Journal of Tropical Ecology (2002) 18:311?325. With 4 ?gures. Copyright ? 2002 Cambridge University Press
DOI:10.1017/S0266467402002237 Printed in the United Kingdom
Fire as a large-scale edge effect in Amazonian
forests
MARK A. COCHRANE* and WILLIAM F. LAURANCE??1
*Basic Science and Remote Sensing Initiative, Michigan State University, East Lansing,
MI 48823, USA (Email: cochrane@bsrsi.msu.edu)
?Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Republic of Panama?
?Biological Dynamics of Forest Fragments Project, National Institute for Amazonian
Research (INPA), C.P. 478, Manaus, AM 69011-970, Brazil
(Accepted 14th July 2001)
ABSTRACT. Amazonian forests are being rapidly cleared, and the remaining
forest fragments appear unusually vulnerable to ?re. This occurs because forest
remnants have dry, ?re-prone edges, are juxtaposed with frequently burned pas-
tures, and are often degraded by selective logging, which increases forest desicca-
tion and fuel loading. Here we demonstrate that in eastern Amazonia, ?res are
operating as a large-scale edge effect in the sense that most ?res originate outside
fragments and penetrate considerable distances into forest interiors. Multi-
temporal analyses of satellite imagery from two frontier areas reveal that ?re
frequency over 12?14-y periods was substantially elevated within at least 2400 m
of forest margins. Application of these data with a mathematical core-area model
suggests that even large forest remnants (up to several hundred thousand ha in
area) could be vulnerable to edge-related ?res. The synergistic interactions of
forest fragmentation, logging and human-ignited ?res pose critical threats to Ama-
zonian forests, particularly in more seasonal areas of the basin.
KEY WORDS: Amazon, Brazil, conservation, carbon emissions, deforestation,
drought, ?re, habitat fragmentation, logging, rain forest, remote sensing, tropical
forest
INTRODUCTION
The Amazon basin contains nearly 60% of the world?s remaining tropical rain
forest, and plays critical roles in maintaining biodiversity, regional hydrology
and climate, and carbon storage (Fearnside 1999). It also has the world?s high-
est absolute rate of forest destruction, averaging roughly 3?4 million ha y-1
(INPE 2001, Whitmore 1997).
1 Corresponding author. Smithsonian Tropical Research Institute, Apartado 2072, Balboa, Republic of
Panama?. Email: laurancew@tivoli.si.edu
311
MARK A . COCHRANE AND WILL IAM F . LAURANCE312
The rapid pace of Amazon deforestation is causing widespread habitat frag-
mentation. By 1988, the area of forest in the Brazilian Amazon that was frag-
mented (< 100 km2 in area) or vulnerable to edge effects (< 1 km from
clearings) was over 150% larger than the total area deforested (Skole & Tucker
1993). Because over 15% of the Brazilian Amazon has now been cleared (INPE
2001), the total area altered by deforestation, fragmentation and edge effects
may comprise a third or more of the region today (Laurance 1998).
Habitat fragmentation affects the ecology of Amazonian forests in many
ways, such as altering the diversity and composition of fragment biotas
(Bierregaard et al. 1992, Lovejoy et al. 1986) and changing ecological processes
such as pollination and nutrient cycling (Didham et al. 1996, Klein 1989).
Recent evidence indicates that fragmentation also alters rain-forest dynamics,
causing markedly elevated rates of tree mortality, damage and canopy-gap
formation (Laurance et al. 1998a), apparently as a result of microclimatic
changes (Kapos 1989), increased wind turbulence and proliferating lianas near
fragment edges (Laurance et al. 2001b). These changes lead to a substantial
loss of living biomass in fragments that may be a signi?cant source of green-
house gas emissions (Laurance et al. 1997, 1998b).
In addition to fragmentation, Amazonian forests are being degraded by vari-
ous other activities, such as human-ignited ?res, legal and illegal logging, gold
mining and over-hunting (Fearnside 1990, Laurance et al. 2001a). These addi-
tional threats may interact additively or synergistically with fragmentation. An
increasing body of evidence, for example, reveals that fragmentation increases
the vulnerability of Amazonian forests to ?re (Cochrane & Schulze 1999,
Cochrane et al. 1999, Gascon et al. 2000, Kauffman & Uhl 1990, Nepstad et al.
1999). This occurs because fragments have relatively dry, ?re-prone edges
(Kapos 1989), are juxtaposed with frequently burned pastures and regrowth
forests (Nepstad et al. 1996, Uhl & Kauffman 1990), and are often degraded
by selective logging, which increases forest desiccation and fuel loading
(Holdsworth & Uhl 1997, Uhl & Kauffman 1990).
