supplementary materials for - science · government plans under pac and pac-2 (the two phases of...
TRANSCRIPT
1
www.sciencemag.org/cgi/content/full/337/6091/228/DC1
Supplementary Materials for
Extinction Debt and Windows of Conservation Opportunity in the Brazilian Amazon
Oliver R. Wearn, Daniel C. Reuman, Robert M. Ewers*
*To whom correspondence should be addressed. E-mail: [email protected]
Published 13 July 2012, Science 337, 228 (2012)
DOI: 10.1126/science.1219013
This PDF file includes:
Materials and Methods Figs. S1 to S8 Tables S1 and S2 References
Other Supplementary Material for this manuscript includes the following: available at www.sciencemag.org/cgi/content/full/337/6091/228/DC1
Movies S1 to S4
1
Extinction Debt and Windows of Conservation Opportunity in the
Brazilian Amazon
Oliver R. WEARN1,2, Daniel C. REUMAN1,3 and Robert M. EWERS1*
1Imperial College London, Silwood Park, Ascot SL5 7PY, UK
2Zoological Society of London, Institute of Zoology, London NW1 4RY, UK
3Rockefeller University, New York 10065, USA
*Corresponding author: Ewers, R. M. ([email protected])
Supplementary Material
2
Materials and Methods
Range overlay maps for Amazonian forest-dependent vertebrates
Geographic range maps for mammals and amphibians were obtained from the respective IUCN
Global Species Assessments (18). These datasets do not include some species categorised as Data
Deficient which are missing range maps, but the vast majority of species occurring in the region
(>98% for both groups) are included. For birds, we used the NatureServe Birds of the Western
Hemisphere dataset (26). Some species in the bird dataset only had point locality data. For poorly
known taxa, points were buffered with a radius of 100 km. For species with larger numbers of
records, minimum convex polygon range maps were used. In some cases there were inconsistencies
between the IUCN taxonomy for birds (which largely follows BirdLife International) and that of
Ridgely et al. (26). In general, BirdLife International uses a more conservative approach with regard
to sub-species versus species designation, and we here followed this by merging the sub-species
range maps together.
Since SAR-based extinction predictions are undermined by incorporating disturbance-tolerant
species, we filtered our species lists at the outset, leaving only the forest-dependent species. We
obtained habitat data for each species from the IUCN Species Information Service (SIS,
www.iucnredlist.org), a database consisting of information collated from published and grey
literature sources, as well as expert consultation and regional workshops. For mammals (n = 204)
and birds (n = 332), forest-dependents were defined as species that have been recorded only from
natural forest (excluding, for example, species known to occur in savanna, wetland, secondary
growth or agricultural land). In the case of amphibians (n = 214), forest species known to occur in
inland wetlands (including, for example, permanent and seasonal pools and streams) were not
excluded, owing to the two-stage life cycle of most amphibians (27). Whilst every bird species had
3
associated habitat info, some amphibians (n = 5) and mammals (n = 7) did not. In these cases, we
assigned a single habitat type based on the species account given or, as necessary, wider literature
sources.
To obtain species richness estimates for each cell, range overlay maps were created in ArcGIS (28),
implementing a Visual Basic for Applications (VBA) script to count only species whose range covered
more than 20% of a grid cell.
Along with the species richness (R
iS ), each grid cell was associated with a value for the area over
which species had been summed ( iA )0( ). Typically, geographic range maps represent the “extent of
occurrence” of a species (18), which may amount to a minimum convex polygon with little regard for
the finer-scale patterns of habitat occurrence. As a result, in most cases iA )0( was simply defined by
the grid spacing (i.e. 50 x 50 km2) but, where all range maps had obviously been clipped to a
geographic feature (such as the Atlantic coastal boundary), iA )0( was lower than the nominal cell
size.
Forest cover time-series data for the Brazilian Amazon
In order to estimate extinction debt and species loss across Brazilian Amazonia, we required high-
resolution maps of forest cover at more or less regular time intervals. Ideally, forest cover time-
series would also extend back to when forests were last in equilibrium.
Forest cover estimates for the recent period (1998 and 2001-2008) were derived from PRODES (the
‘Monitoring Gross Deforestation in the Amazon Project’). As part of PRODES, Instituto Nacional de
Pesquisas Espaciais (INPE) have created a high-resolution, spatially-explicit digitised product on an
4
annual basis since 2000 (following a ‘reference’ map established in 1997). The PRODES product
derives from a mosaic of Landsat Thematic Mapper (TM) images, mostly taken during the height of
the dry season when cloud cover is lowest (typically in August), which is then analysed by a
combination of automated classification (29) and manual photo interpretation. It is then made
publicly available as a categorised land cover map in the format of a 120 m resolution raster (20).
All Landsat-based analyses of deforestation suffer from extensive areas with no data, due to cloud
cover present during each satellite pass. Cloud cover is usually extensive over Amazonia, and occurs
throughout the year; it is especially problematic in the northern Brazilian Amazon (30). We took the
conservative approach of excluding cloud-covered pixels from forest cover estimates (other pixels
with no data were dealt with in the same way). This will ultimately lead to a slight underestimation
of absolute quantities of extinction debt and species loss in some cells (mostly in Amapá state, which
has the greatest cloud coverage), but does not affect the percentage estimates of these quantities.
Moreover, cells with less than 20 % observed forest cover in their pristine state were excluded from
all further analyses. This also removed the large areas of cerrado (and other natural, non-forest
habitats) that the Brazilian Legal Amazon encompasses.
We used the PRODES pixel classes to construct maps of forest cover for 1998 and each year between
2001 and 2008, as well as a “pristine” forest cover map. The “pristine” forest cover map was
ascribed to the year 1970; this year is generally agreed to be the beginning of the post-Columbian,
“modern era” of major deforestation (31, 32) and is therefore suitable as our baseline datum for
when forests are assumed to be at equilibrium. Furthermore, we were able to add intervening data
points using layers of “old” (pre-PRODES) deforestation apportioned to three broad time periods
(1970-1977, 1978-1987 and 1988-1991) (31). Since exact dates could not be attributed to this early
deforestation data, we considered three scenarios in order to capture the resulting uncertainty: (1)
deforestation within each cell was apportioned equally throughout the period considered
5
(“Interpolated”); (2) forest was lost in a single episode at the beginning of the period (“Immediate”),
or (3) all deforestation occurred in a single episode at the very end of the period (“Delayed”). We
consider the first scenario to be the most reasonable. All estimates of extinction debt will be
bounded from above by the Delayed scenario and from below by the Immediate scenario (vice versa
for estimates of species loss).
Scenarios of future deforestation – simulation models and policy targets
‘Business as Usual’ and ‘Governance’ scenarios from SimAmazonia
We used two spatially-explicit scenarios of future deforestation in the Brazilian Amazon – “business
as usual” (BAU) and “governance” (GOV) – as modelled by Soares-Filho et al. (8). These simulations,
collectively called “SimAmazonia”, were originally run beginning with observed forest cover data for
July-September 2001 and predicting forest cover through to 2050. In our models, we used observed
forest cover data up until 2008 and the SimAmazonia predictions from 2009-2050. Since real
deforestation rates underwent an unexpectedly rapid decline in the second half of this decade,
SimAmazonia had overestimated absolute amounts of deforestation in the period 2002-2008 (less so
in the GOV than in the BAU scenario). This discrepancy causes an increase in extinction debt at the
transition between observed and modelled datasets which is artificially rapid.
Under the BAU scenario, deforestation in a sub-region could not exceed a designated 85 % of the
area outside protected areas, and 40 % within protected areas (8). In order to model an
improvement in frontier and protected area governance, as well as increased compliance with
environmental law, maximum deforestation in the GOV scenario was set to 50 % and 0 %,
respectively, outside and inside protected areas and sub-region deforestation rates were
constrained to decrease in an arbitrary logistic manner. Note, even in the GOV scenario, the
6
minimum forest cover outside protected areas still exceeds the 80% that is actually required of
private landowners in the Brazilian Forest Code (8). In the GOV scenario, the protected area network
was also expanded in accordance with plans outlined under ARPA (the “Protected Areas of the
Amazon” programme).
