Where within a geographical range do species survive best? A matter of scale


Chris D Thomas, Department of Biology (Area 18), University of York, PO Box 373, York YO10 5YW, UK. E-mail: cdt2@york.ac.uk



  • 1Opinions differ as to whether declining species are most likely to survive in central or peripheral parts of their distributions. The former pattern is likely to be driven by high extinction risks in peripheral areas; the latter by gradients of extinction risk.
  • 2At a continental scale of analysis, the declining butterfly Euphydryas aurinia survived best in southern and eastern countries within Europe. This was statistically associated with geographical variation in agricultural intensification. At this scale of analysis, there was a gradient of survival, caused by a gradient of agricultural intensification.
  • 3Within England and Wales, survival was greatest in population concentrations, or core areas; that is in 10-km grid squares that were surrounded by other 10-km grid squares that also contained populations of E. aurinia. In the English county of Dorset, populations were also most likely to be found in core areas; that is in habitat patches that were close to other populated habitat patches.
  • 4In this system, there is support for two patterns of decline. At a coarse scale, there is a geographical gradient of habitat degradation, associated with agricultural intensification. But within a region where decline has taken place, populations survive best in core areas, where aggregations of habitat support viable metapopulation dynamics.
  • 5Large-scale geographical patterns of decline towards the periphery (or other locations within) the distribution of a species do not negate the validity of conservation strategies based on core-margin population dynamic principles. Core areas within each country or region represent appropriate targets for conservation action.


Identifying where threatened species survive best is of interest to ecologists and to conservation agencies wishing to prioritise limited resources available for habitat protection and management (e.g. Araújo & Williams, 2000; Araújo et al., 2002; Cabeza & Moilanen, 2003; Cabeza et al., 2004; Moilanen et al., 2005). The traditional view is that populations are likely to be most persistent in central parts of geographical ranges, where species tend to be relatively widespread (and often abundant) and least susceptible to environmental and demographic stochasticity (e.g. Brown, 1984; Thomas et al., 1994; Bourn & Thomas, 2002). It is assumed that relatively high population densities and high frequencies of populations within central areas make it unlikely than any single event (including loss of a given proportion of the habitat) will drive all populations extinct simultaneously. Thus, populations and metapopulations in central areas are predicted to be more persistent.

Some aspects of this hypothesis relate directly to the effects of position in the geographical range (e.g. environmental stochasticity may have greatest impact when a population is already close to its environmental limits; Thomas et al., 1994) and others to the degree to which populations are concentrated in central parts of the geographical distribution (e.g. clusters of large populations have a low extinction risk). For clarity, in this paper, we refer to ‘central’ and ‘peripheral, edge or boundary’ to denote location within a species’ geographical distribution. We refer to ‘core’ and ‘marginal’ areas to describe variation in levels of population concentration.

In contrast to the above expectation, empirical analyses suggest that considerable numbers of declining species have disappeared from their geographical centres, and often survive predominantly at or near one or more of their former range boundaries (Lomolino & Channell, 1995; Channell & Lomolino, 2000a,b). Although there are statistical issues in determining what represents the centre or periphery, and some studies indicate that peripheral areas are more prone to extinction (e.g. Nathan et al., 1996; Donald & Greenwood, 2001), it is nonetheless clear that many threatened species have indeed retreated towards edge, rather than central, parts of their former geographical distributions. There are various possible explanations for why this might arise:

