Linking habitat use to range expansion rates in fragmented
landscapes: a metapopulation approach
Robert J. Wilson, Zoe G. Davies and Chris D. Thomas
R. J. Wilson (R.J.Wilson@exeter.ac.uk), Centre for Ecology and Conservation, Univ. of Exeter, Cornwall Campus, Penryn TR10 9EZ, UK.
? Z. G. Davies, Dept of Animal and Plant Sciences, Univ. of Sheffield, Sheffield S10 2TN, UK. ? C. D. Thomas, Dept of Biology, Univ. of
York, York YO10 5YW, UK.
Temperature increases because of climate change are expected to cause expansions at the high latitude margins of species
distributions, but, in practice, fragmented landscapes act as barriers to colonization for most species. Understanding how
species distributions will shift in response to climate change therefore requires techniques that incorporate the combined
effects of climate and landscape-scale habitat availability on colonization rates. We use a metapopulation model
(Incidence Function Model, IFM) to test effects of fine-scale habitat use on patterns and rates of range expansion by the
butterfly Hesperia comma. At its northern range margin in Britain, this species has increased its breadth of microhabitat
use because of climate warming, leading to increased colonization rates. We validated the IFM by reconstructing
expansions in five habitat networks between 1982 and 2000, before using it to predict metapopulation dynamics over
100 yr, for three scenarios based on observed changes to habitat use. We define the scenarios as ‘‘cold-world’’ (only hot,
south-facing 150?2508 hillsides are deemed warm enough), ‘‘warm-world’’ in which 100?3008 hillsides can be
populated, and ‘‘hot-world’’, where the background climate is warm enough to enable use of all aspects (as increasingly
observed). In the simulations, increased habitat availability in the hot-world scenario led to faster range expansion rates,
and to long-term differences in distribution size and pattern. Thus, fine-scale changes in the distribution of suitable
microclimates led to landscape-scale changes in population size and colonization rate, resulting in coarse-scale changes to
the species distribution. Despite use of a wider range of habitats associated with climate change, H. comma is still expected
to occupy a small fraction of available habitat in 100 yr. The research shows that metapopulation models represent a
potential framework to identify barriers to range expansion, and to predict the effects of environmental change or
conservation interventions on species distributions and persistence.
The capacity of species to survive climate change will
depend on their ability either to adapt in situ to changing
conditions, or to colonize regions that become suitable
outside their current geographic range. Some bioclimate
models predict that future climatically-suitable locations
may show little or no overlap with the current ranges of
many species (Thomas et al. 2004, Fitzpatrick et al. 2008).
Despite reservations regarding the accuracy of such models
(Beale et al. 2008, Trivedi et al. 2008, Kriticos and Leriche
pers. comm.), a wide range of models (Arau ´jo et al. 2005,
Elith et al. 2006) and empirically observed species range
shifts (Parmesan 2006) suggest that the conservation of
many species will depend on their ability to cross landscapes
that have been heavily modified by human activities.
Current evidence suggests that a number of habitat general-
ist or dispersive species have been able to expand their
distributions polewards associated with climate change
(Warren et al. 2001), but that most species (particularly
habitat specialists) are failing to colonize climatically
suitable regions as they become available (Mene ´ndez et al.
2006). Indeed, the distributions of many taxa appear not
to have fully occupied available climate space during the
present interglacial (Arau ´jo and Pearson 2005, Svenning
and Skov 2007, Arau ´jo et al. 2008, Svenning et al. 2008).
To aid the conservation of such species in a changing
climate, methods are required not only to predict the
distribution of future suitable locations (Arau ´jo and New
2007, Beaumont et al. 2007), but also the species traits
(Ward and Masters 2007, Massot et al. 2008) and
landscape or habitat features that will facilitate species
range expansions (Hill et al. 2001, Allouche et al. 2008,
Vos et al. 2008).
One important consequence of climate change is that
species’ regional habitat use may change (Davies et al.
2006). Many species are restricted to hot microhabitats at
their high latitude range margins, but occupy a broader
spectrum of microhabitats nearer the centres of their ranges
(Thomas 1993, Thomas et al. 1999). For such species,
climate change might be expected to increase the breadth of
habitat requirements towards high latitudes, hence increas-
ing habitat availability, population sizes and colonization
rates (Thomas et al. 2001). The important question for
Ecography 33: 73?82, 2010
# 2010 The Authors. Journal compilation # 2010 Ecography
Subject Editors: Nu ´ria Roura-Pascual and Nathan Sanders. Accepted 10 September 2009
S P E C I A L
I S S U E
conservationists is whether the increases in habitat avail-
ability owing to broader habitat use will be sufficient to
allow species to expand their distributions through frag-
In this paper we use a metapopulation model to test how
changes to habitat use because of climate change affect a
species’ distribution at its upper latitude margin. In Britain,
the silver-spotted skipper butterfly Hesperia comma is
restricted to hot microhabitats in calcareous grassland
(Thomas et al. 1986), but its breadth of habitat use has
recently expanded due to changed environmental condi-
tions (warmer summers; Davies et al. 2006). We show how
these changes to habitat use increase habitat availability,
before using the Incidence Function Model (Hanski 1994,
1999) to simulate how increased habitat area and con-
nectivity lead to faster rates of range expansion, and to long-
term differences in distribution size and pattern. In this
system, colonization and extinction dynamics provide a
mechanism linking fine-scale changes in habitat use to
regional-scale changes in species distributions.
