• Open Access

Environmental harshness is positively correlated with intraspecific divergence in mammals and birds



Life on Earth is conspicuously more diverse in the tropics. Although this intriguing geographical pattern has been linked to many biotic and abiotic factors, their relative importance and potential interactions are still poorly understood. The way in which latitudinal changes in ecological conditions influence evolutionary processes is particularly controversial, as there is evidence for both a positive and a negative latitudinal gradient in speciation rates. Here, we identify and address some methodological issues (how patterns are analysed and how latitude is quantified) that could lead to such conflicting results. To address these issues, we assemble a comprehensive data set of the environmental correlates of latitude (including climate, net primary productivity and habitat heterogeneity) and combine it with biological, historical and molecular data to explore global patterns in recent divergence events (subspeciation). Surprisingly, we find that the harsher conditions that typify temperate habitats (lower primary productivity, decreased rainfall and more variable and unpredictable temperatures) are positively correlated with greater subspecies richness in terrestrial mammals and birds. Thus, our findings indicate that intraspecific divergence is greater in regions with lower biodiversity, a pattern that is robust to both sampling variation and latitudinal biases in taxonomic knowledge. We discuss possible causal mechanisms for the link between environmental harshness and subspecies richness (faster rates of evolution, greater likelihood of range discontinuities and more opportunities for divergence) and conclude that this pattern supports recent indications that latitudinal gradients of diversity are maintained by simultaneously higher potentials for both speciation and extinction in temperate than tropical regions.


Although temperate and tropical regions seem to support similar numbers of individuals in many major types of organisms (Enquist & Niklas 2001; Currie et al. 2004), species diversity tends to increase conspicuously around the tropics (Hillebrand 2004). This intriguing geographical pattern has been linked to latitudinal variation in biome area (Terborgh 1973), energy availability (Currie et al. 2004), climatic conditions (Francis & Currie 2003), diversification rates (Weir & Schluter 2007), historical processes (Wiens & Donoghue 2004; Jetz & Fine 2012) and species interactions (Dobzhansky 1950). However, the relative importance and potential interactions between these putative causal drivers are still poorly understood (Wiens 2011).

Evolutionary processes are thought to be particularly important to the formation and maintenance of diversity gradients because speciation and extinction can alter the numbers of species within regions. Consistent with this view, multiple lines of evidence suggest that rates of net diversification (i.e. speciation minus extinction) are highest around the tropics (reviewed in Mittelbach et al. 2007), although exceptions exist (see Soria-Carrasco & Castresana 2012; Jansson et al. 2013). Whether this pattern is a product of latitudinal differences in speciation rates (e.g. Martin & McKay 2004; Cardillo et al. 2005; Jablonski et al. 2006; McKenna & Farrell 2006; Ricklefs 2006; Wright et al. 2006; Mittelbach et al. 2007; Phillimore et al. 2007; Eo et al. 2008; Martin & Tewksbury 2008), differential species turnover (Weir & Schluter 2007, 2008), or both, is a matter of debate. In other words, it is currently unclear whether tropical regions currently harbour greater biodiversity because they are especially fertile grounds for the formation of new species, that is, ‘cradles of diversity’, or because they are less likely to lose the species they already have, that is, ‘museums of diversity’ (Stebbins 1974).

