Identifying the World's Most Climate Change Vulnerable Species: A Systematic Trait-Based Assessment of all Birds, Amphibians and Corals

Climate change will have far-reaching impacts on biodiversity, including increasing extinction rates. Current approaches to quantifying such impacts focus on measuring exposure to climatic change and largely ignore the biological differences between species that may significantly increase or reduce their vulnerability. To address this, we present a framework for assessing three dimensions of climate change vulnerability, namely sensitivity, exposure and adaptive capacity; this draws on species’ biological traits and their modeled exposure to projected climatic changes. In the largest such assessment to date, we applied this approach to each of the world’s birds, amphibians and corals (16,857 species). The resulting assessments identify the species with greatest relative vulnerability to climate change and the geographic areas in which they are concentrated, including the Amazon basin for amphibians and birds, and the central Indo-west Pacific (Coral Triangle) for corals. We found that high concentration areas for species with traits conferring highest sensitivity and lowest adaptive capacity differ from those of highly exposed species, and we identify areas where exposure-based assessments alone may over or under-estimate climate change impacts. We found that 608–851 bird (6–9%), 670–933 amphibian (11–15%), and 47–73 coral species (6–9%) are both highly climate change vulnerable and already threatened with extinction on the IUCN Red List. The remaining highly climate change vulnerable species represent new priorities for conservation. Fewer species are highly climate change vulnerable under lower IPCC SRES emissions scenarios, indicating that reducing greenhouse emissions will reduce climate change driven extinctions. Our study answers the growing call for a more biologically and ecologically inclusive approach to assessing climate change vulnerability. By facilitating independent assessment of the three dimensions of climate change vulnerability, our approach can be used to devise species and area-specific conservation interventions and indices. The priorities we identify will strengthen global strategies to mitigate climate change impacts.

Vulnerability or vulnerability frameworks are used in a variety of contexts to support decision making to address risks to the most vulnerable members of a system to change [1].
The three dimensions that make up vulnerability are usually described as sensitivity (the lack of potential for a species to persist in situ), exposure (the extent to which each species' physical environment will change) and low adaptive capacity (a species' inability to avoid the negative impacts of climate change through dispersal and/or micro-evolutionary change).
This general idea has been suggested for assessing species risks from climate change [2] and is used in some case studies (e.g., [3][4][5]).
A body of work already exists on biological traits associated with vulnerability to extinction due to historic threatening processes [6][7][8][9]. Because climate change poses a new threat, there is little empirical information with which to assess vulnerability, except in a limited way for species that were exposed to relatively rapid climate shifts during the quaternary [10] and rare case studies [11][12][13]. Instead, we used literature review, expert opinion based on expectations from ecological and evolutionary theory to identify traits associated with each dimension of climate change vulnerability (hereafter referred to simply as vulnerability).

Determining the trait sets
Through two workshops and various other consultations, we gathered input from over 30 scientists whose collective expertise covers a broad range of taxonomic groups and ecosystems (see Supporting Discussion, 'Caveats and uncertainty', point 1). Together with extensive literature survey, this process identified more than 90 biological traits that may be associated with species' vulnerability to climate change. The traits were consolidated, firstly according to the three dimensions of vulnerability i.e., sensitivity, exposure and low adaptive capacity, and subsequently into five 'trait sets' for sensitivity, a variable number for exposure, and two for low adaptive capacity, as outlined below.
In the vulnerability framework context, sensitivity is regarded as the lack of potential for a species to persist in situ). Here we describe five components of sensitivity, termed 'trait groups' (adapted from [14]).
A. Specialised habitat and/or microhabitat requirements: Across many studies of both animals and plants, threatened and declining species include a disproportionate number of specialists compared to generalists and of species with extensive geographic ranges [15].
Under a changing climate, most species are likely to face changes in their habitats and microhabitats and those less tightly coupled to specific conditions and requirements are likely to be more resilient. Sensitivity is increased where a species has several life stages, each with different habitat or microhabitat requirements (e.g. water-dependent larval amphibians), or when the habitat or microhabitat to which the species is specialized is particularly vulnerable to climate change impacts (e.g. mangroves, cloud forests or polar habitats). However, in some cases (e.g. deep sea fishes), extreme specialization may allow species to escape the full impacts of competition from native or invading species, so the interaction of such traits with climate change must be considered carefully for each species group assessed. This trait group is not independent of species' low adaptive capacity as habitat and/or microhabitat specialisation also decreases the chances of successful colonisation if species are able to disperse to new climatically suitable areas, (e.g., plants confined to limestone outcrops; cave-roosting bats).
B. Narrow environmental tolerances or thresholds that are likely to be exceeded due to climate change at any stage in the life cycle: The physiology and ecology of many species is tightly coupled to very specific ranges of climatic variables such as temperature, precipitation, pH and carbon dioxide levels, and those with narrow tolerance ranges are particularly vulnerable to climate [16]. Even species with broad environmental tolerances and unspecialized habitat requirements may already be close to thresholds beyond which ecological or physiological function quickly breaks down (e.g., photosynthesis in plants; protein and enzyme function in animals).
C. Dependence on a specific environmental trigger that is likely to be disrupted by climate change: Many species rely on environmental triggers or cues for migration, breeding, egg laying, seed germination, hibernation, spring emergence, and a range of other essential processes. While some cues such as day length and lunar cycles will be unaffected by climate change, others such as rainfall and temperature (including their interacting and cumulative effects) may be severely impacted. Species tend to become vulnerable to changes in the magnitude and timing of these cues when this leads to an uncoupling with resources or other essential ecological processes e.g., early spring warming causes the emergence of a species before its food sources are available. Climate change vulnerability is compounded when different stages of a species' life history or different sexes rely on different cues. D. Dependence on interspecific interactions which are likely to be disrupted by climate change: Many species' interactions with prey, hosts, symbionts, pathogens and competitors will be affected by climate change, either due to the decline or loss of these resource species from the dependent species' ranges or loss of synchronization in phenology. Species dependent on interactions that are vulnerable to disruption by climate change are at risk of extinction, particularly where they have high degree of specialization for the particular resource species and are unlikely to be able to switch to or substitute other species.

E. Rarity:
The inherent vulnerability of small populations to allee effects and catastrophic events, as well as their generally reduced capacity to recover quickly following local extinction events, suggest that many rare species will face greater impacts from climate change than more common and/or widespread species. We consider rare species to be those with small population sizes and those that may be abundant but are geographically highly restricted. In cases where only a small proportion of individuals reproduce (e.g., species with polygynous or polyandrous breeding systems or skewed sex ratios), we use an estimate of effective population size to assess species' rarity, and where species are known to be declining or subject to extreme (greater than ten-fold) fluctuations in population size, we set less conservative population size thresholds. Similarly, thresholds of larger population sizes were used for species with congregatory breeding systems, since they are more likely to experience catastrophic population declines.

