Biological Conservation 143 (2010) 2453–2461

Contents lists available at ScienceDirect

Biological Conservation
journal homepage: www.elsevier.com/locate/biocon

Will future climate change threaten a range restricted endemic species, the
quokka (Setonix brachyurus), in south west Australia?
Lesley Gibson a,*, Asha McNeill b, Paul de Tores a, Adrian Wayne c, Colin Yates b
a
Science Division, Western Australian Department of Environment and Conservation, PO Box 51, Wanneroo, WA 6946, Australia
b
Science Division, Western Australian Department of Environment and Conservation, LMB 104, Bentley Delivery Centre, WA 6983, Australia
c
Science Division, Western Australian Department of Environment and Conservation, Locked Bag 2, Manjimup, WA 6258, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Range-restricted species, such as regional endemics, possess traits that may make them particularly vul-
Received 2 February 2010 nerable to environmental change. The quokka, Setonix brachyurus, is a small macropod, endemic to south-
Received in revised form 31 May 2010 western Australia and two adjacent islands. Climatic factors appear to play a role in defining the distri-
Accepted 9 June 2010
bution of this species. Mainland populations are historically restricted to areas with an annual average
rainfall in excess of 700 mm and their current distribution is almost completely confined within the
1000 mm rainfall isohyet. As such, the predicted increasing aridity of south-western Australia due to cli-
Keywords:
mate change is likely to threaten the continued persistence of the quokka on the mainland. To examine
Climate change
Species distribution model
this possibility, we modelled the distribution of the quokka with Maxent using records of occurrence and
Maxent a combination of historical climate (1961–1990) and habitat variables. Future projections of this distri-
Endemic bution were then examined assuming two simple dispersal scenarios (zero and full migration) and three
Range restricted climate-change scenarios of increasing severity for 2030, 2050 and 2070. The predictive performance of
Marsupial the distribution model generated under historical climate conditions was high (AUC > 0.8), with annual
precipitation contributing the most information to the model. Except for the low-severity climate-change
scenario under the full dispersal assumption, the future projected distribution of quokka was shown to
contract over time. The extent of range contraction tended to increase with the severity of the cli-
mate-change scenario, with the species predicted to lose almost all range by the year 2070 under the
most extreme climate-change scenario. The results indicate the importance of identifying potential ref-
uges for the quokka (i.e. areas where the species is predicted to persist) and defining management strat-
egies to protect these areas from threatening processes.
Ó 2010 Elsevier Ltd. All rights reserved.

1. Introduction
Understanding how species are likely to respond to climate
change is imperative if we are to identify appropriate management
strategies for assisting species to adapt. Climate change can effect a
species’ population and range dynamics in three primary ways: (1)
a species can move in response to changing climate conditions,
tracking appropriate conditions spatially (range shifts, expansions
or contractions); or (2) a species can adapt in situ, either physiolog-
ically, behaviorally or genetically; or (3) failing either of these, as
the living conditions are no longer suitable, the species goes extinct
(Holt, 1990; Martínez-Meyer et al., 2004; Peterson et al., 2005).
Evidence of contractions or shifts in geographic ranges due to re-
cent climate change have been well-documented (e.g. Hickling
et al., 2006; Thomas et al., 2006; Moritz et al., 2008; Rosenzweig
et al., 2008; Thomas, 2010). Models of future distributions under
projected climate-change scenarios, predict dramatic changes for
* Corresponding author. Tel.: +61 8 9405 5152; fax: +61 8 9306 1641.
E-mail address: Lesley.Gibson@dec.wa.gov.au (L. Gibson).
many species (e.g. Williams et al., 2003; Thomas et al., 2004; Beau-
mont et al., 2005).
The ability of a species to withstand and/or recover from an
environmental change (i.e. its resilience) depends on a number of
key life history traits (Isaac et al., 2009). Traits that render a species
particularly vulnerable to environmental perturbations include a
narrow geographic range, limited dispersal capacity, low reproduc-
tive output and a high degree of habitat specialization (Isaac et al.,
2009). Range-restricted species, such as regional endemics, tend to
display most of these traits, and therefore their sensitivity to envi-
ronmental change is likely to be pronounced (Thuiller et al., 2005).
Moreover, contractions in range size of species with already re-
stricted distributions are expected to result in small, highly frag-
mented and isolated populations that are susceptible to losses in
genetic diversity, thereby reducing their capacity to adapt to envi-
ronmental change (Isaac, 2009; Williams et al., 2008).
As a first step in assessing a species’ vulnerability to climate
change, predictions of change in geographic distribution in re-
sponse to alternative scenarios of future climate can be informative
in terms of the range of likely consequences (Araújo et al., 2005;