While it is apparent that fragmented forests are vulnerable to ?res, the
spatial scale of this phenomenon has not yet been quanti?ed. Most ?res origin-
ate in surrounding pastures or regrowth and then burn into fragment interiors,
and thus appear to be operating as a kind of edge effect. To date, edge effects
in Amazonian forests, such as alterations in microclimate, forest dynamics and
faunal communities, have been shown to penetrate from 10?400 m into frag-
ment interiors (Bierregaard et al. 1992, Laurance et al., in press, Lovejoy et al.
1986). Preliminary evidence, however, suggests that some edge-related changes
in tropical forests could penetrate much further than this, perhaps as far as
several km into fragment interiors (Curran et al. 1999, Laurance 2000).
Here we provide evidence that, in two relatively seasonal landscapes of the
eastern Amazon, ?re is operating as such a large-scale edge effect. Using a
simple mathematical model, we demonstrate that even large forest fragments
Fire and forest fragmentation 313
and isolated nature reserves could be vulnerable to edge-related ?res. We
argue that, if fragmented, large expanses of Amazonian forests are likely to be
destroyed or severely degraded by ?re.
METHODS
Dynamics of rain-forest ?res
Natural ?res occur only rarely in undisturbed tropical rain forests
(Goldammer 1990). However, colonists and ranchers use ?res both to clear
forests and maintain pastures, which often burn into adjoining forest blocks.
These surface ?res are deceptively unimpressive, creeping along the ground as
a thin ribbon of ?ames burning through the leaf litter. Except for areas with
unusual fuel structure, the ?res reach only 10 cm in height (Cochrane et al.
1999).
Even during extreme droughts, rain forests maintain high humidity that
makes combustion dif?cult. Many surface ?res burn during the day, only to be
extinguished when relative humidity increases in the evening. Such ?res, how-
ever, can smoulder in old treefalls and standing dead trees for up to several
weeks, until conditions are right for the ?re to continue propagating through
the leaf litter. Surface ?res may cover as little as 150 m distance in a day
(Cochrane & Schulze 1998) but are deadly to trees, because the slow propaga-
tion results in long residence times of the ?ames at the base of encountered
trees. Most rain-forest trees have thin bark (typically < 1 cm thick; Uhl &
Kauffman 1990) and the heat of even small ?res girdles many trees, killing
around 40% of all standing stems (
 10 cm diameter at breast height (dbh);
Cochrane & Schulze 1999).
An initial surface ?re kills mostly smaller trees (< 40 cm dbh), with trees
continuing to die for 2 y or more. The forest canopy becomes fragmented and
the quantity of dead fuels increases as dead leaves and trees begin to fall. Soon,
the forest is far more prone to subsequent ?res, because the diminished canopy
allows rapid drying and the dying trees provide large amounts of combustible
fuel.
Burned forests commonly adjoin ?re-maintained pastures and agricultural
lands. While initial rain-forest ?res may require an extensive drought, sub-
sequent ?res can occur after just a few weeks without rain (Cochrane & Schulze
1999). Forests that burn a second time fare much worse because the ?res are
far more intensive, overwhelming the defences of even larger, thicker-barked
trees. Typically, second ?res kill 40% of the standing trees in all size classes,
comprising around 40% of the standing biomass (Cochrane et al. 1999).
Subsequent ?res are even more likely and severe. During the ?rst several
?res, more fuels are created than destroyed, and a positive feedback results in
which each ?re becomes more probable and intensive. This process can eradic-
ate trees from an area and result in extensive grass invasion. In the absence
of seed sources for grassland or open-woodland species, anthropogenic savanna
MARK A . COCHRANE AND WILL IAM F . LAURANCE314
(sometimes with certain palm species) will result. For regions with a pro-
nounced dry season, this conversion of rain forest to savanna is likely to be
irreversible (Cochrane et al. 1999).
Study areas
The two study areas, Taila?ndia (2470 km2) and Paragominas (1280 km2), are
in the eastern Brazilian Amazon and have extensively fragmented forest cover.
Both areas have similar evergreen terra-?rme forests and pronounced dry sea-
sons, with mean annual rainfall ranging between 1500?2000 mm (Cochrane et
al. 1999, Kauffman et al. 1995). The terrain in both areas is quite ?at and
uniform, but is dissected by a number of streams and small rivers.
Taila?ndia is a new frontier where large-scale deforestation began in the early
1990s when a government-sponsored colonization project was initiated. As is
typical, a ??sh-bone? pattern of deforestation quickly developed as colonists
situated along parallel roads employed slash-and-burn farming (Figure 1). By
1997, 44% of the area?s forests had been destroyed. Paragominas is an older
frontier where logging and cattle ranching predominate (Figure 1). Large-scale
clearing began in the early 1960s with construction of the nearby Bele?m?Brasi-
lia Highway. By 1995, 64% of the area?s forests had been destroyed. Further
information on these study areas is provided in Cochrane (2000).
