Fifty Years of Deforestation and Forest Fragmentation in Madagascar

Article (PDF Available)inEnvironmental Conservation 34(04):325 - 333 · December 2007with 5,469 Reads
DOI: 10.1017/S0376892907004262
Cite this publication
Tropical deforestation is a key contributor to species extinction and climate change, yet the extent of tropical forests and their rate of destruction and degradation through fragmentation remain poorly known. Madagascar's forests are among the most biologically rich and unique in the world but, in spite of longstanding concern about their destruction, past estimates of forest cover and deforestation have varied widely. Analysis of aerial photographs (c. 1953) and Landsat images (c. 1973, c. 1990 and c. 2000) indicates that forest cover decreased by almost 40% from the 1950s to c. 2000, with a reduction in 'core forest' > 1 km from a non-forest edge of almost 80%. This forest destruction and degradation threaten thousands of species with extinction. Country-wide coverage of high-resolution validated forest cover and deforestation data enables the precise monitoring of trends in habitat extent and fragmentation critical for assessment of species' conservation status.
Environmental Conservation 34 (4): 1–9 © 2007 Foundation for Environmental Conservation doi:10.1017/S0376892907004262
Fifty years of deforestation and forest fragmentation in Madagascar
1Conservation International, 2011 Crystal Drive suite 500, Arlington VA 22202, USA, 2Code 923 NASA/GSFC, Greenbelt, Maryland 20771,
USA, and 3Conservation International, BP 5178, Antananarivo 101, Madagascar
Date submitted: 29 June 2006 Date accepted: 26 September 2007
Tropical deforestation is a key contributor to species
extinction and climate change, yet the extent of
tropical forests and their rate of destruction and
degradation through fragmentation remain poorly
known. Madagascar’s forests are among the most
biologically rich and unique in the world but, in spite
of longstanding concern about their destruction, past
estimates of forest cover and deforestation have varied
widely. Analysis of aerial photographs (c. 1953) and
Landsat images (c. 1973, c. 1990, and c. 2000) indicates
that forest cover decreased by almost 40% from the
1950s to c. 2000, with a reduction in ‘core forest’ >1km
from a non-forest edge of almost 80%. This forest
destruction and degradation threaten thousands of
species with extinction. Country-wide coverage of high-
resolution validated forest cover and deforestation data
enables the precise monitoring of trends in habitat
extent and fragmentation critical for assessment of
species’ conservation status.
Keywords: deforestation, forest, fragmentation, Madagascar,
remote sensing
Tropical forests and deforestation
Tropical forests cover less than 10% of Earth’s terrestrial
surface (Mayaux et al. 2005), yet they are thought to host at
least 50% of terrestrial species (Lovejoy 1997) and contain
45% of the above-ground carbon in vegetation (Watson et al.
2000). The annual global deforestation rate of humid tropical
forests is estimated to have been 0.5% between 1990 and
1997, with regional annual rates of up to 5.9% (Achard
et al. 2002). However, large uncertainties still exist, with a
range of ±24% at the 95% confidence level for the global
estimate, and ±56% for Latin America. Achard et al.’s (2002)
estimate for annual deforestation of humid tropical forests at
the global level from 1990–1997 is 23% less than that of the
United Nation’s Food and Agriculture Organization (FAO
2000) from 1990–2000, and 41% less for Africa. Uncertainties
Correspondence: Grady Harper Tel: +1 703 341 2761 Fax: +1 703
979 2514 e-mail:
aside, much forest data produced to date is of limited utility.
FAO forest studies do not produce spatially explicit maps,
which are essential for forest fragmentation analysis, and for
analyses incorporating other spatial datasets. Maps of 1 km
resolution are an improvement, but are still of limited use for
fragmentation analysis and are on a scale that misses much
detail relevant to biodiversity conservation. More precise and
detailed information on the extent and distribution of tropical
forests and their rate of clearing are essential to estimating
threats to biological diversity and carbon emissions.
Deforestation threatens species survival and diminishes
biodiversity by destroying forest habitat, creating forest
fragments too small to maintain viable populations and
increasing ‘edge effects’ at forest/non-forest interfaces (Harris
1984). Edge effects due to fragmentation typically affect
an area several times larger than the forest destruction
itself (Harris 1984; Skole & Tucker 1993), affecting micro-
meteorology over short distances (Kapos 1989), and increasing
exposure to damaging winds (Ferreira & Laurance 1997), fire
frequency (Cochrane 2001), and access for livestock, other
non-forest animals and hunters (Enserink 1999; Cullen et al.
2000). To assess the impact of tropical deforestation on
biological diversity, not only the area deforested, but also the
isolation of forest patches and the area of edge habitat must be
Madagascar’s separation from Africa approximately 165
million years ago, and from India 70 million years
ago (Rakotosamimanana 2003) is reflected in extremely
high biological endemism (Table 1). More than 90% of
Madagascar’s endemic animal species live exclusively in
forest or woodland (Dufils 2003). The tropical forests of
Madagascar are among the highest priority areas in the world
for biodiversity conservation (Myers et al. 2000).
Estimates of forest and woodland cover prior to human
arrival between the fourth and seventh centuries ad have
varied widely, with some arguing that forest covered 90%
or more of the island (Humbert & Cours Darne 1965),
while others argue that it was less (Kull 2000). By 1600
ad, deforestation was reportedly already advanced in the
central highlands, with the use of fire in zebu cattle grazing
and slash-and-burn agriculture playing an important role
(Gade 1996). By the late 19th century, concern about forest
destruction had led to enactment of laws against agricultural
2G. J. Harper et al.
Table 1 Species richness and endemism in Madagascar.
