
Venezuelan Ministry of Environment and Natural Resources
(MARN) (Rodrguez et al., 2010, p. 324). Although there is no
information available on the acquisition dates of the imagery used
in this project, different sources indicate that most of the national
database was produced by visual interpretation of SPOT, Landsat
TM and MSS imagery of 1988 and 1995 with updated information
using imagery from 2000 to 2001 (FRA, 1999; Olea, 2008; PachecoAngulo et al., 2011; Rodrguez et al., 2010, p. 324). This dataset will
be referred as the MARN2000 forest map.
Semi-automated classication methods
A supervised imagery classication methodology was implemented in order to map forests, agriculture (pastures and crops),
and urban land cover types for the entire study area. Locations
representing land cover types were identied by interpretation of
Landsat RGB 453 false color composites (Circa 2010) and vegetation
indices derived from the same Landsat scenes. Criteria used for the
interpretation of Landsat false-color composites is based on
knowledge gained from several eld visits to humid and dry forests
landscapes of the south-western, north-eastern and Andean
regions of Lake Maracaibo Basin. These were carried out during the
2009e2011 time period. Field observations of xeric scrub and dry
forest were achieved by traveling in paved and unpaved roads
within the Machango-Pueblo Viejo sub-basins and the Falcn-Zulia
highway in the eastern coast of Lake Maracaibo. Field observations
of humid forest landscapes were made across the panamerican
highway in the Trujillo, Merida, Tchira State and Zulia State
(Coln-La Fra). For each type of landscape, shades of color representing the spectral response of forest, pasture, urban and agricultural land cover in RGB 453 false color composites, were
identied. Training sites for every land cover type were then digitalized over a mosaic of RGB 453 Landsat composites. A total of
3983 geographical locations of different land cover types over the
study area were collected.

156

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

Fig. 2. Types of forested landscapes considered forest estate by the National Forest Management Law in Venezuela as seen from high resolution imagery: a) continuous native
forests, b) mosaic of native and secondary growth forest, c) secondary growth and native forests in agricultural matrices. Images represent 500  500 m forested landscapes which
corresponds to the spatial resolution of MODIS imagery used in this study.

A 500  500 m vectorial grid was used in order to choose locations representing forests at the minimum mapping unit of MODIS.
The high-resolution imagery available for visualization on Google
Earth (http://earth.google.com) was helpful for the collection of
training sites.
In Venezuela, tropical forests are ofcially dened as native
forests, represented by forest vegetation established by means of
natural ecological succession, with at least 10% occupied by
a canopy of trees, and a minimum area of 0.1 ha (National Forest
Management Law, 2008). The Forest estate denition in this
same law, however, includes not only native forests but also
forest plantations, trees outside forests and vegetation associated
to natural forests (e.g. secondary growth). In order to meet these
criteria, the identication of training sites involved the representation of forest dominated landscapes in LMB with the
exception of forest plantations for commercial use. The forest
concept used in this study includes native forests, secondary
growth and closed canopy layers formed by standing trees with
no understory as a result of human intervention. The types of
forested landscapes represented by training sites are exemplied
in Fig. 2.
A maximum likelihood algorithm was used to classify the
2010 MODIS scene mosaic (Fig. 4a), based on the spectral
signatures derived from training sites. Once general land cover
types were satisfactorily classied, further separation of spectral
mixture for identifying riparian forests in urban and agricultural
landscapes was achieved through unsupervised classication
(Isodata) of the MODIS scene within the agricultural and urban
masks.

For successfully identifying medium-size and small-size


seasonally dry forests fragments within the matrix of agricultural, riparian and xeric scrub vegetation in the lowland plains;
a preliminary vegetation map was produced using unsupervised
classication of a cloud-free Landsat ETM p007/r053 scene from
March 2003 within the Maracaibo dry forest ecoregion. This map
was used as a potential vegetation map within the dry forest
ecoregion. Based on the denition from Sanchez-Azofeifa et al.,
(2005) of dry forest e tropical forest where at least 50% of trees
present are drought deciduous e a method for identifying
phenological patterns within the potential vegetation map was
applied. The method consisted in calculating the standard deviation of Enhanced Vegetation Index (EVI) values from the
2001e2010 time-series of MODIS Vegetation Indices within the
potential vegetation mask. The method allowed discriminating
highly seasonal ecosystems from evergreen riparian and xeric
scrub vegetation by thresholding standard deviation values.
Thresholding was assisted by the use of higher resolution imagery
from Google Earth and interpretation of Landsat Circa 2010 falsecolor composites. The resultant dry forest layer was added to the
nal forest map.
A nal step of exhaustive cross-checking of the produced map
for inconsistencies (areas of obvious misclassication) using the
MODIS imagery and all the ancillary data and high resolution
imagery available, was necessary. Where needed, we performed
manual recoding and vectorial digitalization of misclassied pixels.
Additionally, for the nal map, we considered a minimum mapping
unit of 100-ha (1-km2) as a conservative threshold in order to avoid
including small isolated pixels arisen due to artifacts of the