  • 1The agent of decline itself shows a geographical gradient of intensity. For example, a species threatened by invaders may have been eliminated from accessible areas that are favourable to an invasive species, and may survive only on predator-free offshore islands or in relatively inhospitable habitats (e.g. Clout & Craig, 1995). Alternatively, levels of agricultural activity might be correlated with an environmental gradient (e.g. Laurance et al., 2002). The spread of contagious sources of threat was proposed as the main cause of decline towards peripheral areas put forward by Lomolino and Channell (1995; Channell & Lomolino, 2000a,b).
  • 2Species usually do survive better in central regions (Nathan et al., 1996; Donald & Greenwood, 2001), to such an extent that we only recognise as threatened or endangered those species that have been eliminated from the central parts of their distributions. In this case, the species whose declines have been analysed represent a non-random sample of species.
  • 3The wide range of climate and habitat conditions experienced near range peripheries, and consequent potential diversity of adaptations (García-Ramos & Kirkpatrick, 1997; Alleaume-Benharira et al., 2006), may result in heterogeneous responses of populations at different boundaries to a given threat, whereas central populations may all respond in a similar way to one another. In this case, a single type of threat might eliminate all central populations but only some peripheral ones.
  • 4Some geographical boundaries, particularly coastlines, may represent climatically optimal parts of a species’ distribution. These areas may contain a species’ highest population concentrations, despite being at the geographical periphery of the range.
  • 5Edges could be sharp as defined by, for example, the climate envelope or geological substrate where a species can survive. In such species, most of the distribution could be considered to be environmentally suitable, with only a very narrow edge (Caughley et al., 1988). If this is the case, there is no reason to expect a species to be most abundant close to the geographical centre of its distribution (Sagarin & Gaines, 2002).
  • 6Variation in population concentration may occur at a finer resolution than has been analysed by Lomolino and Channell (1995; Channell & Lomolino, 2000a,b). If every region within a distribution contains core and marginal areas at a scale of metres to tens of kilometres (depending on the habitat and dispersal capacity of the species) (e.g. Thomas & Kunin, 1999; Wilson et al., 2002), then the original prediction that geographically central parts of species distributions are likely to be most persistent may be naïve. The ‘traditional’ view of being common and widespread in central areas but uncommon at the periphery may apply relatively well to widespread species of little global conservation concern, whereas the species that are found on Red Lists may have relatively localised populations (e.g. through habitat specialisation) throughout their ranges.
  • 7Contrary to received wisdom, population-level extinction rates could be naturally low at the range boundaries of many species. Except for the most mobile of species, the natural colonisation rate from populations in central parts of a range towards the edge is likely to be low. If a range edge is the dynamic consequence of extinctions and colonisations (Lennon et al., 1997), and colonisation rates are low, one might only expect to see a species occurring in those geographically peripheral areas where the extinction rate is low (e.g. in locally high-quality habitats). If so, peripheral areas might be expected to be at least as persistent as central ones.

Many of these possible explanations require an understanding of the spatial scale of the patterns of decline. In view of this, we have examined the patterns of decline of the marsh fritillary (checkerspot) butterfly, Euphydryas aurinia (i) throughout its European distribution, at the scale of countries (from van Swaay & Warren, 1999), (ii) throughout England and Wales, at the scale of 10-km grid squares (Asher et al., 2001), and (iii) within a 25 × 25-km area of the English county of Dorset, at the scale of individual habitat patches (100-m resolution; Bulman, 2001; Bulman et al., 2007). The results suggest that, for this species, different processes are important to the survival of species at different spatial scales: local survival within core areas is nested with a large-scale biogeographical gradient of decline, driven by agricultural intensification.


The study system

The range of E. aurinia (Rottemburg), Nymphalidae, extends from Europe, Morocco and Algeria through temperate Asia to Korea (Emmet & Heath, 1990; Tolman & Lewington, 1997). We only deal with the European distribution, which is relatively well known. It declined to between 20% and 50% of its former area of occupancy in Europe over a 25-year period (van Swaay & Warren, 1999). Mainly a meadow butterfly, the principal threats are from agricultural improvement (cultivation and/or addition of fertilisers) and abandonment (low-productivity grasslands become uneconomic to graze in high-intensity agricultural systems; and few or no large wild herbivores are available to replace domestic stock). Thus, habitat is either destroyed or becomes too overgrown for E. aurinia. The butterfly is protected under the 1979 Bern Convention (Annexe II) and the EC Habitats and Species Directive (Annexe II).

