1. Current national and international frameworks for assessing threats to species have not been developed in the context of climate change, and are not framed in a way that recognises new opportunities that arise from climate change.
2. The framework presented here separates the threats and benefits of climate change for individual species. Threat is assessed by the level of climate-related decline within a species’ recently occupied (e.g. pre-1970s) historical distribution, based on observed (e.g. repeat census) and/or projected changes (e.g. modelled bioclimate space). Benefits are assessed in terms of observed and/or projected increases outside the recently occupied historical range.
3. Exacerbating factors (e.g. small population size, low dispersal capacity) that might increase levels of threat or limit expansion in response to climate change are taken into consideration within the framework. Protocols are also used to identify levels of confidence (and hence research and/or monitoring needs) in each species’ assessment.
4. Observed and projected changes are combined into single measures of expected decline and increase, together with associated measures of confidence. We weight risk classifications towards information that is most certain. Each species is then placed in one of six categories (high risk, medium risk, limited impact, equivalent risks & benefits, medium benefit, high benefit) reflecting whether climate change is expected (or has been observed) to cause net declines or increases in the region considered, based on the balance of benefits and threats.
5. We illustrate the feasibility of using the framework by applying it to (i) all British butterflies (N = 58 species) and (ii) an additional sample of British species: 18 species of plants, bats, birds and beetles.
6.Synthesis. Our framework assesses net declines and increases associated with climate change, for individual species. It could be applied at any scale (regional, continental or global distributions of species), and complements existing conservation assessment protocols such as red-listing. Using observed and projected population and/or range data, it is feasible to carry out systematic conservation status assessments that inform the development of monitoring, adaptation measures and conservation management planning for species that are responding to climate change.
Governments and conservation organisations already prioritise the species and ecosystems (general habitat types, vegetation types) that are most in need of conservation action, based on their rarity, current threats and rates of decline; through national and international assessments (e.g. Miller et al. 2007; Mace et al. 2008; Eaton et al. 2009). However, conventional risk frameworks are not particularly appropriate for dealing with (i) species that decline in some regions, but expand into others, a situation that is likely to be common under climate change, and (ii) the long time scales over which species are expected to respond to climate change (Fischlin et al. 2007). In this paper, we concentrate on developing a framework to evaluate species’ responses to climate change, so that both observed and expected responses can be used to inform conservation prioritisation.
Because species will potentially decrease in some areas but increase in others, we develop a methodological framework that separates (i) impacts within regions where a species has traditionally occurred, and where any existing in situ species-specific conservation provision takes place, and (ii) increases outside the species’ former range (e.g. since 1970), where new opportunities for conservation may exist, associated with natural or facilitated range expansion. This framework should help to prioritise resource allocation between works on species that face a perilous future in a changing climate, and projects where tactical use of resources could facilitate the future recovery/spread of species. The framework includes consideration of questions around data adequacy and uncertainties that apply to future climatic changes and the accompanying biological responses. This assessment framework can conceptually be used at any spatial scale, from local to international. We apply it to British taxa to illustrate that it is feasible to undertake such assessments.
Materials and methods
The approach we outline draws inspiration from the IUCN red listing procedure (Mace & Lande 1991, IUCN 2001, 2010; Mace et al. 2008). It is not straight-forward to adapt the IUCN approach to climate change (Thomas et al. 2004; Bomhard et al. 2005; Thuiller et al. 2005; Akçakaya et al. 2006), and hence the framework we present is separate and is intended to sit alongside existing IUCN and National listing procedures. The assessment framework consists of six major stages, each of which contains a series of steps, with the final stage providing an overall classification of observed and projected climate-related declines and increases (Figure 1, Tables 1–6). Our approach also includes information about uncertainties; for example, future projected declines will often be accompanied by lower certainty than information about already observed declines. We weight risk classifications towards information that is most certain.
Table 1. Stage I procedure for observed changes within recent range. If insufficient (single survey only) or no data to assess I.A and I.B, proceed to Stage II (Table 2)
Table 2. Stage II procedure for projected changes within recent range. If no projection available, or model not significant for region considered, proceed to Stage III (Table 3)
Table 3. Stage III procedure for observed increases outside of previous range. If insufficient (e.g. single survey only) or no data, proceed to stage IV (Table 4)
Table 4. Stage IV procedure for potential increases outside of current range. If no projection available, or model is not significant for region considered, proceed to Stage V (Table 5)
Table 5. Stage V procedure to summarise observed and projected changes and confidences
Overall decline (I and II) and expected increase (III and IV)
*Overall scores, rounded to nearest integer; with x.5 values rounded upwards. When only one assessment is possible, use the original decline/expansion score.
†For Score = 3, category = Very High; for Score = 2, category = High; for Score = 1, category = Moderate; for Score = 0, category = Low.
‡Overall confidence score rounded to nearest integer; with x.5 values rounded downwards (all scores <1·5 rounded to 1). When only one assessment is possible, use the original confidence score. |a − c| indicates the absolute difference in value between the (a) and (c) scores.
Table 6. Stage VI summary table for risks and benefits
Stages I and II deal with the assessment of species’ declines within their recent historical ranges (for example, their distributions in the 1970s); and then Stages III and IV consider increases outside these species’ recent historical ranges (including areas of recent colonisation and potential new areas where species may become established in future). Stages I and III assess changes already observed, whereas Stages II and IV consider projected future changes. Each stage contains four similar steps (Tables 1–4); Step A assesses distribution/abundance changes, Step B assesses links with climate, and Step C considers whether exacerbating factors may affect species’ changes. Step D is a combined assessment for that stage based on the sum of the assessment of the rate of change in distribution or population size (A), and any exacerbating factors (C), combined with confidence in the assessment. This results in four levels of threat and benefit; low (score = 0), moderate (1), high (2) and very high (≥3), and three levels of confidence (poor, medium and good; Tables 1–4).
