We begin our analyses by leveraging the highest quality breeding range maps available for birds, inferred with state-of-the-art species distribution models and powered by millions of curated citizen science eBird sightings12,13. At the time of this study, this high-quality dataset was only available for a limited number of species. From that set, we retained only species whose distributions covered at least 20 map cells (50 km × 50 km each, corresponding to geographic distributions exceeding 50,000 km2), because very small samples can lead to noisy and often unstable estimates of kernel densities (see Methods). Additionally, this threshold allowed us to effectively remove small island endemics from our analyses, whose artificially narrow climate niches tend to reflect lack of opportunities as opposed to an inability to evolve. Our final sample comprised 1471 species, spanning the phylogenetic, geographic, and climatic distribution of living birds (Supplementary Figs. 1 and 2). The R code and data needed to replicate our analyses are provided in the supplementary materials.
We characterized the breadth of avian climate niches using the mean, variability, and predictability of local temperature and precipitation cycles. Because these environmental variables tend to be highly correlated at a global scale14, we reduced them first to two composite variables via Principal Component Analysis at a 50 km × 50 km spatial resolution on an equal area projection map (Fig. 1; see Methods and Table S10). Scores for the first resulting principal component, hereafter ‘temperature harshness’, increase in sites with colder, more variable, and less predictable temperatures. Scores for the second component, hereafter ‘xeric harshness’, increase in drier sites with more variable and unpredictable precipitation. For each species in our sample, we extracted the PC scores of all cells contained within its breeding range and used them to compute a 2-D kernel density function of its occupied climates. Accordingly, we define niche ‘breadth’ as the area encompassing a 95% probability of observing novel occurrences of a given species within the climate space (Fig. 1). We note that this metric is more closely associated with realized than fundamental climate niches because the proper quantification of the latter requires experimental approaches that are difficult to achieve at the taxonomic scale of this study4. We also note that specific combinations of temperature and xeric harshness are hereafter referred to as ‘climate types’ and that when we state that a species has a broader climate niche, we specifically imply that it exists in a larger number of such climate types, not that it can withstand greater variation in temperature or precipitation.
The geographic distributions of two representative arctic and montane species are depicted in (a) (purple for the Bohemian waxwing, Bombycilla garrulus, and red for the chestnut-crowned laughingthrush, Trochalopteron erythrocephalum). Correlated variation in the mean, variability, and predictability of local climate patterns was captured via two principal components labelled here as “temperature harshness”, b (where bluer colours depict colder, more variable, and less predictable temperatures) and “xeric harshness”, c (where earthier tones depict fewer, more variable, and less predictable precipitation) The availability of different climate types across the World is depicted through the coloured background in (d), where darker colours indicate a particular climate type tends to occur across a larger area of physical space (light grey indicates climate types that are not available worldwide). Highlighted regions depict the 95% probability of occurrence in climate space (a.k.a., “climate niches”) for the two species in (a) as computed from kernel density estimates. Similarly, coloured dots within these regions depict the respective climate niche centroids, which were used to determine niche “centrality”.
We used Phylogenetic Path Analysis15 to test for direct and indirect relationships among the variables in our set. Our analysis considers the following putative predictors of niche breadth (see Table 1): migration and hand-wing index (likely indicators of a species’ ability to explore space and cover more ground), diet equitability (measured as the Shannon index of percentage use of different diet categories16, which can be interpreted as an indicator of either diet flexibility or the diversity of dietary requirements, see below), niche centrality (measured as the minimum distance from a niche’s centroid to the edge of global climate space), species age (estimated as the time of divergence from nearest relative), brain size (i.e., brain mass, a reasonable proxy for behavioural flexibility and cognitive capacity among birds17,18,19,20,21,22,23,24), and body size (i.e., body mass, another potential indicator of a species’ ability to cover more ground and a likely indicator of competitive ability). Because our sample of brain volumes (N = 716 species) is much smaller than our total sample of climate niches (N = 1471 species), we evaluate first the effects of all other predictors on the full dataset and proceed to use those findings as a template to explore the potential effects of behavioural flexibility on niche breadth. A comparison of the brain sizes included in this study against all those currently available for birds indicates that our sample reasonably covers the range of brain sizes in this clade (Supplementary Fig. 3).
The complete list of directed acyclic graphs included in our initial analysis is provided in Supplementary Table 1 (the R code to run this model is also included as Supplementary Information). The model considers all possible direct effects from the candidate predictors on both niche breadth and niche centrality, as well as the possibility that migration and body size influence niche breadth indirectly through a statistical association with wing shape25 (i.e., Hand-Wing Index). We also consider here the potential effect of body size on migration (i.e., small birds are more likely to migrate than large birds26,27), and of migration and Hand-Wing Index on diet (i.e., higher mobility may allow individuals to access a greater variety of potential resources). Finally, this first model also explores the possibility that more central niche locations facilitate the evolution of wider niche breadths (e.g., by offering fewer constraints to niche expansion). Post hoc analytics15,28 indicate that the web of relationships considered in these analyses properly captures the underlying structure of our data (k = 5, q = 23, C = 12.7, p = 0.242). Additionally, results from replicate analyses on 500 different phylogenetic hypotheses indicate that the findings we report below are robust to phylogenetic uncertainty (Supplementary Table 2).