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

MODIS-Terra surface
reflectance 8-Day 500 m
(January 2010)

MODIS EVI 16-Day


250m time series
2001-2010

Supervised classification

Application of a Landsatderived vegetation mask


(Xeric scrub/Riparian
and dry forests)

Agriculture

Forests

Separation of spectral
mixture
(Unsupervised
classification)

Pastures/Urban
Separation of spectral
mixture
(Unsupervised
classification)

Forest/NonForest

MARN2000/VCF
Forest map

Forest/NonForest

157

EVI standard deviation x


pixel
Thresholding to map dry
forests from other
vegetation

CLM2010
Forest/Non-Forest map

Image differencing:
Change probabilities 0- 100%
classified in four strata
Optimal allocation per stratum and random selection:
24 samples analyzed for deforestation rates from Landsat
TM and ETM+ (circa 1985 circa 2010)
Simple linear regression
model using 2000-2010
change probabilities

Estimates of deforestation rates


at the basin scale
1985-2010

Fig. 3. Schematic of methods used in this study.

classication method and the quality of the imagery and ancillary


data. For practical purposes, the forest map produced using this
methodology will be referred as the CLM2010 forest map.
Validation
We collected an independent sample of 221 points representing
forests and non-forest sites from the Landsat Circa 2010 dataset and
Google Earth high-resolution imagery. The areas selected for the
validation included forested landscapes with combinations of large
continuous and highly fragmented tracts of forests. The selection
was carried out by overlaying a 1 km  1 km grid over each Landsat
scene and generating points randomly within cells that represented 100% forest and 100% non-forest coverage. This approach
was applied in order to reduce the probability of errors typically
derived from comparing coarse-scale resolution data to higherresolution data (Foody, 2002). Finally, we constructed an error or
confusion matrix by comparing the tropical dry forest nal map to

the validation points (Congalton, 1991; Foody, 2002; Stehman &


Czaplewski, 1998). The result allowed us to estimate of overall
accuracy of the map.

Sampling design to quantify deforestation rates


The methodology applied for quantifying deforestation rates is
adapted from the work published by Achard et al. (2002), Hansen,
Roy et al. (2008), Hansen, Shimabukuro et al. (2008), Hansen,
Stehman et al. (2008), Hansen et al. (2009), Broich et al. (2009)
and Stehman et al. (2011). The method consists of three steps: i)
derivation of a change probability map showing general trends of
forest loss, ii) statistical design for selecting 10  10 km sample
blocks, iii) deforestation mapping in 10  10 km sample blocks
using Landsat time series, and iv) derivation of a deforestation
model based on a regression analysis.

158

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

Fig. 4. a) MODIS-Terra surface reectance 8-day 500-m L3 mosaic for January 2010 with very low cloud presence, b) forest map (CLM2010) produced by semi-automated classication of the MODIS scene.

Change probability map (2000e2010)


A change probability map showing general trends of vegetation
cover loss was produced through image differencing of the
MARN2000 forest map and the CLM2010 maps. For the humid
forests in the south-western region within the Colombian side,
where no data was available for 2000, we used the yearly MODIS
500-m Vegetation continuous elds (VCF) product corresponding
to the year 2001 (Hansen et al., 2003). This data is available from
LPDAAC website and it was used as baseline or initial state information in the analysis (Hansen et al., 2009). The MODIS VCF
product provides values of the percentage of vegetation cover per
pixel. Thresholding of the VCF values, aided by visual interpretation
of high resolution imagery, was necessary in order to satisfactorily
map humid forest extent.
The MARN2000/VCF and the CLM2010 datasets represent the best
forest cover data available for the LMB from two dates sufciently
apart to show signicant forest cover change at the basin level.
For change detection between datasets, a 10  10 km vectorial
grid was created within 5 km distance from all MARN2000/VCF
forest edges using Hawths tools (Beyer, 2004) in ArcGIS software.
The amount of forest cover (in km2) represented by MARN2000/VCF
and CLM2010 was then calculated for every 10  10 km cell, as well
as the percentage of change between dates. Given that both datasets
were produced using different classication methods, as well as
different satellite data and ancillary data, estimated changes in forest
cover could be attributed to geolocational errors and mapping
inconsistencies. Therefore, the percentage of change value was
referred as an estimator of the probability of change rather than
a direct estimator of forest loss. Similar to Hansen, Roy et al. (2008)
and Broich et al. (2009), probability of change values were classied
in four strata (low, intermediate, high and very high) using the Jenks
optimization method (natural breaks classication). Blocks with no
change detected were discarded from further analysis.
Nested sampling design
A stratied nested sampling design using optimal allocation was
used to determine the number of 10  10 km samples blocks per