In both scenarios the road network is improved and extended, which is in line with current
government plans under PAC and PAC-2 (the two phases of the “Growth Acceleration Program”).
This includes: (1) the re-paving of the BR-319 to Manaus by 2018; (2) the paving of the Trans-
Amazon (BR-230) between Itaituba and Humaitá by 2025; (3) the completion of the paving of the BR-
364 in Acre; and (4) the construction of a continuously-paved link road between the BR-163 and BR-
364 across northern Mato Grosso by 2025 (Fig. S2).
Targeted deforestation reduction scenarios
On the basis of recent political, societal and economic trends in the Brazilian Amazon, some
commentators have tentatively concluded that the prospects for forest cover, and therefore
biodiversity, in the region may be brighter than they have perhaps ever been since the beginning of
the “modern era” of deforestation (9, 33). This poses a stark contrast to the dramatic scenarios of
future tropical forest loss that have prevailed in the literature for the last three decades (34-36). In
order to model these newly optimistic scenarios of forest loss, we constructed two additional
scenarios to those outlined above. These were implemented quasi-spatially, taking each 50 x 50 km2
grid cell as a homogenous unit of simulation (echoing the scale at which calculations of species loss
and extinction debt would be made).
The Strong Reduction scenario is based on the recent goal – set out by the Brazilian Government – to
achieve an 80% reduction in the deforestation rate by 2020, relative to a ten-year baseline period
7
ending in 2005 (9). For each cell, we calculated this baseline rate, as well as the 80% reduced target
rate for 2020, and interpolated the annual rates that would need to be achieved in each year from
2009 to reach the target in 2020. Note that the basin-wide deforestation rate by 2008 was already
reduced to 66% of the baseline (20), so the reductions from 2009 to 2020 are more modest than the
80% target might suggest. The Brazilian National Plan on Climate Change does not indicate any long-
term commitment to eliminating deforestation; we therefore allowed deforestation after 2020 to
continue in each cell according to the reduced rate that had already been achieved.
The End of Deforestation scenario, on the other hand, ambitiously sets out to end deforestation by
2020, based on a programme of compensation (both direct and indirect) for forest conservation and
increased investment in protected areas (9). This programme, which builds upon the record low
deforestation rate of 7,000 km2 observed for 2009, plans to reduce the baseline in successive stages:
to 5,000 km2 by 2011, 1,500 km2 by 2016 and zero by 2020. We translated these basin-wide absolute
rates into proportional reductions and applied them on a cell-by-cell basis. The 2009 baseline for
each cell, in this case, was derived from the PRODES product for 2009 (20). Thereafter, the transition
to a zero-deforestation Brazilian Amazon was considered complete, and deforestation rates
remained at zero until 2050.
Quantifying extinction debt
Modelling approach
The most common approach taken to identify, and indeed quantify, extinction debt has been to use
SAR-based predictions of equilibrium species richness (4, 37-43). Across systems presumed to be in
equilibrium, SARs are exceptionally well validated and have high generality across taxa, spatial scale
and ecological system (12, 13, 44, 45). In combination with estimates of current species richness,
8
SARs therefore offer a highly practical means of identifying extinction debt at large spatial and
ecological scales. This developed from the early realisation that SARs consistently over-predicted the
observed species losses in the short-term (14, 46, 47), and that this discrepancy could be interpreted
by the persistence of species that were nonetheless committed to extinction (15, 46). Empirical
support for this interpretation has been given by a number of studies (15, 37, 38, 48-50) noting a
correspondence between SAR-based extinction predictions and regional or global threatened
species lists, such as the IUCN Red List.
To model extinction debt through time, we can extend the SAR to incorporate a process of
community relaxation. We know that in the immediate term following habitat loss, individuals of
mobile species concentrate in the remaining habitat fragments, giving rise to so-called “crowding
effects” and species “supersaturation” (51-54). Following this, species richness relaxes to a new
equilibrium over a period of decades to centuries (5, 55). We also know that the shape of the
relaxation curve is much steeper in the initial stages of relaxation and shallower as the system
approaches equilibrium (5, 54, 56-60). These empirical and theoretical findings can be included in a
model which provides instantaneous estimates of species loss and extinction debt, with dynamically
changing habitat area.
Assumptions of the model
We used the familiar power-law form of the SAR (12). We recognise that, though this is by no means
the only model to describe the relationship, and much debate has ensued over this (44, 61-63), it is
the most tested and generally most robust (13,44). Provided it is not used at either very small or
very large scales (less than 1 ha or greater than 107 km2), it will most often provide a good fit to data
(12, 61, 64). We did not favour the endemics-area relationship (EAR) (65), because we were
modelling a community-level process of relaxation in remnant habitat, rather than a sampling
9
process within continuous habitat. Most obviously, these approaches differ in their requirements for
extinction (66): the EAR requires every individual of a species to lose its habitat for extinction to
occur, whilst the SAR incorporates the effects of increased isolation and density-compensation in
smaller habitat patches, such that a species may be driven to extinction before all of its habitat has
been destroyed.
We took a cell-by-cell approach to account for spatial variability in species richness and intensity of
habitat loss (67). Each cell was treated as an independent habitat isolate, akin to the idealised patch
considered in the model’s formulation. We used a relatively coarse-scale grid (2500 km2 cells) to
minimise neighbourhood effects, common to all analyses of this type, but this reduces our ability to
capture effects of within-cell habitat heterogeneity. The cell-by-cell application of SARs to predict
extinction has precedent; for example, it has previously been used to predict losses of primate
species from African countries (38) and plant species from regions of California (67).
Empirical z-values
The exponent z is the slope of the log-log plot of the power-law SAR. Simply, it describes how rapidly
species are lost or gained as, respectively, habitat is lost or sampling size is increased. Employing the
“island analogy” for isolated remnant habitats, a z-value of 0.25 has commonly been used to
predict extinction following habitat loss (2, 4, 14, 15, 23, 33, 37). Rather than using a single mean
value for z, or even an upper and lower estimate, we instead estimated a probability distribution for
z from empirical data (13, 38, 68-72). In turn, this enabled us to capture the effect of parameter
uncertainty on the outputs of our models using Monte Carlo randomisations (see Section Capturing
uncertainty: Monte Carlo simulations).
10
We assembled most of our data for z from a database established for a meta-analysis of SARs (13);
this is the most comprehensive analysis of z-values thus far conducted, using nearly 800 SARs from a
broad diversity of locations, habitats and taxonomic groups. Since we know that at least some of the
wide variation in observed z is systematic (13), we filtered the dataset to leave only those SARs that
were strictly for tropical mammals, birds or amphibians, leaving 34 SARs. Unfortunately, few
amphibian studies remained after this filtering, so we supplemented the data with additional
published SARs located using the ISI Web of Science (www. isiknowledge.com, using the search
term: ["species-area" SAME relationship*] AND [amphibian*]) and other studies that we were aware
of. In total, we used 42 SARs (17, 19 and 6 SARs from mammalian, avian and amphibian taxa,
respectively). Due to the relatively low number of estimates, we could not justify modelling separate
distributions of z for each vertebrate group. The overall bootstrap mean (stratified by study and
taxon) for z was 0.269 (95% CI: 0.224-0.315), and the random draws of z used in Monte Carlo
simulations (Fig. S1) had a mean value of 0.270.
Estimating values of k from the literature
The relaxation constant k is the most important parameter in our model, as it determines the
relative speed at which species losses occur. It can be seen simply as the reduction in absolute
species loss rate that occurs with each extinction. Unlike the case for z, no theory currently exists to
indicate what value it might take, and empirical attempts at its quantification have been rare.
Perhaps prematurely, the “ball-park” findings of a study of bird community relaxation in Kenyan
forest fragments (4) – about 50% species loss in 50 years – have already been adopted in the
literature (73, 74). We attempted to improve upon this rough estimate, ultimately aiming to input a
probability distribution for k into Monte Carlo randomisations, as we did for z (see Section Capturing
uncertainty: Monte Carlo simulations).