The UK was thought to support 5–15% of the European distribution in the 1990s (van Swaay & Warren, 1999). The British distribution declined by approximately 46% (at 10-km resolution) between 1970–1982 and 1995–2004 (Fox et al., 2006). Decline rates measured at such coarse resolution tend to underestimate population-level rates of decline (Thomas & Abery, 1995): populations at monitored sites declined by 73% between 1983 and 2004 (Fox et al., 2006). Even in the UK's population strongholds, colonies are estimated to have been disappearing at a rate of 11.5% per decade (Warren, 1994). Declines in Britain are attributed to the loss and fragmentation of semi-natural grassland and changing grazing patterns (Asher et al., 2001). Considering the two main habitats of E. aurinia in Britain, low-elevation flower-rich grassland has declined by 97% in Britain and Ireland since 1940 and chalk and limestone grassland by 80% over the same time period (Department of the Environment, 1995).

The butterfly is relatively sedentary, and shows a metapopulation structure (Bulman, 2001; Bulman et al., 2007): mark–recapture studies have recorded average movements of between 50 m and 750 m (Porter, 1981; Munguira et al., 1997; Wahlberg, 2000; Wang et al., 2004; Schtickzelle et al., 2005). Very rarely, colonisation may take place over 5 to 20 km (Warren, 1994).

European scale

The 25-year declines were taken from van Swaay and Warren (1999), who collated assessments of the status of E. aurinia from 27 European countries (with no information for a further 11 countries; the distribution in montane north-west Africa was not assessed). Van Swaay and Warren placed declines into categories for each country: decrease by 100%, 75–100%, 50–75%, 25–50% or 15–25%, stable, or increase by 125–200%. Van Swaay and Warren's (1999) fluctuating category was interpreted as equivalent to stable, because there was no overall trend. The midpoint of each category was used for the purposes of analysis (stable and fluctuating scored 0); converted to Arcsin√proportional change. Because E. aurinia shows a Palaearctic distribution, the real geographical centre of the distribution is somewhere in Central Asia and we have no information on the butterfly's status in this region. However, biologically, longitudinal margins are not necessarily very relevant when they are constrained by oceans (which E. aurinia's are), and we are more interested in the latitudinal edges to the distribution, most of which fall within Europe. Therefore, we have analysed the relationship between latitude and latitude2 (to test for non-linearity in the decline pattern). We took the midpoint between the southern and northern limits as the latitude for each country (and corresponding measurements of longitude).

England and Wales

We used data from two distributional Atlases for British butterflies, covering the periods 1970–1982 (Heath et al., 1984) and 1995–1999 (Asher et al., 2001). Data were reported as the presence or absence of the species in 10-km grid cells across Britain. Because the Asher et al. (2001) surveys involved greater search effort than those of Heath et al. (1984), the 1995–1999 data revealed small numbers of E. aurinia populations in some 10-km grid squares where they had not been recorded between 1970 and 1982. Given the overall steep decline of the butterfly and its dispersal capacity, Asher et al. (2001) and we conclude that almost all of these represent populations that were truly present in 1970–1982, but that had been overlooked at that time. Therefore, we assume that these 10-km squares were occupied in 1970–1982 in our analyses. We excluded the butterfly's Scottish distribution from our analyses because of the paucity of historical data; and excluded a known introduction to the county of Lincolnshire (triangle in Fig. 2). We test the hypothesis that the species survived best in core areas by analysing the likelihood that the species would survive (to 1995–1999) in each 10-km square occupied in 1970–1982 as a function of the proportion of nearby 10 km squares (within 20-km buffer area) occupied by E. aurinia in 1970–1982.

Figure 2.

The distribution of Euphydryas aurinia in England and Wales, at 10-km resolution (modified from Asher et al., 2001). The filled triangle represents a documented introduction between the two time periods, and was not included in the analysis.

The county of Dorset

We examined a 25 × 25-km area in which the presence of all potential host plants Succisa pratensis (Moench), Dipsacaceae, was mapped at 100-m resolution. This was converted into an E. aurinia habitat patch network (Bulman, 2001). The butterfly distribution in this area was analysed to evaluate whether the species survived better in core or marginal areas, using Hanski's (1994) index of population connectivity. The Dorset analysis is published in detail elsewhere (Bulman et al., 2007), hence we only provide a brief summary here, for the purposes of comparison with the coarser-resolution analyses.