After removing all non-significant links from the model, we find direct positive effects of migration and niche centrality on niche breadth (Fig. 2A). These findings suggest that niche expansion can be facilitated by easier access to new environments6,29. For example, migratory birds may be more likely to reach new habitats because seasonal movements are occasionally affected by inclement weather and geomagnetic disturbances, sometimes leading individuals astray from their traditional migratory routes30. Similarly, when a species occupies more central positions in niche space, niche breadth may be wider because the colonization of new habitats tends to be less constrained by accessible new climates (i.e., they have easier access to new areas of climate space). It is nevertheless possible that more central niche locations favour the widening of climate niches because they tend to include climate types that are not widely distributed in geographic space and are therefore more effectively exploited in combination with other habitats (note lower density of climate types at the centre of the global niche space, Supplementary Fig. 4)31,32. However, given that niche centrality sets an upper limit to niche breadth (i.e., wide niches are found only in central locations of the climate space, but occupying those locations does not necessarily guarantee a wider niche; Supplementary Fig. 5), the constraint scenario currently appears more likely.
Black arrows represent positive effects whereas red ones indicate negative effects. Dashed lines depict significant effects under the MCC tree that were nevertheless not robust to phylogenetic uncertainty. A, B depict our findings with the full dataset, whereas (C and D) depict follow-up analyses on a reduced dataset to consider the potential effects of brain size (see green rectangles). The four model structures presented here exhibit non-significant C-values, suggesting that they properly fit the underlying data. Mig Migration, HWI Hand-Wing-Index, BS Body Size, Brain Brain Mass, NC Niche Core, NP Niche Periphery.
The analysis of our entire dataset also indicates that body size, diet breadth, and possibly time from divergence are indirectly associated with niche breadth via niche centrality (Fig. 2A). One possible explanation for these patterns is that larger bodies and simpler diets facilitate the occupation of spatially rare habitats (see lower density of climate types near the centre of climate space in Supplementary Fig. 4), where intraspecific competition for space and resources can be intense. Similarly, we see at least two possible explanations for a link between niche centrality and niche breadth. First, it is possible that niches are wider in more central locations of climate space because these climate types tend to be occupied by fewer potential competitors, resulting in ecological release (note that the most biodiverse habitats on Earth tend to occur in or around the yellow outline in Supplementary Fig. 4). In support of this idea, we observe a negative correlation between niche breadth and the number of coexisting bird species in our dataset (Supplementary Fig. 6; pGLS: estimate = −0.18, t = −6.31, DF = 1427, p < 0.001). This correlation reminds us that our metric of niche breadth is more closely associated with the breadth of the realized rather than the fundamental niche. Follow-up experiments testing the effects of ecological interactions at large scales could provide additional insights into this phenomenon. An alternative interpretation for the observed relationship between centrality and niche breadth is that more peripheral climate types in climate space are conducive to the evolution of narrower niches because they cover larger geographic areas and therefore facilitate niche specialization (see red and yellow areas in Supplementary Fig. 4). Consistent with this idea, the mode density across all climate types within a species’ climate niche is negatively correlated with its niche breadth (Supplementary Fig. 7; pGLS: estimate = −0.14, t = −5.17, DF = 1471, p < 0.001). As for the preferential occupation of more central niche locations by younger species, it is possible that the spatial rarity of these habitats leads to higher turnover rates, although we note that this effect is not robust to phylogenetic uncertainty.
We now explore hypotheses on the potential effects of brain size on the breadth of avian climate niches by overlaying these effects onto the well-supported links uncovered by our initial analysis and using the smaller sample of species for which brain size data are currently available (hereafter the “brain size dataset”). In this case, we interpret brain size as a proxy for behavioural flexibility because among birds it is positively correlated with problem solving ability18, foraging innovation20, learning21, memory22, neuron numbers17,23 and particularly with the volume of pallial areas important for general-domain cognition24. Given the allometric scaling of avian brains33, we also consider a potential link between body size and brain size. Similarly, given that larger brain sizes are known to facilitate the occupation of more variable breeding habitats in both space34,35,36 and time34,37, we consider here a potential direct effect of brain size on niche breadth38. Additionally, we explore potential direct effects of diet on brain size (as suggested by39), and of migration and Hand-Wing Index on brain size (given that the high energetic demands of long-distance flight are likely to conflict with the maintenance of large brains26,40,41). Finally, we also consider a potential link between brain size and niche centrality because brain size has been positively linked with the exploration of niche space and the colonization of new habitats34,42. The complete list of acyclic graphs considered in the analysis of our brain size dataset is provided in Supplementary Table 1.
