Introduction

Endemism serves as a measure for the irreplaceability of species because it describes species that occur only in a predefined area (e.g. habitat or country1,2,3) or, more generally, species with restricted ranges4,5,6. Analyses of endemism are a crucial component of global-scale conservation assessments because they identify areas of globally unique biodiversity that are complementary to areas rich in species2,7,8,9,10,11,12,13. Together they form the foundation of understanding global variation in diversity and the efforts required to conserve it7,14.

Current assessments of global endemism are, however, severely biased. Global diversity studies routinely exclude species-poor areas, especially in the southern hemisphere, and most notably the Antarctic region15,16,17,18. While the omission of sites does not affect estimates of alpha diversity (local species richness) in the remaining sites, it invariably alters estimates of endemism because endemism is assessed by comparing the composition of an assemblage with the composition of all other assemblages. In addition, the most commonly used measure for endemism (weighted endemism) is strongly influenced by local species richness4,19. This measure is therefore limited in its suitability for identifying sites with the highest concentration of range-restricted species, that is, the sites whose composition differs most from those of all other sites, and which are therefore most valuable for conservation because they harbour the highest concentration of unique species, functional trait combinations or phylogenetic lineages. Current assessments of global endemism potentially fail to fully represent such areas of high endemism, but the extent of this bias is unknown.

Moreover, if endemism is calculated based on biased evidence and then used to assess the biodiversity conservation value of sites and guide implementation decisions—as is routinely done at continental and global scales10,19,20,21,22—implementation may fail to deliver the intended conservation outcomes. Doing so may instead inadvertently draw attention away from those sites whose communities are most dependent on range-restricted species and which are therefore most vulnerable to conservation threats23,24. These threats include those such as species invasions25, and/or climate change potentially shifting species ranges outside conservation areas26,27 or making current ranges unsuitable28,29,30. Despite the urgency of implementing conservation measures, given rapidly developing threats and declines in biodiversity24,31, the global distribution of irreplaceable diversity is still insufficiently known32.

To address these substantial deficiencies, we analysed global patterns of endemism and diversity of birds for the three key facets of diversity33: taxonomic diversity (i.e. species richness), functional diversity, and phylogenetic diversity, across all global areas using complementarity (also known as corrected weighted endemism4,34 or proportional range rarity35) as a metric36,37,38,39. Complementarity measures how much of a site’s diversity (species, functional trait combinations, or phylogenetic lineages) is shared with other sites; it is independent of richness, and it peaks at sites whose composition differs most from all other sites, thereby providing an accurate picture of endemism as irreplaceability. Complementarity has rarely been analysed on global scales32,35, and less so for all facets of diversity combined.

Specifically (see details in Methods section), we compiled global datasets on avian distributional ranges40, functional traits41, and phylogeny42, and projected range maps onto a grid of 14,640 hexagonal grid cells (size: 11,610 km²) in an equal-area projection (Lambert cylindrical, EPSG:6933). For each grid cell, we determined the presence/absence of each bird species, functional trait combination, and phylogenetic lineage and then calculated alpha diversity (local richness), contribution to gamma diversity (that is, “weighted endemism”), and complementarity. Since contribution to gamma diversity and complementarity are both influenced by the overlap in assemblage composition and, hence, species’ range size, we also determined absolute range size, latitudinal range, longitudinal range, and latitudinal range midpoint for each bird species, functional trait combination, and branch in the phylogeny. For each measure, we calculated median values across latitudinal bands of 5° latitude and then analysed the relationship with latitude using linear models.

In this study, we show that the global pattern of endemism calculated as complementarity differs from global patterns of alpha diversity and contribution to gamma diversity. Complementarity shows a global increase from north to south. This pattern appears to be driven by increasingly smaller and discontinuous landmasses towards the south, which leads to smaller ranges and reduced overlap in community composition in the southern hemisphere.