0006-3207/$ - see front matter Ó 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.biocon.2010.06.011
2454 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461
Heikkinen et al., 2006; Lawler et al., 2006; Yates et al., 2010a).
Species distribution models that statistically relate species geo-
graphic records to historical climate variables, when projected on
changed climate scenarios can provide such predictions, and are
now widely used (reviewed in Elith and Leathwick, 2009).
Here, we use this approach to examine the potential impact of
climate change on the geographical distribution of the quokka
(Setonix brachyurus), a small macropodid marsupial that is endemic
to the south-west of Western Australia (SWWA), including two off-
shore islands (Rottnest and Bald Island) (Fig. 1). This region has a
Mediterranean type climate and supports a large number of ende-
mic species (Hopper et al., 1996; Hopper and Gioia, 2004). Mean
annual temperatures in the SWWA have increased during the
twentieth century and since the 1970’s there has been a significant
decline in autumn and early winter rainfall. The consensus among
Global Climate Models (GCMs) is that temperature will continue to
increase and rainfall will continue to decline in this region (CSIRO,
2007; Bates et al., 2008). Being at the cool, wet end of a hot dry
continent, species occurring in SWWA may be particularly sensi-
tive to climate change, especially because movements to the south
and west are restricted by the coastline, and with the exception of
the Stirling Range which is just 1109 m above sea level, the region
is flat and there is little scope for contraction into cool montane
refuges (Fitzpatrick et al., 2008; Yates et al., 2010a,b).
Climate, particularly rainfall, appears to play a role in defining
the geographic range of the quokka. Historically, mainland quokka
Fig. 1. Location of study region in south west Western Australia, showing the extent of remnant vegetation, the distribution of quokka occurrence records (represented as
circles) and inferred distribution pre-European settlement (within the bold line).
populations were restricted to areas with an annual average rain-
fall of >700 mm (de Tores et al., 2007). Since the 1930s, the species
has declined in its distribution and abundance, and is now almost
completely confined to areas receiving >1000 mm annual rainfall.
Predation by the introduced European red fox (Vulpes vulpes) and
loss of preferred habitat are likely to be the primary causes of this
decline, but the role recent climate change has played is unknown
(de Tores et al., 2007). The quokka is currently listed as vulnerable
in accordance with IUCN Red List Criteria (http://www.iucnredlist.
org/).
In the current study, we examine predicted changes in geo-
graphic range size of the quokka on the mainland in response to
a range of future climate scenarios at 2030, 2050 and 2070. Future
projections are made under the extreme assumptions of no
dispersal and full dispersal (i.e. colonise all locations that are pre-
dicted to become suitable), and taking current land transformation
into account.
2. Method
2.1. Distribution data
The study area was based on the inferred distribution of the
quokka at pre-European settlement in 1829 (i.e. inferred from cur-
rent and historically known location records, as well as subfossil
records – from de Tores et al., 2007) (‘‘study area” in Fig. 1). We ob-
 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461 2455

tained georeferenced quokka occurrences (1905–2007) from a
database held by the Western Australian Department of Environ-
ment and Conservation (DEC), Perth, Australia (see de Tores
et al., 2007). Due to the difficulty of detecting quokkas (Hayward
et al., 2003), absences could not be inferred with any certainty.
Those records for which doubts were raised over their spatial accu-
racy were removed from the dataset prior to analysis. We did not
include island records of quokka occurrence in the analysis as
the resolution of modelled historical and future climate data is
too coarse to differentiate between the islands and surrounding
ocean. An independent dataset based on indirect detection of the
quokka was also made available by DEC and used to evaluate the
model. The latter dataset ranks quokka activity as high, medium
or low based on the occurrence of signs such as characteristic run-
ways through the vegetation, and fresh fecal pellets. Sites of med-
ium to high activity were classified as presence locations for the
purpose of this study.
2.2. Environmental data
Historical climate data (averaging period 1961–1990) were de-
rived from mean minimum (Tmin) and maximum temperature
(Tmax), precipitation (Ptotal) and aerial potential evapo-transpiration
(ET) data layers provided by the Australian Bureau of Meteorology
National Climate Centre. These layers are generated through inter-
polation of mean monthly climate data from climate stations at
0.025° resolution (approximately 2.5  2.5 km in Australia) for
Tmax, Tmin and Ptotal, and 0.1° resolution for ET. Nearest neighbor
resampling was used to align ET to match the other climate layers
at 0.025° resolution. The temperature and precipitation data were
input into Matlab version 6.5.1 to derive 19 bioclimatic variables
using formulas equivalent to those using a monthly time step
ANUCLIM (Houlder et al., 2000).
The indiscriminant use of all available bioclimatic variables in
model-building tends to result in over-fitting and loss off predic-
tive performance of the models (Harrell, 2001), thereby emphasiz-
ing the importance of using expert scientific knowledge in the a
priori selection of variables. Here, experts on quokka ecology were
asked to select and rank the bioclimatic variables they considered
important in terms of defining quokka distribution. Annual precip-
itation (AP), mean maximum temperature of the warmest quarter
(MTHQ) and the sum of ET during the extended summer period
(December to March inclusive, ETDecember–March) were chosen as
the best representative climate predictors. Additionally, a variable
that represented the habitat preference of the quokka for riparian
areas (Hayward et al., 2003) was derived by summing length of
tive GCM resolutions were downscaled by CSIRO using linear inter-
polation to a standard 0.25° resolution to produce patterns of
change for each month and GCM.
To examine the range of uncertainty in future climates, for each
GCM we calculated the average anomalies for monthly Tmin, Tmax
and Ptotal for all combinations of emission scenario, climate sensitiv-
ity and time period for the SWWA. Climate anomalies were visually
compared across combinations and a low, medium and high-impact
scenario was chosen based on the relative severity of the average
predicted anomalies. The three scenarios chosen were: (1) low im-
pact model MIROC-H (Centre for Climate Research, Japan) com-
bined with the B1 emission scenario and low climate sensitivity
(hereafter low-severity), (2) moderate impact model MIROC-M
(Centre for Climate Research, Japan) combined with the A1B emis-
sion scenario and medium climate sensitivity (hereafter medium-
severity), and (3) high impact model CSIRO Mk 3.5 (CSIRO, Austra-
lia) combined with the A1FI emission scenario and high climate
sensitivity (hereafter high-severity) (see Table 1).
For the three climate change severity scenarios, the monthly
Tmin, Tmax, Ptotal, and ET anomalies were converted from 0.25° to
0.025° grid cells using nearest neighbor resampling. The resampled
anomaly grids were added to the 1961–1990 average historical cli-
mate grids to create the Tmin, Tmax, Ptotal, and ET surfaces for the
three scenarios at 2030, 2050 and 2070. These surfaces were then
used to derive annual precipitation, mean maximum temperature
of the warmest quarter, plus ETDecember–March for the three cli-
mate-change scenarios and time periods, using the same methods
described for historical climate data. The three climate-change sce-
narios resulted in increasingly warmer and more arid conditions
across the study area (Table 1, Fig. 2).
2.4. Species distribution modelling
Quokka occurrence records were related to the environmental
variables using Maxent version 3.2.1 (Phillips et al., 2006; Phillips
Table 1
Future climate scenarios to which the quokka distribution was projected.
Low-severity Medium-severity High-severity
Global circulation model MIROC-H MIROC-M CSIRO Mk 3.5
IPCC emission scenario B1 A1B A1FI
Climate sensitivity Low Medium High
Temperature anomalya (°C) 0.8 1.7 4.5
Precipitation anomalya (%) 6 20 32
a
Projected mean change in the mean annual value across the study region by
2070.