11
Ideally, the best data for these purposes would come from contemporary observations of the
relaxation process, from beginning to end. Then, given an assumption of single-episode habitat loss,
solving Eq. 1 is straightforward. After re-arranging terms, we are able to estimate k:
eq
eq
StS
SStk
)(
)0(ln1 (S1)
where t is the time since habitat loss or fragment isolation. In practice, values of S(0) and Seq are
often not observed directly. S(0) can be inferred, however, from suitable historical records, or by
using a space-for-time substitution (i.e. from “control” sites that have not undergone habitat loss).
We can use a suitably parameterised SAR to obtain Seq, so long as data on the areal extent of habitat
are available both before and after habitat loss.
A related parameter used to express the speed of relaxation is the half-life of community species
richness (3, 4). The half-life, th is interchangeable with k via:
2ln1 kth (S2)
Despite their more intuitive nature, we do not use half-lives because the clarity of their definition
becomes lost in the context of a model with continual habitat loss and shifting equilibrium.
We conducted an extensive literature search for suitable studies from which we could calculate k,
using various searches on the ISI Web of Science (www.isiknowledge.com) for articles pertaining to
relaxation, species loss, extinction from fragments or regional extirpations, and following the
12
reference trail where relevant. We supplemented these studies with other published studies we
were aware of.
Acceptable sources of S(0) included: survey work; species checklists (assuming they were sufficiently
specific to the area in question); the application of expert knowledge (often in combination with
other historical evidence), or inferences from present-day “control” sites. A few studies used SARs
(appropriately parameterised) to infer S(0) and, in these cases, a type of “founder effect” on species
richness had been presumed (i.e. instantaneous sampling following habitat loss). However, in most
cases, fragments or landscapes were assumed to be “supersaturated” immediately following habitat
loss, or S(0) had been directly observed. Estimates of current species richness S(t) had to be derived
from survey work of comparable effort and scale; in general, this was often explicitly accounted for,
since most studies had expressly set out to compare S(0) and S(t). Some studies provided two or
more estimates of S(t) at different time points, in which case we took the most recent provided.
Where studies had offered evidence of absent or extremely limited immigration, we followed any
assertions that equilibrial species richness was zero. If the area of original habitat (or the area of the
control fragment) was not given precisely, we used maps or written descriptions to estimate it;
where this was not possible, or fragments were isolated from purportedly “continuous” forest (i.e.
habitat with an area > 1 million ha), we took the initial area to be an arbitrary 100 km2 (75). In such
cases, calculated k-values were not especially sensitive to the initial area estimate, as they almost all
involved drastic reductions in habitat area down to less than 1 km2, meaning Seq was usually close to
zero. The time since habitat isolation, t, was often not equivalent to the time elapsed between
species richness estimates (e.g. the period between surveys), though the latter was more often
provided; where suitable information could be found, we used the former (and the mid-point if t
could only be estimated within bounds). Finally, if the data allowed it, we considered only the subset
of taxa that were forest-dependent and native to the region. We found 53 studies suitable for
estimating values of k (3, 4, 14, 42, 43, 51, 54, 56, 57, 59, 69, 70, 75-115). This included 329 patch-
13
level observations and 45 landscape-level observations (making 309 observations which were
independent in the sense that no two of them used the same taxa from the same location). All told,
238 different habitat isolates (i.e. habitat islands, real islands or habitat-denuded landscapes) were
included, across a diversity of plant, invertebrate and vertebrate taxa. 44% of observations were for
avian taxa.
It has previously been asserted that relaxation rates are dependent on the absolute area of remnant
habitat (4) and, in this way, might be scale-dependent. Since we applied our model of species loss
and extinction debt in the Brazilian Amazon at large scales (in grid cells of 250,000 ha), for the
purposes of the Monte Carlo simulations we excluded all observations of k from small habitat
isolates (< 100 ha). Mean k in this reduced dataset, using an empirically-derived z-value of 0.269 (see
Section Empirical z-values), was 0.0122 (95% CI: 0.0076-0.0194), which corresponds to a half-life of
57 years (from Eq. S2). This is only slightly higher than the “ball-park” half-life often repeated to be
50 years (4, 73, 74), though it is more robustly supported.
Capturing uncertainty: Monte Carlo simulations
In order to explicitly incorporate the uncertainty associated with our parameter estimates into our
models of species loss and extinction debt, we used a Monte Carlo simulation framework. This
framework better reflects the primacy of empirical data than alternative approaches, and uses
available data more effectively than simply inputting an upper and lower estimate for each
parameter.
Using the empirical z-values, we defined a probability density function for mean z by taking stratified
bootstrap samples of the data (n = 10,000) and then fitting a Gaussian curve. Similarly, we fitted a
probability density function to the stratified bootstrap samples of the k-value data: in this case a
14
lognormal density function was used owing to the skew in the distribution. Note that since the
calculation of k usually involves z (in order to calculate Seq; see Eq. S1), we incorporated this
additional uncertainty by taking a random draw of z and re-calculating k before each bootstrap was
taken. We decided to decouple the parameters z and k during Monte Carlo simulations in order to
fully capture the range of uncertainty (otherwise low z-values would always be entered into
randomisations with high k-values, and vice versa). This resulted in a distribution for k (Fig. S1) with
marginally wider confidence intervals than if just a mean value for z had been used.
Random draws from the fitted parameter distributions were then taken in each Monte Carlo
simulation (n = 1000) and used in our model to calculate annual estimates, for each 50 x 50 km2 grid
cell, of species loss and extinction debt for the three vertebrate groups, across the various scenarios
of forest loss. All reported confidence intervals for a given statistic are based on the quantiles of its
distribution, taken from the results across Monte Carlo randomisations.
Model validations
We attempted to validate our model predictions at both the scale of the Brazilian Amazon and at the
scale of individual grid cells. First, we compared our Brazilian Amazon-scale predictions of extinction
debt to the number of species listed as threatened on the Red List (Critical, Endangered or
Vulnerable categories) (18). There are surprisingly few Brazilian Amazon forest-dependent species
listed as threatened – just 35. However, because of the 76 Data Deficient species in the mammals
and amphibians, the ultimate number of threatened species could lie somewhere between 35 and
111. Assuming Data Deficient species will be found to be as threatened as species that have already
been assessed, there will be 38.9 species threatened. This compares with a predicted debt of 35.0
(95% CI: 28.8 - 41.8), which is very close (Fig. 2A).
15
We also attempted to validate our models at the cell level. We obtained bird lists in four gridsquares
in the Brazilian Amazon: Jari (116, 117), Paragominas, Santarem and Alta Floresta (118). Making the
assumption that species range maps gives us an estimate of the species that were present in a given
grid square pre-deforestation, the difference between that estimate and the observed number of
forest-dependent species within those gridsquares in the present day should give an empirical
estimate of local extinction. When we conducted this analysis, we found a positive but non-
significant correlation between predicted and ‘observed’ local extinction across the four grid squares
(r = 0.098, df = 2, p = 0.90). The slope of the relationship is less than 1 suggesting that the validation
exercise shows our model to under-predict extinction relative to ‘observation’. This likely arises
because none of the field data were collected at spatial scales that come close to representing
exhaustive censuses of 50 x 50 km grids at which scale we were modelling. It follows that any species
list is biased towards underestimating the number of species present in a grid square, so this type of
validation is instantly biased towards making it look as though our model under-predicts extinction.
Moreover, all of the gridsquares appeared to gain forest-dependent species, highlighting a problem
in matching the data from the Red List species distribution maps and the field data to estimate
‘observed’ extinction. Part of the problem arises from the two data sources using different
taxonomies (species that are recorded in the field don’t have a species distribution in the Red List
data), and part of the problem arises from the inevitable inaccuracies involved in mapping species
distributions at global scales (some of the ‘new’ species occur in a gridsquare that is outside of their
range according to the distribution maps). Because of these issues with the comparability of the two
datasets, we consider this validation a particularly weak test of the model predictions.