Exposure
These measures reflect the climate change driven environmental pressures on species, based on their geographic locations. For the main results of our study, we consider projected changes in four pressures by 2050, though other pressures, their combinations and alternative time frames could also be used.
A. Sea level rise: Although no global projections of sea level rise are currently available, regional models or surrogate measures such as occurrence in coastal habitat types can be used to assess species' likelihood of threat from sea water inundation due to rising sea levels.
B. Temperature change: Projections of temperature change are typically based on General Circulation Model outputs and interpreted based on the ecosystem occupied by the focal species group (e.g., air temperature for amphibians, sea surface temperature for corals).
Biologically relevant components of temperature change typically include changes in means, variability and/or extremes (magnitude and frequency).

C. Precipitation change:
As for temperature changes, these are typically based on General Circulation Model outputs and biologically relevant components may include changes in means, variability and/or extremes (magnitude and frequency).

D. Elevated atmospheric CO 2 impacts:
While not strictly a climate change phenomenon, we consider this otherwise overlooked potential threat in the general suit of climatechange related impacts. Both direct impacts of elevated CO 2 levels and resulting ocean acidification (e.g., on corals), and indirect impacts (e.g., through changes in competitive relationships between C 3 and C 4 plants) should be considered. Aquatic species are likely to be affected by increased CO 2 absorption by water bodies, the effects of which are projected to be particularly marked in marine ecosystems where ocean acidification and the resulting lowering of calcite and aragonite saturation levels lead to reduced growth and dissolution of organisms with calcium-carbonate exoskeletons or plates, including corals, coccolithophore algae, coralline algae, foraminifera, shellfish and pteropods [17].

Low adaptive capacity
This set of traits reflects the extent to which species have the capacity to reduce the impacts of changes in their immediate environment through dispersal or adaptive change. We define two low adaptive capacity 'trait groups' (adapted from [14]): A. Poor dispersability: In general, the particular set of environmental conditions to which each species is adapted will shift to increasing latitudes and altitudes in response to climate change. Species with low rates or short distances of dispersal (e.g., land snails, ant and rain drop splash dispersed plants) are unlikely to migrate fast enough to keep up with these shifting climatic envelopes and will face increasing extinction risk as their habitats become exposed to progressively greater climatic changes. Even where species could disperse to newly suitable areas, extrinsic barriers may decrease changes of dispersal success. Dispersal barriers may be geographic features such as unsuitable elevations (e.g., species confined to mountain ranges), oceans (e.g., for species on small islands or at the polar tip of a land mass), rivers, and for marine species, ocean currents and temperature gradients; unsuitable habitats and/or anthropogenic transformation may also act as dispersal barriers for habitat specialised species. In this context we describe species as having dispersal barriers both when suitable areas exist but extrinsic factors make them unlikely to reach them, as well as when no newly suitable areas are likely to exist (e.g., for polar species).
B. Poor evolvability: Species' potential for rapid genetic change will determine whether they will be able to undergo evolutionary adaptation at a rate sufficient to keep up with climate driven changes to their environments. Species with low genetic diversity, often indicated by recent bottlenecks in population numbers, potentially face inbreeding depression and generally exhibit lower ranges of both phenotypic and genotypic variation.
As a result, such species tend to have fewer novel characteristics that could facilitate adaptation to the new climatic conditions. Where they exist, direct measures of genetic variability can be supplemented with information on naturalization outside species' native ranges and on the success of any past translocation efforts. Indirect measures of evolvability relate to the speed and output of reproduction and hence the rate at which advantageous novel genotypes could accumulate in populations and species [18].
Evidence suggests that evolutionary adaptation is possible in relatively short time frames (e.g. 5 to 30 years [19]) but for most species with long life cycles (e.g., large animals and many perennial plants), such adaptation is unlikely to keep up with the rate of climate driven changes to their environments.

Selecting appropriate traits and assigning scores
Guided by the trait groups described above, we conducted a second round of expert consultation and through consensus we compiled biological, ecological, physiological and environmental traits that are pertinent for assessing the particular climate change vulnerability of each taxonomic group. The traits selected for birds, amphibians and corals are shown in Tables S1-3 and are discussed in the next section; except for sensitivity trait group C (dependence on environmental triggers or cues) for birds, we were able to represent each of the trait groups with at least one trait for each taxonomic group. Challenges in selecting traits included balancing selection of the most theoretically sound traits with the practicalities of data availability and collection. A further challenge was defining traits in objective and replicable ways and, as far as possible, developing quantitative measures for them.
Species were assigned scores of 'high', 'low/lower' or 'unknown' for each trait, based on a broad range of information sources (discussed below). While in some cases, thresholds of extinction risk were clear (e.g., 'occurs only on mountain tops'), in most cases there is no a priori basis for setting a particular extinction risk threshold. For such traits (e.g., projected temperature change exposure), we arbitrarily selected a threshold of the worst affected 25% of species; those ranked in this group were scored 'high', while the remaining species were assigned scores of 'lower', or 'unknown' where data were lacking. Data on, for example, population sizes, temperature-tolerance thresholds and inter-species interactions, were particularly sparse, necessitating frequent scores of 'unknown' for corresponding species. In some cases where empirical data were unavailable, experts were able to provide information either from unpublished data, their own field knowledge or, where justified, through inference from similar species. For our study, measures of experts' confidence in the data were recorded in most cases, and data that were regarded as particularly uncertain were treated as 'unknown' values in subsequent stages of assessment.
To qualify as highly vulnerable overall, species must have high scores for all three of vulnerability dimensions of sensitivity, exposure and low adaptive capacity. A species scored high under sensitivity if any of the several biological traits in sensitivity trait groups scored high; similarly low adaptive capacity and high exposure were triggered if any single trait in these groups was listed as high (see Fig. S13 for a schematic summary of the logic used to assign species' scores). Uncertainty at the level of unknown trait data is accounted for by calculating scores assuming all unknowns represent high scores (pessimistic scenario) and as 'not high' scores (optimistic scenario) and presenting overall vulnerability results as the range of possible values between these extremes.

Taxonomy and baseline databases
The list of bird species followed BirdLife International (2008), as used by the 2008 IUCN Red List. For amphibians, we followed the taxonomy in Amphibian Species of the World (http://research.amnh.org/vz/herpetology/amphibia/, 2008). Coral species lists were based on the warm-water reef-building corals assessed for the Global Marine Conservation Assessment [20] (obtained from the IUCN Red List), but we excluded 46 species due to unresolved taxonomic problems, and incorporated taxonomic updates to 2010. Although not intended to be a definitive taxonomic source, the IUCN Red List strives to be taxonomically coherent and consistent at all ranks. Higher-level classification follows accepted classifications, but deviates in some respects; further information is available at http://www.iucnredlist.org/technical-documents/information-sources-and-quality. The IUCN Red List (http://www.iucnredlist.org/), BirdLife International's World Bird Database, and AmphibiaWeb (http://amphibiaweb.org/) provided essential information such as distribution maps, habitats and threats, and additional information was gathered from published and unpublished data, online resources, literature and expert knowledge. Where justifiable, we addressed data gaps with experts' inferences and assumptions, though many remain.