stratum to be further analyzed using time-series of Landsat data


(Broich et al., 2009; Stehman et al., 2011). A total sample size
encompassing 5e10% of the total number of blocks was considered
a proper representative and manageable sample size range, based
on the results from similar global and regional studies (Broich et al.,
2009), as well as the computing capacity and man-hours effort
needed.
First, sample sizes were calculated for each stratum based on
Neyman Optimal allocation (Broich et al., 2009; Stehman et al.,
2011). The optimal allocation was determined using per-stratum
standard deviations of the percentage of change of all blocks
within each stratum. Once sample sizes per stratum were calculated, a random selection procedure of 10  10 km sample blocks
per stratum was performed using Hawths tool in ArcGIS.
Landsat deforestation mapping and regression estimation for basinlevel statistics
In addition to the Landsat Circa 2010 dataset, a set of Landsat TM
images acquired through the Global Land Cover Facility with very
low cloud/haze cover scenes (less than 10%) for dates around
1984e1989 (Circa 1985) was used. First, Landsat Circa 1985 and
Landsat Circa 2010 were subset using the area of sample blocks
selected by optimal allocation per stratum. Then, samples were
classied in Forest/Non-forest maps through unsupervised classication (Isodata clustering algorithm). Classication per block was
aided by interpretation of Landsat false color composites and
vegetation indices derived from the same Landsat scenes, high
resolution imagery from Google Earth as well as all ancillary data.
Once forest/non-forest maps were produced for both 1985 and
2010, the annual rate of deforestation (Rd) was calculated per
sample block using the following equation:

Rd%=y

A1  A2
*100
A1 *n

(1)

Where A1 is the forest cover at the initial state (Circa 1985), A2 is the
forest cover at the nal state (Circa 2010) and n is the time (years)

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

between A1 and A2 (Hernndez & Pozzobon, 2004). Four categories


were assigned to deforestation rates: Low (<0.5%/y), Intermediate
(0.5e2%/y), High (2e3.0%/y), Very high (>3%/y). This equation and
categories were used to allow comparability with previous national
deforestation statistics and studies at the landscape-level in
Venezuela (Cataln, 1991, p. 11; Hernndez & Pozzobon, 2004).
Finally, a regression estimator was employed to derive deforestation estimates at the basin level (Hansen et al., 2009). A simple
linear regression model was applied between calculated deforestation rates and correspondent MARN2003eCLM2010 change
probability values. The model was applied to all 377 blocks in order
to derive deforestation rates for the entire study area.
Fig. 3 shows a schematic of the main steps of the methodology
applied for mapping forest cover and estimating deforestation rates.
Results
Forest cover and geographical distribution
The semi-automated classication of MODIS 500-m imagery
allowed us to produce a forest cover map for Lake Maracaibo Basin
using scenes of January 2010 (Fig. 4b). Because the land cover
matrix is very diverse, especially in the lowland plains and submontane areas, pixels with early stages of forests had a higher
probability of being included in the non-forest class (Castro,
Snchez-Azofeifa, & Rivard, 2003; Portillo-Quintero & SnchezAzofeifa, 2010). The map produced is then, mostly representing
old growth, intermediate successional stages, and homogeneous
canopy layers formed by standing trees outside forests.
The forest map was evaluated using a confusion matrix which
showed a high overall accuracy of 90% (Producers accuracy: Forest
82%, Non-Forest 98%; Users accuracy: Forest 97%, Non-Forest 83%).
This product represents the rst forest cover map produced at the
basin level using automated classication methods of satellite
imagery. The map shows that the total extent of forest in Lake
Maracaibo Basin for 2010 is 29,710 km2 representing 38% of the
total terrestrial landmass of the basin. From this extent, 18,343 km2
(61.74%) belongs to Venezuela and 11,367 km2 (38.26%) falls within
the Colombian region.
Table 1 shows the forest extent for the 15 sub-basins forming
the Lake Maracaibo Basin. The Catatumbo river basin, in the southwestern region, contains the largest core area of forest (14,736 km2)
around the lower slopes and lowlands of the Perij and Mrida
cordilleras. Almost 78% of the forests of the Catatumbo river subbasin are located in the Colombian part of the sub-basin. The rest
is scattered along the Venezuela-Colombia border.
At the basin level, most forests are distributed along the western
and southern regions of the basin in the submontane and montane
altitudinal belts of the upper sub-basins. In fact 5069 km2 of forests
are located in the lowland plains of the basin or premontane belt
(<100 masl), 3861 km2 are located in the basimontain belt (>100
and <300 masl), 8166 km2 in the submontane belt (>300 and
<800 masl), and 12,611 km2 in the montane belt (>800 and
2800 masl). Furthermore, a comparison between the Olson and
Dinerstein (2002) delineation of terrestrial ecoregions and the
CLM2010 forest extent shows that 88% of the Maracaibo dry forest
ecoregion, 66% of the Venezuelan Andean montane forests, 58% of
the Catatumbo moist forests and only 6% of the Cordillera Oriental
montane forests have been already converted to anthropogenic
land use (Table 1).
Sampling of deforestation rates
The four strata derived from the comparison between the
MARN2000/VCF and CLM2010 maps were 0e5% (Stratum 1),

159

Table 1
Forest extent in the fteen river sub-basins and four ecoregions of the Lake
Maracaibo Basin.
Sub-basin

Sub-basin
area (km2)

Forest
cover (km2)