16
Supplementary Movies
Movie S1: Historical trends of absolute species loss and extinction debt between 1970 and 2008
(using the “interpolated clearance” scheme of Table 1) and predicted future trends through to 2050
for the Brazilian Amazon under the Business as Usual (BAU) scenario. Spatial patterns of absolute
species loss are represented by the size of squares, with squares reducing in size through time
according to the number of local extinctions predicted to occur. Spatial patterns of absolute
extinction debt are represented by colour, with squares changing from green through to red
according to the number of species committed to local extinction. Large and red squares represent
faunas that are intact but imperilled, and are where the greatest gains could be expected from
conservation actions. Values are cell-wise means across Monte Carlo simulations, and were summed
across birds, mammals and amphibians.
Movie S2: Historical trends of absolute species loss and extinction debt between 1970 and 2008
(using the “interpolated clearance” scheme of Table 1) and predicted future trends through to 2050
for the Brazilian Amazon under the Governance (GOV) scenario. Spatial patterns of absolute species
loss are represented by the size of squares, with squares reducing in size through time according to
the number of local extinctions predicted to occur. Spatial patterns of absolute extinction debt are
represented by colour, with squares changing from green through to red according to the number of
species committed to local extinction. Large and red squares represent faunas that are intact but
imperilled, and are where the greatest gains could be expected from conservation actions. Values
are cell-wise means across Monte Carlo simulations, and were summed across birds, mammals and
amphibians.
Movie S3: Historical trends of absolute species loss and extinction debt between 1970 and 2008
(using the “interpolated clearance” scheme of Table 1) and predicted future trends through to 2050
for the Brazilian Amazon under the Strong Reduction (SR) scenario. Spatial patterns of absolute
species loss are represented by the size of squares, with squares reducing in size through time
according to the number of local extinctions predicted to occur. Spatial patterns of absolute
extinction debt are represented by colour, with squares changing from green through to red
according to the number of species committed to local extinction. Large and red squares represent
faunas that are intact but imperilled, and are where the greatest gains could be expected from
conservation actions. Values are cell-wise means across Monte Carlo simulations, and were summed
across birds, mammals and amphibians.
17
Movie S4: Historical trends of absolute species loss and extinction debt between 1970 and 2008
(using the “interpolated clearance” scheme of Table 1) and predicted future trends through to 2050
for the Brazilian Amazon under the End of Deforestation (EOD) scenario. Spatial patterns of absolute
species loss are represented by the size of squares, with squares reducing in size through time
according to the number of local extinctions predicted to occur. Spatial patterns of absolute
extinction debt are represented by colour, with squares changing from green through to red
according to the number of species committed to local extinction. Large and red squares represent
faunas that are intact but imperilled, and are where the greatest gains could be expected from
conservation actions. Values are cell-wise means across Monte Carlo simulations, and were summed
across birds, mammals and amphibians.
18
Supplementary Tables
Table S1. Absolute species losses per 2500 km2 cell for three forest-dependent vertebrate groups in the
Brazilian Amazon, according to various scenarios of forest loss (see Section Scenarios of future deforestation).
For estimating species loss to 2050, the middle range 1970-2008 scenario (‘Interpolated Clearance’) was
combined with one of the four scenarios of future forest loss. All quantities are summary statistics from Monte
Carlo randomisations (n = 1000). Ranges are the mean minimum and maximum values obtained across
simulations. Median species losses were calculated after rounding estimates for each grid cell to the nearest
species.
Taxon Year Scenario Mean (95% CI) Median (95% CI) Range
Mammals 2008 Immediate Clearance 0.47 (0.30 – 0.71) 0 (0 – 0) 0 – 9.88
Interpolated Clearance 0.42 (0.26 – 0.63) 0 (0 – 0) 0 – 8.34
Delayed Clearance 0.37 (0.23 – 0.57) 0 (0 – 0) 0 – 7.76
2020 Business-as-usual 1.01 (0.63 – 1.51) 0 (0 – 0) 0 – 11.48
Governance 0.93 (0.58 – 1.38) 0 (0 – 0) 0 – 11.32
Strong Reduction 0.82 (0.52 – 1.21) 0 (0 – 0) 0 – 11.28
End of Deforestation 0.78 (0.49 – 1.15) 0 (0 – 0) 0 – 11.28
2050 Business-as-usual 4.00 (2.63 – 5.77) 2 (2 – 4) 0 – 21.26
Governance 2.47 (1.64 – 3.47) 1 (0 – 1) 0 – 19.30
Strong Reduction 2.02 (1.36 – 2.83) 0 (0 – 0) 0 – 17.64
End of Deforestation 1.53 (1.04 – 2.11) 0 (0 – 0) 0 – 17.46
Birds 2008 Immediate Clearance 0.93 (0.58 – 1.39) 0 (0 – 0) 0 – 15.95
Interpolated Clearance 0.83 (0.51 – 1.24) 0 (0 – 0) 0 – 13.73
Delayed Clearance 0.73 (0.45 – 1.11) 0 (0 – 0) 0 – 12.58
2020 Business-as-usual 2.04 (1.27 – 3.06) 0 (0 – 1) 0 – 24.00
Governance 1.87 (1.17 – 2.79) 0 (0 – 1) 0 – 22.33
Strong Reduction 1.63 (1.03 – 2.42) 0 (0 – 1) 0 – 22.30
End of Deforestation 1.56 (0.98 – 2.29) 0 (0 – 0) 0 – 20.62
2050 Business-as-usual 8.55 (5.62 – 12.38) 5 (3 – 7) 0 – 51.82
Governance 5.15 (3.40 – 7.25) 2 (1 – 2) 0 – 43.98
Strong Reduction 4.06 (2.73 – 5.68) 0 (0 – 1) 0 – 51.39
End of Deforestation 3.08 (2.08 – 4.25) 0 (0 – 1) 0 – 36.88
Amphibians 2008 Immediate Clearance 0.20 (0.13 – 0.30) 0 (0 – 0) 0 – 4.07
Interpolated Clearance 0.18 (0.11 – 0.27) 0 (0 – 0) 0 – 3.50 Delayed Clearance 0.16 (0.10 – 0.24) 0 (0 – 0) 0 – 3.07 2020 Business-as-usual 0.46 (0.29 – 0.70) 0 (0 – 0) 0 – 6.56 Governance 0.42 (0.26 – 0.63) 0 (0 – 0) 0 – 6.38 Strong Reduction 0.36 (0.23 – 0.53) 0 (0 – 0) 0 – 6.29 End of Deforestation 0.34 (0.22 – 0.51) 0 (0 – 0) 0 – 5.64 2050 Business-as-usual 2.09 (1.37 – 3.04) 1 (1 – 1) 0 – 16.29 Governance 1.22 (0.81 – 1.73) 0 (0 – 0) 0 – 15.96 Strong Reduction 0.91 (0.61 – 1.27) 0 (0 – 0) 0 – 14.85 End of Deforestation 0.67 (0.45 – 0.94) 0 (0 – 0) 0 – 10.44
19
Table S2. Absolute extinction debt per 2500 km2 cell for three forest-dependent vertebrate groups in the
Brazilian Amazon, according to various scenarios of forest loss (see Section Scenarios of future deforestation).
For estimating extinction debt to 2050, the middle range 1970-2008 scenario (‘Interpolated Clearance’) was
combined with one of the four scenarios of future forest loss. All quantities are summary statistics from Monte
Carlo randomisations (n = 1000). Ranges are the mean minimum and maximum values obtained across
simulations. Median extinction debts were calculated after rounding estimates for each grid cell to the nearest
species.