Stages I and II (observed and projected declines) are weighted according to the level of confidence in each, and then combined to provide one overall assessment of decline within the existing distribution (Stage V; Table 5). In a similar way, Stage III and IV are combined to provide an assessment of benefit to species through expansions of their ranges (Stage V; Table 5). Finally, Stage VI assigns species to an overall risk category based on the balance of threats vs. benefits (Figure 1; Table 6).
Data requirements for assessment of observed trends are similar to those required for conventional conservation status evaluation. Information on potential future change requires projection of a species’ future distribution and/or population size under climate change; these projections could be based on very simple information (see Stage II below), or on the output from complex population and climate models.
Description of the framework
Stage I – observed change within the recent historical range
The observed rate of decline (Stage I, Step A; hereafter denoted I.A) is coupled with an assessment of whether any decline is likely to be linked to climate change (I.B), and whether there are any other exacerbating factors (such as small current range size) that might be expected to increase the level of threat (I.C) (Table 1). These are combined to produce an overall summary of the observed threat (I.D). The ‘recent historical’ distribution should ideally be the distribution prior to recent anthropogenic warming (arguably the 1960s/early 1970s), but in practice this stage will often consider the starting point to be the distribution or population whenever it is first adequately documented.
Step I.A assesses the recent rate of decline, relative to the distribution or population size of the species in the past. Declines may be measured in terms of changes in extent of occurrence (broad geographic range size), area of occupancy (total area supporting populations), or total population size, as in conventional IUCN listing criteria (see IUCN 2001, 2010). The decline rates that trigger a given level of threat in the framework presented here are lower than those normally adopted within IUCN listing criteria, to reflect the long time-scales over which climate change affects species and the length of time it takes to mitigate climate change. Thus, the highest threat category is for species on course to become extinct by 2100 (i.e. losing over 7·5% of the original distribution or population per decade, assuming that the same rate is sustained from 1970 to 2100). In this paper, we apply this to species in the context of regional extirpation rather than global extinction, but the principles are the same. Because of variation in population density in different parts of species’ distributions, and other complicating variables, rates of distribution change would ideally be converted to rates of population decline (Akçakaya et al. 2006); but there will be many cases for which the available data do not make this possible. Species with recent rates of decline expected to generate >50% loss from 1970 to 2100 (loss of 4–7·5% of the original distribution or population per decade) are placed in the second highest category; and those with rates of decline corresponding to ∼10–50% loss from 1970 to 2100 (1–4% loss per decade) are assigned to the third category. The decadal rates are given in Table 1. The rationale for selecting decadal rates that correspond to these levels of change over the period 1970–2100 is because rapid anthropogenic warming is accepted as important from the 1970s onwards, and virtually all emissions and climate scenarios suggest that warming will continue until at least 2100 (IPCC 2007). Note that the longer the period over which data are available to measure declines, the greater will be the accuracy and therefore confidence in the assessment; for species for which data are available for 40 years, from 1970 to 2010, these categories correspond to >30%, 16–30% and 4–16% declines, respectively. If a species is increasing within its recent historical distribution, as a result of climate change, it will be assigned to the lowest threat category (<1% decline per decade).
The IUCN (2001, 2010) provides two options when assessing observed rates of population change; the change per unit time (which we use), or the change over 1–3 generations, with the ‘per generation’ rate being used for species with relatively long generations (up to a maximum of 100 years, even if three generations is longer than 100 years). We have not adopted the per generation alternative in this paper because the rates of change and time scales we discuss are tailored to population changes over a period of up to 130 years (1970–2100), which is already longer than the IUCN cut-off (see Discussion).
Confidence in Step I.A is considered ‘good’ when detailed survey data are available (e.g. population monitoring at multiple locations; repeat surveys preferably spanning 20 years or more, adjusted for any variation in recording effort), and ‘poor’ when less good data are available (e.g. repeat surveys with variable effort or over a shorter time span; qualitative trend data and expert opinion). The published data sources we used for the analyses reported here generally did not present standard errors or 95% confidence intervals for the decline rates (e.g. based on the number of sample locations and variation in population trends among locations; or equivalent error estimates for changes in distribution size). In future iterations of the framework, where assessors are working from raw survey data rather than summarised values from the literature, ‘good’ confidence might, for example, be assigned when a species has a greater than two-thirds chance of being within the specified category, etc. When no trend or change data are available, it is not possible to carry out Stage I of the analysis and the assessor should proceed directly to Stage II.
For all species declining by >1% per decade within their recent historic ranges, Step I.B assesses whether the observed trends are linked to climate change. Assessment scores for Step A are carried forward to Step D only if the observed trends are linked to climate. Thus, a species with a high rate of decline would not be classified in Step D as being at high risk from climate change if declines could be wholly attributed to other threats, such as habitat destruction. The criteria for assessing the confidence in the assessment at Step I.B are shown in Table 1. The criteria for confidence have been stated as explicitly as possible, but a panel of assessors should be asked to balance the strength of various pieces of information that suggest linkage with climate change with the strength of evidence that trends are associated with other factors.