Materials and methods
In Britain, the silver-spotted skipper butterfly Hesperia
comma is restricted to short-turfed chalk grassland, where it
lays eggs exclusively on short tufts (B10 cm) of sheep’s
fescue grass Festuca ovina (Thomas et al. 1986). During the
twentieth century, agricultural intensification, the abandon-
ment of low intensity livestock grazing, and the virtual
elimination of rabbits because of myxomatosis led to a
dramatic reduction in habitat availability for the species.
When a full survey of the British distribution of H. comma
was undertaken in 1982, it had declined to fewer than 70
populations (Thomas et al. 1986). The vast majority of
these refuge populations were in five habitat networks in
south-east England (Surrey, Sussex, Chilterns, Kent, Hamp-
shire; Fig. 1), and on south-facing slopes (90% on aspects of
100?3008; 75% on aspects of 150?2508). At the time, the
species appeared to select the hottest microhabitats for
oviposition, with an estimated optimum sward composition
including 41% bare ground (Thomas et al. 1986). In 1991,
repeat distribution surveys in Surrey and Sussex showed that
a range expansion had begun in Sussex, and that the
maximum colonization event was 8.65 km from the nearest
1982 population (Thomas and Jones 1993).
We carried out a repeat survey of the distribution of
H. comma in 2000, searching for adults and eggs of the
species in all areas of chalk grassland containing suitable
F. ovina plants within 30 km of the 1982 H. comma
distribution, and within 10?15 km of newly-colonized sites
found in 2000. Such a comprehensive search was feasible
because suitable habitat occurs only on unimproved
calcareous grassland, which is restricted to localized chalk
escarpments in south-east England (Fig. 1). We defined
habitat patches as areas of suitable grassland bounded by
continuous woodland or scrub barriers, or by at least 25 m
of unsuitable grassland (Thomas and Jones 1993, Hill et al.
1996), and recorded the area, aspect (8) and vegetation
composition of each habitat patch, and of subdivisions of
habitat patches where contiguous habitats or aspects varied
markedly (Davies et al. 2005). Since 1982, livestock grazing
has been reintroduced to many areas of chalk grassland as
part of conservation programmes, and rabbit numbers have
increased (Davies et al. 2005). In addition, the temperature
during H. comma’s flight period increased by 2.88C over
20 yr as a result of direct temperature changes and
phenological advancement of the butterfly’s flight period
(Wilson et al. 2007). At higher ambient temperatures,
H. comma oviposition rate increases, and oviposition sites
are less restricted to hot microclimates (Davies et al. 2006).
As a result, optimum sward composition for H. comma egg-
laying in 2001 and 2002 included a reduced cover of bare
ground (21%), and by 2000 the species had colonized
north-, east- and west-facing aspects, where bare ground
cover and ambient temperature are typically lower than on
south-facing slopes (Thomas et al. 2001, Davies et al.
2006). In conjunction with these changes, the species
expanded its distribution in Britain, colonizing 179 habitat
patches between 1982 and 2000 (Davies et al. 2005).
Nevertheless, large areas of suitable habitat remained
unoccupied (Fig. 1). In 2002, intensive distribution surveys
were again repeated in the Surrey and Sussex networks,
showing a limited number of colonization events since
The metapopulation model
The Incidence Function Model (IFM) (Hanski 1994) is a
stochastic patch occupancy model based on the assumptions
that 1) extinction risk is inversely related to patch area or
population size, and 2) colonization probability is positively
related to patch connectivity, where connectivity is a
function of the distance to other occupied patches and
their size. For each patch i, annual extinction risk (Ei) is
related to patch area (Ai) by the term Ei?(e/ /Ax
where e and x are species-specific parameters relating
extinction risk to area, and (1?Ci) takes account of
the rescue effect by reducing extinction risk for highly
Figure 1. The distribution of Hesperia comma in 1982 and 2000
in south-east England. Symbols show occupancy status of habitat
patches in five networks: survived (black, occupied 1982 and
2000); colonized (grey, absent 1982, occupied 2000); extinct (grey
triangles, occupied 1982, absent 2000); vacant (white, absent 1982
and 2000). Grey squares show populations introduced between
1982 and 2000. Symbol size exaggerates patch area: small
B0.1 ha, medium 0.1?1 ha, large ?1 ha. Line indicates the coast.
S P E C I A L
I S S U E
connected patches. Annual colonization probability (Ci) is
related to connectivity (Si) by the term Ci? /S2
where y is a species-specific parameter relating colonization
rate to connectivity. Connectivity is estimated as Si?a exp
a negative exponential dispersal kernel, with the proportion
of per generation dispersal over distance d km or greater
corresponding to exp?ad; dijis the distance to patch i from
each other patch j; Ajis the area of each patch j; and b scales
emigration rate to the area of patch j. In this study b was set
to 0.5 to account for the tendency of per capita emigration
to be greater from smaller habitat patches in this and other
species (Hill et al. 1996, Moilanen and Nieminen 2002).
The IFM carries out maximum likelihood estimation of
parameters e, x, y and a for a species based on snapshots of
occupied and vacant habitat patches (Moilanen 1999,
2000). The model was developed to estimate parameters
from steady state metapopulations, where observed occu-
pancy is considered to result from approximately equal
colonization and extinction rates, although parameters may
also be estimated from population turnover without
assuming equilibrium dynamics (Moilanen 2000). The
estimated parameters can then be used to simulate dynamics
for the species in the same or other, potentially non-
equilibrium metapopulations, which may be valuable for
predicting changes to species distributions following envir-
onmental change (Thomas and Hanski 2004).