Results and discussion

Global patterns of diversity and endemism

Alpha diversity peaks in the tropics and generally declines towards higher latitudes in both hemispheres (Figs. 1, 2a). It thus follows the classic pattern of a latitudinal gradient in diversity43, albeit the relationships are slightly bi-modal as alluded to previously44,45, with a local minimum at about 22.5°—more prominently shown in the northern hemisphere and probably caused by the large proportion of desert sites. Contribution to gamma diversity shows a similar decline with latitude, although the decrease is less steep in the southern hemisphere (Figs. 1, 2a). Contribution to gamma diversity is positively correlated with alpha diversity19,36 (species richness, R² = 0.59, P < 0.001; functional diversity, R² = 0.67, P < 0.001; phylogenetic diversity, R² = 0.65, P < 0.001; Supplementary Information, Fig. S1). In contrast, complementarity shows opposing patterns in the two hemispheres. In the northern hemisphere, complementarity decreases from the equator towards higher latitudes, whereas in the southern hemisphere, the pattern is reversed, resulting in a global exponential increase in endemism from north to south (taxonomic complementarity, R² = 0.91, P < 0.001; functional complementarity, R² = 0.91, P < 0.001; phylogenetic complementarity, R² = 0.93, P < 0.001; Figs. 1, 2a and Supplementary Information, Table S1). Complementarity is not correlated with alpha diversity35 (species richness, R² = 0.002, P < 0.001; functional diversity, R² = 0.0001, P = 0.24; phylogenetic diversity, R² = 0.07, P < 0.001; Supplementary Information, Fig. S1).

Fig. 1: Global patterns of avian diversity and endemism.
Fig. 1: Global patterns of avian diversity and endemism.
Full size image

Alpha diversity, contribution to gamma diversity ('weighted endemism') and complementarity ('corrected weighted endemism') for three facets of diversity (taxonomic, functional and phylogenetic diversity) mapped across the world (n = 14,640 grid cells). Cells are coloured according to rank (14,640 highest, 1 lowest value).

Fig. 2: Latitudinal patterns of richness, endemism and range size.
Fig. 2: Latitudinal patterns of richness, endemism and range size.
Full size image

a Rank alpha diversity, contribution to gamma diversity and complementarity against latitude are shown for three facets of diversity (species richness, functional diversity, phylogenetic diversity). Values (n = 13,614 grid cells) are summarised in latitudinal bands of 5° (n = 26 latitudinal bands). Points are coloured according to relative rank and boxplots according to median values. b Absolute range size, latitudinal range and longitudinal range (log-transformed) for units representing the three facets of diversity (species, n = 10,954; functional trait combinations, n = 7929; phylogenetic lineages, n = 19,914). Each unit is assigned to a latitudinal band of 5° (n = 25 latitudinal bands) according to the latitudinal midpoint of its range. Colours represent southern (red) and northern (blue) hemispheres; latitudinal bands inside the tropics are lighter coloured. In the southern hemisphere, ranges decrease towards higher latitudes; in the northern hemisphere, absolute range size and latitudinal range show a peak at mid-latitudes, and longitudinal range increases towards higher latitudes. Boxes show median and 1st and 3rd quartiles; whiskers extend to largest and smallest, respectively, values up to 1.5 x inter-quartile range beyond the quartiles; data points beyond the end of the whiskers are plotted individually.

Global decrease in range size from north to south

The increase in complementarity from north to south is in accordance with increasingly smaller range sizes from north to south (species, R² = 0.74, P < 0.001; functional traits, R² =  0.79, P < 0.001; phylogenetic lineages, R² = 0.83, P < 0.001), smaller latitudinal range sizes from north to south (species, R² = 0.54, P < 0.001; functional traits, R² = 0.39, P < 0.001; phylogenetic lineages, R² = 0.35, P = 0.002), and smaller longitudinal range sizes from north to south (species, R² =  0.72, P < 0.001; functional traits, R² = 0.94, P < 0.001; phylogenetic lineages, R² = 0.89, P < 0.001; Fig. 2b; and Supplementary Information, Table S2). If analysed separately by hemisphere, however, latitudinal range size strongly decreases from the equator towards higher latitudes in the southern hemisphere for functional trait combinations (R² = 0.90, P < 0.001) and phylogenetic lineages (R² = 0.88, P < 0.001), but it shows a hump-shaped relationship with latitude in the northern hemisphere (functional trait combinations, R² = 0.81, P < 0.001, peak at 27.5°N; phylogenetic lineages, R² = 0.79, P < 0.001, peak at 42.5°N).

