Macroevolutionary dynamics of climatic niche space

Abstract

How and why lineages evolve along with niche space as they diversify and adapt to different environments is fundamental to evolution. Progress has been hampered by the difficulties of linking a robust empirical characterization of species niches with flexible evolutionary models that describe their evolution. Consequently, the relative influence of abiotic and biotic factors remains poorly understood. Here, we characterize species' two-dimensional temperature and precipitation niche space occupied (i.e. species niche envelope) as complex geometries and assess their evolution across all Aves using a model that captures heterogeneous evolutionary rates on time-calibrated phylogenies. We find that extant birds coevolved from warm, mesic climatic niches into colder and drier environments and responded to the Cretaceous–Palaeogene (K–Pg) boundary with a dramatic increase in disparity. Contrary to expectations of subsiding rates of niche evolution, our results show that overall rates have increased steadily, with some lineages experiencing exceptionally high evolutionary rates, associated with the colonization of novel niche spaces, and others showing niche stasis. Both competition- and environmental change-driven niche evolution transpire and result in highly heterogeneous rates near the present. Our findings highlight the growing ecological and conservation insights arising from the model-based integration of comprehensive environmental and phylogenetic information.

1. Introduction

Unravelling the evolutionary tempo and mode that over millions of years have brought about the substantial variation in niche space occupied by species worldwide is a foundational goal in biology [1,2]. Climatic niches, the multi-dimensional hypervolume of climatic conditions in which species can persist [3,4], represent a key component of this niche space. While recent years have seen important progress in the establishment of phylogenetic frameworks and the characterization of species’ climatic niches, we still lack a comprehensive understanding of the macroevolutionary dynamics of climatic niche evolution.

Niche divergence, whereby species adapt to a different set of environmental conditions than their ancestors, is prevalent in nature. While under a neutral expectation, lineages would diverge from one another in niche space by random drift, with niche disparity (i.e. the diversity in niches) across lineages increasing linearly with time, several constraints and fundamental mechanisms predict contrasting patterns. For instance, if the total niche space available is finite, then a natural upper bound to dissimilarity among species is expected to be met eventually. Similarly, climatic niche evolution could emerge simply from lineages dispersing and speciating across a landscape with spatial variation in climate [5,6]. Biological constraints, such as those imposed by physiology, mutation or generation time might slow or prevent the spread towards certain regions of niche space [7]. ‘Red Queen hypothesis' [8] explanations emphasize biotic drivers of evolutionary change [9], with competitive dynamics resulting in either spatial exclusion or, among coexisting lineages, an initial acceleration of niche evolution driving niche partitioning [10]. In finite geographical and ecological space, this recurrent spatial and niche partitioning is expected to increasingly hinder niche evolutionary rates as clades diversify and suffer from density-dependence [1,11]. In the absence of such constraints, however, competition among an increasing number of lineages could instead accelerate niche evolutionary rates [12]. Niche shifts might be associated with diversification in other ways. For instance, punctuated equilibrium predicts that bursts of evolutionary change occur at speciation [13]. Considering that most speciation events in terrestrial groups such as tetrapods are allopatric, incipient species will inhabit environments that differ to varying degrees, modifying the climate to which they are exposed and possibly stimulating adaptation [6,14].

Another set of ‘Court Jester' hypotheses argues for environmental changes as the prevalent evolutionary force [9]. For example, environmental fluctuations are expected to speed up evolutionary rates as species adapt to new ecological opportunities [15]. Extreme events such as the Cretaceous–Palaeogene (K–Pg) transition, triggered ubiquitous diversification and caused the preferential extinction or survival of certain phenotypes, acting as ecological filters for surviving lineages [16,17]. It also brought about a sudden worldwide winter and a devastation of forest habitat, with few lineages surviving on spatially scattered refugia and exposed to freezing temperatures and dry weather [18]. The resulting massive extinction, dispersal and population fragmentation, accompanied by the selection for small body sizes, enhanced speciation and adaptive potential rates on the lineages most tolerant to colder, drier conditions that survived and diversified [16,19].