% Forested

Catatumbo
Guasare
Santa Ana
Chama
Motatn
El Palmar
Tucan
Escalante
Apn
Capaz
Pueblo Viejo
Machango
Motatn de los negros
Cocuiza
Cao Carrillo

27,548
5630
6714
5633
5139
4594
2175
4703
3593
1950
3109
1306
1641
2072
365

14,736
2636
2231
1616
1595
1306
1111
966
894
789
722
388
388
320
12

53
47
33
29
31
28
51
21
25
40
23
30
24
15
3

Ecoregion
(from Olson & Dinerstein, 2002)

Ecoregion
area (km2)

Forest
cover (km2)

% Forested

Cordillera oriental montane forests


Catatumbo moist forests
Venezuelan Andes montane forests
Maracaibo dry forests

12,509
20,945
16,517
29,787

11,793
8697
5698
3522

94
42
34
12

>5e17% (Stratum 2), >17e38% (Stratum 3), and >38% (Stratum 4)


change probabilities (Fig. 5a and b). A total of 33 sample blocks
accounting for 8.75% of the total number of blocks (n 377) were
initially analyzed with the allocation to the strata being 10, 13, 9 and
2 sample blocks respectively. During the processing of Landsat
imagery, samples with more than 50% of missing data caused by
cloud and shadow cover and data gaps from the Landsat 7 scan-line
corrector-off (SLC-off) malfunction were discarded. Finally, the
number of samples analyzed for stratum 1e4 was 7, 9, 7 and 1,
respectively. These 24 samples accounted for 6.36% of the total
number of sample blocks. Fig. 5c shows the location of the 24
sample blocks analyzed. Table 2a summarizes the results from
calculating deforestation rates per stratum.
Results for sample blocks analyzed within the low change
probability stratum (Stratum 1) showed a rate of 0.24  0.04%/y,
which indeed falls into the low deforestation rate category
according to Cataln (1991, p. 11). Values within strata 2 and 3
(intermediate and high change probability) were 0.83  0.61%/y
and 0.58  0.57%/y respectively, which in fact, falls into the intermediate deforestation rate category. Only one sample with an
annual deforestation rate of 0.42%/y was registered in the VeryHigh change probability stratum, also in the intermediate deforestation rate category. At the basin level, a mean annual deforestation rate of 0.56  0.24%/y was estimated.
Although mean values differed between strata, reecting only
two levels of deforestation (Table 2b), data obtained within the
intermediate level was highly variable and showed no signicant
statistical difference compared to the low level (ManneWhitney
U 34, p > 0.05). This is due to the fact that a large proportion of
sample blocks showed low deforestation values, even in the areas
were intermediate and high values were predicted (85% in stratum
1; 37% in stratum 2; 71% in stratum 3). Stratication did successfully allocate the areas were intermediate and high deforestation
rates are occurring in the Lake Maracaibo Basin. These were located
in the sub-montane and premontane zones of the Cordillera
Oriental montane forests, Catatumbo moist lowland forests and
Eastern cordillera.
A general spatial distribution of deforestation rates predicted by
the regression estimation model applied to the 377 blocks is shown in
Fig. 5d. The best-tted regression model for the 1985e2010 time

160

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

Fig. 5. a) Map indicating areas of forest change after comparing the MARN2000/VCF and the CLM2010 dataset; b) predicted change probabilities calculated for 10  10 km2 blocks
within 5 km of all mapped forests (n 377). Values were classied in four strata using the Jenks optimization method. c) Sample blocks selected per stratum using Neyman optimal
allocation. Deforestation rates were calculated per sample block using Landsat scenes from 1985 to 2010, d) deforestation rates estimated for 377 blocks based on a simple linear
regression model using calculated deforestation rates and MARN2000eCLM2010 change probability values (R2 0.29, p < 0.05; Rd 0.0263[change prob] 0.1482). The two
sample blocks showing the highest deforestation rates (>2%/y) are highlighted with black line boundaries.

period captured 29% of the variation in Landsat-derived forest cover


change (R2 0.29, p < 0.05, n 20; Rd 0.0263[Change
prob] 0.1482). Modeled deforestation rates range from 0.15%/y to
1.86%/y. The regression model shows a relative accurate distribution
of deforestation pressure across the basin, illustrating higher deforestation rates along Eastern Cordillera (Motatn river sub-basin), the
Central western cordillera (Perij submontane forests) and the lower
slopes of the western cordillera in the Catataumbo river sub-basin.
The model provides a coarse approximation to rates of forest cover
loss in Lake Maracaibo Basin during 1985e2010, but absolute deforestation rates calculated per block might not be necessarily captured
in the spatial model. For example, the model predicts higher deforestation values in the eastern cordillera (Motatn sub-river basin)
when, in fact, Landsat-derived deforestation rates were slightly

higher in the Catatumbo sub-basin lowlands (Mean Annual


Rd 0.89%/y, n 4) than sample blocks analyzed in the Motatn river
sub-basin (Mean Annual Rd 0.52%/y, n 5). Such bias toward the
Motatan river sub-basin was likely produced by higher prior change
probabilities predicted for the sub-basin comparatively to other subbasins. As a complement for the comprehension of deforestation
dynamics in Lake Maracaibo Basin, the two sample blocks that
showed high deforestation rates (>2%/y) with higher-resolution
mapping are marked in Fig. 5d.
Discussion
This work represents the rst synoptic approach to map forest
cover and deforestation in Lake Maracaibo Basin using automated