Important methodological issues complicate the study of latitudinal variation in diversification rates. For example, each species occupies a range of latitudes and exists under a variety of ecological conditions. However, evolutionary studies often reduce the geographical location of each taxon to a single value: the centroid or mode latitude of its breeding distribution. Such simplifying approach can lead to potentially misleading characterizations of the conditions under which taxa exist. For example, globally distributed species may have the same centroids as taxa with narrow tropical ranges, and the centroids of species with disjoint or ring distributions may lie in localities where these species do not even occur. To avoid such obvious pitfalls, some studies have sampled only taxa with restricted latitudinal ranges (e.g. <20° in Phillimore et al. (2007)). Unfortunately, this conservative criterion can lead to a significant loss of valuable data and does not fully capture the striking variation in environmental conditions that is observed even at these smaller geographical scales (e.g. consider differences between Miami, FL, 25.8°N, and Portland, ME, 43.6°N; Rome, 41.8°N, and Oslo, 59.9°N; New Delhi, 28.7°N, and Bangalore, 12.9°N; or Dodoma, −6.2°N, and Pretoria, −25.8°N). In addition, latitudinal bandwidth limits may lead to biased results due to the preferential exclusion of temperate taxa, which tend to have larger breeding distributions than their tropical counterparts (‘Rapoport's rule’, see Stevens 1989). Other methodological issues that complicate the analysis of latitudinal variation in diversification processes relate to the limitations of current approaches for modelling character-state or geography-dependent diversification (e.g. Maddison et al. 2007; FitzJohn 2010; Goldberg et al. 2011). These models explicitly assume that (i) the relationship between character states and diversification rates is identical everywhere across the tree and (ii) that all of the variation in diversification rates in the tree can be explained by the distribution of states in the character of interest. The assumption of process homogeneity is especially likely to be violated when analysing large phylogenetic data sets (see Alfaro et al. 2009; O'Meara 2012), as is typically the case in studies of global patterns of diversity. For example, in mammals and birds (the targets of this study), diversification rates show extreme variation across major clades, and representatives from fast-diversifying and slow-diversifying clades alike can be found in both tropical and temperate regions (Stadler 2011; Jetz et al. 2012). Similarly, given the many biotic and abiotic correlates of latitudinal diversity gradients (see Mittelbach et al. 2007), the assumption that latitude is the sole driver of diversification rates seems highly unsatisfactory.

To address these issues, we explore here the global patterns in evolutionary divergence with an explicitly mechanistic approach. We begin by acknowledging that latitude is a convenient proxy for variables that were historically difficult to measure. Thus, we assemble a broad data set of the environmental correlates of latitude (measured over entire species' distributions) and evaluate their effects directly on the process of evolutionary divergence. Second, given the limitations of current methods for estimating diversification rates on large collections of species, we explore instead the global patterns in recent divergence events within species (i.e. subspecies richness). The idea that subspecies represent instances of incipient speciation has a long history in evolutionary biology (Mayr 1940, 1942, 1982; O'Brien & Mayr 1991). In support of this view, molecular and empirical evidence indicates that avian subspecies tend to reflect populations with divergent evolutionary trajectories in terms of morphology and sexual signals (reviewed in Phillimore et al. 2007; Phillimore 2010), including clear differentiation into distinct phylogenetic clades in ca. 36% of subspecies (Phillimore & Owens 2006).

Our analyses include 2365 species of mammals and 6694 species of birds (missing species largely reflect the intentional exclusion of marine and domesticated lineages) and are informed by the most recent species-level molecular phylogenies available for each group (Bininda-Emonds et al. 2008; Jetz et al. 2012). In addition, they account for the potential effects of well-known correlates of subspecies richness, namely species' age, area of the breeding distributions, adult body size, island dwelling, habitat heterogeneity and historical exposure to glaciation (see Phillimore et al. 2007). We also consider the possibility that the formation of subspecies is less dependent on environmental parameters when species avoid environmental extremes through migration (birds) or hibernation (mammals) and address potential historical biases in the quality of subspecies designations in tropical taxa (Gippoliti & Amori 2007; Tobias et al. 2008).


Data and data sources

All of our data have been deposited in Dryad (http://datadryad.org, doi:10.5061/dryad.sb175). Starting with all extant terrestrial species recognized in the most current taxonomies for mammals and birds, we excluded all domesticated species because their current distribution and subspecies richness are likely to be determined by anthropogenic influences. In addition, we excluded all members of the mammalian superfamily Muroidea because of ‘the unrefined alpha level comprehension over vast geographical regions and the usual dearth of modern infraspecific studies to objectively delimit races and to vouch their distributions' (Musser & Carleton 2005). After removing species for which proper geographical, environmental or biometric data are not currently available, our final sample comprised 2365 of 5020 mammalian species (i.e. 47% of all extant species) and 6694 of 10063 extant avian species (i.e. 67% of all extant species). Subspecies richness (inline image ± SE, mammals: 2.68 ± 0.08; birds: 3.15 ± 0.04) was obtained from Wilson & Reeder (2005) and Clements (2000) through Phillimore et al. (2007). Breeding distributions were obtained from the 2010 IUCN Red List of Threatened Species v. 2010.4. (available at: www.iucnredlist.org. Downloaded 17 Apr 2012), and BirdLife International (available at: www.birdlife.org. Downloaded 17 Apr 2012). Adult body mass data were obtained from the study by Jones et al. (2009) and Dunning (2007). Species' ages were approximated as the length of the corresponding terminal branches in the most recent time-calibrated molecular phylogenies including all extant species in each group (Bininda-Emonds et al. 2008; Jetz et al. 2012).