Core-area model
We used a mathematical ?core-area model? to predict the areal extent of
edge-related ?res in forest fragments of varying sizes and shapes (Laurance &
Yensen 1991). This model provides accurate (> 99%) predictions of the size of
unaffected core-area for any fragment based on three parameters: fragment
area, a fragment-shape index (SI), and the average distance (d) to which edge
effects penetrate into fragment interiors. SI is derived by dividing the frag-
ment?s perimeter-length by that of an equal-sized circle, and varies from 1.0
for perfect circles to 8 or higher for very irregularly shaped fragments. SI is
calculated as SI = P/(2[TA0.5]), where P is the perimeter-length of the fragment
and TA is fragment area. The core area (CA) is given by CA = TA ? affected
area (AA), where AA = 3.55 d SI (TA/104)0.5. AA is overestimated for more-
circular fragments and is therefore adjusted downward by AAadj = (0.265 AA)/
SI0.5 (see Laurance & Yensen 1991 for a detailed explanation of the model).
For both of our study areas, the total area and perimeter-length of all forest
fragments of at least 0.1 ha were determined using georeferenced imagery
from the Landsat Thematic Mapper (TM) satellite, at a 30-m spatial scale.
Calculations were conducted using ArcInfo 8.0.1.
Fire rotation-times
The incidence of subcanopy-surface ?res in the two study areas was assessed
for periods of 12?14 y, from 1984 to the mid-late 1990s, using multi-temporal
analyses of Landsat TM imagery, augmented with extensive ground-truthing
Fire and forest fragmentation 315
Figure 1. Map showing location of study areas. Below, portions of each study area in eastern Amazonia,
showing the complex ??sh-bone? pattern of fragmentation arising from a forest-colonization project
(Taila?ndia) and the somewhat less irregular fragmentation pattern caused by cattle ranching (Paragominas).
Black areas are forest and grey areas are deforested. Each scene shows an area of about 600 km2.
MARK A . COCHRANE AND WILL IAM F . LAURANCE316
and interviews of local residents (Cochrane et al. 1999). As is typical in the
eastern Amazon, our study areas were affected by 2?3 droughts or rainfall
de?cits associated with periodic El Nin?o events (occurring in 1986 and 1991 at
both sites, and in 1997 at Taila?ndia), during which forest burning increased
markedly.
A sub-pixel linear spectral mixture-modelling technique (Cochrane & Souza
1998) was used to detect and classify forests that had been impacted by subcan-
opy surface ?res in the Taila?ndia (1984, 1991, 1993, 1995, 1997) and Paragom-
inas (1984, 1991, 1993, 1995) study areas. Speci?cally, ?re-induced tree mortal-
ity opens the forest canopy, and even though the resulting gaps may be small
they are suf?cient to allow sunlight from more of the forest ?oor and woody
vegetation to be re?ected. These non-photosynthetic vegetation (NPV)
re?ections are a small fraction of the whole scene but the spectral signature
of these substances is signi?cantly higher in ?re-damaged forests than in
undamaged forests. Separation of satellite-imagery information into its frac-
tional components (e.g. vegetation, shade, NPV) allows ?re-damaged forests to
be clearly distinguished for up to 2 y after a ?re (in an earlier study, ground-
truthing revealed that 93% of forests burned < 1 y previously and 71% of
areas burned 1?2 y previously were correctly classi?ed; Cochrane et al. 1999).
A ?ltering technique was used to reduce the likelihood that logged forests
could be misclassi?ed as burned forests (Cochrane & Souza 1998).
In both study areas, ?re rotation-times were calculated as a function of dis-
tance from forest edge (the ?re rotation is the average number of years
required to burn an area under consideration, with the understanding that
some areas may not burn while others may burn more than once during a
cycle; Van Wagner 1978). For each year of the study, all forest was mapped
using classi?ed images at a spatial scale of 30-m pixels. The area of forest in
each edge-distance category (e.g. 0?30 m, 30?60 m, 60?90 m, etc.) was then
calculated, using ArcInfo 8.0.1. Using the mixture-modelling technique
described above, the distribution of surface ?res was also mapped, and the
proportion of pixels that burned in each edge-distance category was calculated.
Data for all years of the study were then combined to yield an average propor-
tion of burned pixels for each edge-distance category. The ?re rotation-time
was simply the inverse of this proportion (for example, if an average of 5% of
all pixels at 0?30 m from the edge burned each year, then the ?re rotation-time
would be 1/0.05 = 20 y).
The distance (d) to which ?re rotation-times near forest edges deviated from
those in forest interiors was estimated as follows: the 95% data distribution
(mean ? 1.96 standard deviations) was calculated for ?re rotation-times in
forest interiors (de?ned by visual inspection as being over 2.5 km from the
nearest edge); and the point at which the observed curve of ?re rotation-times
fell below this range was de?ned as d. The 95% data distribution is an appropri-
ate measure of variability because ?re rotation-times in forest interiors (at
Fire and forest fragmentation 317
Taila?ndia only; see below) did not deviate signi?cantly from normality (Wilk?