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

161

Table 2
LMB mean annual deforestation rates calculated using high resolution imagery: a) per change probabilities strata; b) per level of deforestation according to Cataln (1991).
Strata

Number of blocks

Number of samples

Landsat-derived mean annual deforestation rate (%/y)

Estimated range of valuesa

a)
1
2
3
4

189
125
56
7

7
9
7
1

0.24
0.83
0.58
0.42

0.20e0.28
0.22e1.44
0.01e1.15
e

Strata

b)
Low change probability (stratum 1)
Mid-to-high change probability (stratum 2,3,4)
Total
a

Number
of blocks

Number of
samples

Landsat-derived
mean annual
deforestation rate (%/y)a

Deforestation rate category


according to Cataln (1991)

189
188
377

7
17
24

0.24  0.04
0.70  0.32
0.56  0.24

Low
Intermediate
Intermediate

Estimated ranges are at the 95% condence level.

classication methods and nested sampling of high resolution


imagery. The selection of MODIS 8-day scenes from January 2010
(beginning of the dry season) with very low cloud presence (See
Fig. 4a) allowed to accurately map the distribution of forest cover at
the basin scale. To achieve satisfactory results from supervised
classication, an exhaustive collection of training sites was needed.
The presence of various types of forest vegetation as a result of
a north-south gradient in climatic conditions required a semiautomated classication that included validation, cross-checking
and manual digitalization of forest cover when needed. In the
lowland plains of the basin, the use of an EVI time series was
necessary in order to separate pixels representing dry forests from
riparian vegetation, crops and evergreen xeric scrub. The optimal
accuracy achieved in this map is also a contribution of the use of the
national concept of forest estate as a criterion for selecting training
sites and imagery classication. This concept includes the representation of fragmented forest landscapes, which are inherently
mapped with MODIS at a minimum mapping unit of 0.25 ha (500m pixel). Based on this concept, periodic monitoring forests within
the LMB is methodologically feasible using MODIS 500-m imagery.
Monitoring forests at the scale of the native forests concept within
LMB might require the use of MODIS imagery at a spatial resolution
of 250 m and/or sub-pixel classication methods in order to achieve satisfactory levels of accuracy.
The application of a nested sampling design to estimate deforestation rates generated information on the local patterns of forest
cover change in a statistically-sound and cost-effective way. Results
did not only allowed to distinguish two levels of deforestation rates
(low and intermediate) in the study area and an overall annual rate of
deforestation of 0.56%/y, but it also generated comprehensive
spatially-explicit information on the localization of deforestation
fronts within the basin. Periodic quantication of annual deforestation rates can be highly improved by calculating change probabilities
using the CLM2010 as the initial baseline map and future updated
versions of the CLM2010 map elaborated using similar automated
classication methods. Standard errors of deforestation values within
and among strata can be reduced by increasing the sample size per
stratum. An improvement on the estimation of change probabilities
and an increase in samples blocks per stratum can improve the
coefcient of determination (R2) of the simple linear regression
model estimator, thus providing higher condence estimates for
predicted deforestation rates. Furthermore, the selection of
a minimum proper sample size also needs to be based on affordable
and sustainable costs of man-hour effort and maintenance of
computing capacity in order to ensure its long-term application.
Regarding forest cover and deforestation patterns in Lake Maracaibo Basin, results indicate that forest cover has been reduced
drastically as a result of anthropogenic agricultural and urban