major and minor hydrographic features, including swamps, peren-

nial and non-perennial lakes and watercourses (supplied by DEC), 1600 1961-1990
inside each 0.025° grid cell to create a continuous surface 2070 Low
1400 2070 Medium
(Drainagekm). As future changes in this variable were uncertain,
Mean annual rainfall (mm)

2070 High
we assumed it would remain constant over time. Future decreases 1200
in moisture availability may mean that there is less available quo-
kka habitat than predicted by the model. 1000

800
2.3. Future climate scenarios
600
We represented future climate uncertainty using methods rec-
400
ommended by Australia’s leading climate science organisation
(CSIRO, 2007). Monthly Tmin, Tmax, Ptotal, and ET anomalies (i.e. pro- 200
jected rate of change) for three emission scenarios (B1, A1B and
A1FI), three climate sensitivities (low, medium and high) and three 0
time periods (2030, 2050 and 2070) were provided for eight global 10 12 14 16 18 20 22 24
climate models (GCMs) by the Commonwealth Scientific and Mean annual temperature (C°)
Industrial Research Organisation (CSIRO) from their recent regional Fig. 2. Mean annual rainfall versus mean annual temperature at 1000 points across
analysis of climate change in Australia (CSIRO, 2007; http:// the study region for the period 1961–1990 and three climate severity scenarios in
www.climatechangeinaustralia.gov.au/technical_report.php). Na- 2070.
2456 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461
and Dudík, 2008). Maxent estimates the potential geographic dis-
tribution of a species by finding the probability distribution of
maximum entropy, or closest to uniform, subject to constraints de-
rived from the occurrence data. Specifically, these constraints re-
quire the mean of each environmental variable (or functions
thereof) under the predicted distribution to be close (within some
error bounds) to the empirical average over the observed sample. A
specified study area or ‘background points’ represents the space
over which the maximum entropy probability distribution is de-
fined (Phillips et al., 2006). Maxent has been shown to outperform
various methods for modelling presence-only data (Elith et al.,
2006; Hernandez et al., 2006; Pearson et al., 2007; Phillips et al.,
2006) and performs comparatively well against methods using
presence and absence data (Elith et al., 2006).
Two alternative models were initially generated. The first was
trained on 75% (n = 189) randomly assigned occurrences, with
the remaining 25% (n = 63) used as an evaluation (test) data set
(Fig. 3). In the second, all occurrence records (n = 252) were used
for model training and the independent dataset (n = 178) was used
to test the model. This permitted us to examine the reliability of
the independent dataset in terms of possible biases due to incor-
rect assigning of quokka presence (i.e. the observed activity was
due to another species). For both model types, recommended de-
fault values for the regularisation multiplier (1), convergence
threshold (105) and maximum number of iterations (500)
were accepted as per version 3.2.1 of the software (http://
www.cs.princeton.edu/’schapire/maxent). Duplicate samples (i.e.
two or more records within the same pixel) were also removed
a 40%
Change in range size (%)
20%
0%
-20%
-40%
-60%
-80%
-100%
2030 2050
Low
b 40%
Change in range size (%)
20%
0%
-20%
-40%
-60%
-80%
-100%
2030 2050
Low
Fig. 3. Projected changes in range size of quokkas in 2030, 2050 and 2070 (relative to modelled historical range size) under three climate change severity scenarios, two
dispersal assumptions, and (a) without land transformation and (b) taking land transformation into account.
and 10,000 pixels were drawn randomly from the study area as
background samples. The predictive performance of the two mod-
els was evaluated using the area under the receiver operating char-
acteristic curve (ROC); a plot of true positive cases (or sensitivity)
against corresponding false positive cases (or 1 – specificity) across
a range of threshold values (Fielding and Bell, 1997). The area un-
der the curve (AUC) provides a measure of discrimination ability,
and varies from 0.5 for a model with discrimination ability no bet-
ter than random, to 1.0 for a model with perfect discriminatory
ability (Pearce and Ferrier, 2000). Although this evaluation tech-
nique typically requires presence and absence data, it has also been
applied to presence-only data by substituting the absences with
randomly selected pseudo-absences (Graham and Hijmans, 2006;
Hernandez et al., 2006; Lütolf et al., 2006; Phillips et al., 2006).
But here, the AUC is the probability that a model ranks a randomly
chosen presence site above a random background site (Phillips
et al., 2006).
In this instance, both models had good predictive ability, indi-
cating that the distribution of the quokka can be well described
by the variables selected. As the difference between the AUC scores
for the partitioned model and the model tested using independent
data was minor (AUC: 0.80 versus 0.84, respectively), only the lat-
ter model was projected (as it allowed 100% of the occurrence data
to be used) to the time periods 2030, 2050 and 2070, again using
Maxent.
Model results were imported into ArcGIS version 9.2 (ESRI Inc.,
Redlands, CA, USA), where the relative logistic suitability values
(0–1) from Maxent were converted to predicted presence/absence
Full dispersal
No dispersal
2070 2030 2050 2070 2030 2050 2070
Medium High
Full dispersal
No dispersal
2070 2030 2050 2070 2030 2050 2070
Medium High
 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461
(1/0) using the threshold that maximized training sensitivity plus
specificity under historical climate (Liu et al., 2005).
2.5. Incorporating dispersal, land transformation and land tenure
Two simple dispersal scenarios were used to estimate the per-
centage gain or loss of the quokka’s geographic range: full and no
dispersal. The full dispersal scenario assumes a species can track
its shifting climate envelope, and colonise all locations that are
predicted to become suitable. The no dispersal scenario assumes
a species will persist only in areas where the modelled historical
and future geographic ranges overlap. In cases where no overlap
exists, a species is assumed to go extinct.
Land transformation data were derived from the extent of rem-
nant (uncleared) vegetation polygons for the study region (sup-
plied by DEC corporate database). These were derived from the
1995 Landsat TM satellite imagery and corrected using digital aer-
ial photography for the period 1996–1999 (dates limited by avail-
ability of photography). The remnant vegetation dataset was
converted to a 0.025° grid which aligned the modelled historical
and future quokka predicted distributions. Spatial overlap between
this grid and modelled historical and projected quokka distribu-
tions were calculated by a mathematical addition on a cell by cell
basis for each climate scenario and time period, producing cell
counts of historical and future quokka distributions taking land