2. Results

Here, we use a novel two-dimensional niche framework for phylogenetic analysis to characterize the joint macroevolutionary dynamics of thermal and precipitation niche domains across all extant lineages of birds. In contrast to previous analyses, our approach is the first to assess niche evolution using the full polygon occupied by each species across the two-dimensional temperature and precipitation space [4], which, for clarity, we henceforth call the species climatic niche ‘envelope' (electronic supplementary material). Figure 1b illustrates this difference: for each species, we inform the model with their occupied niche envelope rather than simply using single niche coordinates (say, the average value across each axis). A species niche envelope results from the union among all composing individuals, whereby each individual's envelope is expected to occupy a substantial portion of the species envelope [20]. Scaling up, each species' climatic envelope occupies a substantial portion of available climatic niche space, with a high degree of overlap with other species' envelopes, manifesting the need to take into account the whole of the polygonal surface.

Figure 1.

Niche reconstruction approach illustrated for hummingbirds (Trochilidae). (a–d) Demonstrate, for two species (Heliodoxa aurescens, blue; Heliodoxa branickii, red), the steps from spatial data to climatic niches and their reconstruction. (a) Shows their geographical ranges, corrected for elevation preferences. (b) Characterizes the density of temperature and precipitation conditions experienced over each range grid cell. Left shows raw densities (with lighter colours indicating lower density) and right shows the corresponding 95% Gaussian kernel density estimates (with shading and contour lines corresponding to each 0.05 density level). (c) Climatic niches for all 24 species of a comprising clade mapped as the present-day states across time and (d) with five posterior samples resulting from the niche ancestral estimation (branch colours represent rates as in (g)). (e–g) Estimated climatic niches and their evolutionary rates (expected squared difference in traits change over time, described by the Brownian motion diffusion rate matrix Σ) for all hummingbirds (Trochilidae) shown a single maximum clade credibility (MCC) phylogeny (top), and across all its branches over time (bottom, line corresponds to median, and darker and lighter shading to 50 and 95% quantiles, respectively). Graphs provide the ancestral estimates from the niche space evolution model for (e) posterior median temperature, (f) posterior median precipitation and (g) posterior median branch-specific evolutionary rates. The highlighted subclade in (c) and (d) is marked with a black vertical line and a circle around the shared ancestral node. (Online version in colour.)

Considering climatic niches as envelopes thus prevents the loss of the highly complex multivariate and within-species variation of climatic niches caused by their reduction to coordinate-point estimates (e.g. [21,22]), which compromises ancestral and rate estimates of niche evolution [23]. Specifically, we extracted, for each species, their realized temperature and precipitation during the breeding season and used Gaussian kernel density estimation to delineate their occupied climate niche space as polygonal envelopes in bivariate climate space (figure 1a–c). We then modelled the evolution of species climatic niche by performing inference over polygonal envelopes rather than point coordinates, making use of the ‘relaxed random walk' (RRW) model, a flexible and relaxed model of multivariate trait evolution that allows each branch in the phylogeny to have specific niche evolutionary rates [24]. This model allows the detection of variable evolutionary rates across branches, clades and time, without an a priori expectation of the temporal or taxonomic behaviour of evolutionary rates (figure 1d,g) [24], thus avoiding a smoothing temporal and taxonomic heterogeneity typical for most models of trait evolution, but see other approaches allowing variable rates [25]. For clarity, we consistently use the term evolutionary rate to refer to the expected squared difference in traits change over time (i.e. °C for temperature and mm/month for precipitation), which corresponds to the multivariate Brownian motion diffusion rate matrix Σ (see 'Material and methods'). For the example of hummingbirds (Trochilidae; figure 1e,f), a clade inhabiting a gamut of environments, this phylogenetic climate estimation uncovers that their broad range of occupied conditions extends to ancestral lineages, and that clade-level aggregates remain relatively constant over time.