Taxon Year Scenario Mean (95% CI) Median (95% CI) Range
Mammals 2008 Immediate Clearance 2.22 (1.85 – 2.55) 0 (0 – 0) 0 – 20.81
Interpolated Clearance 2.27 (1.91 – 2.59) 0 (0 – 0) 0 – 21.70
Delayed Clearance 2.32 (1.97 – 2.64) 0 (0 – 0) 0 – 22.16
2020 Business-as-usual 4.92 (4.18 – 5.55) 1 (1 – 2) 0 – 31.03
Governance 3.81 (3.17 – 4.34) 1 (1 – 1) 0 – 30.53
Strong Reduction 2.99 (2.52 – 3.39) 0 (0 – 0) 0 – 40.65
End of Deforestation 2.45 (1.99 – 2.83) 0 (0 – 0) 0 – 32.58
2050 Business-as-usual 11.33 (9.40 – 12.89) 9 (7 – 11) 0 – 48.42
Governance 4.31 (3.21 – 5.24) 2 (1 – 2) 0 – 44.38
Strong Reduction 3.51 (2.73 – 4.17) 0 (0 – 0) 0 – 35.50
End of Deforestation 1.70 (1.12 – 2.22) 0 (0 – 0) 0 – 22.55
Birds 2008 Immediate Clearance 4.51 (3.75 – 5.18) 1 (1 – 1) 0 – 49.06
Interpolated Clearance 4.62 (3.89 – 5.28) 1 (1 – 1) 0 – 51.11
Delayed Clearance 4.71 (3.99 – 5.38) 1 (1 – 1) 0 – 52.72
2020 Business-as-usual 10.34 (8.77 – 11.67) 3 (3 – 4) 0 – 80.41
Governance 7.92 (6.60 – 9.04) 2 (2 – 3) 0 – 69.68
Strong Reduction 6.06 (5.08 – 6.89) 1 (1 – 1) 0 – 104.79
End of Deforestation 4.95 (3.89 – 5.28) 1 (1 – 1) 0 – 55.56
2050 Business-as-usual 25.12 (20.90 – 28.52) 19 (15 – 22) 0 – 117.74
Governance 9.27 (6.94 – 11.24) 3 (2 – 4) 0 – 91.57
Strong Reduction 7.18 (5.58 – 8.54) 1 (0 – 1) 0 – 95.40
End of Deforestation 3.43 (2.24 – 4.49) 0 (0 – 1) 0 – 38.49
Amphibians 2008 Immediate Clearance 1.00 (0.83 – 1.16) 0 (0 – 0) 0 – 13.88
Interpolated Clearance 1.03 (0.86 – 1.18) 0 (0 – 0) 0 – 14.41 Delayed Clearance 1.05 (0.89 – 1.20) 0 (0 – 0) 0 – 14.82 2020 Business-as-usual 2.46 (2.08 – 2.78) 1 (1 – 1) 0 – 25.68 Governance 1.86 (1.55 – 2.13) 0 (0 – 1) 0 – 25.27 Strong Reduction 1.34 (1.12 – 1.53) 0 (0 – 0) 0 – 35.23 End of Deforestation 1.08 (0.86 – 1.18) 0 (0 – 0) 0 – 15.99 2050 Business-as-usual 6.47 (5.43 – 7.33) 4 (3 – 4) 0 – 33.57 Governance 2.33 (1.77 – 2.81) 1 (0 – 1) 0 – 26.88 Strong Reduction 1.63 (1.27 – 1.94) 0 (0 – 0) 0 – 30.14 End of Deforestation 0.75 (0.49 – 0.99) 0 (0 – 0) 0 – 11.09
20
Supplementary Figures
Fig. S1. Empirically-derived input distributions of z and k and the random draws actually used in the Monte
Carlo simulations (n = 1000). In the main plot: the density of points is represented by colour saturation; grey
dashed lines indicate the respective 95% confidence intervals for the empirical z- and k-value data (calculated
by stratified bootstrapping), and axis tick marks correspond to the random draws. The horizontal and vertical
marginal plots show histograms and fitted probability density functions (red dashed lines) for the distributions
of mean z and k.
21
Fig. S2. Significant geographic features mentioned in the text. The base map shows land cover for the region in
2004 (NASA Earth Observatory). The grey outlines show the nine Legal Amazon state borders (AP: Amapá; PA:
Pará; MA: Maranhao; TO: Tocantins; MG: Mato Grosso; RO: Rondônia; AC: Acre; AM: Amazonas; RR: Roraima).
22
Fig. S4. Temporal trajectories of relative species loss and extinction debt under BAU and GOV scenarios of future forest loss (2009-2050), stratified across the nine different
states of the Brazilian Amazon (top panel) and absolute species loss and extinction debt (95% CIs, with medians in parentheses) across forest-dependent mammals, birds and
amphibians (table below). Relative quantities were calculated as a percentage of initial species richness and averaged across 50 x 50 km2 grid cells. Uncertainties arising from
model parameterisation were incorporated into Monte Carlo simulations, with background colour saturation of plots corresponding to the density of simulation runs.
Fig. S3. Temporal trajectories of relative species loss and extinction debt during the modern-era of deforestation in the Brazilian Amazon (1970-2008), stratified across the
nine different state territories (AP: Amapá; PA: Pará; MA: Maranhao; TO: Tocantins; MG: Mato Grosso; RO: Rondônia; AC: Acre; AM: Amazonas; RR: Roraima) and a
landcover map for 2008 (centre). Relative quantities were calculated as a percentage of initial species richness and averaged across 50 x 50 km2 grid cells. Uncertainties
arising from model parameterisation were incorporated into Monte Carlo simulations, with background colour saturation corresponding to the density of simulation runs.
The apparent disjunction in the 2001-2002 extinction debt for Maranhão is an artefact caused by cloud cover (causing an overestimate of 2001 deforestation and an
underestimate for the years prior).
23
Fig. S4. Temporal trajectories of relative species loss and extinction debt under BAU and GOV scenarios of future forest loss (2009-2050), stratified across the nine different
states of the Brazilian Amazon (top panel) and absolute species loss and extinction debt (95% CIs, with medians in parentheses) across forest-dependent mammals, birds and
amphibians (table below). Relative quantities were calculated as a percentage of initial species richness and averaged across 50 x 50 km2 grid cells. Uncertainties arising from
model parameterisation were incorporated into Monte Carlo simulations, with background colour saturation of plots corresponding to the density of simulation runs.
24
Fig. S5. Hotspots of
extinction debt (top 5%
of grid cells) for forest-
dependent mammals,
birds and amphibians in
the Brazilian Amazon
for 2008 (A) and for
2050 under scenarios of
Business as Usual (B)
and improved forest
frontier governance (C).
Current priority areas
are mostly in regions of
lower species richness,
but newly-paved
highways running
through regions of high
species richness will be
the areas of future
extinction in the
Amazon.
25
Fig. S6. Spatial patterns in historical “forest-dependent” (n = 201) mammal species richness (top), and extinction debt (bottom) in the Brazilian Amazon. Current extinction
debt (left) is highest in Maranhão, especially along the BR-316 highway and its intersection with the BR-222. Under the BAU scenario (right), debt accumulation is highest
along the paved BR-319 and Trans-Amazon west of Itaituba, and also in the species-rich state of Roraima. The extensification of Manaus is a focus of debt across both BAU
(right) and GOV (middle) scenarios.
26
Fig. S7. Spatial patterns in historical “forest-dependent” (n = 329) bird species richness (top), and extinction debt (bottom) in the Brazilian Amazon. Current extinction debt
(left) is highest along the BR-316 in Maranhão and in central Rondônia. Under the BAU scenario (right), extinction debt in 2050 is highest along the two major highways
due to be paved in the central Amazon: the BR-319 and Trans-Amazon (BR-230) west of Itaituba. Under the GOV scenario (middle), extinction debt mostly accumulates
around Manaus, but the intersection of the BR-364 and a newly-paved highway across northern Mato Grosso at Ariquemes also becomes a focus of debt.
27
Fig. S8. Spatial patterns in historical “forest-dependent” (n = 176) amphibian species richness (top), and extinction debt (bottom) in the Brazilian Amazon. Current
extinction debt (left) is highest in the western Brazilian Amazon, especially in central Rondônia and around Rio Branco. Under the BAU scenario (right), Rio Branco
continues to be an important focus of debt but, by 2050, the impacts of paving the western section of the BR-364 have caused widespread debt across Acre. Under the
GOV scenario (middle), Manaus is a centre of debt, but Rio Branco and Ariquemes (at the intersection of the BR-364 and a newly-paved highway joining it to the BR-163)
are also peak areas of debt.