Shapiro test, P > 0.50).
RESULTS
Fragment characteristics
We identi?ed a total of 722 forest fragments in our two study areas, with
individual fragments ranging from 0.1 ha to over 57 000 ha in area (Table 1).
Small (< 100 ha) forest patches were abundant, accounting for 91?93% of
all fragments in each study area, but supported only a small proportion
(< 4.5%) of the remaining forest cover at each site.
When shape index (SI) values were plotted against fragment area, two
trends emerged. First, there was a positive relationship between SI and area
(Figure 2), because larger fragments tended to be more irregularly shaped
than smaller fragments (rs > 0.74, P < 0.00001 in both areas; Spearman rank
correlations). This pattern resulted in part from fractal effects (because irregu-
larities in the margins of small fragments tended to be smoothed over using
30-m pixels, thereby causing SI values of smaller fragments to be underestim-
ated relative to those of larger fragments; cf. Li & Reynolds 1994). It does,
nevertheless, re?ect a tendency for most large fragments to be very irregularly
shaped at a spatial scale that is relevant to edge effects in this system. Second,
fragments in Taila?ndia were often more irregularly shaped than those in Para-
gominas (Table 1, Figure 2), re?ecting the complex patterns of forest frag-
mentation caused by government-sponsored colonization projects in the
Amazon.
Core-area model
For Taila?ndia, ?re rotation-times declined sharply near edges, and were
found to deviate from those in forest interiors at a distance of up to 2.35 km
from the forest edge (Figure 3). The core-area model for Taila?ndia was gener-
ated for fragments using shape indices of 2.0, 4.0 and 6.0. This range of values
is realistic but slightly conservative; mean SI values for Taila?ndia ranged from
2.9 to 7.1 (Table 1), indicating that many fragments there were very irregularly
Table 1. Description of forest fragments in two study areas in the eastern Amazon. Taila?ndia (2470 km2)
and Paragominas (1280 km2) had 419 and 303 fragments, respectively.
Fragment size- Per cent of fragments Per cent of forest area Mean fragment shapecategory (ha)
Taila?ndia Paragominas Taila?ndia Paragominas Taila?ndia Paragominas
<1 23.1 27.0 0.1 0.1 1.30 1.29
1?10 50.2 44.4 1.1 0.4 1.42 1.42
10?100 17.5 22.0 3.2 1.6 1.79 1.79
100?1000 6.9 5.5 14.8 5.6 2.33 2.94
1000?10 000 1.7 0.2 35.2 0.9 3.97 3.88
10 000?100 000 0.7 1.0 45.7 91.5 4.16 7.14
MARK A . COCHRANE AND WILL IAM F . LAURANCE318
Figure 2. Relationship between fragment area and shape index (SI) for study areas at Paragominas and
Taila?ndia in the eastern Amazon. Although SI values were somewhat underestimated in small fragments
(because of fractal effects), this analysis reveals that most large fragments were very irregularly shaped at
a spatial scale that is relevant to edge effects in this system.
Fire and forest fragmentation 319
Figure 3. Fire rotation-times as a function of distance from forest edge for Taila?ndia (diamonds) and
Paragominas (triangles). The curves were ?tted with a smoothing function. Dotted lines show the 95% range
of variation (mean ? 1.96 standard deviations) for forest interiors at Taila?ndia (more than 2500 m from
edge).
shaped. The core-area model (Figure 4) suggests that ?re regimes in even large
fragments are being substantially altered. A fragment with an SI of 2.0, for
example, would need to be about 90 000 ha in area to ensure that half of its
total area will be unaffected by edge-related ?res. If the fragment were more
irregularly shaped (SI = 6.0), it would need to be about ten times larger
(890 000 ha) to retain half of its total area in natural condition.
Fragments in Paragominas were generally less irregularly shaped than those
at Taila?ndia, with mean SI values ranging from 2.3 to 4.3 (Table 1). The
fragments in this older frontier, however, were often smaller, and the effects
of ?res even more devastating, than was observed at Taila?ndia. No forest at
Paragominas was further than 2.7 km from the nearest edge, and even at this
distance, predicted ?re rotations were much shorter than those observed in
forest interiors in Taila?ndia (Figure 3). Intact forest interiors have apparently
MARK A . COCHRANE AND WILL IAM F . LAURANCE320
Figure 4. A core-area model showing the predicted impacts of edge-related ?res on fragmented rain forests
in Taila?ndia, Brazil.
vanished from the Paragominas area, and as a result it was not possible to
estimate d and generate a core-area model.
