This study used a metapopulation model incorporating little of the detailed ecological and behavioural attributes of a test species beyond knowledge of the distribution of its habitat and populations in the landscape. Nonetheless, metapopulation parameters estimated from H. comma
's occupancy patterns in one habitat network (Surrey) led to accurate predictions for four other networks both of distance expanded ( and ) and the relative likelihoods that individual patches would be colonized (). When combined with simple information on habitat quality (see Thomas et al. 2001b
), the model captured sufficient essence of the species range expansion that it could estimate species recovery accurately and independently in different landscapes, and identify a landscape where low habitat quality constrained expansion rate.
Relatively simple metapopulation models can predict rates and patterns of range expansion well because they focus on the critical processes of colonization and extinction. If such models are well parametrized, then their predictions may be as accurate as those based on more complex approaches such as individual-based models (Ovaskainen & Hanski 2004
). Predictions are also dependent on a sufficiently accurate measurement of the landscape-scale distribution of suitable habitat. Our assumption that all habitat identified as suitable in 2000 was available for colonization in 1982 was a simplification, and the incorporation of dynamic patterns of habitat availability in metapopulation models could be vital for modelling changes to species distributions (Keymer et al. 2000
), especially in the context of climate change.
In 2000, H. comma
was found to occupy 21
of suitable habitat (10 times its area of occupancy 18 years earlier), in more than 250 discrete populations (an increase from fewer than 70 in 1982) (Davies et al. 2005
). However, although 174 patches were colonized in the five habitat networks, and only two populations went extinct, in each network more patches remained unoccupied in 2000 than had been colonized since 1982 (). An analytical estimation of range expansion by H. comma
through continuous habitat greatly overestimated the average colonization rate of the species in 18 years, whereas metapopulation modelling based on stepping-stone colonization through fragmented habitat networks led to rather accurate predictions of range expansion rates. Thus, limited habitat availability constrained expansion rates substantially, over and above the constraints imposed by the intrinsic dispersal capacity and rate of increase of the species. The fact that habitat fragmentation constrained recovery, even in an expanding species with a landscape-scale conservation programme in place, implies that fragmented landscapes are likely to present an almost insurmountable barrier to the distributional responses of many species to climate change. Continuing the metapopulation simulations for H. comma
for 100 years did not lead to complete patch occupancy in any network: many isolated habitat patches at the margins of each network were occupied in fewer than 50 per cent of simulations, and the section of the North Downs between Surrey and Kent () was never predicted to be colonized naturally (R. J. Wilson 2008, unpublished data).
The results shed light on why, despite pronounced increases in population size on monitored sites (Davies et al. 2005
), H. comma
has failed to expand northwards in recent years to anything like the extent achieved by many British butterflies that have less specialized requirements (Warren et al. 2001
). In this respect, H. comma
shares similarities with many butterflies (Menéndez et al. 2006
) and other taxa (Hickling et al. 2006
) that are expanding their distributions at their poleward range boundaries, but are apparently spreading much more slowly than would be expected if they were keeping track with climate change. Previous modelling studies of the effect of habitat quantity on range expansion rates have concentrated on species with widespread habitats (e.g. Hill et al. 2001
). Although these studies show that habitat availability influences expansion rate, such species are fairly atypical: habitats for most species occupy only a very small percentage of the landscape (Cowley et al. 1999
). These localized species face a far greater challenge responding to climate change.
Modelling the responses of species distributions to habitat availability can identify landscapes where conservation is either most needed or most likely to improve species status (Huxel & Hastings 1999
). Such an approach is potentially applicable to a wide range of systems, assessing the potential for natural expansion in response to habitat restoration, the likelihood that species introductions and reintroductions will succeed and spread, and the potential responses of species to climate warming (Opdam & Wascher 2004
). Habitat fragmentation represents a major impediment to the expansion of species range boundaries as the climate warms (Warren et al. 2001
; Travis 2003
). Many species may only be able to change their distributions in response to climate change in landscapes where there is sufficient density of habitat to allow expansion (Peterson et al. 2002
). Approaches similar to those described here can be used to model rates of expansion in different landscapes, to predict those species that will be able to shift their distributions in response to climate change, and those that will require active conservation management to do so.