Taxon No. of species % Endemic Source
Plants 12 000 85% Gautier & Goodman (2003)
Birds 209 (breeding) 51% Hawkins & Goodman (2003)
Mammals 117 90% Garbutt (1999)
Reptiles 346 >90% Raxworthy (2003)
Amphibians 199 99% Glaw & Vences (2003)
Table 2 Historical estimates of Madagascar’s forest cover. Secondary and degraded forest classes not included in cover estimates from
Humbert and Cours Darne (1965), Faramalala (1988) and the IEFN (1996). Estimates include plantations. Estimate does not include
Source Time of data collection Total forest area (km2) Data source
Blasco (1965), Humbert & Cours
Darne (1965)
1949–1957 159 959 1:40 000 to 1:50 000 aerial
Faramalala (1988)1972–1979 106 370 Landsat MSS photo interpretation
(10% cloud cover)
Green & Sussman (1990) 1984–1985 38 000 (humid forest only) Partial Landsat MSS photo
coverage of eastern forests
IEFN (1996)1990–1994 103 010 Landsat TM5, visual
Mayaux et al. (2000) 1998–1999 101 041 SPOT-4 VEGETATION 1-km
digital analysis
FAO (2000)1990 129 010 Interview
FAO (2000)2000 117 270 Interview
FAO (2000)2000 113 770 Interview
burning (Jarosz 1993), to little effect. A human population
of 17.9 million, growing at 2.8% yr1(UNPF [United
Nations Population Fund] 2004), coupled with widespread
dependence on subsistence agriculture and fuelwood cutting,
makes deforestation in Madagascar difficult to slow. Recent
field studies have confirmed negative effects of deforestation
and forest fragmentation in all of Madagascar’s forested
regions (Langrand & Wilm´
e 1997; Vallan 2000; Watson
et al. 2004).
The first systematic forest map of Madagascar, produced by
F. Blasco based on visual interpretation of aerial photography
from 1949–1957 (Blasco 1965; Humbert & Cours Darne 1965),
reported 159 959 km2of forest and mangrove suffering little or
no degradation (Table 2). Faramalala estimated forest cover in
c. 1973 to be 106 400 km2(Faramalala 1988; IEFN [Inventaire
Ecologique et Forestier National] 1996) based on visual
interpretation of 1:1 000 000 scale Landsat image prints. A
national forest inventory (the Inventaire Ecologique et Forestier
National [IEFN]) based on visual interpretation of Landsat
5 images, indicated 103 000 km2of ‘little- to non-degraded’
forest in 1994 (IEFN 1996). The FAO (2000) reported total
forest cover (including natural forest and plantations) in 1990
to be 129 010 km2and cover in 2000 of 117 270 km2;they
estimated natural forest cover in 2000 to be 113 770 km2(FAO
2000). Using 1-km SPOT Vegetation data, Mayaux et al.
(2000) estimated there was 101 000 km2of primary forest in
The differences among these estimates are caused by
differences in definitions of forest, mapping techniques
and resolution of data used. Work in other regions has
demonstrated that forest cover, fragmentation and clearance
estimates are most accurate when produced from digital
analysis of high-resolution images covering the entire study
area, used in conjunction with ground or aerial verification
(Townshend & Justice 1988; Tucker & Townshend 2000;
Steininger et al. 2001).
For this study, we used orthorectified Landsat images from
NASA’s Geocover project (Tucker et al. 2004) for the 1970s,
c. 1990 and c. 2000, and we digitized the Humbert and
Cours Darne (1965) map from the 1950s. We mapped humid
and dry forest, spiny forest and woodland, mangrove, non-
forest, water cloud/cloud shadow. Similar to previous studies
of tropical forest cover (see for example Skole & Tucker
1993, Steininger et al. 2001), we defined ‘forest’ as areas of
primary vegetation dominated by tree cover at least seven
meters in height, with neighbouring trees crowns touching
or overlapping when in full leaf. In practice, this means
that the canopy is at least 80% closed. ‘Spiny forest and
woodland’ is primary vegetation dominated by closed-canopy
Forest change in Madagascar 3
trees or shrubs in the arid southern and south-western regions
of Madagascar, sometimes as low as two meters in height
in the extreme south. We did not include open-canopy
areas, secondary formations or plantations in our estimates of
forest and woodland areas. Lightly degraded primary forest
and mature secondary forest may be indistinguishable from
primary forest in Landsat imagery. However, we saw little
evidence of forest regeneration in Madagascar. Thus, the
forest classes include virtually all natural forest habitat upon
which 90% of Madagascar’s fauna depend.
Mapping methods
For the 1970s, we used Landsat Multispectral Scanner (MSS)
data from the period 1972–1979, predominantly 1973; for
c. 1990, we used Landsat Thematic Mapper (TM) data from
1989–1996, predominantly 1990; and for c. 2000, we used
Landsat Enhanced Thematic Mapper Plus (ETM+)data
from 1999–2001, predominantly 2000. All dates of satellite
imagery were co-registered to sub-pixel precision to minimize
false change caused by locational inconsistency between dates.
Our analyses were conducted at a 57-m spatial resolution for
the 1970s and at a 28.5-m spatial resolution for c. 1990 and
c. 2000.
The c. 1990 and c. 2000 data were classified together in a
single multi-date image to produce a direct estimate of change.
Classification of multi-date images, rather than classifying
single-date images individually and then combining them to
derive change estimates, reduces false-change errors caused
by differences between image dates in vegetation phenology,
illumination conditions and atmospheric interference.
A supervised methodology was used to classify each two-
date ‘image pair’. Our classification was based on a simple set
of classes: forest, non-forest, water, cloud/shade (no data),
and mangrove. Because we classified two-date images, we had
to train the classification for all observed combinations of these
basic classes (for example forest to forest, forest to non-forest,
non-forest to cloud). For these ‘basic class combinations’ it was
often necessary to create multiple sub-classes or ‘signatures’,
in order to capture the full range of spectral variation in an
image pair. Each of these signatures, in turn, consisted of a
number of polygonal training sites drawn on top of the satellite
image by the classifier. These training sites were identified
through visual interpretation of the satellite imagery informed
by literature research, consultation with biologists familiar
with Madagascar’s landscape, purchased aerial photos and
five days of overflights.