DISCUSSION
Generality and limitations of ?ndings
Our modelling results suggest that habitat fragmentation dramatically
increases the vulnerability of rain forests in eastern Amazonia to anthropogenic
?res, and that such ?res are operating as an edge effect over surprisingly large
spatial scales. Are our ?ndings typical? Both of our study areas are experiencing
intensive exploitation; the Paragominas forests, in particular, have been
severely degraded over the past several decades. Nevertheless, the spatial pat-
terns of forest fragmentation at our two sites are representative of those caused
by ranching and large-scale forest-colonization projects, the two most prevalent
land uses in the Amazon today. In addition, both of our study areas occur in
relatively seasonal areas of the basin. However, about half of the closed-canopy
forests in Brazilian Amazonia experience comparably strong dry seasons, espe-
cially in the eastern, southern and north-central areas of the basin (Nepstad et
Fire and forest fragmentation 321
al. 1994, 1999), and thus could be similarly susceptible to ?re when fragmented.
Analyses of recent Landsat TM imagery suggest that at least 45 000 000 ha of
forest in Brazilian Amazonia (over 13% of the total remaining forest area) are
currently vulnerable to edge-related ?res (Cochrane 2001).
Our study was only 12?14 y in duration ? the period over which satellite
imagery of suf?cient resolution was available ? and our results might be biased
if weather conditions during this interval were unusually dry. During the study
there were moderate El Nin?o droughts in 1986 and 1991 at both sites, and a
strong drought in 1997 at Taila?ndia, and most forest burning occurred during
these drier years (Cochrane 2000, Cochrane et al. 1999). Nevertheless, over the
last century there have been 23 El Nin?o events in the Amazon (Suplee 1999),
and the 1997 drought was comparable to other strong droughts, such as that
in 1983. Thus, droughts or rainfall de?cits appear to be relatively common
occurrences in the eastern Amazon, and the overall weather patterns during
our study did not seem unusual.
Previous studies suggest that, under natural conditions, major ?res are rare
in Amazonian terra-?rme forests, perhaps occurring only once or twice per
millennium on average (Meggers 1994, Piperno & Becker 1996, Saldarriaga &
West 1986, Sanford et al. 1985). At Taila?ndia, however, ?res in intact forest
interiors appeared to occur more frequently than this, with observed burns
implying ?re return-intervals of 100?150 y (Figure 3). Although this pattern
could have resulted from the relatively seasonal nature of the Taila?ndia cli-
mate, we believe a more plausible explanation is that loggers or hunters occa-
sionally ignited ?res in forest interiors. Because our study was relatively short-
term, even a few small ?res in forest interiors could decrease the inferred ?re
rotation-time substantially. If so, then our analysis of edge-related ?re fre-
quency might be conservative, tending to underestimate the distance to which
altered ?re regimes penetrate into forest interiors.
It must be emphasized that our core-area model (Figure 4) predicts only the
effects of habitat fragmentation on ?re frequency, and does not consider the
in?uence of other factors, such as temporal variability in weather and local
topography. Clearly, forest ?res vary in severity and frequency among years;
the estimated ?re rotation-times used in our model (Figure 3) are 12?14-y
averages, and may not be typical of any single year. Topography also in?uences
?re behaviour; trees along rivers and swamps, for example, are less likely to
burn than those in upland areas (e.g. Kellman & Meave 1997). Nevertheless,
?re frequency in our study areas was increased so radically near fragment
edges that local topographic effects were largely obscured. The high incidence
of logging in fragmented Amazonian landscapes (e.g. Nepstad et al. 1999) will
further increase the vulnerability of forest remnants to ?re.
Implications for forest management
Until the past few decades, major ?res have been very rare in Amazonian
forests. As a result, few rain-forest species are adapted for ?re (Goldammer
MARK A . COCHRANE AND WILL IAM F . LAURANCE322
1990, Kauffman & Uhl 1990) and even low-intensity surface ?res that usually
originate in adjoining burned pastures kill many rain-forest trees (Cochrane &
Schulze 1999, Kauffman 1991). Such surface ?res increase ?ne litter, wood
debris and canopy openness, making forests highly vulnerable to intensive wild-
?res in the future. In recurring ?res, up to 98% of all trees become susceptible
to ?re-induced mortality (Cochrane et al. 1999).
An earlier study in Australia suggested that ?re return-intervals of less than
90 y might eliminate rain-forest tree species, while intervals of less than 20 y
could eradicate all but short-lived pioneer trees (Jackson 1968). Our analyses
suggest that forests within 2.35 km of edges at Taila?ndia, and all remaining
forests in Paragominas, had average ?re return-intervals less than that evid-
ently required to maintain rain-forest trees. Of equal concern is that ?re
return-intervals were less than 20 y ? the threshold at which non-pioneer trees
may be largely eradicated ? within 600?700 m of edges at Taila?ndia and at
least 2500 m of edges at Paragominas (Figure 3). These patterns suggest that
rain forests in both study areas will be largely replaced over time by anthropo-
genic savannas and scrubby regrowth. Fire-adapted trees (e.g. Byrsonima spp.,
Curatella spp., certain palms) are unlikely to colonize these areas because there
are no natural savannas or open woodlands nearby. The predicted deforestation
process may be reinforced not only by the tendency for once-burned forest to
become far more vulnerable to subsequent ?res (Cochrane & Schulze 1999,
Cochrane et al. 1999), but also by reductions in rainfall caused by increasing
deforestation (Lean & Warrilow 1989, Shukla et al. 1990) and by the moisture-
trapping effects of smoke caused by extensive forest and pasture ?res
(Rosenfeld 1999).