Each environmental variable (including the mean, variance and predictability of annual precipitation and temperature cycles) was first measured locally for every cell within a species' range and then averaged across the entire breeding distribution to produce a single mean value per species. Precipitation and temperature data were obtained from monthly global maps (0.5 × 0.5 degree cells) for the period of 1901–2009 from the CRU-TS 3.1 Climate Database (Mitchell & Jones 2005). Predictability was measured via Colwell's P (Colwell 1974), an information-theory-based index that captures variation in the onset, intensity and duration of periodic phenomena and ranges from 0 (completely unpredictable) to 1 (fully predictable). Net primary productivity data were obtained from Imhoff et al. (2004). Hibernation data for mammals were obtained from Liow et al. (2009), Turbill et al. (2011) and Feldhamer et al. (2007) unless otherwise stated. Elevation data were obtained from the GTOPO30 data set (available at: lpdaac.usgs.gov/gtopo30/gtopo30.asp. Last accessed 17 April 2012), and mountains were identified based on elevation gradients with slopes equal or higher than 5 degrees (Fig. S1 in Supporting Information). In an effort to be consistent with prior work, we followed Phillimore et al. (2007) when quantifying island dwelling as a binary variable that indicates when 20% or more of the breeding range occurs on islands; habitat heterogeneity as the number of biomes covering at least 5% of a species breeding range (biome data from Olson et al. 2001); and historical exposure to glaciation as a binary variable distinguishing between species that had and had not exhibited more than 20% range overlap with areas glaciated in the last 21 000 years (glaciation coverage from Peltier 1993).


Severe multicollinearity problems (see Farrar & Glauber 1967) are common when multiple bioecological predictors are included in the same statistical model because these variables tend to be strongly correlated with each other. To prevent these problems, we reduced the initial set of continuous predictors to a smaller number of composite orthogonal variables via principal component analysis, PCA. Each variable in the original set was transformed, when required (following Osborne (2002)), and all variables were centred and scaled prior to PCA. For simplicity, we use throughout the PC scores from a single PCA that includes data from mammals and birds (Table 1).

Table 1. Principal components analyses of the continuous bioecological variables in our set of potential predictors of subspecies richness. Data from both mammals and birds are included. Standardized loadings of the main contributors to each component are highlighted in boldface
Bio-ecological variableEnvironmental harshness (PC1)Geographical coverage (PC2)Precipitation unpredictability (PC3)Residual body size (PC4)Uniqueness
ln (Annual variance in temperature) 0.91
Predictability of temperature cycles −0.89 −0.02−
Sqrt (mean precipitation) −0.89 0.130.08−0.050.19
(Mean temperature)2 −0.80
Net primary productivity −0.79 0.210.06−0.150.30
Sqrt (Annual variance in precipitation) −0.75 0.270.350.000.24
Habitat heterogeneity 0.61 0.400.21−0.300.33
ln (Area)0.29 0.78 −0.18−0.350.14
ln (Body size)0.08 0.51 −0.38 0.75 0.02
Predictability of precipitation cycles−0.20−0.13 −0.86 −0.340.09
SS loadings4.761.231.120.92 
% Cumulative variance explained0.480.600.710.80 