European scale

Euphydryas aurinia showed significant geographical gradients of decline, with greater declines in more northern [Arcsin√(proportional decline) vs. latitude, Pearson r = 0.469, d.f. = 26, P = 0.016; Fig. 1a] and western countries [Arcsin√(proportional decline) vs. longitude, Pearson r = –0.417, d.f. = 26, P = 0.034]. Quadratic terms in regressions were not significant for either latitude or longitude. This pattern is consistent with the interpretation that there is a gradient of extinction risk. The butterfly is extinct in the Netherlands, and estimated to have declined by over 50% in Ireland, the UK, Belgium, Germany, Denmark and Poland, forming a swath of extinction across the intensively agricultural north European plain. In contrast, the butterfly was stable in Spain, Albania, Macedonia, Greece, Bulgaria, Slovenia and Estonia, and appears to have increased in Hungary (van Swaay & Warren, 1999).

Figure 1.

Relationship between Euphydryas aurinia decline over 25 years within European countries (van Swaay & Warren, 1999), and (a) the mid-latitude of each country, and (b) the intensity of fertiliser use in 1999 on farmland within each country (European Environment Agency, 2003).

The most plausible explanation for the pattern is the intensity of agricultural use. We used the 1999 fertiliser consumption per unit area of agricultural land (tonnes km−2) as an index of intensification (European Environment Agency, 2003). For the 24 countries that could be included in this analysis, stepwise multiple regression revealed that decline was associated with fertiliser use, and not significantly with latitude or longitude [Arcsin√(proportional decline) = 0.014 + 0.053 (SE 0.012) (Fertiliser Use), T = 4.5, P < 0.001; r = 0.692; Fig. 1b].

England and Wales

Distribution change at 10-km resolution in England and Wales showed a retraction towards the core areas (Fig. 2). Grid squares were more likely to remain populated by E. aurinia from 1970–1982 to 1995–1999 if a high proportion of the surrounding grid cells also contained populations in 1970–1982 [logit survival =–1.8534 + 4.8163 (proportion of neighbours occupied in 20-km buffer area), –2 Log LR = 62.47, d.f. = 1, P < 0.001]. These 10-km grid cells can be regarded as core areas because they (i) were surrounded by other 10-km grid cells that were populated by E. aurinia, and (ii) contained relatively large numbers of E. aurinia populations within them (Spearman rank correlation between the number of 1-km grid cells recorded as occupied by E. aurinia per 10-km grid cell and proportion of occupied 10-km grid cells in buffer area = 0.24, n = 182, P = 0.001).

The county of Dorset

Within the 25 × 25-km study area in Dorset, E. aurinia was present in habitat patches that were relatively large and close to one another (i.e. in core areas) (Fig. 3). Previous analyses showed that connectivity (a weighted measure of proximity to all other populations) was a significant predictor of both probability of colonisation and patch occupancy, taking local habitat quality within individual patches into account (Bulman et al., 2007).

Figure 3.

Distribution of Succisa pratensis, the host plant of Euphydryas aurinia, across the 25 × 25-km Dorset (England) study area in 1998/1999 (modified from Bulman et al., 2007). Patches are scaled by resource area (ha; habitat area × proportional host cover). Black circles indicate patches that were occupied by E. aurinia in 1999 and the white circles show the vacant patches.


These results show that patterns of a species’ survival can be different at different scales of analysis. Patterns of agricultural intensification (and not simply fertiliser use) appear responsible for the Europe-wide decline in the northern part of the distribution: E. aurinia survived best in the less intensively farmed countries in the south and east of the continent. At finer resolutions, although the underlying cause of decline was the same, the aggregated distribution of surviving habitat remnants and the quality of individual patches were increasingly important. These are the locations where the butterfly can exhibit viable metapopulation dynamics (e.g. Hanski, 1994, 1999; Hanski & Ovaskainen, 2000; Bulman et al., 2007). At very coarse scales of analysis, decline towards part of the periphery of the distribution is a reasonable, if approximate, description. At finer scales, the butterfly is surviving best in core habitat patches and landscapes.