A metapopulation model is an appropriate framework
for modelling H. comma’s dynamics in Britain because 1) it
has clearly defined, localized habitat patches; 2) most
populations are small and at some risk of extinction, with
occasional extinctions in 1982?2000 despite generally
increasing population sizes (Thomas and Jones 1993,
Davies et al. 2005); 3) there is limited dispersal between
H. comma populations (Hill et al. 1996), and colonization
rates decrease with distance from populations (Thomas and
Jones 1993, Davies et al. 2005); 4) the probability of
habitat patch occupancy increases with patch area and
connectivity to other populations of the species (Thomas
et al. 1992); 5) the distribution of the species in one
population network in the UK (Surrey) remained rather
stable between 1982 and 2002 (Davies et al. 2005),
allowing us to estimate metapopulation parameters for
this network assuming a steady state.
For IFM parameter estimation we used the Monte Carlo
Markov Chain method (1000 function evaluations in
initiation, 4000 function evaluations in estimation). We
assumed, based on field observations, that the minimum
area which would support a population from one year to the
next without immigration (A0) was 0.02 ha. For this
analysis, we use the parameter estimates of x?0.28, y?
7.26, e?0.34, a?0.45 (Thomas et al. 2001) assuming
that the distribution of the species in Surrey in 2000 (86
occupied and 30 vacant habitat patches) was representative
of a stochastic steady state. We also estimated parameters
using information on turnover in Surrey in 1982, 1991,
2000 and 2002, not assuming a stochastic steady state
(Moilanen 2000). Simulations using these parameters (x?
0.39, y?9.44, e?0.22, a?0.45) are not described in
detail since they gave generally similar results to the
parameters assuming a steady state in Surrey.
j(Moilanen and Nieminen 2002). Parameter a is
We tested the effects of habitat use on H. comma range
expansion rates by simulating metapopulation dynamics
through habitat networks based on three different scenarios
of habitat availability. In a thermally-restricted or ‘‘cold-
world’’ scenario: 1) only areas of habitat with aspects
between 1508 and 2508 were included, corresponding to
75% of occupied habitat patches in 1982. In a ‘‘warm-
world’’ habitat definition, broadly corresponding to that
used in 1982, 2) aspects between 1008 and 3008 were
included, corresponding to 90% of populations in 1982. In
the broadest or ‘‘hot-world’’ habitat definition, correspond-
ing more closely to habitat use in 2000, 3) areas of chalk
grassland on all aspects were included in the simulations.
We first validated the model by running 100 IFM
simulations of 18 yr using the parameters estimated from
the Surrey network, starting with the distribution of the
species in each of the five habitat networks in 1982. To test
whether the model successfully predicted the relative like-
lihood that different patches would be occupied in 2000, we
used the area under the curve (AUC) of the receiver
operating characteristic, a method which allows testing of
the significance with which modelled probabilities correctly
classify cases (Fielding and Bell 1997). In this case, we
calculated whether the proportion of simulations where
each patch was occupied was significantly related to
observed presence or absence of H. comma in 2000 for
each network using SPSS v. 15.0 for Windows. To test
whether the model accurately predicted distance expanded,
we calculated for each network the mean distance to
colonized patches in 2000 from the nearest 1982 popula-
tion. We then calculated the same measure for each 18 yr
simulation (from 1982 to 2000), and tested whether
observed range expansions lay outside the 95% confidence
intervals (2.5th to 97.5th percentiles) of the modelled
expansions. In order to test the longer term consequences of
habitat use for the range dynamics of H. comma, we then
simulated metapopulation dynamics for 100 yr for the three
habitat use scenarios and five habitat networks. To test
whether the distribution pattern used to initiate metapo-
pulation simulations would influence long-term modelled
dynamics, we started simulations both from a) the species
distribution in 1982, and b) the distribution in 2000.
Network habitat availability
Including habitat from all aspects, we identified 1916.6 ha
of suitable habitat for H. comma in the five networks
(Table 1a). Overall, the area of habitat including all aspects
(the ‘‘hot-world’’ scenario) was 1.4 times greater than
habitat availability for the ‘‘warm-world’’ scenario of 100?
3008, and 2.1 times greater than habitat availability for the
‘‘cold-world’’ scenario of 150?2508. Changes in patch
numbers were similar to changes in habitat area: the total
number of suitable patches for all aspects was 1.3 times that
for the 100?3008 scenario, and double that for the 150?
2508 scenario. However, the difference between the three
habitat scenarios differed markedly among networks: there
S P E C I A L
I S S U E
was little difference in habitat availability between scenarios
for the predominantly south-facing Surrey network (most
sites would already have been sufficiently warm in 1982),
whereas in the two networks on the largely north-facing
South Downs escarpment (Sussex and Hampshire), habitat
availability for the 0?3608 scenario was ?3.5 times greater
than for the 150?2508 scenario (Table 1a).
Model validation 1982?2000
Simulations gave good predictions of the relative likelihood
that individual patches would be occupied in 2000 in each
network, with AUC values of 0.85?0.90 for the 0?3608
networks, or of 0.73?0.88 for 100?3008 networks (Table
2). AUC values were higher for the 0?3608 scenario than
for the 100?3008 scenario in all networks except Kent,
although differences were small.