Complementarity as a measure of irreplaceability

By using an endemism measure that is only influenced by the overlap in assemblage composition, but not assemblage richness (alpha diversity), and by including all global areas, we reveal a global north-south gradient in avian endemism with a peak in the southern hemisphere. These findings are in line with continental-scale studies showing a discrepancy between richness and endemism (11,12). On average, southern hemisphere bird assemblages show the highest proportion of irreplaceable, range-restricted species, functional trait combinations and phylogenetic lineages. This pattern for endemism measured as complementarity is strikingly different from that for contribution to gamma diversity (Figs. 1, 2a, 3). In addition to known hotspots of both richness and endemism, such as the tropical Andes, Madagascar, and New Guinea, we reveal endemism peaks at sites with low local richness, such as oceanic islands, and most notably, the sub-Antarctic islands and the Antarctic continent, i.e. sites that are commonly omitted from large-scale analysis of diversity15,16,17,18 because of their low alpha diversity and small landmass (Figs. 1, 3). Their conservation significance, therefore, owes not only to the shear abundance of birds found on them46,47, but also to their global contribution to the retention of the extant avifauna.

Fig. 3: Changes in the relative importance (rank) of sites with regard to their complementarity ('corrected weighted endemism') vs their contribution to gamma diversity ('weighted endemism').
Fig. 3: Changes in the relative importance (rank) of sites with regard to their complementarity ('corrected weighted endemism') vs their contribution to gamma diversity ('weighted endemism').
Full size image

Grid cells (n = 14,640 grid cells) are coloured ranging from maximum increase to maximum decrease in rank differences (rank complementarity—rank contribution to gamma diversity). Antarctica, the sub-Antarctic islands, the Arctic, oceanic islands in the tropics, desert regions and the southern Andes show the largest increase in rank; temperate regions in the Palaearctic show the largest decrease in rank.

Potential drivers of the north–south increase in endemism

Our results reveal a striking difference between hemispheres in endemism and range size. To assess the potential significance of these differences for the conservation of northern-hemisphere and southern-hemisphere species, we explored these differences further. First, we determined if the differences in endemism and range size influence patterns of gamma diversity by calculating taxonomic, functional, and phylogenetic gamma diversity for each latitudinal band of grid cells. Since endemism and range size are influenced by the availability of landmass, we also quantified available landmass, approximated by the number of grid cells per latitudinal band, and then calculated density (i.e. gamma diversity/available landmass) for each latitudinal band of grid cells.

Gamma diversity and available landmass against latitude

Gamma diversity shows similar patterns in both hemispheres, with a peak in the tropics and a decline towards higher latitudes, albeit with a slightly faster decrease outside the tropics in the southern hemisphere (extratropical, species richness, north, R² = 0.98, P < 0.001, south, R² = 0.98, P < 0.001; functional diversity, north, R² =  0.97, P < 0.001, south, R² = 0.99, P < 0.001; phylogenetic diversity, north, R² = 0.97, P < 0.001, south, R² = 0.98, P < 0.001, Fig. 4d). In contrast, available landmass shows strikingly opposing relationships in the two hemispheres, especially outside the tropics (Fig. 4a, b): it continuously increases in the northern hemisphere up until 65°N, but it decreases up until 40°S in the southern hemisphere (Fig. 4b, c), resulting in an overall continuous decrease of available landmass from north to south. This opposing pattern across hemispheres necessarily contradicts the idea48 that the increase in gamma diversity from the poles towards the tropics can be explained by larger available landmass and larger ranges in the tropics; absolute range size, latitudinal range size, and longitudinal range size all generally follow the decrease in available landmass from north to south (Figs. 2b, 4b).

Fig. 4: Landmass and gamma diversity in the southern and northern hemisphere.
Fig. 4: Landmass and gamma diversity in the southern and northern hemisphere.
Full size image

a Extratropical landmasses (green) in the northern hemisphere (above) and southern hemisphere. b Landmass (approximated by number of grid cells) per latitudinal band (n = 252 bands) in the northern (blue) and southern (red) hemisphere against absolute latitude. © EuroGraphics for the administrative boundaries. c Cumulative landmass (number of grid cells) from the equator to the poles in the northern (blue) and southern (red) hemisphere against absolute latitude. d Gamma diversity (above) and density (gamma diversity/landmass) against absolute latitude in the northern (blue) and southern (red) hemisphere. Grid cells (n = 14,640) are divided into latitudinal bands (n = 245 bands) according to their latitudinal midpoint; gamma diversity is calculated for each latitudinal band. Gamma diversity decreases towards higher latitudes in both hemispheres. Density decreases towards higher latitudes in the northern hemisphere (n = 76 extratropical latitudinal bands); in the southern hemisphere outside the tropics (n = 59 latitudinal bands), density peaks at mid-latitudes. Shading around regression lines (mean) represents the 95% confidence interval.