We used Bayesian phylogenetic mixed models, BPMM, with quasi-Poisson error to explore the association between bioecological predictors and the number of subspecies per species. As in phylogenetic generalized least squares regression (PGLS) models (Butler & King 2004), this statistical technique accounts for the different levels of phylogenetic relatedness between species by estimating the variance–covariance structure of the model from an independently derived molecular phylogeny (Hadfield 2010). We began each analysis with a fully parameterized model and proceeded to simplify the list of predictors by excluding nonsignificant terms one at a time, removing interactions first when necessary. Nonsignificance was assessed from credible intervals, CI, that include zero, and MCMC P-values > 0.05 (Hadfield 2010). BPMM were implemented in R (sample code in Data S2 in Supporting Information) through the package MCMCglmm, using uninformative priors and uniformly low levels of belief (Hadfield 2010). Model chains were run for 300 000 iterations with a burn-in of 10 000 iterations and thinning intervals of 50 iterations. To evaluate convergence, we assessed the mixing of MCMC chains visually (Hadfield 2010) and computed formal diagnostics suggested by Geweke (1992) and Heidelberger & Welch (1983) via the R-package ‘coda’ (Plummer et al. 2006). In the analysis of bird data, we accounted for phylogenetic uncertainty by performing analyses across 100 trees chosen at random from the set provided by Jetz et al. (2012). Thus, the statistics we report for bird models reflect averages across these 100 analyses and include the frequency of trees (f) for which MCMC P-values <0.05 (see Jetz et al. 2012).


PCA indicates that several ecological and morphological characters are strongly correlated in terrestrial mammals and birds (Table 1). The first component, which we will refer to as ‘environmental harshness’, captures differences in the level of exposure to drier, less productive environments, with colder, less predictable and more variable annual temperatures (higher scores = harsher environments). The second component, which we have termed ‘geographical coverage’, captures most of the remaining variation in breeding distribution (higher scores = larger areas) and suggests that species that occupy larger areas tend to have larger body sizes and to be present in a greater number of biome types. The third component is almost exclusively associated with the predictability of annual precipitation cycles (note the negative loading in Table 1), and the fourth component captures the remaining variation in body size (higher scores = larger species).

To clarify how these components may ultimately reflect latitudinal patterns, we explored their variation in temperate versus tropical species (Table 2). For this analysis, we classified latitudinal ranges as ‘exclusively tropical’ when a species' entire breeding distribution is contained within the tropics of Cancer and Capricorn (±23.4° latitude), ‘exclusively temperate’ when its entire breeding distribution is outside this region, and ‘mixed distribution’ otherwise. As expected, there is little latitudinal overlap on PC1: temperate mammals and birds occupy significantly drier, less productive environments, with colder, less predictable and more variable annual temperatures than their tropical counterparts (Fig. 1). Temperate species in both groups are also exposed to more unpredictable rainfall patterns, PC3, but the differences are less pronounced in this case than in PC1 (Fig. 1). Latitudinal variation in PC2 and PC4 differs between groups, with breeding distributions being larger in tropical mammals, but not in birds, and body sizes being larger in temperate mammals but smaller in temperate birds (Table 2).

Table 2. Latitudinal differences in bioecological traits. Coefficients reflect the results of multinomial phylogenetic regression models with ‘temperate’ range as the reference category
Bio-ecological variableMammalsBirds
Posterior mean (95% CI)Posterior mean (95% CI) f a
  1. a

    f = frequency of trees for which MCMC P-values < 0.05.

  2. b

    p MCMC < 0.05.

  3. c

    p MCMC < 0.01.

  4. d

    p MCMC < 0.001.

Environmental harshness (PC1)
Intercept1.31 (0.93, 1.63)d1.50 (1.35, 1.66)d1.00
Mixed−0.17 (−0.19, −0.14)d−0.20 (−0.21, −0.18)d1.00
Tropical−0.50 (−0.52, −0.47)d−0.49 (−0.51, −0.48)d1.00
Geographical coverage (PC2)
Intercept0.02 (−0.93, 0.90)0.14 (−0.31, 0.59)0.00
Mixed0.71 (0.63, 0.79)d0.64 (0.59, 0.69)d1.00
Tropical0.25 (0.16, 0.32)d0.04 (−0.01, 0.10)0.00
Precipitation unpredictability (PC3)
Intercept−0.02 (−0.98, 0.99)−0.60 (−1.16, −0.04)b0.76
Mixed0.05 (−0.06, 0.14)0.46 (0.39, 0.53)d1.00
Tropical−0.26 (−0.36, −0.16)d−0.08 (−0.15, −0.01)b0.76
Residual body size (PC4)
Intercept0.91 (−0.12, 2.03)0.72 (0.22, 1.22)c1.00
Mixed−0.41 (−0.49, −0.34)d−0.12 (−0.16, −0.07)d1.00
Tropical−0.25 (−0.32, −0.17)d0.11 (0.06, 0.16)d1.00
Figure 1.