The process of classification was iterative. A set of
signatures representing all observed combinations of basic
classes was created; the classification was run and the resulting
thematic image was inspected for errors with reference to
both dates of the satellite image pair; errors were corrected by
editing the training sites of existing signatures and/or creating
additional signatures. This iterative process continued until
visual inspection of the classification revealed no further
obvious errors. The sub-classes of the final classifications
were recoded into the general classes of the final product.
The recoded classifications were filtered using two passes of
a three-by-three orthogonal neighbourhood majority filter,
followed by a 2-ha ‘eliminate’ function in Erdas Imagine
8.4. The filtered classifications were then joined to produce a
seamless country mosaic.
We classified the 1970s images separately because of their
lower spatial resolution, and different spectral bands and
radiometric sensitivity. This imagery was analysed with
reference to the c. 1990–c. 2000 images and classifications
to minimize the aforementioned sources of error. The 1970s
analysis was merged with the c. 1990–c. 2000 analysis to create
a three-date map of forest cover and deforestation.
We used a number of rules in combining the 1970s data
with the c. 1990–c. 2000 data in order to minimize errors in
the cover and change estimates. For example, areas that were
seen to be forest in either date in the c. 1990–c. 2000 map, but
as non-forest or water in the 1970s map, were recoded to be
forest in the 1970s, because the superior quality of the later
imagery made it more reliable.
Internal geometric differences between the 1950s photo-
based data and the Landsat data of later years made it
impractical to merge the 1950s map with the Landsat-based
map. Thus, only an aggregate numeric forest change estimate
could be made for the 1950s–1970s period, versus the spatially
explicit forest change estimates for the 1970s–c. 1990 and c.
1990–c. 2000 periods.
A bioclimatic mask was used to separate the mapped forest
into humid forest, dry forest and spiny forest; mangrove forest
was already a distinct class (Fig. 1). The bioclimatic mask is
based in part on the Missouri Botanical Gardens ‘Bioclimate
5’ product (Schatz & Lescot 2003), the humid zone being a
combination of the ‘humid’, ‘sub-humid’ and ‘montane’ zones
in Bioclimate 5. We defined the spiny forest and woodland
zone using local expertise, and defined the dry forest zone as
the remainder of the country.
Calculation of deforestation rates
The extent and rate of forest loss in each forest class was
calculated for each two-date time period using only pixels
that were cloud-free in both dates in question. Forest extent
and fragmentation indices were calculated for each of these
four forest classes for each date.
We calculated average annual deforestation rates from the
mosaic map, based on change intervals of 20 years (c. 1953–
c. 1973), 17 years (c. 1973–c. 1990), and 10 years (c. 1990–
c. 2000). Using these average intervals instead of the actual
dates of each image pair classified decreases the temporal
precision of the calculations. But, given the large study area
and the large number of images, we believe the impact on the
accuracy of our figures is small, especially for the c. 1990–
c. 2000 period when most image dates were tightly clustered
around the target dates of 1990 and 2000.
4G. J. Harper et al.
Figure 1 Madagascar forest cover from the 1950s to c. 2000. Forest cover changes from the 1970s to c. 2000 are shown in the main figure,
and forest cover in the 1950s is shown in the lower-right inset.
Forest change in Madagascar 5
Table 3 Madagascar’s known forest cover 1950s–c. 2000. Known forest cover is visible forest plus cloud- or shade-obscured areas that
were visible forest at a later date. Unknown indicates an area obscured by cloud or shadow at a given date, whose land cover type cannot be
deduced with certainty from data from earlier or later dates. n/a =not available. The increase in spiny forest cover from the 1950s to the
1970s seems likely to have been due to differences in forest definition and mapping methods.
Forest Cover Type 1950s aerial photographs 1970s MSS data c. 1990 TM data c. 2000 ETM data
Humid (km2) 87 656 68 760 52 343 41 668
Dry (km2) 42 521 40 277 27 118 24 570
Spiny (km2)29 782 30 298 24 200 21 322
Mangrove (km2) n/a n/a 2396 2261
Total known forest (km2)159 959 141 731 106 057 89 821
Total known forest (%)27.0 23.9 17.9 15.1
Total area of unknown cover type (km2)0 34 433 5003 11 244
Total area of unknown cover type (%)0 5.8 0.8 1.9
Table 4 Madagascar’s deforestation 1950s–c. 2000. Deforestation rates were calculated over average time periods of 20, 17 and 10 years.
Figures for the actual area deforested in the 1950s–1970s are omitted because deforestation was not directly observed for this time range;
percent deforestation rates are calculated from aggregate forest areas. n/a =not available.
Forest Cover Type 1950s–1970s 1970s–c. 1990 c. 1990–c. 2000
Observed deforestation over time interval (km2)
Humid 14 822 3220
Dry 13 116 1982
Spiny 6097 2817
Mangroves n/a 55
Total 34 035 8074
Observed deforestation (%yr1)
Humid 0.6 1.7 0.8
Dry 0.2 1.9 0.7
Spiny -0.1 1.2 1.2
Mangroves n/a n/a 0.2
Total 0.3 1.7 0.9
Fragmentation was represented by two variables: forest patch
size and proximity of forest to non-forest edge. The presence
of clouds required us to make a number of assumptions in
order to provide the most realistic estimates of fragmentation.
Cloud cover can make a large forest patch appear like two
or more smaller patches. To minimize this effect, for a given
date, any areas of cloud cover that were forested in the most
recent previous cloud-free date were counted as forest. This
was done only for the assessment of fragmentation and thus the
total areas for forest and woodland in Tables 3 and 4 differed
from those in Table 5. We defined forest and woodland edge
habitat as areas within a specified distance from non-forest
patches >5ha.
We collected GPS-linked digital photography and video
imagery during five days of low altitude flights in September
2002. We used one subset of these data to assist interpretation
and another to estimate the accuracy of our c. 2000
classification. Based upon our error analysis of 342 areas
distributed among the three forest zones, we estimated 89.5%
accuracy in identification of forest and non-forest. To directly
estimate the error of our deforestation class would have
required two dates of validation data, corresponding to the two
dates of satellite imagery in question, but this was unavailable.