To comprehensively describe the multivariate niche evolutionary history, we estimate cross-branch temperature and precipitation trends (figure 2a–d) and two-dimensional niche occupancy of all ancestral lineages over time (figure 2f; electronic supplementary material, video S1). Across all birds, and only reflecting lineages with extant descendants, our model estimates an ancestral origin of birds in warm, mesic environments with limited disparity across clades throughout the Cretaceous (figure 2a,b,f; electronic supplementary material, figure S3b,c). While the biogeographic origin of modern birds (Neornithes) is hypothesized to be southern (West Gondwana), where our reconstructed environment was prevalent, biogeographic evidence remains ambiguous as to the location or climate inhabited by the most recent common ancestor (MRCA) of modern birds [26]. The climate inferred at the origin of birds and subsequent climate evolution agree with the more modest latitudinal temperature gradient further back in birds’ evolutionary history compared with today, and, possibly, higher extinction rates in colder regions [27,28].

Figure 2.

Evolution and disparity in bivariate climatic niche space in extant birds. (a–d) Show reconstructed temperature and precipitation values and disparity across coexisting branches over time for 100 tree posterior samples. Middle line corresponds to the median and darker and lighter shading to 25–75 and 2.5–97.5% quantiles of branch values, respectively. (e) Expected disparity given a model of strict Brownian motion and tree topology across the full set of 100 empirical birds trees used in this study. (f) Time slices at 100, 60, 40, 20, 10 and 0 Ma of environmental space showing niche filling. The hexagonal grid is drawn by summarizing each lineage's climatic niche as a 95% HPD polygon and colouring each grid cell with the count of overlapping climatic polygons at a given time slice. As a point of reference, the currently occupied realized climatic space for all birds is shown in grey background. (g,h) Range of simulated scenarios of the effect of the K–Pg asteroid winter impact on global land temperatures (g) and precipitation (h) (following Chiarenza et al. [18]). (Online version in colour.)

Coinciding with the aftermath of the K–Pg mass extinction, at around 66.02 Ma, the range of predicted ancestral conditions and their disparity, measured as sample variance in ancestral climatic estimates across contemporary lineages, dramatically increased, with lineages colonizing colder and modestly drier environments (figure 2a,b). This shift is strongly visible in the temporal patterns of niche disparity, measured as sample variance in ancestral climatic estimates across contemporary lineages. We calculated null expectations for niche disparity given the number of species and tree topology during this time under a diffusion process of trait evolution (figure 2e) and under models of bounded evolution (electronic supplementary material, figure S7), which reflect the limited range of climatic conditions available at a given time on Earth. We find that the sudden increase after the K–Pg boundary deviates substantially from these expectations, especially so for temperature. We note that this result emerges only from the signal carried by the evolutionary history of niche evolution across the tree since our model does not explicitly account for episodic events of disparity. This suggests that the observed divergence is not the result of new lineages, but rather of specific colonizations of unoccupied niche spaces. The concurrence of this climatic niche expansion with the K–Pg Boundary is consistent with the ‘big-bang' model of diversification wherein extinction followed by the emergence of new environments and habitats gave rise to new ecological opportunities in surviving lineages [16,19]. Indeed, recent tip-dating efforts on subsections of the bird tree did show an increase in speciation rates after the K–Pg [19], and it is concordant with observations of posterior dispersal towards the cooler Northern Hemisphere [29].

After this rapid and substantial niche expansion around the K–Pg, we find climatic disparity increasing in near-linear fashion, with a final recent sharp increase (later than 4 Ma; figure 2c,d), which is not expected by the number of lineages alone. We hypothesize that this acceleration towards the present is due to the partition of a shared fundamental niche among young radiations that suffer from competitive exclusion following allopatric speciation. Recently, diverged species are expected to retain common environmental tolerances while occupying a subset of their climatic niche given the geographical partition of their ancestor, potentially leading to over-differentiated realized climatic niches [30]. Similarly, observed niche differences at these shorter time scales (later than 4 Ma) might represent differences only in their realized rather than fundamental niche space, which would not necessarily involve divergent adaptation and could result solely from biogeographic processes [6]. Evolutionary changes that result from changes in the realized niche can only be detected over such short time scales and are expected to suffer from the observed higher evolutionary rates than those changes involving the fundamental niche (figure 3a). This difference between the realized and fundamental niche is expected to decrease as lineages adapt to the new conditions. Alternatively, as past environmental fluctuations shaped available climatic space [31], lineages might have been responding evolutionarily to the emergence of novel climates [32]. Indeed, temperate and boreal climates are relatively new with their area expanding towards the present [33]. Finally, while boundary effects alone can create a final sharp increase in disparity (electronic supplementary material, figure S7), the empirical increase is much shallower than that expected under the null model, suggesting that other biological processes are at play.