2
References and Notes
1. F. Achard et al., Determination of deforestation rates of the world’s humid tropical forests. Science 297, 999 (2002). doi:10.1126/science.1070656 Medline
2. S. L. Pimm, G. J. Russell, J. L. Gittleman, T. M. Brooks, The future of biodiversity. Science 269, 347 (1995). doi:10.1126/science.269.5222.347 Medline
3. J. M. Diamond, Biogeographic kinetics: Estimation of relaxation times for avifaunas of southwest pacific islands. Proc. Natl. Acad. Sci. U.S.A. 69, 3199 (1972). doi:10.1073/pnas.69.11.3199 Medline
4. T. M. Brooks, S. L. Pimm, J. O. Oyugi, Time lag between deforestation and bird extinction in tropical forest fragments. Conserv. Biol. 13, 1140 (1999). doi:10.1046/j.1523-1739.1999.98341.x
5. D. Tilman, R. M. May, C. L. Lehman, M. A. Nowak, Habitat destruction and the extinction debt. Nature 371, 65 (1994). doi:10.1038/371065a0
6. W. F. Laurance et al., Environment. The future of the Brazilian Amazon. Science 291, 438 (2001). doi:10.1126/science.291.5503.438 Medline
7. R. A. Mittermeier et al., Wilderness and biodiversity conservation. Proc. Natl. Acad. Sci. U.S.A. 100, 10309 (2003). doi:10.1073/pnas.1732458100 Medline
8. B. S. Soares-Filho et al., Modelling conservation in the Amazon basin. Nature 440, 520 (2006). doi:10.1038/nature04389 Medline
9. D. Nepstad et al., Environment. The end of deforestation in the Brazilian Amazon. Science 326, 1350 (2009). doi:10.1126/science.1182108 Medline
10. Materials and methods are available as supplementary materials on Science Online.
11. J. P. Bird et al., Integrating spatially explicit habitat projections into extinction risk assessments: a reassessment of Amazonian avifauna incorporating projected deforestation. Divers. Distrib. 18, 273 (2011). doi:10.1111/j.1472-4642.2011.00843.x
12. M. L. Rosenzweig, Species Diversity in Space and Time (Cambridge Univ. Press, Cambridge, 1995).
13. S. Drakare, J. J. Lennon, H. Hillebrand, The imprint of the geographical, evolutionary and ecological context on species-area relationships. Ecol. Lett. 9, 215 (2006). doi:10.1111/j.1461-0248.2005.00848.x Medline
14. S. L. Pimm, R. A. Askins, Forest losses predict bird extinctions in eastern North America. Proc. Natl. Acad. Sci. U.S.A. 92, 9343 (1995). doi:10.1073/pnas.92.20.9343 Medline
15. T. Brooks, A. Balmford, Atlantic forest extinctions. Nature 380, 115 (1996). doi:10.1038/380115a0
16. G. Ferraz et al., A large-scale deforestation experiment: Effects of patch area and isolation on Amazon birds. Science 315, 238 (2007). doi:10.1126/science.1133097 Medline
3
17. R. H. MacArthur, E. O. Wilson, The Theory of Island Biogeography (Princeton Univ. Press, Princeton, NJ, 1967).
18. World Conservation Union, 2008 IUCN Red List of Threatened Species (IUCN, 2008); www.iucnredlist.org.
19. G. Ceballos, P. R. Ehrlich, Mammal population losses and the extinction crisis. Science 296, 904 (2002). doi:10.1126/science.1069349 Medline
20. Instituto Nacional de Pesquisas Espaciais (INPE), Projeto PRODES: Monitoramento da Floresta Amazônica Brasileira por Satélite (2009); www.obt.inpe.br/prodes/.
21. C. E. V. Grelle, Predicting extinction of mammals in the Brazilian Amazon. Oryx 39, 347 (2005). doi:10.1017/S0030605305000700
22. S. P. Hubbell et al., Colloquium paper: How many tree species are there in the Amazon and how many of them will go extinct? Proc. Natl. Acad. Sci. U.S.A. 105 (suppl. 1), 11498 (2008). doi:10.1073/pnas.0801915105 Medline
23. K. J. Feeley, M. R. Silman, Extinction risks of Amazonian plant species. Proc. Natl. Acad. Sci. U.S.A. 106, 12382 (2009). doi:10.1073/pnas.0900698106 Medline
24. J. P. Metzger et al., Brazilian law: Full speed in reverse? Science 329, 276 (2010). doi:10.1126/science.329.5989.276-b Medline
25. S. N. Hayashi, C. M. Souza Jr., A. Verissimo, “Mato Grosso SAD alerta” (Belém, Brasil, 2011); www.imazon.org.br/publicacoes/transparencia-florestal/sad-alerta/sad-alerta-mato-grosso-abril-2011.
26. R. S. Ridgely et al., Digital Distribution Maps of the Birds of the Western Hemisphere, version 3.0 (NatureServe, Arlington, VA, 2007).
27. C. G. Becker, C. R. Fonseca, C. F. B. Haddad, R. F. Batista, P. I. Prado, Habitat split and the global decline of amphibians. Science 318, 1775 (2007). doi:10.1126/science.1149374 Medline
28. ESRI, ArcGIS, The Complete Geographical Information System (Redlands, CA, 2007).
29. Y. E. Shimabukuro, G. T. Batista, E. M. K. Mello, J. C. Moreira, V. Duarte, Using shade fraction image segmentation to evaluate deforestation in Landsat Thematic Mapper images of the Amazon Region. Int. J. Remote Sens. 19, 535 (1998). doi:10.1080/014311698216152
30. G. P. Asner, Cloud cover in Landsat observations of the Brazilian Amazon. Int. J. Remote Sens. 22, 3855 (2001). doi:10.1080/01431160010006926
31. D. A. Roberts, M. Keller, J. V. Soares, Studies of land-cover, land-use, and biophysical properties of vegetation in the Large Scale Biosphere Atmosphere experiment in Amazônia. Remote Sens. Environ. 87, 377 (2003). doi:10.1016/j.rse.2003.08.012
32. P. M. Fearnside, Deforestation in Brazilian Amazonia: History, rates, and consequences. Conserv. Biol. 19, 680 (2005). doi:10.1111/j.1523-1739.2005.00697.x
4
33. S. J. Wright, H. C. Muller-Landau, The future of tropical forest species1. Biotropica 38, 287 (2006). doi:10.1111/j.1744-7429.2006.00154.x
34. N. Myers, The Sinking Ark: A New Look at the Problem of Disappearing Species (Pergamon, Oxford, 1979).
35. T. C. Whitmore, in Conservation Biology: An Evolutionary-Ecological Perspective, M. E. Soulé, B. A. Wilcox, Eds. (Sinauer, Sunderland, MA, 1980), pp. 303–318.
