The distribution of living organisms on Earth is spatially structured. Early biogeographers identified the existence of multiple zoogeographical regions, characterized by faunas with homogeneous composition that are separated by biogeographical boundaries. Yet, no study has deciphered the factors shaping the distributions of terrestrial biogeographical boundaries at the global scale. Here, using spatial regression analyses, we show that tectonic movements, sharp changes in climatic conditions and orographic barriers determine extant biogeographical boundaries. These factors lead to abrupt zoogeographical transitions when they act in concert, but their prominence varies across the globe. Clear differences exist among boundaries representing profound or shallow dissimilarities between faunas. Boundaries separating zoogeographical regions with limited divergence occur in areas with abrupt climatic transitions. In contrast, plate tectonics determine the separation between deeply divergent biogeographical realms, particularly in the Old World. Our study reveals the multiple drivers that have shaped the biogeographical regions of the world.
Naturalists have long been fascinated by the variation of life across geographical regions and have described biogeographic areas since the eighteenth century1,
We believe that this lack of knowledge comes from the complex nature and definition of biogeographical boundaries. Indeed, there is no single definition of boundary and they appear to be hierarchically structured and spatially heterogeneous. For instance, Holt et al. recently delineated the zoogeographical regions of the world by integrating species distribution data of terrestrial vertebrates with phylogenetic information11. Measuring the phylogenetic turnover between vertebrate assemblages (taken at 200 km × 200 km resolution) and using a cluster algorithm, they delineated 20 zoogeographical regions of the world that explain most of the variation in biodiversity, while maximizing the phylogenetic dissimilarities between the regions11. Interestingly, the nested nature of the dendrogram created from their cluster analysis also allowed the identification of 11 regions, at a higher level, called realms (Fig. 1)11. However, the position of cut-off points is arbitrary and, along the same dendrogram, if a deeper cut-off of similarity is used, some of the realms collapse, resulting in a smaller number of realms that are mostly consistent with the original maps of Wallace’s realms19 (Fig. 1b). In other words, some boundaries separate highly dissimilar assemblages, while others separate regions with lower dissimilarities (Fig. 1). To refer to this biogeographical hierarchy, since there is no clear accepted terminology, we use the terms shallow, intermediate and deep bioregions and boundaries. Clearly, complex determinants are responsible for this nested structure of biogeographical regions and we argue that some might explain deep bioregion boundaries, while others should be more related to intermediate and shallow boundaries. More specifically, we hypothesize that climatic heterogeneity, orographic barriers, past tectonic history and velocity of past climate change may play a major role in setting biogeographical boundaries. These factors may have a different role in explaining shallow or deep boundaries, as processes acting deeper in the past (for example, plate tectonic movements) may be most important for deep boundaries, while factors representing present-day ecological barriers (for example, climatic heterogeneity) may best explain shallow boundaries.
Climate is a major determinant of the present-day limits of species distributions20 and faunistic turnover is higher between regions with dissimilar environmental features21,22. Therefore, climate could have a major role, for instance, for shallow boundaries18. However, climatic conditions have strongly shifted during the Quaternary period, determining broad-scale changes of species distributions and modifications of assemblages23,
Here, we build on Holt et al.’s zoogeographical regionalization11 by quantitatively measuring the relative importance of the above-mentioned hypotheses across the nested structure of the global regions. First, we used spatial regression models to identify the factors best explaining the occurrence of boundaries. Second, we mapped their spatial heterogeneity, to identify global and regional variation of processes in function of climate and geological history. Third, we explored their relative importance through the nested structure of regions, to assess whether these processes play a consistent role on all the boundaries or whether some are more important for boundaries representing deep or shallow dissimilarity. Finally, we demonstrated the robustness of our conclusions to alternative classifications of zoogeographical regions6,10.
The geographical position of terrestrial biogeographical boundaries was accurately predicted by the spatial models (Supplementary Table 1). When we analysed the factors related to the overall presence of boundaries (all boundaries in Fig. 1), we found support for a joint role of climatic heterogeneity, tectonic movements during the last 65 million years and orographic barriers (Fig. 2 and Supplementary Table 1). Temperature heterogeneity and tectonic movements were the variables with the strongest overall effect size, followed by orographic barriers and heterogeneity of temperature seasonality. We did not detect any relationship between biogeographical boundaries and the velocity of Late Quaternary climate change. Velocity of climate change is strongly related to topography26 (Supplementary Table 2); however, it remained non-significant when altitude was excluded from the model (simultaneous autoregressive model; t-test of the regression coefficient: t2191 = −0.73, P = 0.46).