Step I.C considers whether there are exacerbating factors that might be expected to speed species’ declines, over and above the decline rates already observed in step I.A. These exacerbating factors include small population and/or range size, on the assumption that the demise of a small-range species is more probable than that of a more widely distributed species that has a comparable decline rate. In Table 1, the exacerbating area and population thresholds correspond to those that would qualify as Threatened (Vulnerable or worse) using IUCN criteria (IUCN 2001, 2010). Observed climate-related expansion of a competitor, parasite or predator, or a climate-linked shift in land use or cover, would also qualify as an exacerbating factor, but only if there is a clear indication that this has already started to generate an increased rate of decline, above the average already documented in Step I.A over the whole period considered. Most exacerbating factors are probably already accounted for within the assessment of the species’ observed rate of decline. Only currently observed factors are included here (future exacerbating factors are included in Stage II).
Finally, Step I.D sums the score for decline rate (I.A) and exacerbating factors to generate one overall score, assigning the species to one level of risk on the basis of observed declines; and assesses the overall confidence as in Table 1.
Stage II – projected decline within recent historical range
Stage II assesses projected future changes within the species’ recent historic range using the same four steps outlined earlier (Table 2). Step II.A relies on climate-related projections (e.g. Huntley et al. 2007; Walmsley et al. 2007; Settele et al. 2008) to estimate the expected amount of decline (Table 2). Links to climate (II.B) are already included in models. For species showing projected declines within the recent historical range, Step II.C considers exacerbating factors. Step II.D assesses the overall projected threat in the same way as described in Stage I earlier.
If model projections are not available, simple assessments can be substituted. For example, if a species is restricted to high elevations, adiabatic lapse rates (declining temperature within increasing elevation) may provide a first approximation of the expected rate of loss for a given rate of projected regional warming (cf. Raxworthy et al. 2008; Sekercioglu et al. 2008). Equally, a species that occurs in the hottest and driest part of the study region (and is known also to occur in hotter/drier regions outside the study region) can tentatively be placed within the low threat category if the region under consideration is experiencing both a warming trend and reduced precipitation. Such assessments will carry low confidence.
Good confidence is assigned only to those projections that have been able to predict the changes that have already been observed to take place (Table 2), something that has only rarely been achieved in the literature to date (e.g. Araújo et al. 2005; Walther, Berger & Sykes 2005; Pearman et al. 2008; Vallecillo, Brotons & Thuiller 2009; Willis et al. 2009a; Wilson, Davies & Thomas 2009, 2010; Ficetola et al. 2010), but which is likely to become increasingly feasible as models that combine both climatic and population dynamics are developed (Keith et al. 2008; Anderson et al. 2009) and the period over which observation data are available to parameterise and test such models increases. For example, one could take the distribution of a species in the 1970s, project its 2000s distribution using alternative models (e.g. climate envelope, population dynamic, combined) and then test the capacity of the alternative models to predict the changes since 1970s. This enables assessors to adjudicate the weight that should be assigned to projections that are based on different models. Even so, there is considerable additional uncertainty associated with uncertain future climates and land use patterns.
For the purposes of this paper, we consider only single General Circulation Models and climate scenarios (intermediate levels of expected warming, in each case) because our purpose here is to develop the species assessment approach, rather than to carry out a full analysis for every species. However, assessments should ideally consider a variety of future scenarios and model projections (e.g. using ensembles and regional climate models, and ranges of uncertainty, with specified probabilities of falling within a given risk category). Multiple scenarios will either strengthen (e.g. all projections indicate decline at >7·5% per decade) or undermine (e.g. a species is projected to decline for some scenarios or models but increase for others) confidence in the assessment (Table 2). The range of possible future scenarios, as well as uncertainty about how a species might respond to any single scenario, generally mean that confidence in the projected future status will be low or medium. Hence greater weight will be given to empirically observed decline rates than to projected declines, provided that such data exist (see Stage V). Although the declines are stated in Table 2 as decadal rates, these will normally be recast as percentage declines projected for 2050 or 2100, or some other time period for which future distributions have been projected (see below). This will be particularly appropriate for scenarios that suggest nonlinear climate change (i.e. acceleration or deceleration of warming towards the end of the 21st century), or when the responses of a species are nonlinear even if the underlying rate of climate change is relatively constant.
The level of complexity of projections can vary greatly. Consideration of exacerbating factors (Step II.C) should be limited to factors that were not included in the model projections. Hence, if projections considered under II.A already incorporated information on the initial population size (and stochasticity associated with small population size), this should not be counted again at Step II.C.
Stage III – observed increase outside recent historical range
Stage III considers already observed increases beyond a species’ recent historical range (Table 3). Stage III is equivalent to Stage I in assessing already observed changes, but there is an important difference. In Stage I, all changes are calculated in relation to the species’ original status, such that 10% decline per decade is equivalent to 100% loss after 100 years. In Stage III, by contrast, the species’ initial status is updated every decade (Table 3). This difference is to avoid classifying species with very small initial populations as having very large observed percentage increases over several decades even if their final population size remains small. In practice, assessors should consider the level of expansion observed and the time period over which the increase happened. If the expansion of a species was monitored over 25 years (2·5 decades), then that species would have had to have increased by over 20% (i.e. [100 × 1·0752·5] – 100) to be placed in the highest expansion category, representing decade-on-decade expansion of 7·5% for two and a half decades.
Step III.B assesses the link with climate. Exacerbating and enhancing factors (III.C) are not considered separately for increasing species because non-climatic contributions to the rate of change (e.g. habitat availability, small population sizes limiting propagule number, dispersal rate) should already be represented within the observed expansion rates.