Because fragmentation drastically increased ?re incidence, it appears likely
that the margins of many forest fragments will recede over time, leading to a
progressive ?implosion? of the fragments. Gascon et al. (2000) suggested that
Amazonian fragments of less than 5000 ha will be susceptible to such effects,
but our results imply that much larger fragments (up to several hundred thou-
sand ha) could also be vulnerable to edge-related ?res. An important factor in
this regard is that prevailing land-uses in the Amazon, such as forest-
colonization projects, produce forest remnants that are highly irregular in
shape and thus vulnerable to edge-related ?res and other external disturbances
(Bierregaard et al. 1992, Laurance et al., in press, Lovejoy et al. 1986). Because
forest margins will tend to erode over time, the predictions of our static core-
area model may be conservative, and a dynamic modelling approach would be
useful for predicting longer-term interactions of fragmentation and ?re.
A key implication of our study (see Figure 4) is that even the largest Amazo-
nian nature reserves, if isolated by swaths of deforested land, may be seriously
degraded by ?res. This is especially likely to occur in the extensive areas of the
basin with pronounced dry seasons. In such areas, nature reserves are likely to
require deep buffers (> 3 km wide) of intact forest to ensure they are not
Fire and forest fragmentation 323
chronically degraded by ?res and other edge-related phenomena. Adequate
staf?ng and protection of reserves is also critical; a recent analysis of 86 federal
parks and protected areas in Brazil found that 43% were at high to extreme
risk because of illegal logging, deforestation, colonization, hunting, isolation of
the reserve from other forest areas, and additional forms of encroachment.
More than half of all reserves (54.6%) were judged to have nearly non-existent
management (Ferreira et al. 1999).
Future changes in climate could make the Amazon even more vulnerable to
?re. Some leading global-circulation models suggest that global warming may
increase the frequency of El Nin?os (Timmerman et al. 1999) and warm-weather
events (Mahlman 1997). Such changes, in concert with rapid forest fragmenta-
tion and logging, have the potential to accelerate losses of rain forest across
large expanses of the Amazon basin.
ACKNOWLEDGEMENTS
We thank Francis Putz, Claude Gascon, Philip Fearnside, Jeff Chambers and
three anonymous referees for useful comments, and Christopher Barber for
assistance with data analysis. Support was provided by the NASA-LBA program,
A. W. Mellon Foundation, Basic Science and Remote Sensing Initiative (BSRSI)
of Michigan State University, and U.S. Agency for International Development.
This is publication number 355 in the Biological Dynamics of Forest Fragments
Project technical series and number 01?04 in the BSRSI technical series.
LITERATURE CITED
BIERREGAARD, R. O., LOVEJOY, T. E., KAPOS, V., DOS SANTOS, A. A. & HUTCHINGS, R. W.
1992. The biological dynamics of tropical rain forest fragments. Bioscience 42:859?866.
COCHRANE, M. A. 2000. Forest ?re, deforestation and landcover change in the Brazilian Amazon. Pp.
170?176 in Neuenschwander, L. F., Ryan, K. C., Gollberg, G. E. & Greer, J. D. (eds). Crossing the
millennium: integrating spatial technologies and ecological principles for a new age in ?re management. University
of Idaho, Moscow.
COCHRANE, M. A. 2001. In the line of ?re: Understanding the impacts of tropical forest ?res.
Environment 43:28?38.
COCHRANE, M. A. & SCHULZE, M. D. 1998. Forest ?res in the Brazilian Amazon. Conservation Biology
12:948?950.
COCHRANE, M. A. & SCHULZE, M. D. 1999. Fire as a recurrent event in tropical forests of the
eastern Amazon: effects on forest structure, biomass, and species composition. Biotropica 31:2?16.
COCHRANE, M. A. & SOUZA, C. M. 1998. Linear mixture model classi?cation of burned forests in
the eastern Amazon. International Journal of Remote Sensing 19:3433?3440.
COCHRANE, M. A., ALENCAR, A., SCHULZE, M. D., SOUZA, C. M., NEPSTAD, D. C., LEFEBVRE,
P. & DAVIDSON, E. 1999. Positive feedbacks in the ?re dynamics of closed canopy tropical forests.
Science 284:1832?1835.