The observed mean distance to each patch colonized by
2000 from the nearest patch that was occupied in 1982
ranged from 1.3 km in Surrey to 5.8 km in Sussex
(Table 2). Observed mean colonization distances lay within
the 95% confidence intervals for IFM simulations using the
0?3608 habitat scenario for Surrey, Sussex and Hampshire,
but were overestimated for Kent and underestimated for the
Chilterns (Table 2). Simulations using the 100?3008
habitat scenario underestimated range expansions for
Sussex, Hampshire, and the Chilterns, and overestimated
expansion in Kent. Colonization distance was accurately
predicted in Surrey, for both 100?3008 and 0?3608
scenarios, but this was not surprising because in Surrey
the 0?3608 scenario was only 1% greater in area than the
100?3008 scenario. In contrast, modelled distances ex-
panded over 18 yr were greatly underestimated by the 100?
3008 scenario in Hampshire and Sussex, where 40 and 51%
respectively of potential habitat lay outside the 100?3008
range of aspects. Comparison of modelled occupancy in
Sussex with that observed after 9, 18 and 20 yr from 1982
suggests that the 100?3008 habitat scenario gave rather
accurate predictions of occupancy after 9 yr (1991; Fig. 2b),
but that by 2000 (18 yr) and 2002 (20 yr) inclusion of 0?
3608 habitat was necessary to produce accurate predictions
of occupancy (Fig. 2a). The overestimation of colonization
distance in Kent appears to be related to a lower percentage
cover of larval host plant than in the other networks, and if
this is taken into account then observed colonization
distances fall within the 95% confidence intervals for 0?
3608 simulations (Wilson et al. 2009).
100 year metapopulation simulations
Whether simulations were started with the 1982 or 2000
observed distributions of populations, each network showed
a lower predicted proportion of patches occupied after
100 yr for the 150?2508 habitat scenario than for the 100?
3008 scenario, and for the 100?3008 scenario versus the 0?
3608 scenario (Table 1b). These differences were significant
in all cases (Mann-Whitney U tests for 100 simulations in
each scenario, pB0.001), apart from 150?2508 versus
100?3008 simulations from the 1982 distribution in
Hampshire, where most simulations suffered extinction in
B100 yr. The difference was particularly marked for Sussex,
a landscape that experiences a major increase in habitat
availability with climate warming: on average only 2?3% of
patches were occupied after 100 yr for the 150?2508
simulations, 24?28% were occupied in the 100?3008
simulations, and 67?70% were occupied in the 0?3608
simulations (Table 1b, Fig. 2). In Sussex, the average
number of patches that were occupied after 100 yr in the 0?
3608 scenario was more than five times that for the 100?
3008 scenario (Fig. 2).
Four of the habitat networks also showed significant
differences in simulated occupancy after 100 yr depending
on the starting distribution (1982 versus 2000). This
difference appears to result from 1) the higher risk of
extinction of 1982 metapopulations, which had fewer
populations than in 2000; 2) some observed long-distance
colonizations between 1982 and 2000 which led to
population networks predicted to be persistent in the
long-term. Sussex, Hampshire and the Chilterns all showed
significantly higher occupancy when started with the 2000
distribution, both for 0?3608 and 100?3008 scenarios
(Mann-Whitney U tests for 100 simulations in each
Table 1. Habitat availability and modelled patch occupancy in the five networks under the three habitat use scenarios: ‘‘hot-world’’ (0?3608
aspect), ‘‘warm-world’’ (100?3008) and ‘‘cold-world’’ (150?2508). a) Total habitat area (ha), number of patches in superscript; b) mean
proportion of patches occupied after 100 metapopulation simulations of 100 yr starting from the 1982 distribution of H. comma (superscript
shows patch occupancy starting from the 2000 distribution; values in bold show significant differences from simulations initiated in 1982).
Network Habitat scenario
a) Total habitat area (ha)(n patches)
b) Proportion patch occupancy after 100 yr (start 1982)(start 2000)
S P E C I A L
I S S U E
scenario, pB0.01). Kent showed a significant difference for
the 0?3608 scenario (pB0.001), whereas Surrey showed no
significant differences in occupancy related to the starting
distribution, probably because the butterfly was already
widespread and had limited capacity to spread in this
landscape, and can fully exploit the remaining patches from
either starting scenario. For the 150?2508 simulations,
occupancy did not differ depending on starting conditions,
apart from in Sussex where occupancy was marginally
higher when starting from 2000 (p?0.02).
Simulated UK distributions for H. comma after 100 yr
are shown in Fig. 3, based on 2?2 km distribution
‘‘tetrads’’ (commonly used in regional species distribution
maps). In total, 287 tetrads with suitable habitat were
identified in the five networks, of which 26 were occupied
in 1982 and 96 were occupied in 2000. For the 0?3608
habitat scenario, H. comma’s distribution was predicted to
include 136 tetrads after 100 yr (occupied in 50% or more
of simulations) when starting with the 1982 distribution, or
164 tetrads when starting with the 2000 distribution. The
100?3008 scenario led to a distribution size after 100 yr of
61 tetrads (start 1982) or 81 tetrads (start 2000). The 150?
2508 scenario led to a predicted distribution size of 31
tetrads when starting with either distribution: for this
scenario, the networks in Sussex, Hampshire and the
Chilterns were expected to suffer extinction in ?50% of
simulations (e.g. Fig. 2c, f for Sussex).
Here we show how changes in habitat use and availability,
as may occur in response to climate change, can influence
rates of range expansion through fragmented landscapes.
The results shed light on the processes that govern the size
and pattern of species distributions, and have implications
for the conservation of localized species in a changing
Habitat availability and metapopulation dynamics
We studied the habitat use, distribution and dynamics of
the butterfly Hesperia comma in its five main UK popula-
tion networks. The breadth of habitats used by H. comma
increased between 1982 and 2000 in association with
climate warming (Thomas et al. 2001, Davies et al.