The error in the deforestation class is not the product of the
error for forest and non-forest in the two dates of classification
because we conducted a direct multi-temporal classification.
We believe it is reasonable to assume that the error rate for
the deforestation class was about the same as the error rate for
the forest and non-forest classes.
Supplementary information
A list of image dates used for this analysis, and information
on obtaining full-resolution digital files of the forest cover,
deforestation and fragmentation maps are available via the
internet at
pt?open=512&objID =755&&PageID =127564&mode=2&
In the 1950s, there was 160 000 km2of forest cover in
Madagascar, comprising 55% humid , 26% dry and 19%
spiny forest (Table 3). The c. 2000 data showed a total 89
800 km2of forest , with an estimated accuracy of about 90%.
6G. J. Harper et al.
Table 5 Fragmentation of Madagascar’s forests, 1950s–c. 2000. Fragmentation is measured by distribution of forest area (1) by patch-size,
and (2) by distance of forest from a non-forest edge. All forest types are aggregated in these figures.
1950s 1970s c. 1990 c. 2000
Size class (km2)
0–10 7971 5.0 28 072 19.8 23321 21.7 23 372 23.6
10–50 12 424 7.8 8954 6.3 8442 7.8 7782 7.9
50–100 6078 3.8 4056 2.9 3663 3.4 3820 3.9
100–500 16 206 10.1 11 588 8.2 8982 8.3 9978 10.1
500–1000 5825 3.6 5383 3.8 7392 6.9 8391 8.5
1000–5000 14 780 9.2 26 733 18.9 14 219 13.2 12 293 12.4
5000–10000 0 0 23 308 16.4 25 135 23.3 17 778 18.0
>10000 96 685 60.4 36 637 23.7 16 544 15.4 15 600 15.8
Total 159 969 100 141 732 100 107 698 100 99 015 100
Distance from non-forest edge (m)
0–57 5715 3.6 30 794 21.7 25 272 23.5 24 862 25.1
58–114 5660 3.5 15 454 10.9 12 723 11.8 12 373 12.5
115–257 12 778 8.0 19 453 13.7 15 934 14.8 15 116 15.3
257–513 19 965 12.5 18 818 13.3 15 155 14.1 13 919 14.1
514–998 25 447 15.9 19 118 13.5 14 705 13.7 13 012 13.1
998–2993 45 217 28.3 27 305 19.3 18 441 17.1 15 494 15.6
>2993 45 161 28.2 10 795 7.6 5471 5.1 4244 4.3
Total 159 943 100 141 739 100 107 703 100 99 019 100
An additional area of more than 11 200 km2was obscured by
cloud and of that area, almost 9200 km2were forested in the
most recent previous clear image; the remaining 2100 km2
were cloud obscured in all three dates of satellite imagery.
Cloud cover was generally associated with the humid forest-
covered slopes of the north-eastern mountains. Thus total
forest cover in Madagascar in c. 2000 was in the range 89 800–
101 100 km2, with a probable area of around 99 000 km2.This
forest cover estimate is within 5% of the IEFN (1996) and
Mayaux et al. (2005) estimates (Table 2). Faramalala’s (1988)
estimate, based on data from the early 1970s, is closer to our
c. 1990 estimate than our 1970s estimate.
Average rates of deforestation were 0.3% yr1from the
1950s to the 1970s, 1.7% yr1from the 1970s to c. 1990, and
0.9% yr1from c. 1990 to c. 2000 (Table 4; Fig. 1). The
greatest loss occurred in the humid and dry forests, which
between the 1950s and c. 2000 lost 43% and 41% of their
area, respectively. Spiny forest area decreased 28% over the
same period, but had the highest clearance rate during the
1990s of almost 1.2% yr1.
In the 1990s, the greatest clearance of spiny forest occurred
in the region centred around the city of Toliara. Two
contiguous areas of spiny forest >100 000 ha were cleared
to the north and north-east of the city, along with widespread
smaller patches of deforestation, primarily to the south.
Deforestation rates for the humid and dry forests slowed
during the 1990s. Even so, several contiguous patches of
20 000–50 000 ha were cleared in the dry forests of the west-
central part of the island during the 1990s. The more general
pattern of deforestation in the dry and humid forests was of
small-scale clearance at forest edges.
By the 1950s, over 26% of all forest occurred in patches
<500 km2and over 43% within 1 km of a non-forest edge. By
c. 2000, over 45% of all forest was in patches <500 km2and
over 80% within 1 km of a non-forest edge. A quarter of the
remaining forest in c. 2000 was within 57 m of a non-forest
edge and nearly a quarter was in isolated forest patches of less
than 10 km2(Table 5; Fig. 2). For most parts of the island,
with the exception of the eastern humid forest and south-
western dry and spiny forest, forest patches were <100 km2
(Fig. 1).
Dry forests were by all measures the most fragmented
forest type throughout the study period, and increased in
fragmentation primarily from the 1950s–c. 1990. Despite an
overall decrease in humid forest cover, fragmentation of
the humid forest increased only slightly from the 1970s–c.
2000, whereas spiny forest fragmentation increased
continuously over the study period (Fig. 2).
By the 1950s, only 27% of Madagascar was forested and even
a conservative estimate of pre-human forest cover suggests it
had already lost more than half of its forest cover; the loss
may have been as much as two-thirds, or more. Forest cover
further declined to approximately 16% in c. 2000, a loss of
40% in 50 years. Taking fragmentation into consideration,
the impact was even more dramatic. From the 1950s to
c. 2000, the area of ‘core forest’ (forest >1 km from a non-
forest edge) decreased from >90 000 km2to <20 000 km2.