Figure 3.

Rate of climatic niche evolution over time. (a) Estimated branch-specific evolutionary rates across coexisting branches along time. The horizontal white dashed lined marks rate constancy, in contrast to the empirical pattern (other graph details as in figure 2). (b) Time slices showing niche filling with hexagonal grid cells coloured by the median of the evolutionary rates among the overlapping lineages at a given time slice. (c–e) show median evolutionary rates for lineages intersecting with a hexagonal grid of niche space (see b). (c) relates median lineage rates in each niche-space cell to the time since the first colonization of that cell (i.e. first switch from ‘grey', background, to ‘colour' in (b)). (d) relates for each niche-space cell the changing median rate to the changing count of lineages in that same cell. (e) Compares the median initial rates in each niche-space cell to its present rate (line shows 1 : 1 relationship). (Online version in colour.)

We find that the continuous increase in disparity is paralleled by a general increase in climatic evolutionary rates over time (figure 3a). Rates, however, vary enormously across branches and consistently so over time (figures 3 and 4; electronic supplementary material, figure S3a), implying that at any given time, there are lineages experiencing exceptional high rates and others undergoing very low evolutionary rates. Branch scalars for median rates of bivariate evolution range from about 9 × 10⁻⁵ to 51 in hummingbirds (figure 1g) and from 2 × 10⁻⁶ to 500 in all birds (electronic supplementary material, figure S3a). This seems to conflict with expectations derived from early-burst theory, where high initial rates of trait evolution are followed by stasis as niche space gets progressively occupied [1,11]. However, by simultaneously visualizing evolutionary rates and how niche space has been progressively occupied (figure 3b; electronic supplementary material, video S2), we find that lineages experience high rates of bivariate niche evolution as they move into new niche areas, with rates subsequently waning in descendant lineages not experiencing substantial niche shifts (higher rates are assigned to larger niche shifts, electronic supplementary material, figure S6b,c).

Figure 4.

Variation in reconstructed bivariate temperature and precipitation niche space and associated evolutionary rates across extant birds shown a MCC tree with branches coloured according to median branch evolutionary rate scalar (ϕ_b). For visual clarity, each colour represents a quantile and do not represent equal magnitude. Outer circles represent clade medians for, from inside to outside, evolutionary rates, temperature and precipitation. Silhouettes characterize major clades (holding at least 50 species). (Online version in colour.)

Finally, we partition temperature and precipitation niche space into discrete (hexagonal) units to assess how rate dynamics vary across niche space over time. We find that after a new niche space is colonized and followed by a concomitant increase in the number of lineages, there is a strong slowdown in evolutionary rates (figure 3c,d). Consequently, present rates are lower than initial rates in approximately 84% of cases (figure 3e). Lastly, we find that evolution across bivariate climatic space is not independent, but rather that adaptation to warmer temperatures is associated with humid environments and vice versa (posterior median correlation = 0.51, 95% highest posterior density (HPD) interval = [0.48, 0.54]; electronic supplementary material, figure S3a).

3. Discussion

These results highlight the importance of phylogenetic scale in assessing niche evolution. At small phylogenetic scales, our results are consistent with Simpson's [1] model wherein species experience high evolutionary rates as they enter new adaptive zones but become progressively slower as niche space saturates [34]. When scaling up to broader phylogenetic scales, contemporary subclades undergo recurrent evolutionary phases of niche evolution as they diffuse through climatic niche space, resulting in both high and low evolutionary rates at any moment in time (figures 3 and 4). Across the full avian tree of life, an intriguing picture of replicate filling of environmental niche space emerges, with select lineages retaining their ancestors' niches and others occupying particular sections of temperature and precipitation space over short time scales. While larger clades generally show similar niches over time, reflective of conservatism, the underlying evolutionary rates exhibit strong heterogeneity (figure 4).