The uneven distribution of landmass affects latitudinal patterns of density: in the northern hemisphere, density declines continuously from the tropics to the highest latitudes (species richness, R² = 0.99, P < 0.001; functional diversity, R² = 0.99, P < 0.001; phylogenetic diversity, R² =  0.99, P < 0.001; Fig. 4d), whereas in the southern hemisphere, the relationship is hump-shaped outside the tropics, with a peak at mid-latitudes (species richness, R² =  0.94, P < 0.001; functional diversity, R² = 0.87, P < 0.001; phylogenetic diversity, R² = 0.86, P < 0.001; Fig. 4d); for functional and phylogenetic diversity these peaks represent the highest values observed along the entire latitudinal gradient (Fig. 4d). The decrease of available landmass from north to south corresponds with the observed decrease in range sizes, especially the decrease in longitudinal ranges. At the same absolute latitude, species in the northern hemisphere can expand their ranges over a large area of similar climatic conditions48,49,50, whereas in the southern hemisphere, longitudinal ranges are restricted because landmasses are separated by vast expanses of open ocean (Fig. 4a). The smaller range sizes and the higher isolation of landmasses in the southern hemisphere result in reduced overlap in assemblage composition and, consequently, the higher values of contribution to gamma diversity and complementarity observed in the southern hemisphere.

Relevance for global diversity patterns and conservation

The global north-south increase in endemism markedly departs (Fig. 5) from current understanding of global diversity patterns. In line with recent studies that showed that southern hemisphere ecosystems, especially the sub-Antarctic and Antarctic, are more diverse than previously thought51, we demonstrate that—on average—southern hemisphere assemblages harbour among the highest percentages of irreplaceable, range-restricted species, functional trait combinations and phylogenetic lineages (Fig. 5), which makes these assemblages potentially most vulnerable to changes24,42,52. Similarly, while latitudinal gradients of diversity are similar in both hemispheres43, the diversity is distributed over a much smaller and more disconnected area in the southern hemisphere, leading to a global peak of functional and phylogenetic diversity per area in the mid-latitudes of the southern hemisphere. At the same time, several areas in the southern hemisphere—including Antarctica, the sub-Antarctic islands, oceanic islands in the tropics, and the southern Andes—were among the sites with the highest discrepancy between complementarity (endemism measured as irreplaceability) and contribution to gamma diversity (i.e. weighted endemism). The classic focus on alpha diversity and weighted endemism to identify biodiversity hotspots, hence potentially draws conservation actions away from these highly endemic, but species-poor, assemblages.

Fig. 5: Factors underlying the observed differences in endemism (complementarity) between hemispheres.
Fig. 5: Factors underlying the observed differences in endemism (complementarity) between hemispheres.
Full size image

a, b Extratropical landmasses in the southern hemisphere (S) are smaller and more disconnected than in the northern hemisphere (N; a), resulting in smaller ranges and higher endemism (here exemplified by phylogenetic complementarity) in the southern hemisphere (b). Values for rank phylogenetic complementarity against latitude (n = 14,640 grid cells) are summarised in latitudinal bands of 5° (n = 26 latitudinal bands). Points are coloured according to relative rank and boxplots according to median values. Values for absolute range size (log-transformed) of phylogenetic lineages against latitude (n = 19,914) are summarised in latitudinal bands of 5° (n = 25 latitudinal bands) according to the range midpoint. Colours represent southern (red) and northern (blue) hemispheres; latitudinal bands inside the tropics are lighter coloured. Boxes show median and 1st and 3rd quartiles; whiskers extend to largest and smallest, respectively, values up to 1.5 x inter-quartile range beyond the quartiles; data points beyond the end of the whiskers are plotted individually. c,d Functional and phylogenetic endemism in the southern hemisphere (S) is high because of the occurrence of old phylogenetic lineages with unusual morphologies that have no equivalent in northern high latitudes (N). These include c basal lineages of Australaves (Cariamiformes [green, 2 spp.]) and Passeriformes (e.g. Atrichornithidae [orange, 2 spp.], Menuridae [red, 2 spp.], Acanthisittidae [blue, 2 spp.]) and d the Palaeognathae (Tinamiformes [orange, 47 spp.], Rheiformes, Struthioniformes, and Casuariformes [red, 9 spp.], and Apterygiformes [blue, 5 spp.]), many of which also occupy small ranges. e Several other bird lineages, e.g. penguins (Sphenisciformes [red, 18 spp.]) or albatrosses (Diomedeidae [blue, 22 spp.]) are largely restricted to southern mid and high latitudes, i.e. the Antarctic region. Despite their highly endemic avifauna, sub-Antarctic islands and Antarctica are commonly omitted from global diversity studies. Bird silhouettes by DMD.