transformation into account (i.e. the overlap zone).
The percentage change in geographic range size in each time
period, for both dispersal scenarios, and with and without taking
the extent of remnant vegetation into account was calculated as
[(Rfuture/Rhistorical)  1]100, where R represents the geographic
range of the quokka in hectares.
Land tenure data (supplied by DEC) was converted to a 0.025°
grid and reclassified into State Forest, other land managed by
DEC (e.g. conservation reserves), and private land (not cleared of
vegetation). The area of predicted historical and future range sizes
contained within each of these classes, as well as privately owned
cleared land, was calculated using the same method as remnant
vegetation above.
3. Results
The predictive performance of the model derived using histori-
cal climate data was high with an AUC of 0.84 (tested using inde-
pendent data). Determined by an in-built procedure in Maxent, the
percent contributions of the variables in this model ranked in
decreasing importance were: annual precipitation (72%), mean
maximum temperature of the warmest quarter (11%), sum of eva-
po-transpiration from December to March (10%) and length of
drainage in each cell (7%).
With the exception of the low-severity climate scenario, the fu-
ture distribution of the quokka was predicted to contract over time
for both full and no dispersal assumptions (Fig. 3a). Assuming no
dispersal, a range contraction was also observed over time under
the low-severity scenario, but if we assume the species can shift
to occupy all suitable habitat, an initial small expansion in range
size is predicted, though this trend is reversed by 2070, contracting
to its original size (Fig. 3a). The degree of range contraction also
tends to increase with increasing severity of climate-change sce-
nario. Even allowing for dispersal, by 2070 the range size is pro-
jected to decrease by 53% under the medium-severity climate
scenario, and under the most severe climate scenario, the species
is predicted to essentially lose all range (Fig. 3a). The percent
change in range size under the two dispersal assumptions does
not markedly differ within time periods, except perhaps for the
2457
low-severity climate scenario (Fig. 3a). A southward contraction
in distribution is shown in Figs. 4 and 5.
On re-examination of changes in range size excluding areas pre-
dicted as potentially suitable habitat by the model, but lacking in
native vegetation cover (i.e. transformed for other land uses such
as agriculture and urban development), shows almost the same
trends as above in terms of change in range size (Fig. 3b).
A breakdown of the modelled historical distribution into land
tenure categories shows that the majority of the area predicted
as suitable falls within State Forest and other DEC managed land,
such as conservation reserves, with only a relatively small propor-
tion classified as cleared or private land (Fig. 6). Over time, the
largest proportional loss of suitable habitat is predicted to occur
within State Forest, which is consistent across climate scenarios
and dispersal assumptions (Fig. 6). There also appears to be a gen-
eral trend of range contraction into DEC managed land (other than
State Forest) with all range contained in conservation estate by
2070 under the high-severity climate scenario.
4. Discussion
The results presented here predict that the distribution of the
quokka will contract over time, with the degree of range contrac-
tion increasing with severity of climate-change scenario. Even
assuming the quokka can disperse into all available suitable habi-
tat, the species is predicted to lose almost all range by the year
2070 under the most extreme climate-change scenario. Hughes
et al. (2008) reported a similar result for the riverine rabbit (Bunol-
agus monticularis), a restricted endemic species occurring in the
Karoo region of South Africa, with at least 96% of the current hab-
itat predicted to become unsuitable under the most severe climate-
change scenario. Contractions in range size were also predicted for
several regionally endemic vertebrate species in the Wet Tropics
bioregion of Australia (Williams et al., 2003). In response to a
conservative increase in temperature of 1 °C, for 63 out of the
65 species modelled, Williams et al. (2003) predicted significant
declines in range size (average decrease of 63%) with one species
losing all core range. A study examining climate change impacts
on the distributions of European mammals (both endemic and
widespread species), likewise emphasised the potential severity
of climate change for endemic species, where several species were
predicted to lose all range under low and high climate change
severity scenarios (Levinsky et al., 2007).
The modelling approach used here assumes that the quokka is
in equilibrium with its environment and that the variables in-
cluded in the analysis can adequately define its distribution. Of
the variables considered, precipitation was clearly the most impor-
tant in terms of defining the modelled distribution of the quokka,
which supports the view that the distribution of the quokka was
historically limited to relatively high rainfall areas (de Tores
et al., 2007). There is evidence however that the quokka is physio-
logically able to tolerate some level of aridity. The quokka is a rel-
atively efficient water conserver (Nagy et al., 1990) and
populations currently persist in high densities on the seasonally
arid Rottnest Island (Hayward et al., 2003). Although summer mor-
talities do occur during extended dry periods on Rottnest Island
(Kitchener, 1972; de Tores et al., 2007). We postulate that the asso-
ciation and responses of the quokka to climate on the mainland are
likely to be primarily indirect and as a result of interactions with
other factors, principally vulnerability to introduced predators,
which on the mainland is largely a function of ground-level vege-
tation density (Hayward et al., 2005). Vegetation density within
the mainland range of the quokka is mainly a function of soil and
site moisture, which is mostly a result of climate and topography
(Beard, 1990). Increasing aridity is likely to reduce the density of
the vegetation in current key remnant habitats of the quokka to
2458 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461
Low
M e d iu m
H ig h H is to ric a l 203 0
Fig. 4. Predicted potential distribution on cleared land (light grey) and remnant vegetation (black) under historical conditions, and low, medium and high-severity climate
scenarios assuming full dispersal.
the extent that it no longer affords them protection from predators.
Hughes et al. (2008) implied a similar case for the riverine rabbit, a
species also occupying dense riparian vegetation.
Local quokka populations in the north of their range are small
and widely scattered, occurring in discrete patches of suitable hab-
itat (Hayward et al., 2003), and dispersal between patches appears
to be lacking (Hayward et al., 2004). Hayward et al. (2004) suggest