expansion, especially for the Maracaibo dry forests and the Venezuelan Andean montane forests ecoregions. However, most of this
forest cover change occurred during the 50s and 60s during the
construction of the Pan-American highway that borders all premontane areas of the basin. The only published ofcial rates for the
region are derived from visual interpretation of satellite imagery
and report an annual rate of deforestation of 4.4%/y for the
1977e1982 time period for Zulia state e which represents >50% of
the basin territory (GFW, 2002). For the last 25 years (1985e2010),
our results indicate that deforestation was low (<0.5%/y) in 85% of
the basin, with highly focalized deforestation fronts (intermediateto-high rates, <2.5%/y) in three regions: a) the Motatn river subbasin in the Eastern Cordillera, b) the lower slopes of the Catatumbo river sub-basin and c) the submontane regions of the Apn
and Santa Ana river sub-basins.
In general, deforestation drivers in these regions can be attributed to agricultural, cattle ranching and urban expansion (Strauss,
Fuenmayor, & Romero, 1992, pp. 93e141, www.gobiernoenlinea.
gob.ve). Activities within the Motatn river sub-basin in Trujillo
state are mainly agricultural (vegetables and ower crops) with
areas affected by the construction of roads and infrastructure for
tourism expansion. The submontane regions of the Apn and Santa
Ana river are mostly affected by cattle ranching in the lower slopes
and agricultural activities concentrated toward the Perij cordillera.
Cattle ranching and agriculture also dominate the Catatumbo river
sub-basin. In this area, the reported expansion of illegal crops in the
frontier between the Tibu department (Colombia) and the J.M.
Semprn municipality (Venezuela) might play an important role as
a deforestation driver in the region (Report from the Colombian
National Agency of Narcotics, available at http://www.dne.gov.co/?
idcategoria794).
Future research on the direct causes of deforestation at the
landscape scale (using high-resolution imagery) could help policymakers and local conservation practitioners to understand the
socioeconomic activities specically related to forest loss.
In addition, the results of this study suggest a revision of forest
conservation priorities within the LMB. For example, Llamozas et al.
(2003, p. 555) and Rodrguez et al. (2010, p. 324) report that the
Catatumbo moist forests are critically endangered. While this might
be true for the Venezuelan share of the Catatumbo sub-basin, it is
not an accurate condition for the forests of the whole sub-basin. In
fact, our results indicate that the Catatumbo river sub-basin
contains the largest core area of forest in the basin (14,736 km2)
but 78% of these are located in the Colombian share of the subbasin. This fact is especially important because it underpins the
importance of bi-national management of the Catatumbo river subbasin and the integral study of the Lake Maracaibo Basin beyond
national frontiers.

162

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163

Conclusions
The results of this paper lead the way for understanding current
patterns in socioeconomic drivers of forest clearing in Lake Maracaibo Basin. The study also demonstrated the feasibility of using
alternatives methods to those traditionally used at the local and
national scale in Venezuela. For Lake Maracaibo Basin and
Venezuela in general, the elaboration of forest maps based on the
visual interpretation of satellite imagery as well as the determination of deforestation rates using wall-to-wall approaches of high
resolution imagery, has proved to be time-consuming and nancially unsustainable. For local and national environmental institutions, the application of time and cost-efcient methods guarantees
the long-term generation of information for forest conservation
and management. The methodology here applied can be implemented by small teams of geographers specialized in GIS and
remote sensing using low cost and/or free available satellite data.
An ongoing national project led by the Venezuelan Space Agency
(ABAE) for launching the rst Venezuelan remote sensing satellite
(VRSS-1) in 2012e2013, promises to benet initiatives on forest
cover and deforestation monitoring even more. The VRSS-1 is
designed to provide high resolution optical imagery (VIS-NIR) at
a pixel size of 10-m which would improve the collection of training
and validation sites, and the calculation of deforestation rates from
sample blocks. In the meantime, further research needs to address
the methodological challenges for: a) deriving more accurate forest
cover maps at coarse and high spatial resolutions, b) dening the
minimum proper sample size per stratum and c) setting statistical
methods and quality standards for reporting estimates of annual
deforestation rates. Addressing these questions will help establishing a more accurate and transparent forest cover monitoring
system at sub-national and national scales.

Acknowledgments
This work was carried out as part of the Project Nbr. 1046 of the
Center for Botanical and Agroforestry Studies of the Venezuelan
Institute of Scientic Research (IVIC). We would like to acknowledge the aid from the NGO Fundacin Amigos del Lago (Banco
Occidental de Descuento) in early stages of this work. We thank
Edgar Ola and Miguel Pietrangeli for their comments on the
methods and the analysis of results.

References
Achard, F., DeFries, R., Eva, H., Hansen, M., Mayaux, P., & Stibig, H.-J. (2007).
Pantropical monitoring of deforestation. Environmental Research Letters, 2, 11.
http://dx.doi.org/10.1088/1748-9326/2/4/045022.
Achard, F., Eva, H. D., Stibig, H. J., Mayaux, P., Gallego, J., Richards, T., et al. (2002).
Determination of deforestation rates of the worlds humid tropical forests.
Science, 297, 999e1002.
Achard, F., Stibig, H.-J., Eva, H. D., Lindquist, E., Bouvet, A., Arino, O., et al. (2010).
Estimating tropical deforestation from Earth observation data. Carbon
Management, 1(2), 271e287.
Beyer, H. L. (2004). Hawths analysis tools for ArcGIS. Available at http://www.
spatialecology.com/htools.
Bisbal, F. J. (1988). Impacto humano sobre los hbitats de Venezuela. Interciencia,
13(5), 226e232.
Broich, M., Hansen, M. C., Potapov, P. V., Adusei, B., Lindquist, E. J., & Stehman, S. V.
(2011). Time-series analysis of multi-resolution optical remote sensing imagery
for quantifying forest cover loss in Sumatra and Kalimantan, Indonesia. International Journal of Applied Earth Observation and Geoinformation, 13, 277e291.
Broich, M., Stehman, S. V., Hansen, M. C., Potapov, P. V., & Shimabukuro, Y. E. (2009).
A comparison of sampling designs for estimating deforestation from Landsat
imagery: a case study of the Brazilian Legal Amazon. Remote Sensing of Environment, 113, 2448e2454.
Castro, K., Snchez-Azofeifa, G. A., & Rivard, B. (2003). Monitoring secondary
tropical forest using space-borne data: implications for Central America.
International Journal of Remote Sensing, 24(9), 1853e1894.