Geographically weighted regression (GWR) suggested that relationships between environmental features and boundaries were not homogeneous across the globe (Fig. 3a–d). Overall, temperature heterogeneity best explained the boundaries crossing Eastern Asia, Central America and North America, while heterogeneity of temperature seasonality best explained the boundaries of the Amazonian and Guineo–Congolian regions. Western Eurasia boundaries were best explained by tectonic movements, while orographic barriers best explained the Asiatic boundaries between the Arctico–Siberian, Eurasian, Tibetan and Oriental regions (Fig. 4a). Climatic variables were particularly important to define the boundaries of tropical and subtropical regions. Species turnover is the basis of biogeographical regionalization and is more strongly linked to environmental heterogeneity in the tropics than at the high latitudes21. This probably occurs because the limited short-term climatic variability in the tropics can favour physiological specialization, determining narrower niches and particularly strong responses to climate28.
We then performed sequential analyses on boundaries representing different levels of faunistic dissimilarities. The boundaries representing the shallowest dissimilarities (white lines in Fig. 1) were strongly associated with heterogeneity of temperature seasonality and, to a lesser extent, with orographic barriers (Fig. 2 and Supplementary Fig. 1). Major equatorial regions (Guineo–Congolian and Amazonian) are areas with constant temperature through the year (Supplementary Fig. 2) and their limits, particularly in the south, are strongly related to shifts towards more seasonal climates. This strongly agrees with the idea that limited seasonal variability is a major determinant of the narrow niche of tropical animals28.
When we focused on deeper biogeographical relationships (intermediate bioregions, that is, boundaries among Holt et al.’s realms11), heterogeneity of temperature was the variable with the strongest effect size, followed by plate tectonic movements and orographic barriers (Fig. 2, Supplementary Fig. 1 and Supplementary Table 1). Finally, the deepest biogeographical boundaries were mostly related to plate tectonic motion, with a consistent effect through the boundaries crossing the whole Old World (Figs 2, Fig 3, Fig 4 and Supplementary Table 1). Nevertheless, significant local relationships remained with climatic parameters and orographic barriers (Fig. 3) and the position of the boundary between the Neotropics and the Nearctic corresponded to areas with strong heterogeneity of temperature (Fig. 3e and Fig. 4b). The optimal bandwidth detected by GWRs was 1,000 km in the analysis of shallow boundaries, 1,800 km when focusing on the intermediate boundaries and 4,800 km for deep boundaries. In these spatial regression models, the optimal bandwidth identifies the distance of neighbours to include into local regressions29 and the shorter bandwidths of shallow and intermediate bioregions suggest that more local processes act on the boundaries representing limited dissimilarities.
Our analysis is a first attempt to tease apart the role of multiple factors in shaping zoogeographical boundaries at the global scale, and it shows that multiple factors often interplay to determine major transitions. For instance, past separation of tectonic plates led to long-term isolation and strong dissimilarity of faunas among continents, but biotic interchanges occurred when the movement of some plates brought isolated biotas in contact30,
Conversely, no sharp barriers exist between the Neotropics and the Nearctic; thus, the transition between these two realms is more blurred7,19,33. The northern distribution limit of Neotropical taxa is highly heterogeneous, with some Neotropical families of vertebrates limited to areas south of Panama and others ranging to Texas16. The formation of the Panama isthmus was a complex geological process, with multiple waves of dispersal of terrestrial organisms32,34 and the deepest present-day faunistic transition does not always coincide with the narrowest isthmus or with the point of contact between plates (Uramita suture)16,22,34. The dispersal of organisms between North and South America was probably limited by the interplay between availability of land and suitable environmental conditions32,34 and the transition from tropical to more temperate climates remains the most probable factor limiting biotic homogenization (Figs 3 and 4). A long-standing debate exists on the boundaries of some regions, such as the position of the southern limit of the Nearctic or the existence of the boundaries of the Sino–Japanese region; some of them have been proposed as possible transition zones19,35, even though they harbour many endemic taxa and maintain distinct biotas16,36. Temperature heterogeneity is the strongest correlate of the boundaries of these regions (Figs 3 and 4). Climatic, tectonic and orographic changes are often closely linked, but our results suggest that complex faunistic transitions may be associated with areas where climate does not act jointly with other processes.