Preparing maps of species' distribution ranges
Bioclimatic modelling traditionally relies on the availability of detailed information on points of occurrence (and ideally absence) to 'train' statistical models about focal species' climatic 'requirements' or correlates. The intensive data requirements of these methods limit their application to few taxa and geographical regions, and prevent systematic global-scale assessments. Instead we derive an estimation of species' exposure to climate change by calculating simple metrics of climatic change across refined species' ranges. Species' ranges for birds, amphibians and warm-water reef-building corals have been mapped by experts as part of IUCN Red List assessments and are available at http://www.iucnredlist.org/technicaldocuments/spatial-data; range polygons were available for 81% of birds, 98% of amphibians and 99.9% of corals at the time this component of our analyses was carried out. Range polygons were compiled from a combination of known localities and extrapolation of areas within them that have been assessed by experts as suitable. They represent best estimates of each species' current limits within its historical native range (any introductions are coded accordingly and were excluded from this analysis), but we note that some species will almost certainly occur more or less widely than mapped. Understudied regions include the Andes, most of Central Africa, parts of West Africa, Angola, parts of South and Southeast Asia, and Melanesia [21]. As a result, the biodiversity and potential climate change vulnerability in these regions will be under represented in this study. We also note that, although some of our analyses assume homogeneity of species within distribution ranges, this is unlikely to be the case for most species.
Because IUCN Red List range maps are often generalised polygons, they frequently represent species' Extents of Occurrence (calculated by drawing a polygon around all known places that a species occurs) and thus may include areas not actually occupied by the species and for which climate projections differ. For example, a range polygon may have been drawn around a lowland amphibian's occupied range on either side of a mountain range, or similarly around a coral's range on either side of an ocean. To refine species' ranges for our assessments, we excluded unsuitable within-range habitats and, for terrestrial species, elevations in which the species is known not to occur.
To facilitate processing of the large volume of range data involved, we rasterised range maps at a resolution of 10 minutes (~20x20 km); this is believed to be the scale at which the poorest resolution maps are reliable for each of the three taxonomic groups assessed. A species was regarded as 'present' in a 10 minute grid cell if any part of the underlying range polygon was occupied. For corals, areas of unsuitable habitat were defined as those where any 10 minute grid cell failed to intersected with a coral reef, as defined by ReefBase's global dataset of coral reef locations (www.reefbase.org).
For birds and amphibians, this process was more complex. Habitat affiliations (based on 126 IUCN Red List (2009) habitat categories which include natural and human-transformed habitats in terrestrial, freshwater and marine ecosystems) were obtained from the IUCN Red List database and BirdLife's World Bird Database (and are based on published literature and experts' knowledge), but as these habitat types are not spatially explicit, we cross-referenced them to the Global Land Cover 2000 habitat types (23 categories, including natural and human-transformed habitats and water bodies at1x1 km resolution; http://ies.jrc.ec.europa.eu/global-land-cover-2000), as guided by available literature [22,23].
We aimed to remove only habitats for which we had high confidence of their unsuitability, so we included all expert-listed IUCN habitat types (i.e. those described as 'suitable', 'marginal', or of unknown suitability) and removed none where habitat preferences are not known. We cross-referenced each IUCN habitat type with any potentially similar Global Land Cover 2000 habitat types (e.g., any IUCN forest type triggered all Global Land Cover 2000 forest types). The 1x1 km Global Land Cover 2000 was rasterised into twenty-three 10 minute grids, each representing one of the Global Land Cover 2000 types. For each grid, cells' values represented the percentage of the underlying 1x1 km vector covered by the land cover type in question. The probability of the presence of suitable habitat in each cell of a species' range was calculated as the sum of the percentage presence of all such suitable habitat types; again following a conservative approach, we excluded only cells with zero probability of suitable habitat.
To exclude areas with unsuitable elevations for terrestrial species' ranges, we again used IUCN Red List information on species' individual elevation preferences, comparing these with the U.S. Geological Survey's GTOPO30 global digital elevation model (http://eros.usgs.gov/#/Find_Data/Products_and_Data_Available/gtopo30_info). Species' elevation ranges were buffered to a minimum of 100 metres, and for those with unknown ranges, all elevations were included. The 1x1 km GTOPO30 elevation dataset was rasterised to two 10 minute grids, one containing the maximum elevation and one the minimum value in the underlying vector data. To determine elevation suitability in the cell, we calculated the extent to which each species' elevation range lies between the minimum and maximum elevation for the cell; following the conservative approach, we excluded from species' ranges only cells with no overlap between the species' and cell's elevation ranges.

Birds
The traits, biological information and thresholds used to assess birds' vulnerability are summarised in Table S1.

Sensitivity
The degree of birds' habitat and microhabitat specialisation was estimated using three traits.
The first, termed habitat specialisation, is based on the number of habitats listed in the IUCN Red List (2009) as of major importance (defined as where the habitat is suitable and furthermore is important for the survival of the species, either because it has an absolute requirement for the habitat at some point in its life cycle e.g., for breeding or as a critical food source, or it is the primary habitat or one of two primary habitats within which the species usually occurs or within which most individuals occur), suitable (the species occurs in the habitat regularly or frequently), or as marginal (the species occurs in the habitat only irregularly or infrequently, or only a small proportion of individuals are found in the habitat).
Birds for which only a single habitat was recorded were assessed as of high sensitivity.
Secondly, species were considered dependent on specialised microhabitats if they have a particular requirement for one or more of the following microhabitats: bamboo, vines, tree falls, dead wood, tree hollows [24][25][26][27], rocky outcrops in forests, caves, streams and bromeliads. Lastly, as more detailed data were available on species' dependence on forest specifically, they were assessed as of high sensitivity if they are unable to tolerate forest disturbance. Intolerance was categorised as high for forest specialists characteristic of the interior of undisturbed forest, but that may persist in secondary forest and forest patches if their particular ecological requirements are met. Where such species do occur away from the interior, they are usually less common and are rarely seen in non-forest habitats, and breeding is almost invariably within forest. Species with 'Medium' intolerance are forest generalists that may occur in undisturbed forest but are also regularly found in forest strips, edges and gaps, and tend to be commoner in such situations and in secondary forest than in the interior of intact forest, and breeding is typically within forest. Species with 'Low' intolerance are often recorded in forest, but are not dependent on it, and are almost always more common in non-forest habitats where most individuals breed. Intolerance was coded as unknown for species that occur or probably occur in forest but for which their degree of dependency on it is unknown.
Because empirical evidence of bird species' environmental tolerances is sparse, we use the range of historical temperatures and precipitation levels tolerated by the species across its historical range as a proxy. Based on the Worldclim global dataset's 1950-2000 monthly means for terrestrial areas (excluding Antarctica) at 10 minute resolution [28] (http://www.worldclim.org), we calculated the average absolute deviation across all cells in each species' refined range, for each of the 1975 (mean 1950-2000) monthly means, producing two measures, one for precipitation and one for temperature, that represent tolerance of variability both seasonally and spatially. The average absolute deviation (AAD) is a summary statistic of dispersion, and, for a data set {x1, x2, ..., xn}, AAD is defined [29] as: In our calculations, each x represents a monthly mean for a cell in a species' refined range.
Species were ranked according to their AAD scores and the 25% with the narrowest values for temperature and/or precipitation were regarded as of highest sensitivity.
In a minority of cases (3.9% of birds e.g., for marine and Antarctic birds and some small island species), less than 80% of species' ranges were covered by the Worldclim dataset. As a result, we also calculated all species' rankings based on modelled HadCM3 global projections (supplied by the U.K. Meteorology Office) for 1975 (mean 1961-1990), downscaled to 10 minutes using a cubic spline; for species with <80% of their ranges covered above, exposure scores were based on these rankings, and thresholds for AAD temperature and precipitation were 1.24 o C and 44.02 mm respectively.
Species with high dependence on very few (typically <5) species of ants, termites, insects, bees or wasps were assessed as having high vulnerability to potentially declining positive interactions with other species. Rare species were defined as those with estimated total population sizes (from BirdLife's World Bird Database) of fewer than 10,000 individuals, or those where the total population size numbers less than 20,000 and sensitivity to threatening processes is heightened due to skewed sex ratios (males to female ratio of ≤0.4 or ≥0.6), polygynous or polyandrous breeding systems, cooperative breeding systems, or declining or extremely fluctuating populations (fluctuations >10-fold). Cooperative breeding systems include lekking, as well as those that regularly or seasonally congregate at particular sites, and then disperse over a wide area. It also includes species that breed colonially (e.g., Southern Royal Albatross), congregate during migration (e.g., European Honey-buzzard) or during the non-breeding season (e.g., Snow Goose). At least 1% of the global population must be found at one or more sites to qualify, and hence this excludes species that congregate to breed, feed or move in numbers that are small relative to the global population (e.g., Little Swift).