Distribution of species with exclusively tropical (red), exclusively temperate (black) and mixed (gray) breeding distributions along the axes of environmental harshness and precipitation unpredictability. Mammalian data are shown in (A) environmental harshness and (B) precipitation unpredictability. Avian data are shown in (C) environmental harshness and (D) precipitation unpredictability.

Table 3 summarizes our findings on the potential bioecological correlates of subspecies richness. Our results indicate that intraspecific geographical divergence is influenced by similar factors in mammals and birds. In both groups, subspecies richness is greater in island dwellers as well as in species that have larger breeding distributions, smaller body size and have been historically exposed to glaciation (Fig. 2). Subspecies richness also increases with harsher environmental conditions, PC1 (Fig. 2), although this effect is not seen among birds that avoid climatic extremes through migration (see interaction term in Table 3, Fig 2E). Finally, subspecies richness peaks at intermediate rainfall unpredictability values in both mammals and birds (Fig. 2F). Spatial sensitivity analyses indicate that these results are robust to variation in environmental conditions across species' ranges in both taxonomic groups (see Data S1 and Tables S1–S2 in Supporting Information).

Table 3. Summary of results for Bayesian phylogenetic mixed models of subspecies richness in extant terrestrial mammals (excluding Muroidea) and birds
Posterior mean (95% CI)Posterior mean (95% CI) f b
  1. a

    N.S, not significant.

  2. b

    f = frequency of trees for which MCMC P-values < 0.05.

  3. c

    Hibernation in mammals, migration in birds.

  4. d

    p MCMC < 0.05.

  5. e

    p MCMC < 0.01.

  6. f

    p MCMC < 0.001.

Intercept−1.16 (−2.66, 0.33)−1.27 (−1.92, 0.63) f1.00
Dissected by mountains0.89 (0.70, 1.08)f0.37 (0.29, 0.44)f1.00
Environmental shieldingcN.S.1.05 (0.35, 1.75)e1.00
GlaciationN.S.0.17 (0.00, 0.34)d0.60
Island dwelling0.33 (0.01, 0.01)d0.66 (0.55, 0.77)f1.00
ln(species age)N.S.N.S.0.00
ln(environmental harshness, PC1)1.15 (0.79, 1.52)f0.74 (0.55, 0.92)f1.00
Geographical coverage, PC20.84 (0.71, 0.98)f0.67 (0.59, 0.74)f1.00
Precipitation unpredictability, PC3−0.24 (−0.36, −0.12)f−0.04 (−0.10, 0.01)0.00
PC32−0.06 (−0.11, −0.01)d−0.03 (−0.06, −0.01)d1.00
Residual body size, PC4−0.17 (−0.31, −0.03)d−0.46 (−0.54, −0.38)f1.00
ln(environmental harshness)d Environmental shieldingN.S.−0.88 (−1.35, −0.42)f1.00
Precipitation unpredictabilityd Environmental shieldingN.S.N.S.0.00
Figure 2.

Bioecological correlates of subspecies richness in extant terrestrial mammals and birds. Individual species are depicted as points and lines reflect fits from phylogenetically informed Poisson regression models (computed at the average value for each of the other predictors in the model). Variables on the x-axes were derived via principal components analysis. Higher scores for environmental harshness (A, E) reflect increased exposure to drier, less productive environments, with colder, less predictable and more variable temperatures. Higher rainfall unpredictability scores (B, F) imply exposure to cycles with greater interyear variation in duration, onset and extent of the rainy season. Higher scores for geographical coverage (C, G), reflect larger, more heterogeneous breeding distributions and larger body sizes. Residual body size scores (D, H) reflect the remaining variation in body size. Panel E includes regression lines fitted for migrants (dashed line) and nonmigrants (solid line). Because ln(0) is not defined, ln (environmental harshness) was computed as ln(PC1(x) - min(PC1) + 1) following Osborne(2002).