The area in patches of >100 km2decreased by more than
half. The slowing rate of deforestation in the humid and dry
Forest change in Madagascar 7
Figure 2 Trends in fragmentation of Madagascar’s forest and woodland from the 1950s to c. 2000. Fragmentation is measured by
distribution of forest area (1) by patch-size, and (2) by distance of forest from a non-forest edge. (a) Percentage of forest in patches >500 km2
in area. (b) Percentage of forest in patches <100 km2area. (c) Percentage of forest >1 km from a forest edge. (d) Percentage of forest <250 m
from a forest edge.
forests after c. 1990 is encouraging, but the deforestation rates
among all forest types are still disturbing, given their small
remaining area and fragmented state.
These results demonstrate extensive loss and degradation
of the forest habitat on which 90% of Madagascar’s fauna
depend. Given the probable lag-time of species extinction
following habitat destruction (Brooks et al. 1999; Cowlishaw
1999), it is likely that many species are living on ‘borrowed
time’. These results emphasize the need for redoubled forest
conservation efforts in Madagascar. We suggest (1) halting
further primary forest clearance as soon as possible, and
(2) exploring the potential of strategically located forest
restoration efforts for mitigation of species extinctions.
The data here contribute to the goal of halting deforestation
by providing precise information necessary to study the causes
of deforestation in Madagascar. Only with such understanding
may effective policy be formulated. These data may also
be used to help prioritize forest conservation activities by
identifying forest habitat critical to biodiversity. Critical forest
habitat may be identified by combining these forest data with
the range polygons of forest-dependent species, protected
area polygons, and spatial analyses of levels of endemism
and number of threatened species. Similar analysis, with the
addition of data on forest fragmentation and secondary forest
cover, may be used to identify strategic locations for forest
restoration. We do not know how much forest has been lost
since 2000. We recommend regular updates to these data
to enable more rapid and adaptive response to deforestation
threats in Madagascar.
This work was supported by Conservation International (CI),
the Center for Biological Conservation (CBC) in Madagascar,
and NASA’s ‘Mission to Planet Earth’. We thank the CBC
8G. J. Harper et al.
for support for the aerial surveys; Franc¸ois Blasco of the
University of Toulouse for use of the 1950s maps, and
Sara Musinsky, Leanne Miller and Minnie Wong of CI for
assistance with the work.
Achard, F., Eva, H.D., Stibig, H.J., Mayaux, P., Gallego, J.,
Richards, T. & Malingreau, J.P. (2002) Determination of
deforestation rates of the world’s humid tropical forests. Science
297(5583): 999–1002.
Blasco, F. (1965) Notice de la carte de Madagascar. In: Notice de
la carte de Madagascar, ed. H. Humbert & G. Cours Darne,
pp. 4–13. Puducherry, India: Section Scientifique et Technique
de L’Institut Francais de Pondich´
Brooks, T.M., Pimm, S.L. & Oyugi, J.O. (1999) Time lag between
deforestation and bird extinction in tropical forest fragments.
Conservation Biology 13(5): 1140–1150.
Cochrane, M.A. (2001) Synergistic interactions between habitat
fragmentation and fire in evergreen tropical forests. Conservation
Biology 15(6): 1515–1521.
Cowlishaw, G. (1999) predicting the pattern of decline of African
primate diversity: an extinction debt from historical deforestation.
Conservation Biology 13(5): 1183–1193.
Cullen, L., Bodmer, R.E. & Padua, C.V. (2000) Effects of hunting
in habitat fragments of the Atlantic forests, Brazil. Biological
Conservation 95(1): 49–56.
Dufils, J.M. (2003) Remaining forest cover. In: The Natural History
of Madagascar, ed. S.M. Goodman & J.P. Benstead, pp. 88–96.
Chicago, IL, USA: University of Chicago Press.
Enserink, M. (1999) Biological invaders sweep in. Science 285(5435):
FAO (2000) Global Forest Resources Assessment 2000. Rome, Italy:
Food and Agriculture Organization of the United Nations.
Faramalala, M.H. (1988) Etude de la v´
etation de Madagascar `
l’aide des donn´
ees spatiales. Doctoral Thesis, L’Universite Paul
Sabatier, Toulouse, France.
Ferreira, L.V. & Laurance, W.F. (1997) Effects of forest
fragmentation on mortality and damage of selected trees in Central
Amazonia. Conservation Biology 11(3): 797–801.
Gade, D.W. (1996) Deforestation and its effects in highland
Madagascar. Mountain Research and Development 16(2): 101–116.
Garbutt, N. (1999) Mammals of Madagascar.NewHaven,CT,USA:
Yale University Press.
Gautier, L. & Goodman, S.M. (2003) Introduction to the flora of
Madagascar. In: The Natural History of Madagascar,ed.S.M.
Goodman & J.P. Benstead, pp. 229–250. Chicago, IL, USA:
University of Chicago Press.
Glaw, F. & Vences, M. (2003) Introduction to amphibians. In:
The Natural History of Madagascar, ed. S.M. Goodman & J.P.
Benstead, pp. 883–898. Chicago, IL, USA: University of Chicago
Green, G.M. & Sussman, R.W. (1990) Deforestation history of the
eastern rain forests of Madagascar from satellite images. Science
248(4952): 212–215.
Harris, L.D. (1984) The Fragmented Forest: Island Biogeographic
Theory and the Preservation of Biological Diversity. Chicago, IL,
USA: The University of Chicago Press.
Hawkins, A.F.A. & Goodman, S.M. (2003) In: The Natural
History of Madagascar, ed. S.M. Goodman & J.P. Benstead,
pp. 1019–1044. Chicago, IL, USA: University of Chicago Press.
Humbert H. & Cours Darne G., eds (1965) Notice de la
Carte Madagascar. Puducherry, India: Section Scientifique et
Technique de L’Institut Francais de Pondich´
IEFN (1996) Inventaire Ecologique Forestier National. Antananarivo,
Madagascar: Minist`
ere de l’Environnement.
Jarosz, L. (1993) Defining and explaining tropical deforestation:
shifting cultivation and population-growth in colonial Madagascar
(1896–1940). Economic Geography 69(4): 366–379.