that predation by the introduced red fox on quokkas is suppressing
population booms on the mainland which would otherwise drive
dispersal. This apparent limited dispersal capability of the quokka
suggests that of the two dispersal options modelled here, the
assumption of no dispersal is better supported. However, the pres-
ence or absence of dispersal in the modelling process did not
appreciably alter the outcome in terms of changes in range size.
This suggests the ability of the quokka to disperse may not be
important in the context of responses to climate change. The small
difference between the dispersal assumption outcomes is probably
explained by the predicted contractions in range size, rather than
expansions or shifts to new areas. The pattern observed for the
low-severity climate scenario was an exception, with small expan-
sions (but less than 11%) observed in both 2030 and 2050 (negligi-
ble difference by 2070) assuming full dispersal, but range
contractions assuming no dispersal.
Historically, quokka declines have been linked to loss of habitat
due to the transformation of land for urban development and agri-
culture (de Tores et al., 2007). The modelled historical distribution
of the quokka largely overlaps areas of intact vegetation (Figs. 4
2050 207 0
and 5), with only 17% of the total area predicted as suitable already
cleared of vegetation (Fig. 6). This area predominantly occurs on
the western-most margin of the modelled distribution, an area
where the quokka is known from only one isolated location (Sin-
clair and Hyder, 2009). Consideration of land transformation, by
recalculating range size minus the area cleared of vegetation, re-
sulted in only minor differences to those of the original predictions.
This again is probably because the distribution is almost always
predicted to contract, and within this range, vegetation cover is rel-
atively intact. Here, we only considered land transformation up to
1999, while the quokka occurrence records extend up to 2007. Re-
cent and future transformations of land, may mean our predictions
of range contraction are conservative, and less suitable habitat is or
will be available than predicted under historical and future cli-
mates. Also, our analysis does not consider the influence of distur-
bance regimes such as fire and biotic factors such as predation, and
the interaction of climate change with these. The increasing inten-
sity, frequency and extent of wild fires that are a potential conse-
quence of climate change (Steffen et al., 2009), are likely to
further reduce the amount of quokka habitat available.
Another source of uncertainty unaccounted for in the analysis,
is the choice of a single modelling algorithm. Different algorithms
can result in different projections for a number of reasons (see Elith
and Graham, 2009). Exploration of these differences can provide
insights on model uncertainty (Araújo and New, 2006). However,
deciding which model best approximates reality is problematic
(Elith and Graham, 2009), especially when projecting future distri-
 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461
Historical 2030
Low
Medium
High
Fig. 5. Predicted potential distribution on cleared land (light grey) and remnant vegetation (black) under historical conditions, and low, medium and high-severity climate
scenarios assuming no dispersal.
butions of species in response to climate change. In this case, the
availability of presence-only data restricted the choice of algo-
rithm, and as the model performance of Maxent has been shown
to be superior to other presence-only approaches (Elith et al.,
2006; Hernandez et al., 2006; Pearson et al., 2007; Phillips et al.,
2006), it was an obvious choice. Using this single method also al-
lowed a more straightforward examination of uncertainty due to
the selection of climate-change scenario by examining the range
of projections across predicted severity of change.
Conservation strategies aiming to maximise habitat connectiv-
ity by protecting and enhancing dispersal corridors may provide
little benefit to the quokka given that it is unlikely to migrate by
natural dispersal. More benefit is likely to be gained by concentrat-
ing effort on identifying and protecting refuges, such as ‘core’ hab-
itat where quokkas are predicted to persist over time. The trend of
range contraction into land managed by DEC (i.e. largely conserva-
tion estate), provides some flexibility in terms of developing man-
agement strategies to protect the habitat in these core refugial
areas. Additionally, mitigation of other key threatening processes,
such as the impacts of predators, may significantly help to increase
the resilience of the quokka to climate change, not only by reduc-
ing the direct effect of predation on them, but also by removing the
potential barrier to dispersal that predators seem to confer (Hay-
ward et al., 2004).
Assessing a species’ vulnerability to the effects of climate
change is a necessary first step in developing management strate-
gies for enhancing adaptation and resilience (Williams et al., 2008).
Restricted range, endemic species are particularly susceptible to
2459
2050 2070
climate change and should be allocated high priority in regional
assessments of vulnerability and conservation planning in re-
sponse to climate change. In the SWWA, further examples deserv-
ing such attention include the western ringtail possum
(Pseudocheirus occidentalis), western false pipistrelle (Falsistrellus
mckenziei) and the honey possum (Tarsipes rostratus), although
the latter species will probably be less effected due to its wider dis-
tribution. The species distribution modelling approach used here
gives a useful first approximation of the potential response of the
quokka to a range of climate-change scenarios. As global green-
house gas emissions have been rising faster than projected, aver-
age global temperature increases of 4 °C above pre-industrial
levels are now anticipated (Anderson and Bows, 2008), and this
corresponds with the high-severity climate-change scenario used
in the current study. Under this scenario the quokka is predicted
to lose essentially all range by 2070. Therefore the risk of cli-
mate-driven extinction of the quokka on the mainland is predicted
to be highly likely, particularly if there is no effective mitigation. A
refinement of the analytical approach used here will require the
incorporation of population dynamics (e.g. dispersal mechanism),
biotic interactions (e.g. predation) and physiological tolerances
(e.g. energy and water balance) in a mechanistic modelling frame-
work (e.g. Keith et al., 2008; Anderson et al., 2009; Kearney and
Porter, 2009; Zurell et al., 2009; Huntley et al., 2010) to more real-
istically predict the response of the quokka to climate change and
determine the relative importance of other factors. To date the
parameters required to feed into this modelling approach are lack-
ing or limited. Addressing these limitations should be the aim of
2460 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461