2006). Habitat area in the five networks increased by
1.4 times (and the number of patches by 1.3 times) because
of an increase in the range of aspects used, from 100?3008
in 1982 to 0?3608 in 2000. Greater habitat availability
should lead to lower rates of local extinction due to larger
population sizes; and to higher rates of colonization,
because areas of suitable habitat are closer together, and
more and larger populations are available to act as sources of
colonists. To model these effects, we applied the Incidence
Function Model (IFM) (Hanski 1994, 1999) to the
dynamics of H. comma. The model gave a relatively good
fit to patterns and rates of range expansion between 1982
and 2000, capturing the differences in colonization dis-
tances among networks varying in habitat area and
connectivity (see also Wilson et al. 2009). Furthermore,
Table 2. Validation of metapopulation simulations against observed distribution change in 1982?2000. Average expansion distance is the mean distance to colonized patches from the nearest patch
occupied in 1982. Modelled expansions show the median and 95% confidence intervals of 100 Incidence Function Model (IFM) 18 yr simulations; 95% CIs of modelled expansion are shown in bold
where they include the observed expansion distance. AUC is calculated for observed patch occupancy in 2000 against proportion occupancy in the 100 simulations.
Patch occupancy, 2000
Average expansion distance (km)
AUC of modelled patch occupancy
S P E C I A L
I S S U E
the model showed how rates of range expansion accelerated
through H. comma’s population networks associated with
wider habitat availability, as demonstrated by the under-
estimation of empirically observed expansion rates by
metapopulation simulations using the 100?3008 habitat
scenario (Table 2). Our estimated changes to habitat area
assume that all aspects were managed appropriately for
H. comma in 1982 and 2000: in practice, much habitat
in 1982 was not managed appropriately for H. comma,
and subsequent introduction of grazing management as part
of conservation programmes and agri-environment schemes
has been necessary for the species to exploit potential
habitats (Davies et al. 2005). These changes to habitat
management may partly explain why observed expansion
rate in Sussex in 1991 was overestimated by the simulations
The consequences of greater habitat availability and
faster range expansion rates for long-term distribution
patterns were investigated using 100 yr metapopulation
simulations, from H. comma’s distribution either in 1982 or
2000. Broader habitat use led to a higher proportion of
patches occupied after 100 yr (ca 1.5 times more for 0?3608
versus the 100?3008 scenario; Table 1), which equated to
approximately double the number of patches occupied,
and double the overall number of 2?2 km ‘‘tetrads’’ in
H. comma’s distribution (Fig. 3). Thus, metapopulation
dynamics provided the mechanism linking changes in
habitat use to relatively large-scale changes to the species
Habitat fragmentation and species range shifts
Dispersal failure is potentially hugely important in deter-
mining the consequences of climate change for biodiversity.
The complete failure of species to expand their distributions
Start 2000 Start 1982
Number of patches occupied
0 2040 6080100
0 204060 80100
0 204060 801000 2040 60 80 100
0 20 406080 100
0 204060 80100
Figure 2. Modelled patch occupancy over 100 yr in Sussex based on different habitat availability and starting distributions. Panels a?c
show simulations starting from the 1982 distribution; d?f show simulations from the 2000 distribution. Habitat availability includes all
available aspects (a, d), 100?3008 aspects (b, e), or 150?2508 aspects (c, f). Solid line shows median patch occupancy from 100 IFM
simulations, dashed lines show 95% confidence intervals (lower interval reaches zero occupancy in b, c and f). Diamonds in panels a and b
show observed occupancy in 1991 (9 yr since 1982), 2000 (18 yr) and 2002 (20 yr).
S P E C I A L
I S S U E
into new regions could approximately double extinction
rates from climate change, relative to rates that assume full
colonization (Thomas et al. 2004). Research on butterflies
(Warren et al. 2001, Mene ´ndez et al. 2006) and birds
(Julliard et al. 2004) has already shown that more wide-
ranging or habitat generalist species have been able to shift
their distributions in response to climate change much more
readily than sedentary species with localized habitat
distributions. Our study species H. comma is a habitat-
specialist that has been able to expand its distribution
(Thomas et al. 2001, Davies et al. 2006). However, despite
increasing habitat availability because of climate change,
conservation management and agri-environment schemes
(Davies et al. 2005), H. comma’s range expansion rates have
been constrained by habitat fragmentation (Wilson et al.
2009). In this paper, we show that even in the long term the
species is only expected to colonize a fraction of its available
habitat. The most optimistic ‘‘hot-world’’ metapopulation
simulations (with the broadest habitat requirements) sug-
gest that after 100 yr only about a half of existing suitable
habitat patches would be occupied by the species (49% if
starting with the 1982 distribution; 56% if starting with the
2000 distribution; or 47% of 2?2 km tetrads containing
suitable habitat starting from 1982, versus 57% starting
from 2000). Should regional habitat requirements become
narrower, then the proportion of suitable habitat occupied
by the species would likewise be expected to reduce (Fig. 3).
Our results should be considered as indicative of the
potential differences in realized range expansions among
landscapes differing in topography (slope and aspect) and
level of habitat fragmentation, rather than as absolute
predictions of range expansion rates. It has been shown
that insect dispersal rates may increase related to warmer
temperatures (Nieminen 1996, Sparks et al. 2005), poten-
tially leading to faster colonization rates. Furthermore,
increased dispersal ability may be selected for in newly
established populations (Hanski et al. 2006) and at the
expanding front of species distributions (Niemela and
Spence 1991, Thomas et al. 2001, Simmons and Thomas
2004). In the case of H. comma, individuals in the
expanding Sussex network show larger thorax:abdomen
size ratios than in the more stable Surrey network (Hill et al.