The boundaries across Eurasia (for example, between the Palearctic and the Saharan region, and between the Sino–Japanese and the Oriental regions) were strongly related to tectonic movements, that is, the recent contact between the Eurasian, Arabian and Indian plates37, a pattern well recognized in the biogeographical literature16,38,39. The importance of tectonic movements was particularly clear in western Asia (Fig. 3c). In this region, the boundary between the Saharan and the Eurasian bioregions matches the limits of the Arabian plate well, which remained isolated from Eurasia until the Miocene epoch37,38. The formation of major mountain chains (for example, the Zagros Mountains) after the collision between Arabia and Eurasia, and the harsh climatic conditions, probably contributed to the strong differentiation between the Arabian and Eurasian faunas16. The GWR analysis performed on all boundaries taken together suggested that tectonic movements have a very broad influence over western Eurasia, with apparent effects spanning northwards up to the Urals (Fig. 3c). However, this is probably an artefact of GWR analysis, which, in this case, overestimated the influence of tectonics across space, probably because of the very strong local effect of the movements of the Arabian plate. There is indeed no global effect of tectonics on shallow boundaries (such as the one between the Eurasian and the Arctico–Siberian plates; Fig. 2). Furthermore, no tectonic movements occurred inside the Eurasian plate during the last 100 million years37 (Supplementary Fig. 4) and the boundary between the Eurasian and the Arctico–Siberian plates was clearly unrelated to tectonic movements if analysed separately (Supplementary Fig. 1).
Boundaries in eastern Asia and between the bioregions of central-northern America were related to the presence of a strong temperature gradient (Fig. 3a). Regional-scale analyses on eastern Asia yielded a similar pattern and showed that the interplay between present-day climate and elevational gradients is a strong determinant of zoogeographical boundaries in this area39. He et al. suggested that orographic barriers and tectonics were the most probable determinants of biogeographical structure in western China, while the transition from tropical to temperate and continental climates was a major determinant of the regionalization in eastern China39, which corroborates our findings.
Here, we focused on the biogeographical boundaries proposed by Holt et al.11. Alternative biogeographical structures have been proposed using both qualitative and quantitative approaches6,10,12,
We built on Holt et al.’s maps of biogeographical regions11 that we converted into a raster grid at a 200 km resolution (Mollweide equal-area projection; see Supplementary Figs 2 and 4 for Earth maps at this resolution), a scale generally appropriate for global analyses of species distribution42,43. The ‘terrestrial’ biogeographical boundaries were defined as the boundaries between zoogeographical regions that were not separated by the sea at this resolution (Fig. 1). A cell was considered to be on the boundary if a nearby cell belonged to a different zoogeographical region or realm (depending on the analysis). A few boundaries were represented by narrow sea straits that are not evident at the 200 km resolution (Gibraltar, Djibouti and La Pérouse Straits; see Fig. 1 and Supplementary Fig. 2) and were also considered among the analysed boundaries.