Post hoc evaluation of model statistics indicates that the links retained from our initial analysis are insufficient for describing the relationships observed within the brain size dataset (k = 11, q = 25, C = 41.9, p = 0.006). Specifically, d-separation metrics indicate that three previously dismissed effects (i.e., those of body size, Hand-Wing Index, and diet equitability on niche breadth, see Supplementary Table 3) acquire additional relevance when we account for the potential effects of brain size. After adding these links back and confirming proper model fit (k = 8, q = 28, C = 22, p = 0.143), we find qualitatively similar trends to those reported earlier (compare Fig. 2A and C).
Many of the links observed in our initial analysis remain significant when considering only the brain size dataset but are no longer robust to phylogenetic uncertainty (Fig. 2C and Supplementary Table 4). We attribute these differences to a lack of power driven by a combination of comparatively weak effects (see relevant coefficients in Fig. 2A) and a severely reduced sample size (i.e., brain size data is available for only 716 species out of 1471 species). Such caveat aside, we can conclude from the new results that brain size is directly correlated with a variety of factors and that, just as we observed with several other variables in our earlier analysis, it is indirectly associated with niche breadth via niche centrality (Fig. 2C). Additionally, the new model indicates that although brain size has a similarly sized effect on niche centrality than body size, this effect goes in the opposite direction. Given the allometric scaling of the brain, this finding indicates that when either body or brain size are used independently to predict niche breadth, their effects may be difficult to detect (e.g., we see no effect of body size when the brain size data set is used but brain size is excluded from the analysis: Supplementary Fig. 8) or could at least be underestimated (compare coefficients for body size in Fig. 2A and C).
As in the earlier model, we also find that diet equitability is negatively associated with niche breadth. Although this relationship is estimated to be a direct effect in the new model, we note that this finding is still consistent with the notion that more diverse dietary requirements can reduce the number of habitats that are suitable for a species. Alternatively, this finding is also consistent with the idea that broader diets enable climate niche specialization by facilitating site fidelity43. We also find now that brain size and flying ability (i.e., HWI in Fig. 2C) have negative effects on niche centrality and/or niche breadth. This finding is somewhat unexpected because these traits are widely thought to facilitate the use of a greater diversity of habitats and are therefore expected to promote broader climate niches. Acknowledging from the onset that some of these links are not phylogenetically robust and may therefore be simple statistical noise (Fig. 2C), we offer here a potential explanation for them in case larger samples can more fully support them in the future. Specifically, we note again that the most geographically widespread climate types on Earth tend to occur in relatively tight clusters at the periphery of climate space. Thus, brain size could be negatively correlated with niche centrality because it facilitates the occupation of more extreme and more seasonal environments34,44 (see the north temperate region—right high density cluster in climate space—and the Old World deserts—top cluster—in Supplementary Fig. 9), and/or because they can help buffer individuals against interspecific competition in species-rich habitats like tropical forests, savannahs and grasslands45,46,47 (left cluster in Supplementary Fig. 4). As noted with diet breadth, these two possibilities are also consistent with the general idea that traits that facilitate site fidelity can ultimately contribute to climate niche specialization. Similarly, flying ability could now be negatively correlated with niche breadth because it is particularly useful in geographically widespread habitats, which happen to be tightly clustered into small areas of climate niche space.
The observation that the most common climate types on Earth occupy relatively small areas in climate space suggests that there is a potential mismatch between the area covered by geographic distributions and the breadths of climate niches. To more fully evaluate this possibility, we compared these two parameters and discovered that while there is a generally positive association between them (PGLS: Estimate = 0.58, t = 27.28, DF = 1471, p < 0.001, pseudo-R-square = 0.36; Supplementary Fig. 10), there are at least two notable exceptions. Specifically, Arctic lineages tend to have small climate niche breadths and very large geographic distributions whereas montane ones tend to exhibit the opposite pattern (e.g., Fig. 1d). These exceptions are noteworthy because these two critical habitats are experiencing faster climate change than all other regions in our planet48,49, and because large geographic distributions are commonly assumed to indicate lower vulnerability to this phenomenon50. Thus, our analysis suggests that the current practice of using the extent of geographic distributions as an indicator of vulnerability to climate change could underestimate risk in arctic species and overestimate it in montane ones (particularly in mid-elevation lineages).