Impact of environmental change in the two hemispheres

The observed hemispheric differences in endemism and available landmass suggest potential differences in the way in which northern and southern species might respond to environmental change. For instance, the smaller area and higher discontinuity of available landmass in the extratropical latitudes of the southern hemisphere might affect the ability of southern species to shift their ranges53. The majority of landmass in the southern hemisphere consists of the southern tips of continents (Fig. 4), isolated by vast expanses of ocean, whereas in the northern hemisphere, extensive areas of landmass are connected across latitudes and longitudes48,49,50, and the two largest separated blocks of land (Palaearctic and Nearctic) are only separated by the comparatively narrow Bering Strait54,55,56. Moreover, for birds with range limits at the southern tip of continents (e.g. Fig. 5), the nearest poleward landmasses are the sub-Antarctic islands or the Antarctic continent, which have climatic conditions that make them unsuitable as breeding grounds for most birds. If environmental conditions in the current ranges become unsuitable28,29, many southern-hemisphere species might therefore not be able to reach suitable new breeding areas. In addition, due to the isolation of landmasses, it is not uncommon for southern-hemisphere species to have their most closely related species on a different continent2, and local extinctions are therefore less likely to be replaced by the functionally or phylogenetically most-closely related species.

To date, most studies on the potential effects of environmental changes on species assemblages are based on data from the northern hemisphere57, and it is not clear in which way species assemblages in the two hemispheres might differ in their responses to climatic changes. Our results demonstrate that, in addition to protecting high-richness areas, conservation actions should be aimed towards protecting the most-highly unique ecosystems that are vulnerable due to their high dependence on endemics, that is, their high dependence on irreplaceable species, functional trait combinations, and evolutionary lineages, and which are most prevalent in the southern hemisphere.

Methods

Species distributions

We used data on avian ranges from the IUCN40. We re-projected all ranges into a Lambert cylindrical equal-area projection with standard parallels at 30° N and 30° S (EPSG 6933), and then mapped the distribution in a hexagonal grid of 14,640 cells with a size of 11,610 km2, roughly equal to 1°  × 1° at the equator, the smallest suitable size for global analyses58. We included extant birds in their native and reintroduced ranges, and we excluded passage and uncertain parts of their ranges. We only included the part of the ranges that included land. Our dataset included ranges for 10,954 bird species. For each bird species, we determined range size and the latitudinal midpoint of its distribution. We used latitudes 23.5°N and 23.5°S as cut-offs to divide all grid cells into three groups according to their latitudinal midpoint (north, 7631 grid cells: tropics, 5548 grid cells; south, 1461 grid cells).

Alpha and gamma diversity

Taxonomic diversity

We measured taxonomic alpha diversity as species richness, i.e. the number of bird species found in each site; we measured taxonomic gamma diversity as the total number of bird species.