Given historical biases in research effort and taxonomic sampling, it is possible that the number of subspecies per species is currently underestimated in some tropical birds (Tobias et al. 2008) and mammals (Gippoliti & Amori 2007). To evaluate the extent to which this potential bias may have influenced our findings, we replicated our analyses using only data from resident north-temperate species (as in Weir & Schluter 2008). As seen with the global data set, the effects of environmental harshness on subspecies richness are also significant among north-temperate birds and mammals (Fig. 3 and Table S3 in Supporting Information; BPMM for mammals: β for PC1 = 0.45, CI = 0.09 to 0.87, p MCMC = 0.023; BPMM for birds: β for PC1 = 0.53, CI = 0.16 to 0.90, p MCMC = 0.005, f = 1.00). However, we found that the effects of rainfall unpredictability are not significant in either of these groups, which is not surprising given that geographical variation in precipitation levels is relatively minor above the tropic of Cancer (see Fig. 2D in Jetz & Rubenstein 2011).

Figure 3.

Variation in subspecies richness as a function of environmental harshness in (A) north-temperate mammals, (B) north-temperate birds, (C) Carnivora and (D) Sciuridae. Individual species are depicted as points and lines reflect fits from phylogenetically informed Poisson regression models that account for other potential covariates of subspecies richness (computed at the average value for each of the other predictors in the model).

Because the robustness of subspecies characterization is particularly problematic in mammalian taxonomy (e.g. see Musser & Carleton 2005), we also replicated our analyses using only well-characterized mammalian groups at two different taxonomic levels. First, we analysed three of the best-known clades: Carnivora, Primates and Artiodactyla (N = 541 species). As in the analyses presented above, we found that environmental harshness exhibits a significant positive effect on subspecies richness when considering these three groups together and accounting for phylogenetic effects and other potential covariates of diversity (BPMM: β for ln(PC1) = 0.61, CI = 0.16 to 1.08, p MCMC = 0.007). Separate analyses for each clade indicates that environmental harshness has significant effects in Carnivora (Fig. 3C, BPMM: β for ln(PC1) = 1.05, CI = 0.38–1.70, p MCMC = 0.001), but not in Primates or Artiodactyla. This finding is not surprising given that Primates and Artiodactyla occur primarily at tropical latitudes where variation in the environmental contributors to PC1 is minimal. Perhaps more interesting is that we see no significant effects of rainfall unpredictability in this sample whether we analyse the three clades combined or separately. An additional sensitivity analysis on Sciuridae, the speciose, well-known and globally distributed family of squirrels (N = 193 species), confirms this result. Here too, environmental harshness is a significant correlate of subspecies diversification (Fig. 3D, BPMM: β for ln(PC1) = 1.27, CI = 0.79–1.76, p MCMC < 0.001), but rainfall unpredictability is not.


Our findings are consistent with the hypothesis that physiology and ecology interact to promote latitudinal differences in the evolution of biological diversity (Janzen 1967). Importantly, the correlates of subspecies richness are remarkably similar in mammals and birds, and their effects are robust to accommodations for potential sampling biases and for biases related to incomplete knowledge of tropical taxonomy. Although it is clear from Primates and Artiodactyls that details may vary for particular groups (especially in predominantly tropical clades), our results indicate that at broader taxonomic scales, (i) there is a strong positive association between environmental harshness and the formation of subspecies and (ii) that this effect can generate some temperate–tropical differences in the evolution of intraspecific divergence.

Our key finding is that subspecies richness is greater when environmental conditions become more extreme as is increasingly the case in higher latitude habitats. Although this pattern is consistent with prior findings in New World mammals and birds (Weir & Schluter 2007), it is nevertheless surprising given that most indirect evidence suggests that the tropics, not the temperate regions of the world, are hotbeds for biological diversity (Mittelbach et al. 2007). Furthermore, greater subspecies richness in exclusively temperate taxa could be considered suspicious because the historically better taxonomic knowledge of temperate regions could have increased our chances of recognizing interpopulation differences (and of labelling these populations as different subspecies), artificially resulting in the same geographical pattern. Regarding these concerns, we note that (i) the positive effect of environmental harshness on subspecies richness is strong even if only north-temperate taxa are included in the analysis (i.e. the region of the world we presumably know best) and that (ii) the same result is obtained when only particularly well-studied and globally distributed groups such as Carnivora or Sciuridae are considered. Thus, the higher propensity for speciation in temperate regions could be reconciled with the reality of higher tropical diversity by considering that net diversity is the product of both speciation and extinction and that the gradients we see today may be generated by higher species turnover (i.e. simultaneously high speciation and extinction rates) in temperate than tropical regions (Schluter & Weir 2007; Weir & Schluter 2007, 2008). This possibility appears to be well supported by the abundant anecdotal evidence linking environmental harshness with higher probabilities of extinction (e.g. see Hardie & Hutchings 2010).