It is important to consider at this point that every analysis presented so far was performed by superimposing individual climate niches unto the global climate space. Given that land masses are sometimes separated by large bodies of water, it is nevertheless possible that some species do not have geographic access to climate types that would appear to be accessible in climate space. To investigate this issue, we repeated our analyses using distributional and climatic data exclusively from the continuous land masses of the Americas (i.e., North, Central and South America). Because the patterns recovered through these analyses are qualitatively identical to those in Fig. 2 (see Supplementary Fig. 11), we conclude that geographic constraints on the availability of new climate types are not a major limiting factor in our global analysis. A possible reason for the consistency of our findings across scales is that, overall, the Americas offer a similar range of climates to that observed in the entire climate space covered by our global sample.
Having established the general drivers of niche breadth we now consider a common phenomenon: species tend not to be uniformly distributed in climate space. To better understand these habitat-use asymmetries, we repeated the above analyses substituting climate niche breadth with the breadths of the niche’s “core” and “periphery”, acknowledging in our path analysis that larger niche cores could be associated with larger niche peripheries. Operationally, we defined niche core as the area that captures a 50% probability of occurrence in climate space, and niche periphery as the remaining portion of the total niche (i.e., the area contained within the 50% and 95% contour density lines of the entire niche). As in our earlier models, post hoc statistics indicate that the proposed links properly describe the underlying structure of our data (k = 5, q = 31, C = 12.7, p = 0.242) and that consideration of phylogenetic uncertainty does not fundamentally alter the nature of our findings (Supplementary Table 5).
We acknowledge here that one possible drawback of defining niche core and periphery based solely on species’ occurrences is that these characterizations do not account for intrinsic differences in the availability of different climate types (e.g., some climate types may seem “peripheral” not because they are suboptimal, but because they cover small areas, which prevents a more frequent observation of species in them). We addressed this issue by also considering an alternative definition for niche core that includes all climate types in which the expected frequency of observing a species is equal or higher than the frequency of occurrence of the climate type itself. This alternative definition has its own downfalls as it could label even frequently used habitats as “peripheral” if the climate type in question happens to be exceptionally widespread. Reassuringly, though, our results are qualitatively similar regardless of the definition we use for core and periphery (see Supplementary Fig. 12).
Once all non-significant links are removed from the new model (Fig. 2B), we observe that, as expected, larger core areas are correlated with larger niche peripheries. This finding suggests that when species are well-suited to cope with a greater variety of environmental conditions, they are also capable of exploiting a greater diversity of either suboptimal or difficult-to-find habitats. We also find positive direct effects of migration and niche centrality on both niche core and niche periphery, which supports the common perception that opportunity and recurring long-distance seasonal movements can promote niche expansion through increased exposure to new habitats. Both of these predictors had stronger effects on the core than the periphery (Fig. 2), suggesting that they exert their influence by affecting the range of conditions that a species is most likely to encounter in daily life.
Including brain size in our analysis of niche core and periphery leads to qualitatively similar patterns to those observed for niche breadth, albeit with some loss of robustness to phylogenetic uncertainty (k = 10, q = 35, C = 22.2, p = 0.332; see Supplementary Table 6). The new analyses further indicate that the negative effect we had detected earlier for brain size on niche breadth is likely to come specifically from its effects on the niche core. As noted for migration and niche centrality, this finding suggests that having large brains tends to facilitate the occupation of primary habitats that are geographically widespread but relatively uniform in terms of climate types (see Supplementary Fig. 4). We also note that the effect of diet on niche breadth that was estimated earlier to be indirect is now a direct negative effect on the breadth of the niche periphery (Fig. 2D). Once more, this result can be seen as additional support for the notion that a wider array of dietary requirements can make it more difficult to deal with suboptimal conditions.
In conclusion, we have shown here that the evolution of avian climate niches has been shaped by a complex web of interconnected factors, resulting in patterns of association that sometimes challenge existing views on how individual variables relate to niche breadth. In the context of conservation and management, the latter observation highlights the importance of multivariable analyses in climate risk assessment because they show that independent assessment of different risk-factors (a commonly used practice50) may yield contradictory findings that could obfuscate decision making. Our findings also underscore the critical role played by the structure of the available climate niche space in shaping niche breadth. Specifically, many of the patterns we report here appear to be reasonably well explained by the fact that the three most geographically widespread habitats on Earth (namely the Holarctic, the suite of tropical savannahs, and the mid- to low-latitude deserts) include a relatively small number of climate types that happen to be clustered at the periphery of climate niche space. This deceptively simple observation is particularly important for conservation because it shows that large geographic distributions do not always result in broader climate niches and it implies that traditional risk factors, such as population size and breeding range can sometimes severely misestimate vulnerability to rapid climate change (particularly in Arctic species). Overall, our analyses highlight the importance of investigating complex eco-evolutionary phenomena from a variety of perspectives and using appropriate statistical tools that embrace, rather than overlook, the complexities of the natural world.