Our results suggest that, while populating niche spaces decreases subsequent evolutionary rates at short time scales, sufficient environmental opportunity has been available for young radiations to colonize new (or recycle old) adaptive zones, thus maintaining (and increasing) a set of high rates across time. These results are remarkably consistent with simulated scenarios where inter-specific competition increasingly accelerates niche evolutionary rates as lineages diversify in an effectively boundless environmental space [12]. Likewise, our results are in line with previous analyses of morphological evolution across time which have not detected any slowdown as lineage richness increases [25,35]. Considering the bounded nature of Earth's temperature and precipitation and the observation that rates of climatic evolution increase together with temperature and precipitation disparity, either more time is needed for this radiation to face density effects, or most clades are not moving into new climatic spaces but recolonizing already occupied environmental space. Alternatively, an increased availability of new environmental space could explain accelerated evolutionary rates as lineages adapt to previously unavailable climates [33].

Different realized climatic niches result only from dissimilar spatial distributions, yet the latter does not assure the former. Disregarding the interplay between biogeographic dynamics, biotic interactions and environmental adaptation hinders a mechanistic understanding of how abiotic and biotic factors underlie, if at all, the observed patterns of climatic niche evolution. If biotic interactions, such as competitive dynamics have been determinant, the seemingly relative ease at which species can overlap across climatic niche space could be explained by spatial segregation as well as differentiation along other ecological and behavioural axes that facilitate coexistence at finer scales [10,36]. Likewise, environmental adaptation dynamics might also play a role in constraining dispersal and thus climatic niche evolution for lineages following niche expansion [37,38]. Finally, climatic niche evolution could emerge from speciation, fragmentation and dispersal of lineages along with climatic heterogeneous landscapes [6]. More mechanistic models that can test for interactions across these interdependent processes are needed to advance our understanding of the drivers behind climatic niche evolution.

The evolutionary rates presented in this paper encompass a significantly larger range, by orders of magnitude than those previously reported in the literature (figure 4, electronic supplementary material). Very fast niche shifts are commonly reported over ecological time scales (but see [39]), but rarely emerge from fossil and phylogenetic analyses at evolutionary time scales [30,40]. Notably, a methodological dependence exists when a single evolutionary rate for a tree is used (smoothing underlying rate variation), and ancestor-descendant change is divided by the corresponding branch length: as the interval at which change is measured increases, estimated rates tend to decrease, particularly in methods using the strict Brownian motion or similar models where rates of change are proportional to branch length [21,41]. Our RRW model overcomes these methodological constraints by explicitly untangling branch lengths from evolutionary rates (explaining <0.01 of the variation in climatic rates, slope = − 0.21, s.e. = 0.012, electronic supplementary material, figure S6a). By explicitly considering bivariate climatic space, our analyses also minimize biases that arise from collapsing the multi-dimensional and multi-population nature of species’ climatic niches to single points, as typical for earlier rate estimates [23]. Most of our recent evolutionary rate estimates are, on average, higher than those previously suggested (median 95% temperature rates range = 3.43 − 1177.9 °C² My⁻¹), yet, even the maximum median branch rates remain below projected changes of climate change (electronic supplementary material). Interpreting these past estimates of evolutionary adaptation to novel climates as a signal for future adaptability to projected rates of temperature and precipitation change thus presents an overall adverse picture. Nonetheless, responses will be highly species and habitat-specific, and, coupled with the large rate of heterogeneity observed, might offer hope and highlight the potential for some lineages much more than others to cope with impending change.