Exposure
Since no global projections of sea level rise are available, we used habitat types as a proxy for high exposure to sea level rise impacts. Mangroves, intertidal salt marshes, coastal freshwater, brackish or saline lakes and lagoons, marine lakes, coastal caves, intertidal shorelines (including rocks, beaches, flats and tide pools), sea cliffs, rocky offshore islands, and coastal sand dunes were regarded as at high risk from sea level rise. Species were considered to have high exposure if they occur exclusively in one or more of these habitats (with the habitat ranked as suitable or of major importance for the species), or in these and only one other habitat.
To estimate which species will be most exposed to future changes in temperature, we calculated, firstly, the absolute difference between mean projected historical temperatures across each species ' range in 1975 (1961-1990 average of mean annual temperature) and the mean projected temperature across the same range for 2050 (2041-2060 average of mean annual temperature). Secondly, to incorporate projected changes in temperature variability, we calculated the absolute difference in projected AAD (i.e. a measure of variability across all cells and months; see above section on determining environmental tolerances to assess sensitivity for birds for details) between 1975 (based on mean monthly temperatures from   match with the fine-scale coastal boundaries used to map species. Additionally, the dataset above does not include marine areas or Antarctica. We found that 10.4% of birds, namely those restricted to small oceanic islands, with narrow coastal distributions, in Antarctica or with largely marine ranges, had less than 80% of their ranges covered by the above GCM model ensemble. As a result, we also calculated all species' rankings based on HadCM3 projections for scenario A1B (supplied by the U.K. Meteorology Office) for 1975 (mean 1961-1990) and 2050 (mean 2046-2055), downscaled to 10 minutes using a cubic spline, which cover all land and marine areas. Exposure scores for the species with <80% overlap with the GCM ensemble projections were based on these rankings. Temperature and precipitation thresholds for all of the above exposure measures are shown in Table S1.

Low adaptive capacity
We estimated bird species' intrinsic dispersal ability using published or unpublished data on known mean and maximum dispersal (usually from studies involving ringing or marking nestlings and then recording the distance to where they first breed). Estimates were placed in logarithmic bands, and extrapolated from close relatives where no direct estimates were available. According to Malcolm et al. [32], required migration rates of ≥1 km per year were relatively common in all models if species were to remain within their bioclimatic envelopes, so we selected a threshold of 1 km per year, below which species were considered to have low adaptive capacity. To include species whose climate change driven migration might be extrinsically limited by dispersal barriers, we assigned low adaptive capacity scores to species with distribution ranges entirely within approximately 2000 m of a mountain-top or described as having "mountain-top" distribution, on small islands with maximum altitudes <500 m, and those with ranges within c.10º latitude from the polar edge of a land mass and within which ≥20% of current vegetation type is projected to disappear under doubled CO 2 levels.
Species with poor evolvability were identified in three ways. Information on species' genetic diversity is rare, but 69 species were reported in published studies to either have gone through a genetic bottleneck and/or have low genetic diversity. Slow turnover of generations was assessed based on species' generation lengths, estimated according to the IUCN Red List Guidelines [33]. We have no empirical reference point for a threshold of low adaptive capacity for this trait, so following our established methodology, we selected the ~25% of species with the longest generation lengths, resulting in a threshold of 6 years. Similarly, the 37.96% of species with the lowest reproductive output (mean annual clutch sizes ≤2) provided the closest threshold to 25% for data categories available.

Amphibians
The traits, biological information and thresholds used to assess amphibians' vulnerability are summarised in Table S2.

Sensitivity
Amphibians' habitat and microhabitat specialisation was assessed based on two traits. As for birds, habitat specialisation was assessed according to the number of IUCN Red List habitats listed for the species; species occurring in only one habitat were considered of high sensitivity, while those with 2-33 habitat types were considered of 'not high' sensitivity.
Species' microhabitat dependencies were considered to confer high sensitivity if species are larval developers and dependent on freshwater microhabitats (based on the IUCN Red List (2008)). Forests are anticipated to buffer the climate change impacts on freshwater microhabitats, so species occurring in forests were excluded.
Narrow temperature and precipitation tolerances were measured in the same way as those for birds. Only one species had <80% of its range covered by the Worldclim dataset, so the HadCM3 1975 modelled climate was not used for amphibians.
Although amphibians are likely to be affected by a range of climate change driven disruptions in environmental triggers, insufficient data are available to systematically assess the group. Based on literature, expert knowledge and phylogenetic inference, we were, however, able to identify species dependent on the particularly vulnerable cue of rainfall or increased water availability for their mass (often termed 'explosive') breeding. This excludes species buffered by occurring in forests, and typically includes mud-aestivating grassland representatives of the frog genera Hyperolius, Litoria and Leptodactylus.
Amphibians' interspecies interactions with the pathogenic chytrid fungus Batrachochytrium dendrobatidis have been associated with population declines and extinction around the world [34][35][36][37][38]. One of the leading explanatory hypotheses proposes that the physiological stress caused by changing climates has a synergistic negative effect in combination with chytridiomycosis ( [36,39,40], but see [41,42]), while climatic changes appear to facilitate the fungus' expansion into new areas [43]. We considered species to be subject to high sensitivity to increasing negative interactions with chytrid where (i) a chytridiomycosisimplicated decline or threat has already been recorded or is suspected (i.e. according to the IUCN Red List (2008), experts have listed threat by native or alien pathogens in the past, present and/or future); (ii) they are considered to be experiencing enigmatic decline [34]; or (iii) where future infection is probable and could potentially cause decline. The extent to which chytrid infection causes negative impacts on species appears to have a phylogenetic basis [44], so making the assumption that chytrid will be globally ubiquitous by 2050, we considered species to have high probability of future infection under (iii) where they are in a genus with a recorded non-benign infection, are freshwater dependent and occur in subtropical or tropical forests, shrublands or grasslands.