CURRAN, L. M., CANIAGO, I., PAOLI, G., ASTIANTI, D., KUSNETI, M., LEIGHTON, M., NIRARITA,
C. & HAERUMAN, H. 1999. Impact of El Nin?o and logging on canopy tree recruitment in Borneo.
Science 286:2184?2188.
DIDHAM, R. K., GHAZOUL, J., STORK, N. E. & DAVIS, A. J. 1996. Insects in fragmented forests:
a functional approach. Trends in Ecology and Evolution 11:255?260.
FEARNSIDE, P. M. 1990. Environmental destruction in the Amazon. Pp. 179?225 in Goodman, D. &
Hall, A. (eds). The future of Amazonia: destruction or sustainable development? MacMillan, London.
FEARNSIDE, P. M. 1999. Biodiversity as an environmental service in Brazil?s Amazonian forests: risks,
value and conservation. Environmental Conservation 26:305?321.
MARK A . COCHRANE AND WILL IAM F . LAURANCE324
FERREIRA, L. V., DE SA?, R. M. L., BUSCHBACHER, R., BATMANIAN, G., BENSURAN, B. R. &
COSTA, K. L. 1999. A?reas protegidas ou espac?os ameac?ados? World Wide Fund for Nature, Bras??lia.
GASCON, C., WILLIAMSON, G. B. & DA FONSECA, G. A. B. 2000. Receding edges and vanishing
reserves. Science 288:1356?1358.
GOLDAMMER, J. G. (ed). 1990. Fire in the tropical biota. Springer-Verlag, New York.
HOLDSWORTH, A. R. & UHL, C. 1997. Fire in the eastern Amazonian logged rain forest and the
potential for ?re reduction. Ecological Applications 7:713?725.
INPE 2001. Deforestation estimates in the Brazilian Amazon, 2000. National Institute for Space Research
(INPE), Sa?o Jose dos Campos, Brazil.
JACKSON, W. D. 1968. Fire, air, water and earth: an elemental ecology of Tasmania. Proceedings of the
Ecological Society of Australia 2:9?16.
KAPOS, V. 1989. Effects of isolation on the water status of forest patches in the Brazilian Amazon.
Journal of Tropical Ecology 5:173?185.
KAUFFMAN, J. B. 1991. Survival by sprouting following ?re in the tropical forests of the eastern
Amazon. Biotropica 23:219?224.
KAUFFMAN, J. B. & UHL, C. 1990. Interactions of anthropogenic activities, ?re, and rain forests in
the Amazon Basin. Pp. 117?134 in Goldammer, J. (ed). Fire in the tropical biota. Springer-Verlag, New
York.
KAUFFMAN, J. B., CUMMINGS, D. L., WARD, D. E. & BABBITT, R. 1995. Fire in the Brazilian
Amazon: biomass, nutrient pools, and losses in slashed primary forests. Oecologia 104:397?408.
KELLMAN, M. & MEAVE, J. 1997. Fire in the tropical gallery forests of Belize. Journal of Biogeography
24:23?34.
KLEIN, B. C. 1989. Effects of forest fragmentation on dung and carrion beetle communities in central
Amazonia. Ecology 70:1715?1725.
LAURANCE, W. F. 1998. A crisis in the making: responses of Amazonian forests to land use and
climate change. Trends in Ecology and Evolution 13:411?415.
LAURANCE, W. F. 2000. Do edge effects occur over large spatial scales? Trends in Ecology and Evolution
15:134?135.
LAURANCE, W. F. & YENSEN, E. 1991. Predicting the impacts of edge effects in fragmented habitats.
Biological Conservation 55:77?92.
LAURANCE, W. F., LAURANCE, S. G., FERREIRA, L. V., RANKIN-DE MERONA, J. M., GASCON,
C. & LOVEJOY, T. E. 1997. Biomass collapse in Amazonian forest fragments. Science 278:1117?1118.
LAURANCE, W. F., FERREIRA, L. V., RANKIN-DE MERONA, J. M. & LAURANCE, S. G. 1998a.
Rain forest fragmentation and the dynamics of Amazonian tree communities. Ecology 79:2032?2040.
LAURANCE, W. F., LAURANCE, S. G. & DELAMONICA, P. 1998b. Tropical forest fragmentation and
greenhouse gas emissions. Forest Ecology and Management 110:173?180.
LAURANCE, W. F., COCHRANE, M. A., BERGEN, S., FEARNSIDE, P. M., DELAMONICA, P.,
D?ANGELO, S., BARBER, C. & FERNANDES, T. 2001a. The future of the Brazilian Amazon. Science
291:438?439.
LAURANCE, W. F., PEREZ-SALICRUP, D., DELAMONICA, P., FEARNSIDE, P. M., D?ANGELO, S.,
JEROZOLINSKI, A., POHL, L. & LOVEJOY, T. E. 2001b. Rain forest fragmentation and the
structure of Amazonian liana communities. Ecology 82:105?116.