Stage IV – projected increase outside recent historical range
Stage IV uses model-based projections of the future potential distributions of species (as in Stage II) to assess potential increases that are expected to take place outside the species’ current (recent past) range, in the absence of deliberate human intervention, such as translocations or possible future management to create new habitat or increase habitat connectivity (Table 4).
As before, the good confidence (✓) in Step IV.A will be for models that accurately predict changes that have already been observed (e.g. Araújo et al. 2005; Walther, Berger & Sykes 2005; Pearman et al. 2008; Vallecillo, Brotons & Thuiller 2009; Willis et al. 2009a; Wilson, Davies & Thomas 2009, 2010; Ficetola et al. 2010) and when alternative models and scenarios project comparable changes; i.e. models that accurately predict changes to distributions through time. Medium confidence (∼) would be appropriate for models where the original data have been separated into training (model building) data, and then the predictions of the model provide a good fit to the remaining testing data (but for which there is no test of predicted changes through time); i.e. models that accurately predict current distributions in space. Low (x) confidence will be appropriate for models that have only been tested against the data used to build the models. Step IV.B is not relevant because links with climate are assumed in these models. If it is not known whether a species’ distribution is truly linked to climate, this should result in projections being assigned low confidence in Step IV.A.
Exacerbating factors are considered in Step IV.C if the model is a simple bioclimate model. It is not possible to develop precise measures of what values of dispersal ability or habitat availability, for example, should be used adjudicate ‘Low’ in Table 4. ‘Low’ should apply for constraints that will result in a species achieving a lower level of expansion (lower expansion category) than predicted by the models in Step IV.A (e.g. Willis et al. 2009b). For example, if low dispersal rate is expected to slow the decadal expansion rates to such an extent that the expected increase would now be in a lower expansion rate category than before, then low dispersal rate should be invoked as an exacerbating factor (e.g. reduction of expected increase from 10% per decade to 5% per decade). If habitat availability is expected to be lower (as a percentage cover of the landscape) in the new range than in the previous range, sufficient to reduce the potential area occupied into a lower category, then this should be invoked as an exacerbating constraint. For example, if a bioclimate model projected a 6% increase in suitable climate space per decade, but the required geology was twice as rare in this new area as in the species’ existing distribution, then the realised increase is expected to be only 3%, and low habitat availability should be invoked. No separate assessment of exacerbating factors is needed if a model has already included species-specific information on dispersal, habitat availability within regions of potential colonisation, and interactions with other species (e.g. the potential distribution and dynamics of a host plant, in the case of an herbivorous insect). Deliberate attempts to increase colonisation rates (e.g. translocations or possible future management to increase habitat connectivity) or the area of available habitat within the new range might be considered as the reverse of exacerbating factors, but would usually be brought in as a response to the initial risk assessment, rather than as part of the risk assessment itself.
Stage V – summarising decreases and increases
Stage V uses the information obtained in Stages I–IV to compute an overall measure of climate-related decline within the existing range (by combining Stage I and II information) and one measure of climate-related expansion (from Stages III and IV) (Table 5). Both these overall scores are based on the average scores for the two stages, weighted towards the stage score with the greatest confidence, using equations in Table 5. The overall scores are rounded to the nearest integer. Stage V results in each species being placed into one overall climate-related threat category for its existing range (κ in Table 5) with an associated confidence category (λ), and one overall increase category (μ, with associated confidence ν).
This is best understood through a numerical example. Take a species with observed threat class a = 1 (moderate from Table 1; and associated confidence is classed as b = 2, medium) and modelled threat of c = 2 (high from Table 2; and associated confidence is classed as d = 1, poor). Using the equation (a × b + c × d)/(b + d) (from Table 5), the overall score is then (1 × 2 + 2 × 1)/(2 + 1), equalling 1·33, which rounds to κ = 1. Thus, the overall threat class to this species within its existing range is 1 (moderate), weighting the conclusion towards the empirically observed trend because there was greater confidence in this. The confidence in this overall assessment, using the equation (b + d − |a – c|)/2 (from Table 5), would then be (2 + 1 – |1 – 2|)/2 (from Table 5). This equates to (3 – 1)/2 = 1, so λ = 1, poor confidence. Note that the uncertainty here is whether the species should be assigned to moderate threat, based on empirical observations, or high threat, based on models; so poor confidence should not be interpreted as not knowing whether the species is expected to decline at all but which specific category it should be placed in (see Discussion). It is inevitable that there will often be poor confidence in assessments; but to take no action before confidence is high (e.g. by the middle of the 21st century) may be too late.
Stage VI – overall risk categories
Stage VI places species into one of six categories representing species with High climate change Risk (red); Medium Risk (amber); Limited Change (pale green); equivalent Risks & Benefits (bright green); and those for which benefits outweigh risks (Medium Benefit, blue; High Benefit, purple) (Table 6). Depending on the position of a species in Table 6, the requirement for adaptation measures in response to climate change will differ and different conservation actions may be more likely to be appropriate.
Applying the framework to selected British species
We used two test groups of British species, based on publicly available data, to assess the practicality of the procedure. Test case 1 consisted of all British butterfly species; Test case 2 was a non-random selection of species across a variety of taxonomic groups, which varied in the quantity and type of information available, to evaluate whether the approach was applicable more widely.