a 1,000,000
Cleared land
Predicted range size (hectares) Private vegetation
800,000 State Forest

Conservation reserve

600,000

400,000

200,000

0
Historical 2030 2050 2070 Historical 2030 2050 2070 Historical 2030 2050 2070
Low Medium High

b 1,000,000
Cleared land
Predicted range size (hectares)

Private vegetation
800,000
State Forest
Conservation reserve
600,000

400,000

200,000

0
Historical 2030 2050 2070 Historical 2030 2050 2070 Historical 2030 2050 2070

Low Medium High

Fig. 6. Breakdown of land ownership of predicted potential range under historical conditions, and low, medium and high-severity climate scenarios, and assuming (a) full and
(b) no dispersal.

future studies. Additionally, the spatial resolution of the climate
data used here is more appropriate for the development of regional
management strategies. For application at local scales, refinement
of the spatial resolution is required.
Acknowledgments
We thank Graeme Liddlelow (DEC) for supplying the indepen-
dent data set and Karlene Bain (DEC) for providing advice regard-
ing quokka ecology. We are grateful to Linda Beaumont and
Andres Roubicek at Macquarie University for providing us with
the Matlab scripts for calculating bioclimatic variables. The study
was supported by the Western Australian Department of Environ-
ment and Conservation. Insightful comments from three anony-
mous reviewers improved the manuscript.
References
Anderson, K., Bows, A., 2008. Reframing the climate change challenge in light of
post-2000 emission trends. Philosophical Transactions of the Royal Society A
366, 3863–3882.
Anderson, B.J., Akçakaya, H.R., Araújo, M.B., Fordham, D.A., Martínez-Meyer, E.,
Thuiller, W., Brook, B.W., 2009. Dynamics of range margins for meta-
populations under climate change. Proceedings of the Royal Society B 276,
1415–1420.
Araújo, M.B., New, M., 2006. Ensemble forecasting of species distributions. Trends in
Ecology and Evolution 22, 42–47.
Araújo, M.B., Whittaker, R.J., Ladle, R.J., Erhard, M., 2005. Reducing uncertainty in
projections of extinction risk from climate change. Global Ecology and
Biogeography 14, 529–538.
Bates, B.C., Hope, P., Ryan, B., Smith, I., Charles, S., 2008. Key findings from the Indian
Ocean Climate Initiative and their impact on policy development in Australia.
Climatic Change 89, 339–354.
Beard, J.S., 1990. Plant Life of Western Australia. Kangaroo Press, Australia.
Beaumont, L.J., Hughes, L., Poulsen, M., 2005. Predicting species distributions:
use of climate parameters in BIOCLIM and its impact on predictions of
species’ current and future distributions. Ecological Modelling 186, 250–
269.
CSIRO, 2007. Climate Change in Australia – Technical Report 2007. CSIRO, Bureau of
Meteorology, Melbourne, Australia.
de Tores, P.J., Hayward, M.W., Dillon, M.J., Brazell, R.I., 2007. Review of the
distribution, causes for the decline and recommendations for management of
the quokka, Setonix brachyurus (Macropodidae: Marsupialia). An endemic
macropodid marsupial from south-west Western Australia. Conservation
Science Western Australia 6, 13–73.
Elith, J., Graham, C., 2009. Do they? How do they? Why do they differ?–on finding
reasons for differing performances of species distribution models. Ecography
32, 66–77.
 L. Gibson et al. / Biological Conservation 143 (2010) 2453–2461
Elith, J., Leathwick, J.R., 2009. Species distribution models: ecological explanation
and prediction across space and time. Annual Review of Ecology, Evolution and
Systematics 40, 677–697.
Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J.,
Huettman, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A.,
Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T.,
Philips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J.,
Williams, S., Wisz, M.S., Zimmermann, N.E., 2006. Novel methods improve
prediction of species’ distributions from occurrence data. Ecography 29, 129–
151.
Fielding, A.H., Bell, J.F., 1997. A review of methods for the assessment of prediction
errors in conservation presence/absence models. Environmental Conservation
24, 38–49.
Fitzpatrick, M.C., Gove, A.D., Sanders, N.J., Dunn, R.R., 2008. Climate change, plant
migration, and range collapse in a global biodiversity hotspot: the Banksia
(Proteaceae) of Western Australia. Global Change Biology 14, 1337–1352.
Graham, C.H., Hijmans, R.J., 2006. A comparison of methods for mapping species
ranges and species richness. Global Ecology and Biogeography 15, 578–587.