Kapos, V. (1989) Effects of isolation on the water status of forest
patches in the Brazilian Amazon. Journal of Tropical Ecology 5(2):
Kull, C.A. (2000) Deforestation, erosion, and fire: degradation myths
in the environmental history of Madagascar. Environment and
History 6(4): 423–450.
Langrand, O. & Wilm´
e, L. (1997) Effects of forest fragmentation on
extinction patterns of the endemic avifauna on the central high
plateau of Madagascar. In: Natural Change and Human Impact in
Madagascar, ed. S.M. Goodman & B.D. Patterson, pp. 280–308.
Washington, DC, USA: Smithsonian Institution Press.
Lovejoy, T.E. (1997) Biodiversity: what is it? In: Biodiversity II.
Understanding and Protecting Our Biological Resources,ed.M.L.
Reaka-Kudla, D.E. Wilson & E.O. Wilson, pp. 7–14. Washington,
DC, USA: Joseph Henry Press.
Mayaux, P., Gond, V. & Bartholome, E. (2000) A near-real
time forest-cover map of Madagascar derived from SPOT-4
VEGETATION data. International Journal of Remote Sensing
21(16): 3139–3144.
Mayaux, P., Holmgren, P., Achard, F., Eva, H., Stibig, H.-J. &
Branthomme, A. (2005) Tropical forest cover change in the 1990s
and options for future monitoring. Philosophical Transactions of
the Royal Society of London Series B-Biological Sciences 360(1454):
Myers, N., Mittermeier, R.A., Mittermeier, C.G., da Fonseca,
G.A.B. & Kent, J. (2000) Biodiversity hotspots for conservation
priorities. Nature 403(6772): 853–858.
Rakotosamimanana, B. (2003) Foreword. In: The Natural History of
Madagascar, ed. S.M. Goodman & J.P. Benstead, p. xi. Chicago,
IL, USA: University of Chicago Press.
Raxworthy, C.J. (2003) Introduction to the reptiles. In: The Natural
History of Madagascar, ed. S.M. Goodman & J.P. Benstead,
pp. 934–949. Chicago, IL, USA: University of Chicago Press.
Schatz, G.E. & Lescot, M. (2003) Gazetteer to Malagasy
botanical collecting localities [www document]. URL http://
Skole, D. & Tucker, C.J. (1993) Tropical deforestation and habitat
fragmentation in the Amazon: satellite data from 1978 to 1988.
Science 260(5116): 1905–1910.
Steininger, M.K., Tucker, C.J., Townshend, J.R.G., Killeen, T.J.,
Desch, A., Bell, V. & Ersts, P. (2001) Tropical deforestation in the
Bolivian Amazon. Environmental Conservation 28(2): 127–134.
Townshend, J.R.G. & Justice, C.O. (1988) Selecting the spatial
resolution of satellite sensors required for global monitoring of
land transformations. International Journal of Remote Sensing 9(2):
Tucker, C.J. & Townshend, J.R.G. (2000) Strategies for monitoring
tropical deforestation using satellite data. International Journal of
Remote Sensing 21(6–7): 1461–1471.
Tucker, C.J., Grant, D.M. & Dykstra, J.D. (2004) NASA’s global
orthorectified Landsat data set. Photogrammetric Engineering and
Remote Sensing 70(3): 313–322.
Forest change in Madagascar 9
UNPF (2004) State of World Population 2004. New York, NY, USA:
United Nations Population Fund.
Vallan, D. (2000) Influence of forest fragmentation on amphibian
diversity in the nature reserve of Ambohitantely, highland
Madagascar. Biological Conservation 96(1): 31–43.
Watson, J.E.M., Whittaker, R.J. & Dawson, T.P. (2004) Avifaunal
responses to habitat fragmentation in the threatened littoral forests
of south-eastern Madagascar. Journal of Biogeography 31: 1791–
Watson, R.T., Noble, I.R., Bolin, B., Ravindranath, N.H.,
Verardo, D.J. & Dokken, D.J., eds (2000) Land Use, Land-
Use Change, and Forestry: Summary for Policy Makers.
Geneva, Switzerland: Intergovernmental Panel on Climate
  • Article
    The Asian toad, Duttaphrynus melanostictus, was accidentally introduced to Toamasina (eastern Madagascar) around 2010, and since then has spread at a substantial rate across a larger area. This study documents the expansion of the invasive range of this species, calculates the invasion spread rate, and provides estimates of toad abundance and habitat preferences. Updates of the distribution range revealed a fivefold increase of the invaded area over 3 years, and a doubling of the rate of spread, showing a shift of the invasion towards the North-West, most probably because of the absence of ecological barriers. We used N-mixture models to estimate toad abundance on the basis of repeated count data of six areas in Toamasina and its surrounding countryside. Toad distribution shows heterogeneous density across the distribution range, with an average abundance of 184 toads ha⁻¹ (95% CI 132–263). The toad’s abundance was highest in sites with the presence of organic waste, and was negatively related to the density of road networks in the proximity of study sites. The rapid expansion of the Asian toad in the Toamasina region suggests that this toad is an increasing threat for Madagascar. We propose immediate management actions that could limit the spread of alien toads in this megadiverse country.
  • Article
    Full-text available
    Deforestation and forest degradation are major drivers of global environmental change and tropical forests are subjected to unprecedented pressures from both. For most tropical zones, deforestation rates are averaged across entire countries, often without highlighting regional differentiation. There are also very few estimates of forest degradation, either averaged or localized for the tropics. We quantified regional and country-wide changes in deforestation and forest degradation rates for Madagascar from Landsat temporal data (in two intervals, 1994–2002 and 2002–2014). To our knowledge, this is the first country-wide estimate of forest degradation for Madagascar. We also performed an intensity analysis to categorize the magnitude and speed of transitions between forest, vegetation matrix, cultivated land and exposed surface. We found significant regional heterogeneity in deforestation and forest degradation. Deforestation rates decreased annually in lowland evergreen moist forest by �0.24% and in all other vegetation zones. Forest degradation rates had annual increases in the same period in lowland evergreen moist forest (0.09%), littoral forest (0.06%) but decreased in medium altitude moist evergreen forest (�0.25%), dry deciduous forest (�0.23%) and scelrophyllous woodland (�0.61%) in the same period. Despite these regional differences, higher rates of deforestation and forest degradation were consistently driven by rapid and large-sized conversions of largely intact forest to cultivated lands and exposed surfaces, most of which occurred between 1994 and 2002. These results suggest that while targeted conservation projects may have reduced forest degradation rates in some areas (e.g. medium altitude moist evergreen forest), the drivers of land cover change remain intense in relatively neglected regions. We advocate a more balanced approach to future conservation initiatives, one recognizing that deforestation and forest degradation, particularly in tropical Africa, are often driven by region-specific conditions and therefore require conservation policies tailored for local conditions.