LAURANCE, W. F., LOVEJOY, T. E., VASCONCELOS, H. L., BRUNA, E., DIDHAM, R. K.,
STOUFFER, P. C., GASCON, C., BIERREGAARD, R. O., LAURANCE, S. G. & SAMPIAO, E. In
press. Ecosystem decay of Amazonian forest fragments, a 22-year investigation. Conservation Biology.
LEAN, J. & WARRILOW, D. A. 1989. Simulation of the regional climatic impact of Amazon deforestation.
Nature 342:411?413.
LI, H. & REYNOLDS, J. F. 1994. A simulation experiment to quantify spatial heterogeneity in categorical
maps. Ecology 75:2446?2455.
LOVEJOY, T. E., BIERREGAARD, R. O., RYLANDS, A. B., MALCOLM, J. R., QUINTELA, C. E.,
HARPER, L. H., BROWN, K. S., POWELL, A. H., POWELL, G. V. N., SCHUBART, H. & HAYS,
M. 1986. Edge and other effects of isolation on Amazon forest fragments. Pp. 257?285 in Soule?, M.
E. (ed). Conservation biology: the science of scarcity and diversity. Sinauer, Sunderland.
MAHLMAN, J. D. 1997. Uncertainties in projections of human-caused climate warming. Science
278:1416?1417.
MEGGERS, B. J. 1994. Archeological evidence for the impact of mega-Nin?o events on Amazonian
forests during the past two millennia. Climate Change 28:321?338.
NEPSTAD, D. C., CARVALHO, C. R., DAVIDSON, E. A., JIPP, P. H., LEFEBVRE, P. A., NEGREIROS,
P., SILVA, E. D., STONE, T. A., TRUMBORE, S. E. & VIEIRA, S. 1994. The role of deep roots
in the hydrological cycles of Amazonian forests and pastures. Nature 372:666?669.
NEPSTAD, D. C., MOUTINHO, P. R., UHL, C., VIEIRA, I. C. & DA SILVA, J. M. C. 1996. The
ecological importance of forest remnants in an eastern Amazonian frontier landscape. Pp. 133?150
Fire and forest fragmentation 325
in Schelhas, J. & Greenberg, R. (eds). Forest patches in tropical landscapes. Island Press, Washington,
DC.
NEPSTAD, D. C., VERISSIMO, A., ALENCAR, A., NOBRE, C., LIMA, C., LEFEBVRE, P.,
SCHLESINGER, P., POTTER, C., MOUTINHO, P., MENDOZA, E., COCHRANE, M. & BROOKS,
V. 1999. Large-scale impoverishment of Amazonian forests by logging and ?re. Nature 398:505?508.
PIPERNO, D. R. & BECKER, P. 1996. Vegetational history of a site in the central Amazon Basin
derived from phytolith and charcoal records from natural soils. Quaternary Research 45:202?209.
ROSENFELD, D. 1999. TRMM observed ?rst direct evidence of smoke from forest ?res inhibiting
rainfall. Geophysical Research Letters 26:3105?3108.
SALDARRIAGA, J. & WEST, D. C. 1986. Holocene ?res in the northern Amazon basin. Quaternary
Research 26:358?366.
SANFORD, R. L., SALDARRIAGA, J., CLARK, K., UHL, C. & HERRERA, R. 1985. Amazon rain-forest
?res. Science 227:53?55.
SHUKLA, J., NOBRE, C. A. & SELLERS, P. 1990. Amazon deforestation and climate change. Science
247:1322?1325.
SKOLE, D. & TUCKER, C. J. 1993. Tropical deforestation and habitat fragmentation in the Amazon:
satellite data from 1978 to 1988. Science 260:1905?1910.
SUPLEE, C. 1999. El Nin?o/La Nin?a: nature?s vicious cycle. National Geographic Magazine, March, pp. 21?34.
TIMMERMAN, A., OBERHUBER, J., BACHER, A., ESCH, M., LATIF, M. & ROECKNER, E. 1999.
Increased El Nin?o frequency in a climate model forced by future greenhouse warming. Nature
398:694?697.
UHL, C. & KAUFFMAN, J. B. 1990. Deforestation effects on ?re susceptibility and the potential
response of the tree species to ?re in the rain forest of the eastern Amazon. Ecology 71:437?449.
VAN WAGNER, C. E. 1978. Age-class distribution and the forest ?re cycle. Canadian Journal of Forest
Research 8:220?227.
WHITMORE, T. C. 1997. Tropical forest disturbance, disappearance, and species loss. Pp. 3?12 in
Laurance, W. F. & Bierregaard, R. O. (eds). Tropical forest remnants: ecology, management, and conservation
of fragmented communities. University of Chicago Press, Chicago.