Despite the methodological advances, our results, like all work based purely on extant taxa, remain limited by their ignorance about extinct species and palaeoclimatic information, which are not represented in our reconstruction of climatic niche evolution, and thus embody a high degree of uncertainty as one looks deeper into the past. Comprehensive fossil information could prevent potential biases in our climatic estimates if extinction probability is associated with species' climatic niches, and it would provide germane information about past rate dynamics and ancestral states [42]. For instance, what would seem like available niche space using extant species-only reconstructions could actually have been occupied by now extinct clades [35], most significantly so across the K–Pg boundary where substantial diversity was lost after the mass extinction [43]. This issue also affects the interpretation of our estimated evolutionary rates since the inclusion of extinct lineages would subdivide a branch (and its associated rate) and, since the total amount of niche divergence to explain can only be equal or larger, evolutionary rates should either increase or balance a decrease in one branch with an increase in another. Similarly, like all multivariate models of trait evolution, we assume that available climatic space is limitless (with precipitation being non-negative), yet available climatic conditions during birds' evolutionary history have had bounds that fluctuated substantially [31]. These fluctuations impacted species distributions differently, generating widespread dispersal, fragmentation and contractions, with substantial repercussions to their climatic niches [9]. Non-absorbing bounds, as expected from the evolution across climatic niche space, pose an upper bound to inferred evolutionary rates because, after some time, the amount of evolutionary change measured cannot be larger than the breadth of available niche space [21]. Similarly, species evolving near the boundary of climatic conditions could have their rates of adaptation not be as fast as they could be given unbounded space. Indeed, the steady increase in evolutionary rates towards the present could also respond to the surge of available cold, dry climatic conditions lineages could occupy. As sampling issues of the fossil record ameliorate, its inclusion can dramatically improve inference [42]. We expect that sufficient fossil evidence in the future combined with improved models of palaeoclimatic conditions [18,31] will provide an important test, and update, of our phylogenetic work on extant species and will at that point further advance our understanding of biological climatic niche dynamics.

Our analyses used comprehensively characterized bivariate climatic niches, phylogenetic information with uncertainty propagation and a flexible model of evolution to identify dramatic among-branch variation and consistent increase in evolutionary rates throughout the radiation of birds. Our results suggest an evolutionary model of recurrent events where lineages move into new niche spaces at high evolutionary rates followed by relative stasis as species diversify within. This process seems to have continued at an increasing rate even when niche space is bounded, suggesting that species can recolonize already occupied expanses of niche space at a steady pace, at least for the duration of the Aves radiation. Why some lineages maintain and conserve their ecological niche while others quickly adapt to new optima across environmental space might be due to different abiotic or biotic factors, including rare extreme events (such as the K–Pg boundary), physiology, coexistence dynamics and biogeographic history. Usually regarded as antipodes, our results support an interplay between the Red Queen, by agreeing with expectations from ubiquitous competition-driven niche evolution, and Court Jester hypotheses, by the significant effect of environmental catastrophes. We find that complex niche dynamics reiterate consistently throughout the evolutionary history of birds, but overall, estimated rates are low compared to what might be required to adapt to impending future changes. With ever-increasing rates of climatic change forecast for the planet, further research expanding a multi-dimensional understanding of niche evolution to other groups and phylogenetic scales is urgently needed. Increasingly comprehensive and robust environmental and phylogenetic data and models offer new opportunities for understanding the evolutionary dynamics at macroevolutionary scales, informing biogeography, ecology and conservation.

4. Material and methods

(a) . Avian realized climatic data

We extracted climatic niche data for all 9993 bird species, following the taxonomy of Jetz et al. [44], using the following procedure. First, we obtained expert distribution range maps for each species. These range maps have been validated at a resolution of approximately 110 km [45], describing species climatic distributions at a relatively low resolution if one considers, in particular, strong climatic gradients, such as in mountainous regions. To increase the accuracy of our climatic niche envelopes, we, therefore, incorporated elevation distributions, thereby significantly enhancing the resolution of presence–absence information in expert distribution ranges [46]. We used a recently compiled dataset that includes data for species that inhabit mountainous regions (approx. 86%) and, whenever available, uses mountain region-specific elevational information, for full description see Quintero & Jetz [46]. The remaining species do not inhabit strong elevational gradients and thus their niche accuracy should not be compromised by it. A global digital elevation model (DEM) was obtained at a resolution of 10 min latitude/longitude from New et al. [47]. We then refined the expert distribution ranges by removing all cells that had an elevation value outside from where the species occurs.