36. P. R. Ehrlich, A. H. Ehrlich, Healing the Planet (Addison-Wesley, New York, 1991).
37. T. M. Brooks, S. L. Pimm, N. J. Collar, Deforestation predicts the number of threatened birds in insular Southeast Asia. La extension de las deforestaciones predice el numero de aves amenazadas de extincion en las islas del Sureste de Asia. Conserv. Biol. 11, 382 (1997). doi:10.1046/j.1523-1739.1997.95493.x
38. G. Cowlishaw, Predicting the pattern of decline of African primate diversity: An extinction debt from historical deforestation. Conserv. Biol. 13, 1183 (1999). doi:10.1046/j.1523-1739.1999.98433.x
39. C. E. V. Grelle et al., Prediction of threatened tetrapods based on the species–area relationship in Atlantic forest, Brazil. J. Zool. (London) 265, 359 (2005). doi:10.1017/S0952836905006461
40. A. Báldi, J. Vörös, Extinction debt of Hungarian reserves: A historical perspective. Basic Appl. Ecol. 7, 289 (2006). doi:10.1016/j.baae.2005.09.005
41. A. Helm, I. Hanski, M. Pärtel, Slow response of plant species richness to habitat loss and fragmentation. Ecol. Lett. 9, 72 (2006). Medline
42. K. Piessens, M. Hermy, Does the heathland flora in north-western Belgium show an extinction debt? Biol. Conserv. 132, 382 (2006). doi:10.1016/j.biocon.2006.04.032
43. I. Hanski, H. Koivulehto, A. Cameron, P. Rahagalala, Deforestation and apparent extinctions of endemic forest beetles in Madagascar. Biol. Lett. 3, 344 (2007). doi:10.1098/rsbl.2007.0043 Medline
44. E. F. Connor, E. D. McCoy, The statistics and biology of the species-area relationship. Am. Nat. 113, 791 (1979). doi:10.1086/283438
45. M. V. Lomolino, Ecology’s most general, yet protean 1 pattern: The species-area relationship. J. Biogeogr. 27, 17 (2000). doi:10.1046/j.1365-2699.2000.00377.x
46. V. H. Heywood, G. M. Mace, R. M. May, S. N. Stuart, Uncertainties in extinction rates. Nature 368, 105 (1994). doi:10.1038/368105a0
47. S. Budiansky, Extinction or miscalculation? Nature 370, 105 (1994). doi:10.1038/370105a0
48. R. Dial, Extinction or miscalculation? Nature 370, 104 (1994). doi:10.1038/370104b0
49. T. Brooks, J. Tobias, A. Balmford, Deforestation and bird extinctions in the Atlantic forest. Anim. Conserv. 2, 211 (1999). doi:10.1111/j.1469-1795.1999.tb00067.x
5
50. C. E. V. Grelle, G. A. B. Fonseca, M. T. Fonseca, L. P. Costa, The question of scale in threat analysis: A case study with Brazilian mammals. Anim. Conserv. 2, 149 (1999). doi:10.1111/j.1469-1795.1999.tb00060.x
51. B. A. Wilcox, Supersaturated island faunas: A species-age relationship for lizards on post-pleistocene land-bridge islands. Science 199, 996 (1978). doi:10.1126/science.199.4332.996 Medline
52. R. O. Bierregaard, T. E. Lovejoy, Acta Amazon. 19, ••• (1989).
53. D. M. Debinski, R. D. Holt, A survey and overview of habitat fragmentation experiments. Conserv. Biol. 14, 342 (2000). doi:10.1046/j.1523-1739.2000.98081.x
54. G. Ferraz et al., Rates of species loss from Amazonian forest fragments. Proc. Natl. Acad. Sci. U.S.A. 100, 14069 (2003). doi:10.1073/pnas.2336195100 Medline
55. C. Loehle, B.-L. Li, Habitat destruction and the extinction debt revisited. Ecol. Appl. 6, 784 (1996). doi:10.2307/2269483
56. M. E. Soule et al., Reconstructed dynamics of rapid extinctions of Chaparral-Requiring birds in urban habitat islands. Conserv. Biol. 2, 75 (1988). doi:10.1111/j.1523-1739.1988.tb00337.x
57. W. D. Robinson, Long-term changes in the avifauna of Barro Colorado Island, Panama, a tropical forest isolate. Conserv. Biol. 13, 85 (1999). doi:10.1046/j.1523-1739.1999.97492.x
58. K. R. Crooks, A. V. Suarez, D. T. Bolger, M. E. Soulé, Conserv. Biol. 15, 159 (2001).
59. Z. Chocholoušková, P. Pyšek, Changes in composition and structure of urban flora over 120 years: A case study of the city of Plzeň. Flora 198, 366 (2003). doi:10.1078/0367-2530-00109
60. P. C. Stouffer, C. Strong, L. N. Naka, Twenty years of understorey bird extinctions from Amazonian rain forest fragments: Consistent trends and landscape-mediated dynamics. Divers. Distrib. 15, 88 (2009). doi:10.1111/j.1472-4642.2008.00497.x
61. F. L. He, P. Legendre, On species-area relations. Am. Nat. 148, 719 (1996). doi:10.1086/285950
62. E. Tjørve, Shapes and functions of species-area curves: A review of possible models. J. Biogeogr. 30, 827 (2003). doi:10.1046/j.1365-2699.2003.00877.x
63. F. Guilhaumon, O. Gimenez, K. J. Gaston, D. Mouillot, Taxonomic and regional uncertainty in species-area relationships and the identification of richness hotspots. Proc. Natl. Acad. Sci. U.S.A. 105, 15458 (2008). doi:10.1073/pnas.0803610105 Medline
64. M. Williamson, K. J. Gaston, W. M. Lonsdale, The species-area relationship does not have an asymptote! J. Biogeogr. 28, 827 (2001). doi:10.1046/j.1365-2699.2001.00603.x
65. F. He, S. P. Hubbell, Species-area relationships always overestimate extinction rates from habitat loss. Nature 473, 368 (2011). doi:10.1038/nature09985 Medline
6
66. T. M. Brooks, Extinctions: Consider all species. Nature 474, 284 (2011). doi:10.1038/474284b Medline
67. E. W. Seabloom, A. P. Dobson, D. M. Stoms, Extinction rates under nonrandom patterns of habitat loss. Proc. Natl. Acad. Sci. U.S.A. 99, 11229 (2002). doi:10.1073/pnas.162064899 Medline
68. O. Langrand, W. Wilmé, in Natural Change and Human Impact in Madagascar, S. M. Goodman, B. D. Patterson, Eds. (Smithsonian Institution Press, Washington, DC, 1997), pp. 280–305.
69. F. Michalski, C. A. Peres, Disturbance-mediated mammal persistence and abundance-area relationships in Amazonian forest fragments. Conserv. Biol. 21, 1626 (2007). Medline
70. W. D. Newmark, Insularization of Tanzanian parks and the local extinction of large mammals. Conserv. Biol. 10, 1549 (1996). doi:10.1046/j.1523-1739.1996.10061549.x
71. R. E. Ricklefs, I. J. Lovette, The roles of island area per se and habitat diversity in the species-area relationships of four Lesser Antillean faunal groups. J. Anim. Ecol. 68, 1142 (1999). doi:10.1046/j.1365-2656.1999.00358.x
72. B. L. Zimmerman, R. O. Bierregaard, Relevance of the equilibrium theory of island biogeography and species-area relations to conservation with a case from Amazonia. J. Biogeogr. 13, 133 (1986). doi:10.2307/2844988
73. S. L. Pimm, P. Raven, Biodiversity. Extinction by numbers. Nature 403, 843 (2000). doi:10.1038/35002708 Medline
74. R. Dirzo, P. H. Raven, Global state of biodiversity and loss. Annu. Rev. Environ. Resour. 28, 137 (2003). doi:10.1146/annurev.energy.28.050302.105532
75. R. K. Didham, P. M. Hammond, J. H. Lawton, P. Eggleton, N. E. Stork, Beetle species responses to tropical forest fragmentation. Ecol. Monogr. 68, 295 (1998). doi:10.1890/0012-9615(1998)068[0295:BSRTTF]2.0.CO;2
76. D. T. Bolger et al., Response of rodents to habitat fragmentation in coastal southern California. Ecol. Appl. 7, 552 (1997). doi:10.1890/1051-0761(1997)007[0552:RORTHF]2.0.CO;2
77. B. W. Brook, N. S. Sodhi, P. K. L. Ng, Catastrophic extinctions follow deforestation in Singapore. Nature 424, 420 (2003). doi:10.1038/nature01795 Medline
78. J. H. Brown, Mammals on mountaintops: Nonequilibrium insular biogeography. Am. Nat. 105, 467 (1971). doi:10.1086/282738