Harrell, F.E., 2001. Regression modelling strategies with applications to linear
models, logistic regression and survival analysis. Springer Series in Statistics.
Springer-Verlag, New York.
Hayward, M.W., de Tores, P.J., Dillon, M.J., Fox, B.J., 2003. Local population structure
of a naturally occurring metapopulation of the quokka (Setonix brachyurus
Macropodidae: Marsupialia). Biological Conservation 110, 343–355.
Hayward, M.W., de Tores, P.J., Augee, M.L., Fox, B.J., Banks, P.B., 2004. Home range
movements of the quokka Setonix brachyurus (Macropodidae: Marsupialia), and
its impacts on the viability of the metapopulation on the Australian mainland.
Journal of Zoology, London 263, 219–228.
Hayward, M.W., de Tores, P.J., Banks, P.B., 2005. Habitat use of the quokka, Setonix
brachyurus (Macropodidae: Marsupialia) in the northern jarrah forest of
Australia. Journal of Mammalogy 86, 683–688.
Heikkinen, R.K., Luoto, M., Araujo, M.B., Virkkala, R., Thuiller, W., Sykes, M.T., 2006.
Methods and uncertainties in bioclimatic envelope modelling under climate
change. Progress in Physical Geography 30, 751–777.
Hernandez, P.C., Graham, C., Master, L., Albert, D., 2006. The effect of sample size
and species characteristics on performance of different species distribution
modeling methods. Ecography 29, 773–785.
Hickling, R., Roy, D.B., Hill, J.K., Fox, R., Thomas, C.D., 2006. The distributions of a
wide range of taxonomic groups are expanding polewards. Global Change
Biology 12, 450–455.
Holt, R.D., 1990. The microevolutionary consequences of climate change. Trends in
Ecology and Evolution 5, 311–315.
Hopper, S.D., Gioia, P., 2004. The Southwest Australian Florisitic Region: evolution
and conservation of a global hotspot of biodiversity. Annual Review of Ecology
and Systematics 35, 623–650.
Hopper, S.D., Harvey, M.S., Chappill, J.A., Main, A.R., Main, B.Y., 1996. The Western
Australian biota as Gondwanan Heritage – a review. In: Hopper, S.D., Chappill,
J.A., Harvey, M.S., Georeg, A.S. (Eds.), Gondwanan Heritage: Past Present and
Future of the Western Australian Biota. Surrey Beatty and Sons, Chipping
Norton, NSW, pp. 1–46.
Houlder, D., Hutchinson, M., Nix, H., McMahon, J., 2000. ANUCLIM User Guide,
http://fennerschool.anu.edu.au/publications/software/anuclim.php. Centre for
Resource and Environmental Studies, Australian National University, Canberra.
Hughes, G.O., Thuiller, W., Midgley, G.F., Collins, K., 2008. Environmental change
hastens the demise of the critically endangered riverine rabbit (Bunolagus
monticularis). Biological Conservation 141, 23–34.
Huntley, B., Barnard, P., Altwegg, R., Chambers, L., Coetzee, B.W.T., Gibson, L.,
Hockey, P.A.R., Hole, D.G., Midgley, G.F., Underhill, L.G., Willis, S.G., 2010.
Beyond bioclimatic envelopes: dynamic species’ range and abundance
modelling in the context of climate change. Ecography 33, 1–6.
Isaac, J.L., 2009. Effects of climate change on life history: implications for extinction
risk in mammals. Endangered Species Research 7, 115–123.
Isaac, J.L., Vanderwal, J.J., Johnson, C.N., Williams, S.E., 2009. Resistance and
resilience. quantifying relative extinction risk in a diverse assemblage of
tropical mammals. Diversity and Distributions 15, 280–288.
Kearney, M., Porter, W., 2009. Mechanistic niche modeling: combining physiological
and spatial data to predict species’ ranges. Ecology Letters 12, 334–350.
Keith, D.A., Akçakaya, H.R., Thuiller, W., Midgley, G.F., Pearson, R.G., Phillips, S.J.,
Regan, H.M., Araujo, M.B., Rebelo, T.G., 2008. Predicting extinction risks under
climate change: coupling stochastic population models with dynamic
bioclimatic habitat models. Biology Letters 4, 560–563.
Kitchener, D.J., 1972. The importance of shelter to the quokka, Setonix brachyurus
(Marsupialia), on Rottnest Island. Australian Journal of Zoology 20, 281–299.
2461
Lawler, J.J., White, D., Neilson, R.P., Blaustein, A.R., 2006. Predicting climate-induced
range shifts: model differences and model reliability. Global Change Biology 12,
1568–1584.
Levinsky, I., Skov, F., Svenning, J.C., Rahbek, C., 2007. Potential impacts of climate
change on the distributions and diversity patterns of European mammals.
Biodiversity and Conservation 16, 3803–3816.
Liu, L., Berry, P.M., Dawson, T.P., Pearson, R.G., 2005. Selecting thresholds of
occurrence in the prediction of species distributions. Ecography 28, 385–393.