1999), suggesting that more dispersive forms may indeed
have been selected for by the process of range expansion.
Nevertheless, parameters estimated from the Surrey net-
work gave relatively good fits to expansion rates and
patterns in the other metapopulations (Wilson et al.
2009). Habitat availability at the landscape scale may still
be critical in explaining range expansion rates, since a
sufficient density of habitat may be necessary to promote
colonization in the first place, and therefore to select for
The differences among the five networks in observed and
modelled rates of range expansion, and in future predicted
0002 t r a tS 289 1 t ra tS
N≥ ≥50%= 136
N≥ ≥50%= 164
N≥ ≥50%= 61
N≥ ≥50%= 81
N≥ ≥50%= 31N≥ ≥50%= 31
Figure 3. Modelled 100 yr expansions by Hesperia comma based on different habitat availability and starting distributions. Maps show the
modelled species distribution in 2?2 km squares from 100 IFM simulations. Symbols show: black ? occupied in ?50 simulations;
grey ? occupied in 1?50 simulations; white ? habitat present but never occupied. Panels a?c show modelled expansions from the 1982
distribution, d?f show modelled expansions from the 2000 distribution. Habitat availability includes all available aspects (a, d), 100?3008
aspects (b, e), or 150?2508 aspects (c, f). (e) X shows a persistent part of the Sussex network colonized between 1982 and 2000.
S P E C I A L
I S S U E
distributions under the three habitat scenarios, show that
there is an important landscape context to species responses
to climate change. The metapopulation in one landscape
(Surrey) was persistent under all three habitat scenarios, and
showed no significant differences in long term distribution
patterns related to the distribution used to initiate simula-
tions (1982 versus 2000). In contrast, the Sussex network
showed significant differences in long term occupancy
among all three habitat scenarios, and between 1982 and
2000 starting conditions; while the Hampshire metapopu-
lation was only predicted to be persistent for 0?3608 and
100?3008 scenarios started from 2000. There is a relatively
small proportion of habitat in Sussex and Hampshire on
south-facing slopes, hence the ability to use a wider range of
aspects has been critical in allowing H. comma to expand its
distribution from the regional refuge populations (probably
one population in each network in the late 1970s or early
1980s; Thomas et al. 1986). In contrast, the majority of
habitat in the Surrey network is south-facing; persistence in
this network seems likely, as long as environmental change
does not render south-facing slopes unsuitable. Moreover,
the proportion of habitat occupied by H. comma remains
relatively high in Surrey, because although the overall area
of habitat is quite small, patches are densely distributed over
a relatively restricted area.
Two scenarios could lead to metapopulation declines in
Surrey: reduced habitat suitability if south-facing slopes
become too hot or dry under climate change, or a reduction
in suitable warm microclimates for larval growth because of
increasing vegetative growth related to longer growing
seasons or nitrogen deposition (WallisDeVries and Van
Swaay 2006). In fact, south-facing slopes may be relatively
resistant to the latter type of vegetation change because of
frequent exposure to drought conditions (Bennie et al.
2006). Simulations suggest that the Surrey metapopulation
may be unlikely to expand because of a low density of
suitable habitat at its eastern and western extremes (Fig. 1,
3). Nevertheless, a site at the eastern edge of the Surrey
network (11.4 km from the nearest 1982 population) was
colonized in 2002, suggesting that there is scope for further
expansion in this landscape. Metapopulation simulations
using parameters derived from the Surrey network should
be treated with some caution since they rely on the
assumption that the Surrey network in 2000 was at
colonization/extinction equilibrium. Furthermore, the un-
certain future interactions between climate, habitat condi-
tion and habitat use mean that long-term changes to species
distributions may be very difficult to predict with any
degree of certainty.
Simulations suggest that in many fragmented landscapes,
expanding metapopulations do not approach equilibria
gradually (e.g. for Sussex see Fig. 2a, d). Instead, high
occupancy is achieved quickly among habitat patches close
to the starting distribution. Subsequently, relatively rare
long-distance colonization events result in colonization of
other persistent habitat sub-networks which can achieve
high levels of occupancy, leading to what appear as step-
changes in metapopulation occupancy (e.g. Fig. 2a, step-
changes ca 20 yr and ca 60 yr into the simulation). Such
relatively rare colonization events can be important in
determining large-scale species distributions, and probably
play a key role in explaining why simulations started from
the 2000 distribution achieved greater patch occupancy
than those started from the more restricted 1982 distribu-
tion. Indeed, there is a risk that models such as that used
here might underestimate rates of range expansion because
of the role played by rare long-distance colonization events.
Refuge populations for species may reflect historical
climate or habitat requirements (Petit and Burel 1998,
Helm et al. 2006), and may not necessarily represent the
most efficient locations for seeding range expansions into
habitats which are suitable now or in the future. The
differences in simulation outcomes starting from 1982
versus 2000 distributions show that historical factors can
influence long-term distribution size and pattern, and could
constrain conservation efforts to facilitate species range
shifts in response to climate change. In Sussex, for example,
the observed 1982?2000 range expansion resulted in
H. comma colonizing a part of the habitat-network that is
persistent under the 100?3008 scenario (Fig. 3e). As a
result, 100?3008 simulations started with the 2000 dis-
tribution achieved significantly higher occupancy than those
started with the 1982 distribution (this was not the case in
Surrey or Kent, where additional persistent sub-networks
had not been colonized between 1982 and 2000). In such
circumstances, population translocations could play a part
in facilitating species range shifts (Hoegh-Guldberg et al.