Test case 1 – British butterflies
We assessed all British butterfly species (N = 58 species) using published sources on the observed distribution and population trends since 1970s (Asher et al. 2001; Fox et al. 2006). Estimates of observed decreases in range size (Stage I) were calculated from 10-km-grid squares that were recorded as occupied in 1970–1982, but absent in 1995–2004; a reliable (high confidence) estimate of range decline because recording effort was higher in the latter period than the former. Increases in range size (Stage III) were the values taken from Fox et al. (2006) that included adjustment for variation in recording effort between time periods, and hence the published values were treated as having high confidence.
Published maps of projected future distributions from Settele et al. (2008), using projections for 2080 for their intermediate level of climate change (BAMBU – A2), were the basis for estimating potential decreases (Stage II) and increases (Stage IV) of species between ∼1980 and 2080 (comparing projected present and projected future distributions). Decreases projected over 100 years in Stage II were taken to be losses of >75% (score 3), 40–75% (score 2), 10–40% (score 1) and <10% (score 0) for the four decline rate categories. Critical values for the increase categories became: score 3 for >100% increase (as an incremental increase of 7·5% per decade for 10 decades corresponds to a 106% increase; i.e. [100 × 1·07510] − 100 = 106); 2 for 50–100% increase; 1 for 10–50%; 0 for <10% increase. The projected (modelled) increases and decreases of species in Britain were assessed from published maps (Settele et al. 2008). When working from published maps, a grid (e.g. GB 10-km National Grid) can be superimposed across the enlarged map, and the assessor can tally-up numbers of cells showing stasis, appearance and disappearance. Data from Fox et al. (2006) were used to assess observed increases (Stage III) over 24 years. Critical values for empirically observed increase categories were: three for >19% increase (incremental increase of 7·5% per decade for 2·4 decades corresponds to a 19% increase; i.e. 1·0752·4 =1·19); 2 for 10–19%; 1 for 2·4–10%; 0 for <2·4%.
The second set of species represents a non-random sample of species, and they are included simply to evaluate the feasibility of carrying out assessments for a wider range of taxa. Species were considered if (i) they were included in the UK MONARCH phase 3 project, which used bioclimate models to assess the potential future distributions of selected species within Britain (Walmsley et al. 2007), and (ii) if appropriate distribution data were also available from the UK’s National Biodiversity Network Gateway (http://www.nbn.org.uk/), which provides online access to species data (allowing us to separate areas of decline within the existing range, and expansion beyond it). Data were available for assessing three bat species, one beetle and 12 plant species. For all but one of these species, distribution changes were considered for a 20-year period of change between 1970–1989 and 1990–2009. For the Stag beetle Lucanus cervus, we analysed change between 1990–1999 and 2000–2009. We assigned critical values for declines in the same way as for butterflies.
We also included two British bird species for which recent population/abundance trends and climate-related research had been published (Beale et al. 2006; Pearce-Higgins et al. 2007). Other bird species were not considered because a new national distribution atlas is being compiled, and it seems more appropriate to wait for such data to become available before proceeding with further assessments. The time periods over which population and distribution trend data were considered were 1990–2002 for Black grouse Tetrao tetrix (Pearce-Higgins et al. 2007), and 1952–2004 (exact years varying among sites) for Ring ouzel Turdus torquatus (Beale et al. 2006). Projected distribution changes were taken from Huntley et al. (2007), for the HADCM3 GCM and the modest B2 emissions scenario (which is expected to result in slightly higher levels of warming than the B1 scenario). Information regarding habitat requirements, causes of decline and dispersal ability, was sourced from Gibbons, Reid & Chapman (1993) along with online species information pages provided by the British Trust for Ornithology (http://www.bto.org/birdfacts/indexa_all.htm) and the Royal Society for the Protection of Birds (http://www.rspb.org.uk/wildlife/birdguide/name/a/); both accessed in December 2009.
Because the above species represent a non-random sample (e.g. the two bird species are relatively northern species with respect to their current distributions within Britain), nothing should be read into the frequencies of species likely to suffer or benefit from climate change within Britain. In contrast, such conclusions are reasonable for the butterflies, for which we included all regularly breeding British species.
Test case results
Most British butterfly species are more widespread in the south than in northern Britain, and hence a warming climate might be expected to result increases in many species (Warren et al. 2001; Hill et al. 2002). The three (out of the 58) species assessed as being at high or very high risk of decline within their existing British ranges (scotch argus Erebia aethiops, mountain ringlet Erebia epiphron, northern brown argus Plebeius (Aricia) artaxerxes; Table 7) are all northerly and/or montane species. Three species (large heath Coenonympha tullia, pearl-bordered fritillary Boloria euphrosyne, small pearl-bordered fritillary B. selene) were classified as having Moderate threat within their existing distributions (with poor confidence in the assessments) based on modelled retreats rather than observed climate-related declines. The remaining 52 species were classified as low threat.
Table 7. Summary table of risks and benefits for British butterfly species. Bold indicates assessments with medium or good confidence (confidence in risks for Red and Amber species, confidence in benefits for Blue and Purple species, confidence in both risks and benefits for Green species). Names of the other genera are given in the Supporting Information
Nearly half of the species (24) were identified as having Moderate (e.g. purple emperor Apatura iris), High (e.g. silver-spotted skipper Hesperia comma) or Very High (e.g. brown argus Plebeius (Aricia) agestis) benefits from climate change. All but two of these species are southerly distributed; they are expanding already and/or can be expected to shift northwards with climate warming. The remaining 27 species were assessed as low threat and low benefit from climate; they are not projected to expand to any great extent beyond their existing ranges, and are not threatened within their existing ranges (Table 7).