Cataln, A. (1991). El Proceso de Deforestacin en Venezuela entre 1975e1988.


Venezuela: Ministerio del Ambiente y de los Recursos Naturales
RenovablesCaracas.
Congalton, R. (1991). A review of assessing the accuracy of classication of remotely
sensed data. Remote Sensing of Environment, 37, 35e46.
DeFries, R., Achard, F., Brown, S., Herold, M., Murdiyarso, D., Schlamadinger, B., et al.
(2007). Earth observations for estimating greenhouse gas emissions from
deforestation in developing countries. Environmental Science & Policy, 10(4),
385e394.
Eckert, S., Ratsimba, H., Rakotondrasoa, L., Rajoelison, L., & Ehrensperger, A. (2011).
Deforestation and forest degradation monitoring and assessment of biomass
and carbon stock of lowland rainforest in the Analanjirofo region, Madagascar.
Forest Ecology and Management, 262, 1996e2007.
Eva, H. D., Belward, A. S., De Miranda, E. E., Di Bella, C. M., Gond, V., Huber, O., et al.
(2004). A land cover map of South America. Global Change Biology, 10(5), 731e744.
FAO. (2011). State of the Worlds forests 2011. Rome: Food and Agriculture Organization of the United Nations.
FAO/JRC. (2011). Global forest land use change from 1990 to 2005: Initial results from
a global remote sensing survey. Rome: Food and Agriculture Organization of the
United Nations/Joint Research Center. Available online at http://www.fao.org/
forestry/fra/remotesensingsurvey/en/.
Foody, G. (2002). Status of land cover classication accuracy assessment. Remote
Sensing of Environment, 80, 185e201.
Forner, C., Blaser, J., Jotzo, F., & Robledo, C. (2006). Keeping the forest for the
climates sake: avoiding deforestation in developing countries under the
UNFCCC. Climate Policy, 6(3), 275e294.
FRA. (1999). Memorias del Taller sobre el programa de evaluacin de los recursos
forestales en once pases latinoamericanos e FRA Working paper 10. Food and
Agriculture Organization of the United Nations. Available online at http://www.
fao.org/docrep/007/ad102s/AD102S00.htm#TopOfPage.
Fuller, D. O. (2006). Tropical forest monitoring and remote sensing: a new era of transparency in forest governance? Singapore Journal of Tropical Geography, 27, 15e29.
Gardner, W. S., Cavaletto, J. F., Bootsma, H. A., Lavrentyev, P. J., & Troncone, F. (1998).
Nitrogen cycling rates and light effects in tropical Lake Maracaibo, Venezuela.
Limnology and Oceanography, 43(8), 1814e1825.
Giri, C., & Jenkins, C. (2005). Land cover mapping of Greater Mesoamerica using
MODIS data. Canadian Journal of Remote Sensing, 31(4), 274e282.
Global Forest Watch. (2002). The state of Venezuelas forests: A case study of the
Guayana region. Fundacin Polar.
Hansen, M. C., DeFries, R., Townshend, J., Carroll, M., Dimiceli, C., & Sohlberg, R.
(2003). Global percent tree cover at a spatial resolution of 500 meters: First results
of the MODIS vegetation continuous elds algorithmIn Earth Interactions, Vol. 7,
Paper No. 10.
Hansen, M., Roy, D., Lindquist, E., Adusei, B., Justice, C., & Altstatt, A. (2008).
A method for integrating MODIS and Landsat data for systematic monitoring of
forest cover and change in the Congo Basin. Remote Sensing of Environment, 112,
2495e2513.
Hansen, M. C., Shimabukuro, Y., Potapov, P., & Pittman, K. (2008). Comparing annual
MODIS and PRODES forest cover change data for advancing monitoring of
Brazilian forest cover. Remote Sensing of Environment, 112, 3784e3793.
Hansen, M., Stehman, S., Potapov, P., Arunarwati, B., Stolle, F., & Pittman, K. (2009).
Quantifying changes in the rates of forest clearing in Indonesia from 1990 to
2005 using remotely sensed data sets. Environmental Research Letters, 4.
Hansen, M., Stehman, S., Potapov, P., Loveland, T., Townshend, J., DeFries, R., et al.
(2008). Humid tropical forest clearing from 2000 to 2005 quantied by using
multitemporal and multiresolution remotely sensed data. Proceedings of the
National Academy of Sciences, 105, 9439e9444.
Hernndez, E., & Pozzobon, E. (2004). Tasas de deforestacin en cuatro cuencas
montaosas del occidente de Venezuela. Revista Forestal Venezolana, 46, 35e42.
INPE. (2006). Monitoramento da Floresta Amazonica Brasileira por Satelite, Projeto
PRODES. Available online at http://www.obt.inpe.br/prodes/index.html.
Jensen, J. (2007). Remote sensing of environment: An earth resource perspective (2nd
ed.).In Prentice hall series in geographic information science, USA: Pearson
Prentice Hall.
Justice, C. O., Vermote, E., Townshend, J. R. G., Defries, R., Roy, D. P., Hall, D. K., et al.
(1998). The moderate resolution imaging spectroradiometer (MODIS): land
remote sensing for global change research. IEEE Transactions on Geosciences and
Remote Sensing, 36(4).
Laurance, W. (1999). Reections on the tropical deforestation crisis. Biological
Conservation, 91, 109e117.
Laurance, W., & Bierregaard, R. (1997). A crisis in the making. In W. Laurance, &
R. Bierregaard (Eds.), Tropical forest remnants: Ecology, management and
conservation of fragmented communities (pp. 612). USA: The University of Chicago Press.
Llamozas, S., Duno de Stefano, R., Meier, W., Riina, R., Stauffer, F., Aymard, G., et al.
(2003). Libro Rojo de la Flora Venezolana. Caracas: Provita, Fundacin Polar,
Fundacin Instituto Botnico de Venezuela Dr. Tobas Lasser.
Medina, E., & Barboza, F. (2006). Lagunas costeras del lago de Maracaibo: Distribucin, estatus y perspectivas de conservacin. Ecotropicos, 19(2), 128e139.
National Forest Management Law. (2008). Ley de bosques y gestin forestal. Gaceta
Ocial N 38.946 del 5 de junio de 2008. Available online at http://www.
minamb.gob.ve.
Olea, E. (2008). Diseo de una metodologa de anlisis SIG para evaluacin ambiental
de gasoductos en reas terrestres. MSc thesis in Geomatics. Universidad Central
De Venezuela.