2008), and approaches similar to that taken here could
assist in identifying habitat networks where species intro-
ductions would be more likely to result in range expansions.
There is also a message for projects aiming to restore habitat
connectivity. The climate-dependent nature of habitat
associations may result in climate-driven changes (increases
and decreases) to habitat availability which are much larger
than those brought about by conservation programmes for
particular species and landscapes. Therefore, such schemes
need to be carefully targeted if they are to achieve real
increases in species’ expansion rates.
The majority of species across the world exhibit very small
areas of occupancy (Gaston 1994, Kunin and Gaston
1997), as exemplified by the many butterfly species that
are restricted to a few tens of km2or less in Britain (Cowley
et al. 1999). Many, and probably most, such localized
species perceive the environment as heterogeneous, with
small areas of suitable habitat surrounded by much larger
areas which are unsuitable for reproduction (Sobero ´n pers.
comm.). Species of this type shift their distributions by
colonizing from one patch of suitable habitat to another,
rather than expanding as a continuous front. Even if such
species do not exhibit metapopulation dynamics at equili-
brium, they are likely to do so during climate change, which
is expected to render some existing habitats unsuitable and
other previously-unsuitable locations available for coloniza-
tion. Thus, metapopulation models could provide a suitable
framework for the analysis of range shifts by many habitat
specialist species. Our research on the butterfly Hesperia
comma shows how metapopulation models can link
fine-scale changes in habitat use to large-scale changes in
species distributions. The results of long-term metapopula-
tion simulations show how metapopulations in different
S P E C I A L
I S S U E
landscapes may respond differently to climate-related
changes in habitat use, because of regional differences in
habitat area, configuration, and type (in this case, aspect).
The implication is that it will be more straightforward to
adapt conservation to climate change in some landscapes
than in others, and that species-specific responses to
adaptation programmes may differ markedly (see also Vos
et al. 2008).
Acknowledgements ? N. Roura-Pascual, J. Hortal, N. Sanders and
two anonymous referees provided helpful comments on the text.
H. Burton, K. Ericson, P. Ewin, R. Fox, S. Glencross, A.
Goodhand, S. Hanna, D. Hoare, C. Holloway, R. Leaper, J.
Mellings, and A. Moilanen assisted with fieldwork and analysis.
Funding was provided by the UK Natural Environment Research
Council, EU TMR FRAGLAND and English Nature.
Allouche, O. et al. 2008. Incorporating distance constraints into
species distribution models. ? J. Appl. Ecol. 45: 599?609.
Arau ´jo, M. B. and Pearson, R. G. 2005. Equilibrium of species’
distributions with climate. ? Ecography 28: 693?695.
Arau ´jo, M. B. and New, M. 2007. Ensemble forecasting of species
distributions. ? Trends Ecol. Evol. 22: 42?47.
Arau ´jo, M. B. et al. 2005. Reducing uncertainty in projections of
extinction risk from climate change. ? Global Ecol. Biogeogr.
Arau ´jo, M. B. et al. 2008. Quaternary climate changes explain
diversity among reptiles and amphibians. ? Ecography 31: 8?
Beale, C. M. et al. 2008. Opening the climate envelope reveals no
macroscale associations with climate in European birds.
? Proc. Nat. Acad. Sci. USA 105: 14908?14912.
Beaumont, L. J. et al. 2007. Where will species go? Incorporating
new advances in climate modelling into projections of species
distributions. ? Global Change Biol. 13: 1368?1385.
Bennie, J. J. et al. 2006. Influence of slope and aspect on long-term
vegetation change in British chalk grasslands. ? J. Ecol. 94:
Cowley, M. J. R. et al. 1999. Flight areas of British butterflies:
assessing species status and decline. ? Proc. R. Soc. B 266:
Davies, Z. G. et al. 2005. The re-expansion and improving status
of the silver-spotted skipper butterfly (Hesperia comma) in
Britain: a metapopulation success story. ? Biol. Conserv. 124:
Davies, Z. G. et al. 2006. Changing habitat associations of a
thermally constrained species, the silver-spotted skipper
butterfly, in response to climate warming. ? J. Anim. Ecol.
Elith, J. et al. 2006. Novel methods improve prediction of species’
distributions from occurrence data. ? Ecography 29: 129?151.
Fielding, A. H. and Bell, J. F. 1997. A review of methods for the
assessment of prediction errors in conservation presence/
absence models. ? Environ. Conserv. 24: 38?49.
Fitzpatrick, M. C. et al. 2008. Climate change, plant migration,
and range collapse in a global biodiversity hotspot: the Banksia
(Proteaceae) of Western Australia. ? Global Change Biol. 14:
Gaston, K. J. 1994. Rarity. ? Chapman and Hall.
Hanski, I. 1994. A practical model of metapopulation dynamics.
? J. Anim. Ecol. 63: 151?162.
Hanski, I. 1999. Metapopulation ecology. ? Oxford Univ. Press.
Hanski, I. et al. 2006. Dispersal-related life-history trade-offs in a
butterfly metapopulation. ? J. Anim. Ecol. 75: 91?100.
Helm, A. et al. 2006. Slow response of plant species richness to
habitat loss and fragmentation. ? Ecol. Lett. 9: 72?77.
Hill, J. K. et al. 1996. Effects of habitat patch size and isolation on
dispersal by Hesperia comma butterflies: implications for
metapopulation structure. ? J. Anim. Ecol. 65: 725?735.