Species that are classified as having low risk and low benefit fall into two groups. The first are those species that are already widespread in Britain and therefore have limited scope to expand. Just over half of the species are in this category; e.g. small heath Coenonympha pamphilus, green-veined white Pieris napi. The second group are southerly distributed species for which there is either little or no additional habitat/host plant likely to be available in the north (e.g. chalk-hill blue Polyommatus coridon), or they are unlikely to reach it without conservation intervention (e.g. heath fritillary Melitaea athalia). These species will probably increase within the regions where they currently occur but are unlikely to achieve major latitudinal range extensions; e.g. adonis blue Polyommatus bellargus is increasing locally, but it is not as yet extending its distribution at a broader geographical extent (Fox et al. 2006). Species of this type that do in future expand would be reclassified as benefiting in subsequent assessments.
Confidence levels for assessments of most species were higher for observed trends, based on existing monitoring, than they were for projected changes based on climate envelope models (Appendix 1). A total of 13 of the 58 study species had ‘poor’ confidence in the assessments of their declines, of which four species (C. tullia, B. euphrosyne, B. selene, E. aethiops) were classified as having Medium or High Risk from climate change, and hence they become priorities for further research and monitoring. Half of the species (29) had ‘poor’ confidence in assessments of increases; many were southern species for which there was uncertainty about future habitat availability in the areas of potential range expansion (and species’ abilities to colonise it), or where current rapid expansions were at odds with the longer-term projected climate suitability of Britain for these species. The confidence levels shown by the bold text in Tables 7 and 8 show the species for which there is medium or good confidence for the ‘relevant’ assessment (based on confidence in the risk for declining species and based on confidence in the benefits for expanding species).
Table 8. Summary table for exemplar species of other taxa. Bold indicates assessments with medium or good confidence (confidence in risks for Red and Amber species, confidence in benefits for Blue and Purple species, confidence in both risks and benefits for Green species). Names of the other genera are given in the Supporting Information
A higher proportion of the second set of species, which included more northern/montane species, was assessed as declining in response to climate change than was the case for the butterflies. There were four High Risk, two Medium Risk, four Low Risk & Benefit, four Medium Benefit and four High Benefit species (Table 8). All four species classified as having High climate-related risk have predominantly northern and/or montane distributions (e.g. ring ouzel T. torquatus, twinflower Linnaea borealis). The species showing potential to benefit are southerly distributed species, within Britain, that can be expected to expand their ranges northwards under a warmer climate (e.g. greater horseshoe bat Rhinolophus ferrumequinum, stag beetle L. cervus, broad-leaved cudweed Filago pyramidata).
Several threatened northern plants and birds that were classed as at risk were also assigned low confidence in their decline assessments (e.g. black grouse T. tetrix; oblong woodsia fern Woodsia elvensis), identifying the need for further monitoring and research. All of the species expected to increase their British ranges to a Moderate (four species) or Very High extent (four species; Table 8) were assigned poor confidence in this assessment. Uncertainty in assessments of the responses of these southern species mainly relates to whether they would be able to spread across human-dominated landscapes, and find suitable habitats to the north of their current distributions. Less complete data for plants, bats and the stag beetle meant that confidence in assessments for these other taxa were on average lower than for the butterflies.
It was possible to apply the climate change risk and benefit framework using existing published and web-based information, and this resulted in the identification of potential winners and losers from climate change. While our results apply only to the British distributions of these species, the framework should be applicable to much larger geographic areas, or to the entire distributions of individual species. The constraints mainly relate to data availability, but in that respect the current framework is little different from the requirements of current Red Listing procedures.
Once the lengthy process of developing the methodological approach had been completed, it was then possible to carry out the classification of individual species relatively quickly, generally taking about an hour per species. Given that the future distributions of all species had already been modelled and compilations of population and distribution trends were available, it was possible to analyse the 58 butterfly species in less than 2 weeks. Had we started from a situation whereby population and trend data had to be assembled and analysed from raw values, the procedure would have taken longer. However, such data need to be assembled for other types of conservation assessment (e.g. red listing, species action plans) that are already carried out, so we do not regard this as an additional requirement, provided that this risk and benefit framework is used in parallel with other existing arrangements. Of course, the time taken would be much larger were it necessary to model the possible future distribution of each species from scratch. However, this is becoming increasingly quick and feasible, so may not be a major constraint in the near future. The time and resources required to carry out these assessments so as to identify priorities appears to be modest relative to those required for conservation itself.
The framework is expected to be most robust for those species with good observed trends data and tested model projections, but it was also possible to apply when certain types of information were lacking. The wood white butterfly Leptidea sinapis was not distinguished until recently from its close congener, L. reali. As a result, the European distributions of the two species are not known, and future projections are not available for the two species separately (Settele et al. 2008). However, it was still possible to make a preliminary assessment for L. sinapis (the only one of the two species present in Britain) based on observed trends within Britain (Appendix 1). For the plant Norwegian mugwort Artemisia norvegica, observed trend data were not available from within its existing British range, and it was too localised to assess any decline based on 10-km-grid resolution squares. Hence, the assessment for this species ignored Stage I (Appendix 1).
Globally, the main issue is likely to be a lack of suitable trend data. IUCN Red Listing commonly classifies low density or localised species as threatened, based on distribution or population size criteria, in the absence of trend data (Mace et al. 2008), whereas the approach we have outlined requires either observed or projected trend data (preferably both). Simple projections of expected declines and expansions could be based on the elevational or latitudinal ranges of species in the first instance (Raxworthy et al. 2008; Sekercioglu et al. 2008), and it is possible to assess the amount of climate-related vegetation change expected within the distribution of any species, based on the output of global vegetation models (Malcolm et al. 2006; Jetz, Wilcove & Dobson 2007). It is also increasingly feasible to develop acceptable distribution models for taxa and regions for which data are sparse, provided that at least some accurate ‘presence data’ are available (Elith et al. 2006; Kremen et al. 2008).