C.A. Portillo-Quintero et al. / Applied Geography 35 (2012) 152e163


Olson, D. M., & Dinerstein, E. (2002). The global 200: priority ecoregions for global
conservation. Annals of the Missouri Botanical Garden, 89(2), 199e224.
Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Powell, G. V. N.,
Underwood, E. C., et al. (2001). Terrestrial ecoregions of the world: a new map
of life on earth. BioScience, 51, 933e938.
Pacheco-Angulo, C., Aguado Suarez, I., & Mollicone, D. (2011). Dinmica de la
deforestacin en Venezuela: anlisis de los cambios a partir de mapas histricos. Interciencia, 36(8), 578e586.
Peres, C. A. (2011). Conservation in sustainable-use tropical forest reserves.
Conservation Biology, 25(6), 1124e1129.
Portillo-Quintero, C., & Snchez-Azofeifa, G. A. (2010). Extent and conservation
of tropical dry forests in the Americas. Biological Conservation, 143(1),
144e155.
Potapov, P., Hansen, M. C., Stehman, S. V., Loveland, T. R., & Pittman, K. (2008).
Combining MODIS and Landsat imagery to estimate and map boreal forest
cover loss. Remote Sensing of Environment, 112, 3708e3719.
Rivas, Z., Sanchez, J., Troncone, F., Marquez, R., Ledo De Medina, H., Colina, M., et al.
(2005). Nitrgeno y fsforo totales de los ros tributarios al sistema lago de
Maracaibo, Venezuela. Interciencia, 34(5), 308e314.
Rodrguez, J. P., Rojas-Surez, F., & Hernndez, D. G. (2010). Libro Rojo de los Ecosistemas
Terrestres de Venezuela. Caracas, Venezuela: Provita, Shell Venezuela, Lenovo.

163

Snchez-Azofeifa, G. A., Quesada, M., Rodrguez, J. P., Nassar, J. M., Stoner, K. E.,
Castillo, A., et al. (2005). Research priorities for neotropical dry forests. Biotropica, 37(4), 477e485.
Stehman, R., & Czaplewski, R. (1998). Design and analysis for thematic map accuracy assessment: fundamental principles. Remote Sensing of Environment, 64,
331e344.
Stehman, S. V., Hansen, M. C., Broich, M., & Potapov, P. V. (2011). Adapting a global
stratied random sample for regional estimation of forest cover change derived
from satellite imagery. Remote Sensing of Environment, 115(2), 650e658.
Stehman, S. V., Sohl, T. L., & Loveland, T. R. (2003). Statistical sampling to characterize land-cover change in the U. S. Geological Survey land-cover trends
project. Remote Sensing of Environment, 86, 517e529.
Strauss, E., Fuenmayor, W., & Romero, J. (1992). Sntesis municipal (2nd ed.).In Atlas
del Estado Zulia.
Tucker, C. J., & Townshend, J. R. G. (2000). Strategies for monitoring tropical
deforestation using satellite data. International Journal of Remote Sensing,
21(6e7), 1461e1471.
Viloria, A. (2002). Episodios en la naturaleza limtrofe. Maracaibo, Cali: Universidad
Catlica Cecilio Acosta/Imgenes de la Naturaleza.
WWF. (2002). Terrestrial ecoregions reports. World Wildlife Fund. Available at http://
www.worldwildlife.org/wildworld/proles/terrestrial_nt.html.

Вам также может понравиться