Climatic data were extracted at a resolution of 10 min longitude/latitude from New et al. [47]. The use of a 10-fold higher resolution for climatic data compared to that at which planar range maps have been validated (i.e. approx. 1° longitude/latitude at the Equator) is justified by the incorporation of three-dimensional presence–absence information [46]. For each grid cell within the refined species breeding distributions, we extracted the average annual temperature and average monthly precipitation. For migratory species, we only considered the temperature and precipitation during the breeding months when extracting climatic information (i.e. April to October for northern migrants and October to April for austral migrants). Two-hundred thirty species (approx. 2.3% of the total) were left without climatic information. Of these, 144 species inhabit small islands where climatic information was not available in New et al. [47], and we thus used the climatic characterization of islands with an area > 1 km² from Weigelt et al. [48] instead.

The remaining 86 species (approx. 0.86% of the total) are narrow-ranged species for which the elevation refinement procedure described above reduced their range completely. The cause was the resolution of the DEM used, which represents a summary average over deviant values within grid cells at a 10 min grain. Thus, we used the 30 arc-seconds resolution NASA/NGA Shuttle Radar Topography Mission DEM to refine again these species according to their elevational range. The use of a higher resolution layer for these species is justified by its necessity: our information on where these species occur is at a higher resolution than that of our previous layers after incorporating the species' elevational ranges. To obtain average annual temperature and monthly precipitation information at this resolution, we used data from Climatologies at high resolution for Earth's land surface areas [49]. Using higher resolution (30 arc-seconds) for these 0.86% of species with narrow elevational ranges (compared to the 10 min latitude/longitude resolution from New et al. [47] used for most species), helps avoid averaging over the steep environmental gradients that they inhabit.

(b) . Estimation of avian bivariate climatic envelopes

Our final climatic dataset for each bird species is a collection of m bivariate climate coordinates that make up a matrix ${\mathbf{C}}_{(m,2)}$. Given that our climatic evolution model (see below) assumes that the reconstructed traits do not have bounds, we added one unit to the precipitation data and log-transformed. This is only done to precipitation data since in contrast to temperature data, it cannot be negative. We mapped ${\mathbf{C}}_{(m,2)}$ in climatic space and used bivariate Gaussian kernel density estimation using the kde2d() function of the MASS package for R [50,51]. We assumed a common bandwidth across all species so that each niche coordinate has the same weight, independent on the range size of the species (and thus of the resulting number of climatic coordinates m). Optimal bandwidth selection is usually performed by minimizing the discrepancy between the estimator ${\widehat{\hspace{0.17em}f}}_{\mathrm{\Xi }}(x,y)$ and the true density $f_{\mathrm{\Xi }}(x,y)$ [52]. We attempted to use optimizers that minimize a measure of discrepancy such as the mean integrated square error (MISE); however, the large amount of data points (>6.8 × 10⁷ environmental coordinates) made the computation impractical. Thus, we first visually inspected the use of different kernel bandwidths and decided to use a 1% of the total axis range across all species striking a good balance between boundary delimitation and complexity for species with large m and cohesion for species with small m. Across all species, the annual temperature range is [− 23.55, 34.7] in °C and the average monthly precipitation range is [0.0, 7.33] in ln(mm month^–1), resulting in a bandwidth of approx. 0.58 and approx. 0.07, respectively. We then explored two commonly used approximations to an optimal kernel bandwidth when assuming that $f_{\mathrm{\Xi }}(x,y)$ is a multivariate normal, table 4.1 in.Silverman [52]. The resulting bandwidths where similar to those estimated by visual inspection (Silverman's rule: approx. 1.5 and approx. 0.13 for temperature and precipitation, respectively; Scott's rule: approx. 0.47 and approx. 0.04 for temperature and precipitation, respectively). Since the resulting polygons are comparable, but the 1% bandwidth estimates lie in between the other approximations, we only used the envelopes D that resulted from using the visually inspected bandwidth. Further details on the considerations of our niche delimitation are given in the electronic supplementary material. Finally, to gauge the consequences of this decision, we evaluated the impact of using different bandwidths on evolutionary rates (electronic supplementary material).