There were very few unexpected outcomes in the classification of butterfly species, indicating that the framework delivers intuitively sensible assessments (as judged by experts on the relevant taxa), when there are adequate data. Initial concerns whether the quality of data and of projections would be adequate to carry out any kind of assessment were not realised.
Evaluating the degree of uncertainty in the assessments of benefits and risks in the framework should help to identify future monitoring and research needs. Two fritillary butterflies, B. euphrosyne and B. selene, were placed in the Medium Risk category, which seemed surprising, given their relatively wide British distributions. This risk classification depended on future climate change scenarios rather than on currently observed trends, and confidence in the assessment is poor. Such unexpected outcomes would indicate that further investigation is required, to identify whether they are truly threatened.
Uncertainty often arose because of the absence, for most species, of analyses that adequately apportion changes in population and distribution between climate change and other drivers; this uncertainty is likely to remain for some time to come, not least because the effects of different drivers often combine and interact. However, the uncertainty would in most cases only have shifted a species to an adjacent risk category, rather than switching it from ‘at risk’ to ‘benefiting’ or vice versa. From a conservation perspective, this broader definition of uncertainty may be more useful than certainty over assigning a species to one of our narrower categories.
We did not consider uncertainty associated with different emissions scenarios, General Circulation Models or distribution (bioclimate/niche) models (e.g. Diniz et al. 2009; Buisson et al. 2010). Low certainty would be ascribed when only one scenario is considered, or when a species is predicted to increase for some scenarios/models and decrease for others; in which case, projection information would likely be discarded in favour of empirical observation. Uncertainty in projections will remain high for the foreseeable future because the future climate is uncertain, whereas data for already observed changes should improve; we expect assessments to be weighted towards observed trends as time proceeds.
Establishment (or continuation) of monitoring schemes that adequately sample environmental gradients is likely to represent the most effective means of reducing future uncertainty, so as to (i) improve observational data (Stages I and III), and (ii) be available to test the abilities of alternative models to predict changes that have already been observed (Stages II and IV). In addition, the development and adoption of robust methods to account for differences in recording effort between surveys will help to reduce uncertainty in observed trends. This may be especially relevant for global datasets with large spatial and temporal variation in recorder effort (e.g. GBIF, http://www.gbif.org/).
The whole procedure requires the input of expert opinion, especially at the beginning and end of the assessments. During the initial stages, advice is needed on which sources of potential data are available and most robust, and should be used as the primary inputs to the assessments. Expert opinion is also likely to be important in evaluating the factors and species attributes that are likely to be important ‘exacerbating factors’ for a particular group, and defining appropriate levels for them. Defining exacerbating factors more robustly is an area for further consideration in future iterations of the framework. Once the framework has been applied using available data, both the risk assessments and the confidence assignations should be reviewed by a panel of experts comprised of people with knowledge of the taxa involved, as well as those experienced in climate change projections. Normally, any recommended changes would be achieved by the expert-review panel agreeing to change values within Stages I–IV (e.g. recognising additional constraints, or taking a different view on the confidence assigned to projections), with the altered final designations contingent on these decisions.
In any framework, a number of the decisions have to be made for pragmatic reasons – when an alternative would have been possible. For example, we used a linear decline rate (relative to the original distribution or population size) but an incremental expansion rate (change relative to the status at the beginning of each successive decade), to avoid classifying very rare species that expand as showing extremely high percentage increases. However, other authors might in future choose to evaluate the consequences of alternative ways of assessing expansion and decline rates. For the 58 British butterflies, we did carry out an additional analysis using percentage expansions in relation to the original range size of the species. This alternative way of measuring the rate of increase resulted in the benefit category being changed for two species at Stage III and for five different species at Stage IV (in each case moving species from High to Very High benefit). However, for no species did this alter the final categories into which species were placed in Stages V and VI. Therefore, this detail did not affect the classifications of the species we considered.
The time-scale over which declines and increases are measured can also be debated. The IUCN (2001, 2010) have taken the view that different time-scales are appropriate for species with different generation times; for example, defining as ‘Vulnerable’ species showing a decline of over 10% per decade, or 10% in three generations (to a maximum duration of 100 years considered), whichever is the longer. For long-lived species, such as trees that might be able to survive without recruitment for several 100 years in climatically unsuitable regions, it might be worth considering even longer time-scales than we have in this paper. On the other hand, it could be argued that such species are genuinely not so endangered, and that there is a longer period available in which to ensure their conservation. Because we already consider changes over periods of up to 130 years (1970–2100), which is longer than any time period currently considered by IUCN (see Methods; Stage I description), we do not think that there is any great benefit in scaling changes by generation time in the current context.
None of the species considered showed both climate-related threats within its existing range and climate-related benefits (Tables 7 and 8). This pattern will be typical for analyses that are conducted for regions of similar geographic area to Britain, where species at the climatic margins of their geographic distributions will tend either to decline (montane/northern species) or to increase (lowland/southern species). At a continental scale, the majority of species might experience both threats (declines at their low elevation/latitude range boundaries) and benefits (expansion at their high elevation/latitude boundaries) from climate change (Parmesan et al. 1999). For example, the silver-washed fritillary Argynnis paphia and gatekeeper Pyronia tithonus butterflies are expanding at the cool edges of their ranges in Britain as a result of climate change, and they are shown as benefiting from climate change within Britain in Table 7; but they are simultaneously retreating at the hot edges of their distributions in Spain (Wilson et al. 2005). Likewise, analyses for regions that are more topographically diverse than Britain will also have more species with both threats (at low elevation) and benefits (expansion to higher elevations) from climate.