Functional diversity

We measured functional diversity as the diversity of functional trait combinations of birds. We selected eight morphological traits59,60: culmen length, beak length from tip to nares, beak depth, beak width, tarsus length, wing length (carpal joint to wingtip), Kipp’s distance (carpal joint to tip of outermost secondary), and tail length. We obtained trait measurements from AVONET41. Our dataset included 10,953 species (we excluded Gallirallus lafresnayanus because there are no trait data available for this species). We used principal coordinates analysis (PCoA) to project all bird species into one common four-dimensional trait space, where they were arranged according to the differences in their trait combinations. Since the position in trait space represents a species’ niche position but not its niche size61,62, we assigned a volume around each bird species with a radius equivalent to 1/15 of the length of the shortest PCoA axis, which is a conservative approach (using 1/10 lead to very similar results). This approach is similar to calculating functional diversity as cumulative trait probability densities with fixed standard deviation (TPDc)63, with virtually identical FD values (R2 = 0.99–1.0), but it is much faster and facilitates the inclusion of more species and trait dimensions. We measured functional alpha diversity as the volume of trait space occupied by the bird species present in a geographic grid cell, ignoring the overlap between the volumes of individual species as well as any 'empty' regions in the trait space62. Likewise, we measured functional gamma diversity as the volume of trait space occupied by all bird species from the global dataset.

Phylogenetic diversity

We obtained full phylogenetic trees based on the Hackett backbone64 from birdtree.org42. Since bird taxonomy differs between the trait dataset and the phylogenetic dataset, we used the supplementary tables from AVONET41 to merge the phylogenetic data and the IUCN ranges. Our phylogenetic dataset included 9841 species. We measured phylogenetic alpha and gamma diversity as Faith’s PD65, the combined length of all branches that connect the bird species present in a grid cell (alpha), or all bird species in the dataset (gamma). We repeated all analyses that involved phylogenetic data using 30 different trees, all of which yielded virtually identical results (Supplementary Information, Fig. S2).

Endemism

We quantified local endemism with two measures: contribution to gamma diversity and complementarity. Both are derived from measures for comparing the diversity of species’ functional roles in ecological networks36,62 and species’ contribution to functional and phylogenetic diversity38, which can also be applied to large-scale comparisons of diversity across communities39.

Contribution to gamma diversity (weighted endemism)

Contribution to gamma diversity represents the contribution of each site to global gamma diversity36,38,62. It is calculated by weighting local alpha diversity according to how often each element (species, functional trait combination, branch in the phylogeny) is found in other sites (i.e. a species found in n sites contributes 1/n to gamma in each of the communities in which it occurs). These local weighted alpha diversities add up to gamma; the fraction between weighted alpha/gamma represents the contribution to gamma. Contribution to gamma diversity is strongly correlated with local alpha diversity (Supplementary Information, Fig. S1).

Complementarity

Complementarity measures the degree to which the elements of the local alpha diversity (species, functional trait combinations, branches in the phylogeny) are shared with other sites36,39. It is calculated as weighted alpha diversity/alpha diversity36.

Although derived differently, contributions to gamma diversity and complementarity are conceptually similar to the measures of weighted endemism and corrected weighted endemism4,34,66,67, provided that they are measured across sites of equal size (e.g. a grid). Contribution to gamma diversity is also equivalent to range(-size) rarity35,68; complementarity is equivalent to proportional range rarity35. Since complementarity is independent of local alpha diversity (Supplementary Information, Fig. S1), it is a more suitable measure for the irreplaceability of a site or region than contribution to gamma diversity.

Relationships with latitude, available landmass

Complementarity and range size against latitude

We divided all grid cells into latitudinal bands of 5° from 60° S to 70° N (n = 26 bands) and summarised median taxonomic, functional and phylogenetic complementarity and median absolute, latitudinal, and longitudinal range size for each band. We used linear regression models to relate median values to latitude.

Gamma diversity against latitude

We divided all grid cells into latitudinal bands according to their latitudinal midpoint (n = 252 bands). We calculated taxonomic, functional, and phylogenetic gamma diversity per band, but we removed six latitudinal bands with fewer than four grid cells. For each band, we also quantified available landmass, approximated by the number of grid cells. We excluded grid cells south of 67°S because only a tiny fraction of their landmass is ice-free, and an inclusion would therefore misrepresent the relationship between gamma diversity and available landmass (an analysis including the cells south of 67°S, however, led to almost identical results). We used linear models to test the relationship between gamma diversity and latitude and the relationship between density (gamma diversity/available landmass) and latitude.

Software

For the analyses, we used R version 4.2169 and the R packages dplyr70, ggplot271, viridis72, sf73, rnaturalearth74, giscoR75, vegan76, pdist77, ape78, and picante79.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.