  • Article
    Glucocorticoids are metabolic byproducts of animals' physiological responses to ecological or social challenges and are thought to represent an adaptive response allowing beneficial responses to short-term challenges. Glucocorticoid metabolites (GCs) can be assayed non-invasively through faeces and therefore can be a useful tool to gauge the health of populations experiencing natural and/or anthropogenic stressors. However, the response of GCs to anthropogenic stressors varies, with both higher and lower GC levels reported. Here, we describe variation in GC secretion within eight diademed sifaka (Propithecus diadema) groups across 1 year. These groups span a gradient of anthropogenic habitat disturbance, including groups in continuous forest ('CONT') and disturbed fragments ('FRAG'), and indicators of health suggest that FRAG groups are negatively impacted by habitat disturbance. We monitored phenology, used focal animal follows to quantify diet and collected faeces (n = 547) from which we quantified GC content using enzyme immunoassay. All groups showed elevated lean-season GCs, but with a single, brief peak. GCs were inversely correlated with feeding time. No overall effect of habitat (CONT vs. FRAG) was found, but the lean-season peak was significantly higher in CONT groups. There was a significant season*age-sex interaction; adult females had an attenuated lean-season response compared with groupmates. The observed lean-season 'challenge' is consistent with previous lemur studies, as well as mammals in general. Low and largely invariable GC levels in FRAG, within the context of observed health and nutritional declines, suggest that FRAG groups employ a strategy whereby the adrenal response to stressors is downregulated. More research is needed to contextualize our observations of GC variation and health on an individual level, both in terms of corroborating evidence for ecological and social stressors, and longer-term quantification of reproductive success and fitness. Cite as: Tecot SR, Irwin MT, Raharison J-L (2019) Faecal glucocorticoid metabolite profiles in diademed sifakas increase during seasonal fruit scarcity with interactive effects of age/sex class and habitat degradation.
  • Article
    Full-text available
    2019): Contrasting tree-cover loss and subsequent land cover in two neotropical forest regions: sample-based assessment of the Mexican Yucatán and Argentine Chaco, Journal of Land Use Science To link to this article: https://doi. ABSTRACT The neotropical-forest's northern and southern extremes, covering the Mexican Yucatán and the Argentine Chaco, have among the highest rates of recent tree-cover loss in the biome. This study contrasts the character of loss in these regions, estimating proportions of types of loss and subsequent land cover. It is based on two-stage probability sampling design and field and satellite-image surveys. All estimates include uncertainties , which could be further reduced via model-assisted estimation or additional sampling. This approach can be replicated in other regions to estimate types of loss and associated land cover from a definitive, in-situ perspective. The character of loss in the two areas differed greatly. That in the Yucatán was 54% temporary, mostly under fallow or selectively logged, while that in the Chaco was 85% permanent, split nearly equally between crops and pasture. These data contribute to a quantitative basis for studies of socioeconomic drivers of neotropical deforestation. ARTICLE HISTORY
  • Article
    Conversation organizations in developing countries are beginning to target human health. In Madagascar, half of all children are malnourished, due to food insecurity and poor diet diversity. The purpose of this study was to assess food choice and barriers to diet diversity. Primary caregivers were recruited to participate in a cross-sectional exploratory study on food choices, food beliefs, and economic practices through a survey and focus group discussions. Important factors identified when shopping and preparing foods were cleanliness and health. Energy-dense foods were important, but diet diversity was not. Food cost, availability, and health were identified as concerns and important in decision-making. Understanding food choices is vital for developing programs targeting human health.
  • Article
    Full-text available
    It is essential to monitor both the mangrove coverage and the encroachment of anthropogenic activities to assess the evolution of these highly valuable and threatened ecosystems. Using the grid technique, Landsat images of three estuaries in Pernambuco, northeast Brazil, were analyzed to track changes in land use over the past three decades. This study is the first time the grid technique has been used as a precise method to quantify and localize the mangrove coverage changes at local scale. We found that the technique was a synthetical and cost-effective way of observing land-use changes over the study period, and its precision was evident for local-scale study. An increasing trend of mangrove coverage was observed and can be related to the salinization. This increase in salinization is a result of anthropogenic activities and climate change. We also found that differences in geometrical properties influenced the spatiotemporal patterns of mangroves. To mitigate adequate- ly the negative impacts induced by anthropization and climate change, the characteristics of mangrove forest configuration and human activities should therefore both be considered.