Lütolf, M., Kienast, F., Guisan, A., 2006. The ghost of past species occurrences:
improving species distribution models for presence-only data. Journal of
Applied Ecology 43, 802–815.
Martínez-Meyer, E., Peterson, A.T., Hargrove, W.W., 2004. Ecological niches as stable
distributional constraints on mammal species, with implications for Pleistocene
extinctions and climate change projections for biodiversity. Global Ecology and
Biogeography 13, 305–314.
Moritz, C., Patton, J.L., Conroy, C.J., Parra, J.L., White, G.C., Beissinger, S.R., 2008.
Impact of a century of climate change on small-mammal communities in
Yosemite national Park, USA. Science 322, 261–264.
Nagy, K.A., Bradley, A.J., Morris, K.D., 1990. Field metabolic rates, water fluxes and
feeding rates of quokkas, Setonix brachyurus and tammars, Macropus eugenii, in
Western Australia. Australian Journal of Zoology 37, 553–560.
Pearce, J., Ferrier, S., 2000. Evaluating the predictive performance of habitat models
developed using logistic regression. Ecological Modelling 133, 225–245.
Pearson, R.G., Raxworthy, C.J., Nakamura, M., Peterson, A.T., 2007. Predicting species
distributions from small numbers of occurrence records: a test case using
cryptic geckos in Madagascar. Journal of Biogeography 34, 102–117.
Peterson, A.T., Tian, H., Martínez-Meyer, E., Soberó, J., Sánchez-Cordero, V., Huntley,
B., 2005. Modeling distributional shifts of individual species and biomes. In:
Lovejoy, T.E., Hannah, L. (Eds.), Climate Change and Biodiversity. Yale University
Press, New Haven and London, pp. 211–228.
Phillips, S.J., Dudík, M., 2008. Modeling of species distributions with Maxent: new
extensions and a comprehensive evaluation. Ecography 31, 161–175.
Phillips, S.J., Anderson, R.P., Schapire, R.E., 2006. Maximum entropy modeling of
species geographic distributions. Ecological Modeling 190, 231–259.
Rosenzweig, C., Karoly, D., Vicarelli, M., Neofotis, P., Wu, Q., Casassa, G., Menzel, A.,
Root, T.L., Estrella, N., Seguin, B., Tryjanowski, P., Liu, C., Rawlins, S., Imeson, A.,
2008. Attributing physical and biological impacts to anthropogenic climate
change. Nature 453, 353–357.
Sinclair, E.A., Hyder, B.M., 2009. Surviving quokka (Setonix brachyurus) population
on the Swan Coastal Plain, Western Australia. Australian Mammalogy 31, 67–
69.
Steffen, W., Burbidge, A.A., Hughes, L., Kitchling, R., Lindenmayer, D., Musgrave, W.,
Stafford Smith, M., Werner, P.A., 2009. Australia’s Biodiversity and Climate
Change. Technical Synthesis. Department of Climate Change, Commonwealth of
Australia.
Thomas, C.D., 2010. Climate, climate change and range boundaries. Diversity and
Distributions 16, 488–495.
Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M., Beaumont, L.M., Collingham,
Y.C., Erasmus, B.F.N., Ferreira de Siqueira, M., Grainger, A., Hannah, L., Hughes,
L., Huntley, B., van Jaarsveld, A.S., Midgley, G.F., Miles, L., Ortega-Huerta, M.A.,
Peterson, A.T., Philips, O.L., Williams, S.E., 2004. Extinction risk from climate
change. Nature 427, 145–148.
Thomas, C.D., Franco, A.M.A., Hill, J.K., 2006. Range retractions and extinctions in the
face of climate warming. Trends in Ecology and Evolution 21, 415–416.
Thuiller, W., Lavorel, S., Araújo, M.B., 2005. Niche properties and geographical
extent as predictors of species sensitivity to climate change. Global Ecology and
Biogeography 14, 347–357.
Williams, S.E., Bolitho, E.E., Fox, S., 2003. Climate change in Australian tropical
rainforests: an impending environmental catastrophe. Proceedings of the Royal
Society of London B 270, 1887–1892.
Williams, S.E., Shoo, L.P., Isaac, J.L., Hoffmann, A.A., Langham, G., 2008. Towards an
integrated framework for assessing the vulnerability of species to climate
change. PLoS Biology 12, 2621–2626.
Yates, C.J., Elith, J., Latimer, A., Le Maitre, D.C., Midgley, G.F., Schurr, F.M., West, A.,
2010a. Projecting climate change impacts on species distributions in
megadiverse South African Cape and Southwest Australian Floristic Regions –
opportunities and challenges. Austral Ecology 35, 374–391.
Yates, C.J., McNeill, A., Elith, J., Midgley, G.F., 2010b. Assessing the impacts of climate
change and land transformation on Banksia in the South West Australian
Floristic Region. Diversity and Distributions 16, 187–201.
Zurell, D., Jeltsch, F., Dormann, C.F., Schröder, B., 2009. Static species distribution
models in dynamically changing systems: how good can predictions really be?
Ecography 32, 733–744.