Exposure
Amphibian exposure to sea level rise and temperature and precipitation changes was estimated in the same way as for birds. The coarse scale of the original GCM data resulted in poor coverage of the ranges of some small island and coastal species. 2.6% of amphibians had <80% of their ranges covered by the GCM ensemble projections for terrestrial areas (those on small oceanic islands and/or with narrow coastal ranges) and these, like birds, were assessed using rankings based on HadCM3 global projections. Temperature and precipitation thresholds for all of the above exposure measures are shown in Table S2.

Low adaptive capacity
We considered species to have low intrinsic dispersal capacity if they are not known to have become established outside their natural ranges, are not associated with flowing water, and have small ranges (≤ 4,000 km2). Since there is no empirical threshold for what constitutes a small range size, we identified the 25% of species with smallest ranges, in combination with the other characteristics of low intrinsic dispersal capacity described above.
Extrinsic dispersal barriers were identified for species that occur exclusively on mountaintops, small islands, at polar edges of land masses and/or at polar edges of suitable natural habitat. Species were considered to have low reproductive capacity and hence poor evolvability where they are have low annual reproductive output (≤50 offspring (where known) or they are viviparous).

Corals
The traits, biological information and thresholds used to assess corals' vulnerability are summarised in Table S3.

Sensitivity
We assessed corals' habitat and microhabitat specialisation using two traits. We defined coral habitats as: barrier and patch reefs; atolls; fringing reefs; incipient, submerged and nonaccreting reefs; and non-reefal rocky shores. The first three types each have subtypes: outer upper reef slope; outer lower reef slope; inter reef channel; spur and groove; outer reef crest; outer reef flat; inner reef flat; reef lagoon; back reef slope; and back reef crest. These describe a total of 32 habitats, the definitions of which accompany deposited data. The detailed nature of these habitats meant that almost all species occur in multiple habitat types; since an empirical threshold for high specificity to these habitats is not available, we assessed the ~25% of species with the fewest habits (<13) as of high sensitivity. We consider depth range (maximum known depth minus minimum known depth) to be a component of microhabitat specialisation and, lacking an empirically-based threshold, we selected the ~25% of species' with greatest depth specificity (depth ranges ≤14m) to provide a relative estimate of this characteristic.
We used corals' reproductive strategy as a proxy for larval temperature tolerance. Because coral larvae must undergo dispersal via the water column, and broadcast spawners in particular require fertilisation and larval development near the sea surface, these corals are likely to be more at risk from climate change associated changes in sea surface temperatures and irradiance than those that are able reproduce by budding or fragmentation. We therefore scored species known to reproduce by means of only broadcast spawning and/or brooding as of high sensitivity to climate change. Secondly, we used evidence (published or observational) of past high temperature mortality of > 30% of local population on a reef or reef tract (typically inferred from smaller sample sizes) as a proxy for the magnitude of adult coral colonies' temperature tolerances. Lastly, because the impacts of increasing sea surface temperatures, irradiance and storms are known to attenuate depth, we considered species occurring exclusively above 20 m depth to have high sensitivity relative to those with ranges where such impacts are buffered by depth.
While some corals species have azooxanthellate colonies that are not dependent on dinoflagellate algae, that vast majority (>99%) of reef-building corals form obligate symbioses with Zooxanthellae algae [45]. The relationship between corals and their Zooxanthellae is a rapidly expanding field of research, and although massive advances have been made in recent years, the highly complex physiological relationship between the host coral and its endosymbiont Zooxanthellae, and the extreme challenges in Zooxanthallae taxonomy and identification, including inconsistency between researchers, leave large unknowns in our understanding of coral bleaching. Because certain clades of the Zoothanthellae genus Symbiodinium, including clades D, C1 and C15, are known to have relatively higher temperature tolerances [46,47] and be less vulnerable to bleaching, we used the most recent published and grey literature to record associations between each coral species and types of Symbiodinium, including clades A (2 types), B (20 types), C (59 types), D (6 types), F, and G (i.e. a total of 89 individual Symbiodinium types).
While relatively heat-tolerant Zooxanthellae types can confer an advantage to symbiont corals under high temperatures, their lower photosynthetic efficiency under typical, favourable temperature conditions confers a disadvantage due to resulting lower energy reserves [48,49] and slower growth [50]. Some coral colonies are known to experience changes in the relative and absolute abundance of different Zooxanthellae clades and/or types (that were already present within that coral colony) over time, a phenomenon often referred to as Zooxanthellae 'shuffling'. Typically one clade may be dominant and the others may be present at low to very low abundance [51]. Their presence at low abundance can facilitate shuffling [52], generally after a bleaching episode. We regard shuffling potential as likely to provide the flexibility needed for coral colonies to both survive high temperature events, and to retain a sufficiently rapid growth potential under favourable temperatures to compensate for ongoing erosion. We considered coral species to be capable of shuffling if a single colony sample has been found to harbour more than one Zooxanthellae clade or type simultaneously, a phenomenon reported for 55 species to date.
In conclusion, we regarded coral species as of high sensitivity to disruption of Zooxanthellae symbioses where these interactions are obligatory and where species are either not known to have temperature tolerant Zooxanthellae types D, C1 or C15, or where these clades are present but colonies are not known to 'shuffle' to more photosynthetically efficient types under favourable temperatures. Ongoing research is likely to add to the numbers of temperature tolerant types, the species known to harbour them, and reports of Zooxanthellae shuffling. We believe the logic applied to identifying corals highly sensitive to disruption of Zooxanthellae is justifiable and, based on the information available at the time of this assessment, this trait identifies 92.7% of corals as of high sensitivity to climate change.
We defined rare species as those occurring in geographically restricted areas (for example the Hawaiian Islands, Chagos Archipelago, Japan or parts of Arabia), as well as those that are typically sparsely distributed across their geographic ranges. The vast majority of subpopulations of virtually all reef-building coral species have not been adequately censused by researchers, and no overarching, detailed, quantitative data on actual abundances of metapopulations are available to assess global rarity. In the absence of such data, Veron [45] and our own published and unpublished datasets on local abundance estimates from multiple sites at more than 30 different locations including the Red Sea and other areas of Arabia, Madagascar, India, Thailand, E and W Australia, Vietnam, China, Indonesia, Philippines, Micronesia, Papua New Guinea, Solomon Islands and Fiji [45,[53][54][55][56][57][58] provided the basis for our qualitative assessments.