Hill, J. K. et al. 1999. Flight morphology in fragmented
populations of a rare British butterfly, Hesperia comma.
? Biol. Conserv. 87: 277?283.
Hill, J. K. et al. 2001. Impacts of landscape structure on butterfly
range expansion. ? Ecol. Lett. 4: 313?321.
Hoegh-Guldberg, O. et al. 2008. Assisted colonization and rapid
climate change. ? Science 321: 345?346.
Julliard, R. et al. 2004. Common birds facing global changes: what
makes a species at risk? ? Global Change Biol. 10: 148?154.
Kunin, W. E. and Gaston, K. J. 1997. The biology of rarity.
? Chapman and Hall.
Massot, M. et al. 2008. Climate warming, dispersal inhibition and
extinction risk. ? Global Change Biol. 14: 461?469.
Mene ´ndez, R. et al. 2006. Species richness changes lag behind
climate change. ? Proc. R. Soc. B 273: 1465?1470.
Moilanen, A. 1999. Patch occupancy models of metapopulation
dynamics: efficient parameter estimation using implicit statis-
tical inference. ? Ecology 80: 1031?1043.
Moilanen, A. 2000. The equilibrium assumption in estimating the
parameters of metapopulation models. ? J. Anim. Ecol. 69:
Moilanen, A. and Nieminen, M. 2002. Simple connectivity
measures for metapopulation studies. ? Ecology 83: 1131?
Niemela, J. and Spence, J. R. 1991. Distribution and abundance
of an exotic ground-beetle (Carabidae): a test of community
impact. ? Oikos 62: 351?359.
Nieminen, M. 1996. Migration of moth species in a network of
small islands. ? Oecologia 108: 643?651.
Parmesan, C. 2006. Ecological and evolutionary responses to
recent climate change. ? Annu. Rev. Ecol. Evol. Syst. 37: 637?
Petit, S. and Burel, F. 1998. Effects of landscape dynamics on the
metapopulation of a ground beetle (Coleoptera, Carabidae) in
a hedgerow network. ? Agric. Ecosyst. Environ. 69: 243?252.
Simmons, A. D. and Thomas, C. D. 2004. Changes in dispersal
during species’ range expansions. ? Am. Nat. 164: 378?395.
Sparks, T. H. et al. 2005. The influence of temperature on
migration of Lepidoptera into Britain. ? Global Change Biol.
Svenning, J.-C. and Skov, F. 2007. Ice age legacies in the
geographical distribution of tree species richness in Europe.
? Global Ecol. Biogeogr. 16: 234?245.
Svenning, J.-C. et al. 2008. Postglacial dispersal limitation
of widespread forest plant species in nemoral Europe.
? Ecography 31: 316?326.
Thomas, C. D. and Jones, T. M. 1993. Partial recovery of a
skipper butterfly (Hesperia comma) from population refuges:
lessons for conservation in a fragmented landscape. ? J. Anim.
Ecol. 62: 472?481.
Thomas, C. D. and Hanski, I. 2004. Metapopulation dynamics in
changing environments: butterfly responses to habitat and
climate change. ? In: Hanski, I. and Gaggiotti, O. E. (eds),
Ecology, genetics and evolution of metapopulations. Elsevier,
Thomas, C. D. et al. 1992. Distributions of occupied and vacant
butterfly habitats in fragmented landscapes. ? Oecologia 92:
Thomas, C. D. et al. 2001. Ecological and evolutionary processes
at expanding range margins. ? Nature 411: 577?581.
S P E C I A L
I S S U E
Thomas, C. D. et al. 2004. Extinction risk from climate change. Download full-text
? Nature 427: 145?148.
Thomas, J. A. 1993. Holocene climate changes and warm man-
made refugia may explain why a sixth of British butterflies
possess unnatural early-successional habitats. ? Ecography 16:
Thomas, J. A. et al. 1986. Ecology and declining status of the
silver-spotted skipper butterfly (Hesperia comma) in Britain.
? J. Appl. Ecol. 23: 365?380.
Thomas, J. A. et al. 1999. Intraspecific variation in habitat
availability among ectothermic animals near their climatic
limits and their centres of range. ? Funct. Ecol. 13: 55?64.
Trivedi, M. et al. 2008. Spatial scale affects bioclimate model
projections of climate change impacts on mountain plants.
? Global Change Biol. 14: 1089?1103.
Vos, C. C. et al. 2008. Adapting landscapes to climate change:
examples of climate-proof ecosystem networks and priority
adaptation zones. ? J. Appl. Ecol. 45: 1722?1731.
WallisDeVries, M. and Van Swaay, C. A. M. 2006. Global
warming and excess nitrogen may induce butterfly decline by
microclimatic cooling. ? Global Change Biol. 12: 1620?1626.
Ward, N. L. and Masters, G. J. 2007. Linking climate change and
species invasion: an illustration using insect herbivores.
? Global Change Biol. 13: 1605?1615.
Warren, M. S. et al. 2001. Rapid responses of British butterflies to
opposing forces of climate and habitat change. ? Nature 414:
Wilson, R. J. et al. 2007. Insects and climate change: processes,
patterns and implications for conservation. ? In: Stewart, A. J. A.
et al. (eds), Insect conservation biology. CABI, pp. 245?279.
Wilson, R. J. et al. 2009. Modelling the effect of habitat
fragmentation on range expansion in a butterfly. ? Proc. R.
Soc. B 276: 1391?1534.
S P E C I A L
I S S U E