We see at least three possible explanations for the positive association between environmental harshness and subspecies richness. First, the stronger natural and/or sexual selection typically associated with harsher environmental conditions (see Botero et al. 2009) is likely to translate into faster rates of evolution (see Estes & Arnold 2007). Thus, other things being equal, populations in more extreme environments could conceivably diverge more quickly and may therefore be more likely to evolve into different subspecies or reproductively isolated lineages than those exposed to milder conditions. Alternatively, more extreme environments could simply provide more frequent opportunities to diverge in allopatry (Hua & Wiens 2013) because they are more likely to exhibit the type of extreme weather events that can result in local extinction and range discontinuities (see Hawkins et al. 2006; Jablonski et al. 2006; Schluter & Weir 2007). A related possibility is that more extreme environments offer more opportunities for diversification, for example through unexploited niches, because communities are less likely to achieve or maintain equilibrium under high rates of local extinction (see Rabosky 2009). The potentially opportunistic nature of evolutionary divergence is supported by the fact that subspecies richness did not simply increase with species age in any of our models (see Rabosky 2009).

At a global scale, our results indicate that characterizing the association between latitude and subspecies richness as a unidirectional gradient is overly simplistic and potentially misleading (see also Jetz et al. 2012 for a similar point of view). Specifically, different correlates of latitude appear to influence intraspecific divergence in different ways, a detail that can be overlooked when we refer to the collection of latitudinal effects as a single latitudinal gradient. For example, latitudinal patterns in body size and breeding distribution suggest lower temperate than tropical subspecies richness in mammals, whereas variation in environmental harshness suggest the opposite trend. The interaction between different environmental correlates of latitude could therefore potentially explain why studies that include different species and/or involve different regions of Earth (where the relative importance of these environmental parameters may differ) have arrived at different conclusions regarding the role of latitude on speciation rates (see Jablonski et al. 2006; McKenna & Farrell 2006; Wright et al. 2006; Mittelbach et al. 2007; Weir & Schluter 2007, 2008; Brochmann & Brysting 2008; Tobias et al. 2008).

We conclude that several key environmental correlates of latitude predict the evolution of intraspecific divergence in birds and mammals and that they are likely to explain some geographical differences in the potential for speciation. Similarly, we conclude that global patterns of subspecies richness are consistent with the notion that temperate regions are currently important producers of biological diversity (Schluter & Weir 2007; Weir & Schluter 2007, 2008) and suggest that future studies may benefit from a greater appreciation for the potential role of extinction and species turnover on the formation and maintenance of global patterns of diversity.


We thank A. Phillimore, N. Kane, P. Nosil, B. Langerhans, D. Kissling, D. Rabosky and CD Cadena for comments on earlier versions of this manuscript; L.H. Liow and K. Lintulaakso for assistance with mammalian hibernation data and R. Stepp for GIS layers with slope values for elevation. The National Evolutionary Synthesis Center, NESCent, fostered this collaboration through partial funding and logistic support to CB, RD and RS (#EF-0905606 and short-term visiting scholar fellowship to RD). This work was also supported by the US National Science Foundation (Safran: IOS #0717421 and DEB CAREER #1149942; McCain: DEB 0949601), Postdoctoral Fellowships from NESCent (CAB) and North Carolina State University (CAB) and by Grant/Cooperative Agreement # G10AC00624 from the United States Geological Survey (CAB).

C.A.B. analysed the data and wrote the article; all authors contributed data and commented on the final manuscript draft.

Data accessibility

Data Archival Location: Dryad (http://datadryad.org), doi:10.5061/dryad.sb175