Overall, the framework we present here proved to be practical. Despite the many uncertainties, the broad classification of species’ prospects appears sufficiently robust to consider the implications for conservation.
Implications for conservation
We regard the approach as a first step towards identifying species that could benefit from a range of adaptation measures. Climate change represents an additional potential threat to species within their existing distributions, but it also provides new opportunities for the conservation of species beyond their current ranges.
The species assessed as High or Very High Risk are all northern/montane species, which are expected to decline within their existing ranges and have little or no capacity to expand elsewhere within Britain (Tables 7 and 8). Whether conservation effort is worthwhile for such species will depend on the species’ status elsewhere, on whether British genotypes/lineages are unique, and hence whether Britain is important to a species’ European and global conservation. The only feasible in situ conservation measures for these species would be to concentrate efforts (e.g. reduce or remove other threats; maximise habitat quality; enhance habitat heterogeneity, Oliver et al. 2010) in those parts of Britain and microhabitats (e.g. cool, north-facing cliffs at high elevation) that are expected to remain climatically suitable for longest. The extent to which declining northern species are affected by change climate vs. other drivers (e.g. land use change) differs among species (Franco et al. 2006), and so the practicality of alleviating other pressures will vary. Although translocation to Scandinavia is a further possibility, most (but not all) of the species that are restricted to northern parts of Britain are already found there.
The majority of species that have southern distributions within Britain are classified as High or Medium Benefit species (Purple and Blue; Tables 7 and 8). Continuation of existing conservation management (if required for other reasons) may be sufficient for most of these species. However, monitoring is appropriate, especially for species with poor confidence assessments. New conservation opportunities may exist to increase the speed of expansion, and this would be desirable for species that are declining elsewhere in their ranges, for example, in southern Europe (Parmesan et al. 1999; Wilson et al. 2005). This last point identifies the need to complement national assessments with broader assessments of the species’ distributions (in this case throughout Europe, at least).
The priority species to target for enhancing range expansion are likely to be those which: (i) are otherwise rare, threatened or even regionally extinct, for which climate change provides new opportunities (e.g. potential to reintroduce the black-veined white Aporia crataegi to Britain; Carroll et al. 2009), (ii) are known or predicted to decline further south in Europe (e.g. Dartford warbler Sylvia undata; Wotton et al. 2009), (iii) have rates of expansion that are substantially reduced by exacerbating factors that could potentially be lessened through conservation measures, and (iv) show modelled rates of expansion (Stage IV) that are much faster than those observed (Stage III), again implying that there may be additional constraints that could be addressed through conservation actions (e.g. Glanville fritillary Melitaea cinxia).
The same priorities apply to species at Low risk from climate, but with Low benefits (top-right cells in Tables 6–8), as some of these species were classified as Low benefit from climate change only because negative exacerbating factors prevent unaided expansions. Actions that may promote range expansions of these species (Hopkins et al. 2007; Huntley 2007) may include (i) management of existing populations/habitats to increase the number of individuals available to colonise new sites, (ii) combined with increasing the quantity and quality of target habitats (e.g. Self 2005), (iii) managing non-breeding habitats to increase the likelihood that colonists will successfully reach new breeding habitats and, (iv), translocating individuals to sites beyond the current colonisation range (e.g. Hoegh-Guldberg et al. 2008; Hodgson et al. 2009; Willis et al. 2009b).
None of the species we assessed fell into categories where there were both substantial (Moderate, High, Very High) risks and equivalent levels of benefits (bright green diagonal in Table 6). A species with Very High risks and Very High benefits would likely require closer monitoring and potentially more conservation intervention than a species with lower risks and benefits, even if the overall net status of both species was expected to be similar; for example, to ensure that projected increases beyond the existing range were realised.
Particular types of conservation option tend to follow from where a species falls within the matrix of threats and benefits, but which options are actually prioritised will depend on a variety of additional factors associated with the species, habitat type and region under consideration. It is feasible to alleviate non-climatic pressures for some northern/montane species, but not for others. New habitat can be created to facilitate the expansions of some southern species, but may be prohibitively expensive for species with exacting requirements. Hence, no single ‘best’ conservation option necessarily applies to all species with a given threat/benefit combination, nor necessarily to all areas of the range for any given species. Therefore, the framework should be used to inform conservation strategies and help develop priorities, but additional information will be required to identify specific conservation actions, on a case by case basis.
In summary, classification of species according to their potential future conservation status under a changing climate is feasible, based on observed trend data and/or modelled projections. Even when data are limited and confidence in the assessments is low, the assessment process is useful because it helps to identify knowledge gaps. It can inform monitoring of climate change impacts, development of adaptation measures for habitat and species conservation and help identify those species that will be beneficiaries of wider landscape-scale conservation actions.
The approach reported here was developed during workshops supported by UKPopNet (funded by NERC and Natural England). We thank Keith Kirby, Roger Street and Richard Smithers for their contributions to the workshops, and Wendy Foden, Keith Kirby, Georgina Mace, Wilfried Thuiller and an anonymous referee for their helpful comments on the manuscript. We also thank the thousands of volunteers who recorded species distributions and abundances; on which the analyses are based.