(c) . Trait evolution for niche envelopes

To model the evolution of climatic niche envelopes, we used the bivariate RRW described in Lemey et al. [24] and Pybus et al. [53] along a phylogenetic tree. This model relaxes a time-homogeneous Brownian diffusion process by multiplying tree-wide evolutionary rates with a scalar ϕ_b for each branch b. Specifically, let Σ be the infinitesimal tree-wide covariance matrix for bivariate diffusion, then the bivariate rate for branch b is rescaled to Σϕ_b, such that the process can vary across branches [24]. We denote $\mathit{\Sigma }{\varphi }_b={\mathit{\Sigma }}_b=\left(\begin{array}{cc}{\sigma }_{\mathrm{T},b}^2& {\sigma }_{\mathrm{T},\mathrm{P},b}\\ {\sigma }_{\mathrm{T},\mathrm{P},b}& {\sigma }_{\mathrm{P},b}^2\end{array}\right)$. We assume that ϕ_b follow a log-normal prior distribution with mean of 1 and estimated variance σ_B [24,53].

Methods that allow to specify polygonal envelopes instead of points as trait values at the tips have recently been developed [23,54]. Here, we numerically integrate over polygonal envelopes following [54] using Markov chain Monte Carlo (MCMC) through a uniform transition kernel over the polygonal envelope. To further accommodate the size of our dataset, we substantially improved the computational efficiency with which BEAST evaluates the model posterior density if proposed trait values fall within the polygonal domains. We accomplished 8−10-fold total run-time improvement through a two-step density evaluation and then memoization, an approach that most MCMC simulators in phylogenetics, including BEAST, have employed to minimize computationally expensive sequence data likelihood evaluations. However, memoization has remained ineffective for the relatively cheap and few prior density evaluations in traditional phylogenetic problems. At a scale with thousands of tips, evaluating if Metropolis–Hasting proposed trait values fall within specific polygonal domains, returning a non-zero density, or fall outside, returning zero density, becomes rate-limiting. The two-step procedure first quickly checks if values fall within the minimal hyper-rectangle that encloses each domain. Only when this condition holds does the procedure then more slowly evaluate if the values fall within the domain itself. Finally, memoization of the last set of values that fall within the domain avoid costly re-computation when proposed states are rejected in generating the Markov chain. These extensions are now available in BEAST v. 1.10 [55].

(d) . Evolutionary analysis of avian climatic niche envelopes

Phylogenetic information was based on Jetz et al. [44], currently the only source with all extant bird species included as tip taxa in a cohesive phylogenetic framework that uniquely enables the presented clade-wide niche analyses. To properly propagate the phylogenetic uncertainty inherent in this tree, we randomly sampled 100 phylogenetic trees from the posterior distribution and cycle through this set of posterior trees during the MCMC. Ideally, one would perform joint estimation of the tree and RRW model, but this remains computationally unfeasible. For each species we used their polygonal climatic envelope as tip values for trait evolution. Given that our data consist of the realized climatic niche for all species, ignoring dispersal limitations, biotic interactions and behavioural climatic regulation, we acknowledge that our evolutionary analysis do not reflect the evolution of the fundamental climatic niche (electronic supplementary material). MCMC convergence analysis is available in the electronic supplementary material.

Data accessibility

Data and code used in this paper are available at Dryad Digital Repository: https://doi.org/10.5061/dryad.b5mkkwhg2 [56] and includes the following files. (i) source_datasets.txt: text data describing all the publicly available datasets used to reconstruct species climatic niche envelopes used. (ii) extract_climatic_niches.r: R code used to extract bird climatic information and create their climatic niche envelopes. (iii) Beast_run.xml: xml file uses to run the analysis of climatic niche evolution. (iv) MCCtree.tre: Maximum Clade Credibility tree holding the posterior information for the climatic niche evolution analysis, including per branch median and posterior highest posterior density intervals for temperature, precipitation and rates of evolution.

Electronic supplementary material is available online [57].

Authors' contributions

I.Q.: conceptualization, data curation, formal analysis, investigation, methodology, project administration, software, visualization, writing—original draft, writing—review and editing; M.A.S.: software, writing—review and editing; W.J.: resources, supervision, validation, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This project has received funding from the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement no. 897225 for IQ. This work was partially supported by the National Institutes of Health (R01 AI153044). W.J. acknowledges support from NSF DEB-1441737.