  • Chapter
    Primate Research and Conservation in the Anthropocene - edited by Alison M. Behie January 2019
  • Article
    Full-text available
    Cambridge Core - Ecology and Conservation - Primate Research and Conservation in the Anthropocene - edited by Alison M. Behie
  • Article
    Full-text available
    Landscape fragmentation is one of the greatest threats to environments globally, affecting all living organisms within fragments at many assembly levels. Despite such general assumptions, very few reports exist on the diversity patterns of Amazonian social wasps (Polistinae) in response to fragmentation. Our study aims at addressing how the community patterns of social wasp change from continuous to fragmented landscapes in the Amazon rainforest. We collected social wasps for a year (11 520 trap‐hour effort) in a continuous forest and a highly disturbed urban forest fragment, located in Manaus, Brazil. We attracted wasps using sardine and orange juice traps, placed in 20 sampling plots. We collected 2742 wasps from 29 species. Agelaia pallipes, Angiopolybia pallens and Agelaia fulvofasciata were the most abundant species. We detected significant differences in overall abundance, species richness and species composition of social wasps between landscapes. Conversely, no difference in species diversity (effective number of species) was found between habitats. Despite the low quantity of exclusive species, the numbers of most species declined sharply in the fragmented habitat, while for one species (Polybia rejecta), abundance increased. Regarding the percentile of species contribution to dissimilarity between landscapes, five out of the first six species were highly abundant (>100 individuals), thus demonstrating that the observed pattern was very likely driven by common species. Additionally, our findings highlight that fragmentation threatens abundant species as much as rare species. We hope that our results encourage the use of social wasps as models in basic and applied ecological surveys and further efforts for conservation of pristine forest areas in the tropical region of South America.
  • Conference Paper
    Landsat satellite images from the mid-1980s and early 1990s were used to map tropical forest extent and deforestation in similar to 800,000 km(2) of Amazonian Bolivia. Forest cover extent, including tropical deciduous forest, totaled 472,000 km(2) while the area of natural non-forest formations totaled 298,000 km(2). The area deforested totaled 15,000 km(2) in the middle 1980s and 28,800 km(2) by the early 1990s. The rate of tropical deforestation in the >1,000 mm y(-1) precipitation forest zone of Bolivia was 2,200 km(2) y(-1) from 1985-1986 to 1992-1994. We document a spatially-concentrated "deforestation zone" in Santa Cruz Department where >60% of the Bolivian deforestation is occurring at an accelerating rate in areas of tropical deciduous dry forest.
  • Article
    NASA has sponsored the creation of an orthorectified and geodetically accurate global land data set of Landsat Multispectral Scanner, Thematic Mapper, and Enhanced Thematic Mapper data, from the 1970s, circa 1990, and circa 2000, respectively, to support a variety of scientific studies and educational purposes. This is the first time a geodetically accurate global compendium of orthorectified multi-epoch digital satellite data at the 30- to 80-m spatial scale spanning 30 years has been produced for use by the international scientific and educational communities. We describe data selection, orthorectification, accuracy, access, and other aspects of these data.
  • Article
    Full-text available
    The distributions of forest and deforestation throughout the tropics are poorly known despite their importance to regional biodiversity and global climate and biodiversity. Deforestation estimates based on surveys or sampling have large errors, and high-resolution, wall-to-wall mapping of tropical forests is necessary to assess the impacts of fragmentation. Landsat satellite images from the mid-1980s and early 1990s were thus used to map closed-canopy tropical forest extent and anthropogenic deforestation in an approximately 700 000 km2 area of Amazonian Bolivia with precipitation >1000 mm yr1 precipitation below 1500 m elevation and north of 19° S, was 1529 km2 yr1986 to 1992deforestation zone’ in Santa Cruz where >60% of the Bolivian deforestation has occurred. These results indicate that the rate of deforestation in Bolivia has been rapid despite a relatively small human population, and, as in Brazil, clearance has concentrated in the more deciduous forests.
  • Article
    Madagascar's highland region once was covered with an evergreen forest dominated by about 20 endemic tree species. This has been permanently replaced by a floristically impoverished steppe vegetation on ferrolitic soils. Human intervention has caused this deforestation, aided by a notable failure of highland forests to spontaneously regenerate. Beginning about A.D.600, Indonesian settlers began removing forests in this region to create swidden fields. After A.D.1000, zebu cattle introduced from Africa provided a strong motive for islanders to increase grassland at the expense of forest. By A.D.1600 the highland forest cover had mostly disappeared except for localized tract, patches, and strips. The historical record and contemporary observations point to the importance of fire in destroying the forest. This landscape conversion has had devastating outcomes. Tree and humus removal has led to massive erosion, floods, water shortages, and faunal extinctions or endangerments. Agricultural and pastoral production on upland soils is marginal. Reestablishment of long-term ecological stability to this region would require an end to burning as a form of pasture management, changes in communal land tenure, and reforestation of the highlands with native tree species.
  • Article
    The growing prevalence of fragmentation and fire in tropical forests makes it imperative to quant fy changes in these disturbances and to understand the ways in which they interact across the landscape. I used a multitemporal series of Landsat images to study the incidence and coincidence of fire and fragmentation in two areas of Para state in the eastern Brazilian Amazon,, Tailandia and Paragominase. In both areas, deforestation and forest fires were quantified for time series of 6-10years. The Tailandia study area typifies a landscape with The herringbone pattern of government-settled colonists, and the Paragominas area is dominated by large cattle ranches. In both areas, over 90% of the forests affected by fire were associated with form est edges. Although most burned forest occurred within 500 m of forest edges, some fires occurred in deep forest, several kilometers from any edge. The obvious synergism between forest fragmentation and fire poses serious risks to tropical ecosystems and has important implications for land management.
  • Article
    Full-text available
    Measuring the aerial extent of tropical deforestation for other than localized areas requires the use of satellite data. We present evidence to show that an accurate determination of tropical deforestation is very di Å cult to achieve by a ' random sampling' analysis of Landsat or similar high spatial resolution data unless a very high percentage of the area to be studied is sampled. In order to achieve a Landsat-derived deforestation estimate within Ô 20% of the actual deforestation amount 90% of the time, 37 of 40 scenes, 55 of 61 scenes and 37 of 45 scenes were required for Bolivia, Colombia and Peru respectively.
  • Article
    PREDICTING INVASIONSGlobalization may be the mantra of the new economy, but for the environment it may spell disaster, as international travel and trade are creating a tidal wave of exotic species that is transforming ecosystems worldwide. If ecologists could identify likely invaders, governments could simply restrict imports of those treacherous species, and managers could mercilessly weed them out or trap them. But the few predictive models are still hotly debated, and they apply to only a narrow range of organisms at best.