Exposure
To estimate which corals will be most exposed to climate change impacts, we calculated, firstly, their probability of experiencing bleaching, and secondly, the proportion of their ranges exposed to levels of ocean acidification beyond which no corals are currently known to exist. Mass bleaching events are commonly predicted based on the accumulation of sea surface temperatures in excess of a local climatological maximum. For example, mass bleaching is expected to be severe and lead to some coral mortality when the accumulation of 'degree heating months' (DHM) exceeds 2 o C-month [59]. Because corals can recover from mass bleaching events if intervals between bleaching events are of sufficient durations, we use each species' mean frequency of severe bleaching (DHM>2 o C) across its range as a metric of mortality-causing bleaching exposure. Corals' projected exposure to ocean acidification was calculated based on projections of ocean aragonite saturation [61], low levels of which are known to reduce their growth rates, disrupt metabolic processes and, at particularly low rates, lead to dissolution [17]. While the aragonite saturation states (Ω aragonite) of 3.5 [61] and 3.25 [62] have been proposed as thresholds below which almost no reefs currently occur, for this study we used an optimistic threshold of 3, levels below which are described as "extremely marginal" by Guinotte et al. [63].
To represent the SRES A1B scenario for 2050, we used spatially explicit projections of aragonite saturation levels, corresponding to atmospheric CO 2 concentrations of 550ppm, created by Cao and Caldeira [61]. We down-scaled these from 2.5 x 3.75 degrees to 10 minutes using a cubic spline. For both bleaching frequency and aragonite saturation projections, the downscaling of data from coarse scales to 10 minutes resulted in a poor coastline definition and non-overlapping of surfaces with some reefs, particularly in small or narrow marine areas such as the Red Sea and Persian Gulf. Where ≤50% of a species' range was not covered by the surfaces (9.0% of species for bleaching frequency and 8.6% for aragonite saturation), the species was assessed as unknown for the respective trait. Thresholds for the above exposure measures are shown in Table S3.

Low adaptive capacity
We used larval competency, specifically the maximum time known for successful larval settlement, as a proxy for species' intrinsic dispersal capacities. Some coral species' larvae can survive up to several months in the water column, potentially being transported enormous distances if no appropriate habitat for settlement is available and environmental conditions in the plankton are suitable. We categorised maximum settlement time into five categories (<7 days, 7-14 days, 14-30 days, >30 days, and unknown); lacking an empirical basis for a threshold, we selected a cut-off of <14 days to settlement which identified the worst 14.0%, the closest possible to the 'worst 25%' used elsewhere in this study. For species whose larval competency times to settlement are unknown, the 'typical' values for their particular sexual reproductive modes were used to infer dispersal distances, and for species that are both brooders and spawners, and for which specific larval competency data were unavailable, the typical competency period of the brooding mode were used. For species whose reproductive mode(s) are also unknown, the major mode of either their congeners or confamilials was assumed where appropriate.
While ocean currents can provide an excellent vector of dispersal, together with ocean temperature, they can also be barriers to dispersal (e.g., [64,65]  Corals' evolvability was estimated, firstly, based on each species' generational turnover, as estimated by typical colony longevity. Age of colonies can be difficult to determine accurately, although there are relatively consistent relations between growth rate, colony size and age in some species, and coring of some large massive corals has provided independent minimum age estimates. For others, fragmentation, injury, disease and other factors can confound such relations. Furthermore, fragmentation and budding produce clones that can ensure that the same genotypes persist on reefs for millennia. Such genotypic ages are not considered here, and we focus on the age of 'individual' colonies. Because many species are widespread and researchers have examined only a very small fraction of the populations in detail, it is not possible to assign a maximum age definitively to colonies of any species. As a result, we used broad categories of colony longevity, namely <10 years, 11-50 years, 50-100 years, >100 years, and unknown, and selected species with colonies typically living more than 50 years (1.6% of species) as those with low adaptive capacity according to this trait.
Corals with slow growth rates tend to have lower reproductive capacity on average, because colony size and reproductive output are related. Conversely, faster growing species attain larger size and, in some species at least, reach reproductive maturity sooner than their slower growing counterparts, and hence contribute more rapidly to the gene pool [66]. Field studies over the past century have established coral colonies' growth rates for species representing most of the main growth form categories. These vary with environment and phenotype, so we have assigned species to broad growth rate categories (<10mm yr -1 , 11-30 mm yr -1 , 31-100mm yr -1 , >100mm yr -1 , unknown) making inferences based on growth form and phylogeny where specific growth rate data are lacking. Once again, without an empiricallybased threshold for this trait, we identify the ~25% of species in the slowest growth categories.

Plotting areas of greatest concentrations of vulnerable species
Here we use bivariate plots to highlight the relationship between vulnerability dimensions based on biological traits (i.e. sensitivity and low adaptive capacity) and exposure, since this largely a function of how much climatic change is projected in the geographical area in which a species occurs (Fig. 2). Bivariate plots were produced by dividing per cell frequencies of (i) species that are both sensitive and of low adaptive capacity and (ii) exposed species into 10 classes based on Jenks natural breaks. These classes were used as coordinates on a 10 x 10 grid, with the biological trait-derived dimensions on the y-axis and exposure on the x-axis.
Each grid cell was assigned a colour which graduated from muted colours for low frequencies to highly saturated colours representing extreme values (blue for sensitivity and low adaptive capacity (i), yellow for climatic exposure (ii) and purple for areas with high numbers of both groups; see legend for Fig. 2). Each grid cell of the global map was assigned a colour value according to the projected frequency of species in these groups, thereby illustrating spatial covariation between the two variables of interest [67]. Areas of greatest concentrations of species in groups (i) and (ii), as well as of their overlap, and are described in Tables S8-9.

List threat statuses
We used the IUCN Red List (2008) as a basis for establishing each species' level of extinction risk. The IUCN Red List Categories and Criteria are the most widely accepted system for classifying species' extinction risks [68][69][70]. populations, subpopulations and subspecies. As a result, the extinction risk category reflects the overall status of the species, which may, for example, be of Least Concern despite some populations/subspecies being at risk [21]. In particular cases, separate assessments of subspecies and/or populations are carried out, but these are not included in the analyses presented in this paper.
Although climate change is frequently listed as a threat during red listing, no species of birds, amphibians or corals were listed as threatened solely or principally due to climate change. As a result, we included all species in our comparison between threatened and vulnerable species. For each of birds, amphibians and corals, we used a chi-square test to compare the numbers of species that were threatened and vulnerable, threatened and not vulnerable, vulnerable and not threatened, and neither threatened nor vulnerable (see Table S10). To show areas containing greatest concentrations of 'threatened', 'vulnerable' and 'threatened and vulnerable' species, we used bivariate plots (as described above in Supporting Methods section 'Plotting areas of greatest concentrations of vulnerable species'), which are shown in Fig. 3 and Fig. S6 and described in Tables S11-12.