We considered four processes that might be related to the probability that a given world cell represents biogeographical boundaries: (1) areas of high climatic heterogeneity (climatic barriers); (2) orographic barriers; (3) tectonic separation; and (4) instability of past climate. The climatic heterogeneity hypothesis proposes that boundaries correspond to areas where climatic parameters show strong spatial turnover (heterogeneity among neighbouring cells). We considered the heterogeneity for four climatic variables: annual mean absolute temperature, temperature seasonality, annual summed precipitation and precipitation seasonality; all climatic variables were extracted from the WorldClim dataset44 up-scaled at a 200 km resolution. These variables represent both average conditions and their variability across the year, and are simple major determinants of vertebrate distribution45. Furthermore, mean annual temperature and precipitation seasonality are enough to explain most of the climatic variation at the global scale21 and other important variables (for example, summer and winter temperatures) are strongly related to linear combinations of the four climatic parameters considered in our analyses (Supplementary Table 4). To measure climate heterogeneity, for each cell, we calculated the coefficient of variation between the focal cell and its neighbouring cells, using a queen connection scheme. Therefore, the values at a given cell are higher if the cell is strongly different from its neighbours (Supplementary Fig. 4). To test for the orographic barrier hypothesis, we calculated the mean absolute difference between the altitude of each cell and its neighbouring cells. To test for the potential effect of past climatic change or stability, for each cell we calculated the average velocity of climate change since the last glacial maximum26. Past climate change from the Cenozoic could also probably explain present-day biogeographical structure. However, given that paleoclimatic reconstructions are still unable to reliably reproduce deep past climates46,
We used spatially explicit regression models to assess the factors that may explain the position of biogeographical boundaries. We first analysed the factors related to the overall presence of boundaries (all boundaries in Fig. 1; global analysis). The dependent variable was whether a grid cell was in contact with a terrestrial biogeographical boundary (Y/N; Fig. 1), while the seven environmental variables, scaled to mean = 0 and variance = 1, were the independent variables. We then performed three analyses to assess the factors related to boundaries representing different values of phylogenetic turnover: shallow phylogenetic turnover (boundaries between shallow bioregions but not between realms; white lines in Fig. 1), deep turnover (boundaries between intermediate and deep bioregions, that is, Holt et al.’s realms11) and very deep turnover (boundaries between deep bioregions, that is, Wallace’s realms4). These analyses were performed to assess the relative importance of variables identified by the global analysis in determining boundaries representing specific levels of turnover; therefore, we used variables significant in the global analysis as independent variables. Each analysis was limited to within 1,000 km from the target biogeographical boundaries, to avoid an excessive number of zeros.
The residuals of preliminary ordinary least squares regression showed significant spatial autocorrelation (global analysis: Moran’s I = 0.357; analysis on shallow boundaries: I = 0.374; analysis on intermediate boundaries: I = 0.361; and analysis on deep boundaries: I = 0.366; all analyses P < 0.001) and failure in taking into account spatial autocorrelation may bias the result of regression analyses51. Therefore, we used simultaneous autoregressive spatial (SAR) models with binomial error distribution to identify the environmental features related to the occurrence of biogeographical boundaries. SAR models are spatially explicit regression techniques that deal with spatial autocorrelation; in our models, spatial autocorrelation was incorporated in the error term using neighbourhood matrices (SARERR). SARERR are considered among the best-performing approaches to spatial regression51,
SAR models provide one single coefficient per each independent variable, representing the overall relationship (global analysis), but biogeographical and ecological relationships can often vary as a function of the location, showing strong spatial heterogeneity63. We thus used GWR analysis to assess the spatial heterogeneity of relationships between environmental features and boundaries. GWR analysis is an exploratory technique that pinpoints where non-stationarity occurs within the geographical space; that is, where locally-weighted regression coefficients deviate from their global values. If the local coefficients vary across space, this may be considered as an indication of non-stationarity29. GWR analysis was performed after the SARERR analyses, considering variables significant in SARERR. We used a binomial model and standardized independent variables. The best bandwidth was identified through a fixed Gaussian kernel; to identify the best bandwidth, we built all the models with bandwidths from 5,000 to 1,000 km at intervals of 200 km, and selected the one with lowest corrected Akaike information criterion. GWR was run using the software GWR4.0.80 (ref. 64); local significance of GWR was adjusted for multiple testing following ref. 65.
The data and the scripts that support the findings of this study are available from the corresponding author on request.
How to cite this article: Ficetola, G. F., Mazel, F. & Thuiller, W. Global determinants of zoogeographical boundaries. Nat. Ecol. Evol. 1, 0089 (2017).
We thank S. Ramdhani for providing high-resolution maps of bioregions. The research leading to these results has received funding from the European Research Council under the European Community’s Seven Framework Programme FP7/2007–2013 Grant Agreement no. 281422 (TEEMBIO). All authors belong to the Laboratoire d’Écologie Alpine, which is part of Labex OSUG@2020 (ANR10 LABX56).
Supplementary Figures 1–5, Supplementary Tables 1–5, Supplementary Discussion, Supplementary References.