79. M. Castelletta, N. S. Sodhi, R. Subaraj, Heavy extinctions of forest avifauna in Singapore: Lessons for biodiversity conservation in Southeast Asia. Conserv. Biol. 14, 1870 (2000). doi:10.1046/j.1523-1739.2000.99285.x
80. A. G. Chiarello, Effects of fragmentation of the Atlantic forest on mammal communities in south-eastern Brazil. Biol. Conserv. 89, 71 (1999). doi:10.1016/S0006-3207(98)00130-X
7
81. M. B. Christiansen, E. Pitter, Species loss in a forest bird community near Lagoa Santa in southeastern Brazil. Biol. Conserv. 80, 23 (1997). doi:10.1016/S0006-3207(96)00073-0
82. G. C. Daily, G. Ceballos, J. Pacheco, G. Suzán, A. Sánchez-Azofeifa, Countryside biogeography of neotropical mammals: Conservation opportunities in agricultural landscapes of Costa Rica. Conserv. Biol. 17, 1814 (2003). doi:10.1111/j.1523-1739.2003.00298.x
83. J. M. Diamond, K. D. Bishop, S. V. Balen, Bird survival in an isolated Javan woodland: Island or mirror? Conserv. Biol. 1, 132 (1987). doi:10.1111/j.1523-1739.1987.tb00022.x
84. B. Drayton, R. B. Primack, Plant species lost in an isolated conservation area in metropolitan Boston from 1894 to 1993. Conserv. Biol. 10, 30 (1996). doi:10.1046/j.1523-1739.1996.10010030.x
85. K. B. H. Er, J. L. Innes, K. Martin, B. Klinkenberg, Forest loss with urbanization predicts bird extirpations in Vancouver. Biol. Conserv. 126, 410 (2005). doi:10.1016/j.biocon.2005.06.023
86. F. Escobar, G. Halffter, Á. Solís, V. Halffter, D. Navarrete, Temporal shifts in dung beetle community structure within a protected area of tropical wet forest: A 35-year study and its implications for long-term conservation. J. Appl. Ecol. 45, 1584 (2008). doi:10.1111/j.1365-2664.2008.01551.x
87. S. M. Goodman, D. Rakotondravony, The effects of forest fragmentation and isolation on insectivorous small mammals (Lipotyphla) on the central high plateau of Madagascar. J. Zool. (Lond.) 250, 193 (2000). doi:10.1111/j.1469-7998.2000.tb01069.x
88. J. R. Karr, Avian extinction on Barro Colorado Island, Panama: A reassessment. Am. Nat. 119, 220 (1982). doi:10.1086/283904
89. G. H. Kattan, H. Alvarez-López, M. Giraldo, Forest fragmentation and bird extinctions: San Antonio eighty years later. Conserv. Biol. 8, 138 (1994). doi:10.1046/j.1523-1739.1994.08010138.x
90. T. Larsen, Forest butterflies in West Africa have resisted extinction… so far (Lepidoptera: Papilionoidea and Hesperioidea). Biodivers. Conserv. 17, 2833 (2008). doi:10.1007/s10531-008-9399-z
91. W. F. Laurance, Ecological correlates of extinction proneness in Australian tropical rain forest mammals. Conserv. Biol. 5, 79 (1991). doi:10.1111/j.1523-1739.1991.tb00390.x
92. M. K. Leach, T. J. Givnish, Ecological determinants of species loss in remnant prairies. Science 273, 1555 (1996). doi:10.1126/science.273.5281.1555
93. C. F. Leck, Auk 96, 343 (1979).
94. A. C. Lees, C. A. Peres, Rapid avifaunal collapse along the Amazonian deforestation frontier. Biol. Conserv. 133, 198 (2006). doi:10.1016/j.biocon.2006.06.005
8
95. A. J. Lynam, I. Billick, Differential responses of small mammals to fragmentation in a Thailand tropical forest. Biol. Conserv. 91, 191 (1999). doi:10.1016/S0006-3207(99)00082-8
96. J. MacHunter, W. Wright, R. Loyn, P. Rayment, Bird declines over 22 years in forest remnants in southeastern Australia: Evidence of faunal relaxation? Can. J. For. Res. 36, 2756 (2006). doi:10.1139/x06-159
97. P. Magsalay, T. Brooks, G. Dutson, R. Timmins, Extinction and conservation on Cebu. Nature 373, 294 (1995). doi:10.1038/373294a0
98. W. D. Newmark, A land-bridge island perspective on mammalian extinctions in western North American parks. Nature 325, 430 (1987). doi:10.1038/325430a0 Medline
99. W. D. Newmark, Tropical forest fragmentation and the local extinction of understory birds in the eastern Usambara Mountains, Tanzania. Conserv. Biol. 5, 67 (1991). doi:10.1111/j.1523-1739.1991.tb00389.x
100. W. D. Newmark, Extinction of mammal populations in western North American national parks. Conserv. Biol. 9, 512 (1995). doi:10.1046/j.1523-1739.1995.09030512.x
101. L. M. Renjifo, Composition changes in a subandean avifauna after long-term forest fragmentation. Conserv. Biol. 13, 1124 (1999). doi:10.1046/j.1523-1739.1999.98311.x
102. R. Ribon, J. E. Simon, G. T. de Mattos, Bird extinctions in Atlantic forest fragments of the Viçosa region, Southeastern Brazil. Conserv. Biol. 17, 1827 (2003). doi:10.1111/j.1523-1739.2003.00377.x
103. A. D. Richman, T. J. Case, T. D. Schwaner, Natural and unnatural extinction rates of reptiles on islands. Am. Nat. 131, 611 (1988). doi:10.1086/284810
104. N. S. Sodhi, T. M. Lee, L. Pin Koh, R. R. Dunn, A century of avifaunal turnover in a small tropical rainforest fragment. Anim. Conserv. 8, 217 (2005). doi:10.1017/S1367943005001927
105. N. S. Sodhi, T. M. Lee, L. P. Koh, D. M. Prawiradilaga, Long-term avifaunal impoverishment in an isolated tropical woodlot. Conserv. Biol. 20, 772 (2006). doi:10.1111/j.1523-1739.2006.00363.x Medline
106. J. A. Stratford, P. C. Stouffer, Local extinctions of terrestrial insectivorous birds in a fragmented landscape near Manaus, Brazil. Conserv. Biol. 13, 1416 (1999). doi:10.1046/j.1523-1739.1999.98494.x
107. J. Terborgh, Preservation of natural diversity: The problem of extinction prone species. Bioscience 24, 715 (1974). doi:10.2307/1297090
108. C. R. Trainor, Changes in bird species composition on a remote and well-forested Wallacean Island, South-East Asia. Biol. Conserv. 140, 373 (2007). doi:10.1016/j.biocon.2007.08.022
9
109. I. M. Turner, Species loss in fragments of tropical rain forest: A review of the evidence. J. Appl. Ecol. 33, 200 (1996). doi:10.2307/2404743
110. I. M. Turner, K. S. Chua, J. S. Y. Ong, B. C. Soong, H. T. W. Tan, A century of plant species loss from an isolated fragment of lowland tropical rain forest. Conserv. Biol. 10, 1229 (1996). doi:10.1046/j.1523-1739.1996.10041229.x
111. I. M. Turner et al., A study of plant species extinction in Singapore: Lessons for the conservation of tropical biodiversity. Conserv. Biol. 8, 705 (1994). doi:10.1046/j.1523-1739.1994.08030705.x
112. D. Vallan, Influence of forest fragmentation on amphibian diversity in the nature reserve of Ambohitantely, highland Madagascar. Biol. Conserv. 96, 31 (2000). doi:10.1016/S0006-3207(00)00041-0
113. S. Van der Veken, K. Verheyen, M. Hermy, Plant species loss in an urban area (Turnhout, Belgium) from 1880 to 1999 and its environmental determinants. Flora 199, 516 (2004). doi:10.1078/0367-2530-00180
114. E. O. Willis, Populations and local extinctions of birds on Barro Colorado Island, Panama. Ecol. Monogr. 44, 153 (1974). doi:10.2307/1942309
115. E. O. Willis, E. Eisenmann, A revised list of birds of Barro Colorado Island, Panamá. Smithson. Contrib. Zool. 291, 1 (1979). doi:10.5479/si.00810282.291
116. J. Barlow, L. A. M. Mestre, T. A. Gardner, C. A. Peres, “Diagnóstico socioeconomico e ambiental do Projecto REDD + Amapá - A report to Arvorar Soluções Florestais LTDA” (Belém, Brasil, 2012).
117. J. Barlow, L. A. M. Mestre, T. A. Gardner, C. A. Peres, The value of primary, secondary and plantation forests for Amazonian birds. Biol. Conserv. 136, 212 (2007). doi:10.1016/j.biocon.2006.11.021
118. A. C. Lees, thesis, University of East Anglia, Norwich, UK (2008).