Calculating numbers of vulnerable species under different emissions scenarios and time frames
To investigate the roles that differing concentrations of atmospheric greenhouse gas emissions could have on species' vulnerability, we compared the numbers of vulnerable species presented in previous analyses in this paper (i.e. based on the 'midrange' A1B scenario for 2050; terrestrial species ranges) with those calculated for A1B for 2090, B1 (low emissions) for 2050 and 2090, and A2 (high emissions) for 2050 and 2090. These three standard scenarios represent a range of the possible future scenarios explored by the IPCC [71,72].
We applied the threshold values identified for each exposure variable using the baseline scenario (A1B for 2050; threshold values shown in Tables S1-3) to the same four variables under the alternative scenarios and timeframes discussed above, and recalculated exposure and vulnerability scores accordingly. For example, under the baseline scenario, amphibians were regarded as highly exposed to changes in mean temperatures where the absolute changes in mean temperatures between 1975 and 2050 are ≥ 2.96 o C. We used the same threshold of 2.96 o C to classify species as exposed under the A1B 2090 scenario, and since projected temperature changes are generally greater, more species qualified as exposed under this trait. Recalculating overall exposure and then vulnerability based on these results yielded larger numbers of vulnerable species overall.
As expected, the high (A2) and low (B1) scenarios for 2090 yielded higher and lower numbers of vulnerable species than the midrange (A1B) for 2090, except for corals under a pessimistic scenario, where A2 and A1B produced the same number of vulnerable species.
We note that A2 produced fewer vulnerable species than A1B at 2050, reflecting correspondingly higher mean global temperatures and precipitation for A1B relative to A2 for

Assessing the influence of each trait on overall vulnerability
To explore the relative contribution of each trait to overall vulnerability, we calculated the number of species and the size of geographic priority area uniquely identified by each biological trait, for each taxonomic group. We present these results in Tables S13-15 and rank traits according to their relative contributions to both numbers of vulnerable species and size of geographic areas containing vulnerable species. We find, firstly, that traits contributing most to numbers of vulnerable species are, in many cases, not the same as those contributing most to the geographic priorities identified. Secondly, we find that highest ranking traits are generally not consistent across taxonomic groups. There is no overlap between birds and amphibians in the highest ranking traits for the identification of unique geographic areas. For birds, 'narrow temperature tolerances' ranks highly, as do 'low reproductive output' and 'geographical barriers to dispersal'. For amphibians, 'changes in mean temperature', 'narrow precipitation ranges' and 'slow turnover' play the greatest roles in uniquely identifying regions of high vulnerability. These analyses provide useful information to inform prioritisation of ongoing trait data collection for birds, amphibians and corals. For example, 11% of birds qualified as vulnerable due to relatively small population sizes, but no species were uniquely identified by this trait, suggesting that it is not a priority for further data collection. We note, however, that because (like several other traits) it contains a number of unknown values, and even a few species or regions identified could be of particular significance, we do not suggest dropping any traits altogether at this stage.

Assessing the influence of trait thresholds and other sources of uncertainty on overall vulnerability
We distinguished four types of traits, each of which required distinct threshold selection approaches. Firstly, where species' tolerance thresholds are clearly established and widely accepted in the peer-reviewed scientific literature (e.g., ocean temperature conditions at which coral bleaching occurs), we referenced and used these. The second trait threshold type applies to data that are binary and where independent, widely accepted categorisations are available (e.g., occurrence only on islands; occurrence in only one habitat type). We regard these thresholds as objective and do not consider them to be a significant source of uncertainty in vulnerability assessments.
The third trait threshold type, used when no binary or established thresholds were known and where trait data were continuous or categorical, involved selecting the worst affected 25% of species (e.g., temperature and precipitation change tolerances), or the species in categories with a break closest to 25% (e.g., generation length for birds; depth ranges for corals). The fourth threshold type was used for traits where sufficient information and/or experience were available for experts to believe that they could defensibly set thresholds for heightened vulnerability (e.g. exposure to sea level rise based on habitat affiliations; inherent rate of dispersal required for birds based on projections in the literature [73]). Because these thresholds could introduce subjectivity into assessments, we explored sensitivity of vulnerability scores to shifting them to higher and lower values. In Tables S16-18 we identify the traits for which these 'percentage thresholds' (marked as (P), blue text) and 'expert thresholds' (marked as (E), green text) were used and examined their influence on overall vulnerability scores (see Tables S19-21).
We found that shifting percentage thresholds by 10% (i.e., to a more lenient 35% and a stricter 15%) changes the numbers of vulnerable species by only +11% to -12% for birds and +9% to -9% for amphibians, suggesting that these groups are relatively robust to the percentage thresholds selected. For corals, however, changes of +29% and -8% suggest that threshold choices play a larger role. Shifting expert threshold had an even lower impact on numbers of vulnerable species, shifting them by +1% to -1% for birds, +7% to -9% for amphibians and +3% and -0% for corals. In conclusion, the broad range of sensitivity analyses conducted shows that missing data, choice of traits and their thresholds and expert judgement all introduce a degree of uncertainty into vulnerability assessments. We find, however, that the geographic priorities identified by our approach are notably robust to this uncertainty, strengthening confidence in the main results of this paper. By presenting results as ranges of possible numbers of vulnerable species under different scenarios, emphasising repeatedly that scores are relative measures, and conducting sensitivity analyses on all main possible sources of uncertainty, we believe that we have dealt responsibly with the uncertainty inherent in assessments of future impacts of climate change on complex biological systems, and that the results presented provide the best assessments possible given available data and knowledge.

Opportunities for validation of the framework
Vulnerability assessments, including the one we present, should be empirically evaluated to determine whether they produce robust ecological or conservation assessments of the impacts of climate change. At this stage, however, this remains challenging. A body of ad hoc observations of climate change impacts on species is emerging (as summarised in [73][74][75]) but the use of these studies for testing global assessments such as ours has several serious limitations. They cover only a small fraction of our study's species and generally address a limited range of possible climate change impacts (typically distribution range shifts and phenological changes), ignoring a broad range of other possible impacts that our approach considers. Such studies have strong geographic and ecosystem biases (typically towards Northern Hemisphere temperate regions); species in other regions and ecosystems may not respond in the same way. They tend to demonstrate population changes rather than the species-scale responses we project, and are based on non-standardised surveying methods.
Lastly, existing observation studies represent a non-random subset of climate change responses and due to publication bias, are likely to under-represent species of lower vulnerability to climate change (e.g. those of high latent risk, and many potential persisters and potential adapters (as identified in Fig. 1)).
Another possibility for validation, particularly of trait selection, is to examine species' past responses to climatic changes, as evidenced in the paleorecord. We plan to explore this avenue of research, but are aware that it suffers from many of the limitations described above for ad hoc observational studies. Cross-referencing results of our assessment with others based on, for example, species distribution models [76,77], dynamic global vegetation models [32] and novel and disappearing climates [78,79] provides a further avenue for investigation, but since such models are simply alternative predictions with their own limitations and assumptions, also often unvalidated, results will need to be interpreted with caution. We note that outputs from other approaches could be incorporated into our assessment framework. For example, global vegetation models could be used to assess a species' exposure to habitat changes, and projections of species' range shifts could inform assessments of the likelihood of a species' successful dispersal in response to climate change.
Finally, we propose that the most effective and reliable means of gathering observational data for validating this and other approaches is through standardised monitoring schemes that adequately sample environmental gradients [5]. Several such schemes are being established (e.g., [80][81][82]) and we believe that, with further expansion, this approach can deliver the information needed to effectively validate climate change vulnerability assessments. We emphasize the need for immediate and ongoing support and expansion of standardised monitoring schemes globally.