The various explanations and patterns of decline that have been discussed in the literature are compatible with one another, provided one recognises the appropriate scale at which each process dominates patterns of survival. Here, we found support for the hypotheses that large-scale decline patterns are driven by gradients of environmental change (hypothesis 1) and that regions contain both core and marginal areas of populations within them (hypothesis 6). Euphydryas aurinia also appears to show greater diversity in its local habitat adaptations in the south than north of the continent (Singer et al., 2002; hypothesis 3), but we have no direct evidence whether this contributes to the latitudinal gradient of extinction rate.

Araújo and Williams (2001) suggested that if extinction is caused mainly by demographic factors, then core areas (population concentrations) should be targeted, but geographically peripheral areas might be more important if the main drivers are extrinsic factors. Our study suggests that both are important, because both extrinsic drivers and demographic factors combine to determine the likelihood of population persistence (Caughley, 1994). At a continental scale, conservation organisations might be seeking to influence policies that will reduce rates of agricultural intensification or abandonment of low productivity lands: the economic forces of agricultural change operate at international scales. Within countries or states, while the same policies are relevant, the specific protection of core landscapes is likely to be most important if one is to ensure long-term demographic persistence of the target species. Within these regions, conservation efforts should focus on maintaining habitat characteristics that maximise population productivity within individual habitat patches (Thomas et al., 2001), as well as maintaining a sufficient quantity of habitat at the landscape-scale to ensure metapopulation persistence (Bulman et al., 2007). Different scales of conservation effort – within habitat protection and management, maintenance of networks of habitat patches across counties or countries, and action to minimise socio-economic pressures leading to habitat loss – are equally valid, and may all have to be achieved to ensure success.

Gradients of extinction pressure are probably common, associated with invasion gradients, climatic gradients (e.g. affecting patterns of land use along environmental gradients), and political and socio-economic gradients. However, the existence of forces leading to extinction gradients does not mean that survival is only possible in the peripheral areas of species’ distributions. Conservation at range boundaries is often the last resort once species have already been eliminated from central regions; maintaining intact core habitats and landscapes within geographically central parts of the distribution might have been preferable, had the decline been foreseen. The reality is that our foresight is limited, especially when dealing with the conservation of insects. A strategy of maintaining core habitats and populations within regions, but doing so in regions throughout the geographical distribution of a species, would seem to be an appropriate risk-averse strategy for minimising species-level extinctions, given that current and future gradients of threat are at least partially unpredictable. Achieving this is difficult for invertebrates in data-poor regions. On the other hand, habitat-driven probability of occurrence and population concentration surfaces can be combined with reserve-selection approaches to identify provisional priority areas (Cabeza et al., 2004; Moilanen et al., 2005), and these can form the focus of subsequent survey work to identify whether these locations are truly important for conservation.

Climate change is increasingly becoming a factor driving extinction gradients (e.g. Parmesan, 1996; Wilson et al., 2005; Franco et al., 2006). In this case, we still think that there are benefits in at least initially aiming to retain core areas throughout species’ geographical distributions (Hampe & Petit, 2005). The individualistic responses of species to different aspects of climate (e.g. summer vs. winter temperature, precipitation vs. temperature) and to climate-driven changes in biological communities mean that not all species will shift their distributions in the same direction as one another: reliable predictions of the future responses of species are not yet available, and are unlikely to be forthcoming for the majority of insect species in the foreseeable future. The exact level of future warming, details of future precipitation changes, and how vegetation (and hence the distributions of plant-feeding insects) will respond to carbon dioxide concentration changes remain uncertain (Intergovernmental Panel on Climate Change, 2001, 2007). In the event that Earth's temperature returns to pre-industrial levels within the next few hundred years, it would potentially be disastrous for biodiversity if the only surviving populations of most species were then at or beyond the current cold margins of their ranges. With the current state of knowledge, attempting to conserve population concentrations in a number of locations scattered throughout a species’ geographical range would appear to be a prudent approach.


We thank the Countryside Council for Wales (especially Adrian Fowles) and English Nature (especially David Sheppard) for supporting this work, David Blakeley for providing the latitude and longitude values, and Josef Settele and an anonymous referee for comments on the manuscript.