Caveats and uncertainty
Since the results of this assessment are, at this stage, largely unvalidated, we note some important caveats to our methods. These are necessary to consider when interpreting the results, but also form priority areas for new research.
1. We acknowledge that experts' judgements can be subject to certain biases [83], but emphasise their value, particularly where timely decisions are needed in the face of novel, future or uncertain situations [84], for example for IPCC assessment reports and the IUCN Red List.
2. The selected trait threshold we chose (25%) is arbitrary and is unlikely to represent any real limit to species' tolerances. It simply highlights the top scoring species as a basis for analysis. Sensitivity of results to this threshold is explored in section 'Assessing the influence of trait thresholds on overall vulnerability' above, but ideally the threshold would be updated or validated through observations and or experiments of the way in which climate change and traits interact (e.g., [85,86]). When interpreting the absolute values of the percentages for each group, it is important to recognise that these simply represent the degree of overlap between sensitivity, low adaptive capacity and exposure within the taxonomic group (e.g., highest overlap in birds (24-50% of species highly vulnerable) vs. lower overlap for corals (15-32% highly vulnerable)). It is particularly important to emphasise that comparisons between the percentages of high vulnerability species cannot be interpreted to represent any real differences in vulnerability between taxonomic groups.
3. Our framework's scoring system is based on the assumption that species have multiple pathways to extinction; traits were selected and scores calibrated such that a 'high' score on any single e.g. sensitivity trait would result in the species being ranked as 'sensitive' overall. As anthropogenic climate change progresses, the range, species-specificity and frequencies of extinction pathways (no doubt including some not yet identified) will become apparent, but at this point, we believe it is premature to rank one trait as more important than another or exclude any that have been identified as possibilities. We acknowledge that this simple, equally-weighted combination of traits and trait groups fails to account for their potentially differing importance in conferring climate change vulnerability, but we are unable to quantify or justify relative trait weightings. 4. In practice, the biological traits are likely to interact with each other and with environmental change in non-linear ways, and there will be thresholds and abrupt state changes as a result. These effects are likely to be very specific and context-dependent and the only way to develop an understanding will be through detailed field studies over many years with a great deal of relevant climate and environmental information. This is simply going to be impossible for many species, but the availability of a few such studies [11,85,87] and the deployment of more mechanistic models (e.g., [88,89]) should start to support more sophisticated approaches than the very broad brush approach we use here. 5. Our approach does not specify the relationship between vulnerability scores and the risk of extinction. Although our analysis shows vulnerability to be correlated with extinction risk (as determined by the IUCN Red List TM Criteria) within a taxonomic group, it is not possible to equate vulnerability with a specific level of threat, and the relationship between vulnerability and extinction risk may be different for different groups. Results may be interpreted, for example, to predict which bird species and geographic regions will be at relatively higher risk of climate driven extinction than others, but not to quantify this risk, nor to compare birds' vulnerability with that of amphibians or corals. Our exposure modelling suggests that corals, in particular, are likely to face a much higher risk of extinction than the other taxonomic groups, though this is not reflected in the results of this study. 6. We recognize that climate will have positive effects on many species. In fact many species are already benefitting from climate change especially in temperate areas [5], and to date most range shifts recorded have resulted in range expansions more than range contraction [90]. However, our framework does not attempt to incorporate thiswe are interested in identifying species at risk from climate change. Graphs show the percentages of each family's species that are highly sensitive vs. of low adaptive capacity (A-C), sensitive vs. exposed (D-F), and of low adaptive capacity vs.

Birds Amphibians Corals
exposed (H-J) for birds, amphibians and corals respectively.     Table 2 and          Table S8: Summary of geographic focal areas (identified in Figure 2 (A, C and E)) that contain high total numbers of species that are (i) highly sensitive and of low adaptive capacity, (ii) highly exposed, and both (i) and (ii).

Supporting Tables
[High sensitivity and low adaptive capacity] and [high exposure] (purple in Fig. 2) High sensitivity and low adaptive capacity only (blue in Fig. 2) High exposure only (yellow in Fig. 2 Figure 2 (B, D and F)) that contain high proportions of species, relative to species richness, that are (i) highly sensitive and of low adaptive capacity, (ii) highly exposed and both (i) and (ii).

[High sensitivity and low adaptive capacity] and [high exposure]
(purple in Fig. 2) High sensitivity and low adaptive capacity only (blue in Fig. 2) High exposure only (yellow in Fig. 2 Table S11: Summary of the geographic focal areas identified in Figure 3 that contain high total numbers of species that are threatened (according to the IUCN Red List TM ), highly climate change vulnerable and high numbers of both. Threatened and vulnerable (purple in Fig. 3) Vulnerable only (yellow in Fig. 3) Threatened only (blue in Fig. 3   S13: Summary of the numbers of species and size of geographic area uniquely identified by each of the biological traits used to assess overall climate change vulnerability of birds. Traits highlighted in yellow identify the five most influential traits for uniquely identifying numbers of species and those in red text identify these traits for geographic areas. Trait and trait group descriptions are shortened versions; full titles are shown in Table S1.     Table 2 and Figure 2) and stricter thresholds. Thresholds for traits indicated with a (P) and highlighted in blue were selected based on arbitrary percentage thresholds (35%, 25% and 15%) while those indicated by an (E) and highlighted in green were selected based on experts' judgements. All results shown are based on an optimistic scenario for 2050 under the A1B emission scenario.   Table 2 and Figure 2) and stricter thresholds. Thresholds for traits indicated with a (P) and highlighted in blue were selected based on arbitrary percentage thresholds (35%, 25% and 15%) while those indicated by an (E) and highlighted in green were selected based on experts' judgements. All results shown are based on an optimistic scenario for 2050 under the A1B emission scenario.   Table 2 and Figure 2) and stricter thresholds. Thresholds for traits indicated with a (P) and highlighted in blue were selected based on arbitrary percentage thresholds (35%, 25% and 15%) while those indicated by an (E) and highlighted in green were selected based on experts' judgements. All results shown are based on an optimistic scenario for 2050 under the A1B emission scenario.