Determinants of orchid species diversity in world islands.

Abstract

Kreft et al. (2008) presented a global analysis of factors relating to differences in species numbers among 488 island and 970 mainland floras. They tested the relationship between island characteristics (area, isolation, topography, climate and geology) and species richness using traditional and spatial models. They found that area was the strongest determinant of island species number, followed by isolation, temperature and precipitation. Altitude and island geology have shown relatively weak albeit significant effects. Yet drivers of regional diversity may depart from these global patterns, and may also be taxon specific (Patiño et al., 2014). Here we examine regional and global patterns of orchid diversity among island groups. Ackerman et al. (2007) presented a comprehensive analysis of orchid species diversity in the Caribbean. Their data included 49 islands and 728 species. They found a strong relationship between the number of species and area explaining 46% of the variability. However, maximum altitude (a measure of habitat diversity on the island) was the best predictor of species richness and accounted for 79% of the variability. Most papers, including the two mentioned earlier, assume that the relationship between logarithms of the number of species present and area is linear, although other relationships are also possible (e.g. linear relationship between nonlogged values or nonlinear relationships). Here we compare four different models describing the dependence of species richness on area, maximum altitude in the island and its latitude, using the data on orchid species richness from a global review of islands. Besides the commonly used factors (area, latitude), we also use maximum altitude, because it was the most important factor for orchids in Ackerman et al. (2007). Area and altitude of the highest peak are significantly positively correlated with the number of orchid species present on an island. Significance of latitude not surprisingly disappears when the latitudinal extent of an archipelago is small. We show that the best multivariate model for prediction of orchid species richness is the classical species–area relationship combined with linear dependence of the log number of species on altitude of the highest peak and latitude. The numbers of orchid species (S) for 117 islands from eight regions in the world: (1) Caribbean, (2) Western Pacific and Oceania, (3) Western Indian Ocean, (4) Mediterranean, (5) North Atlantic, (6) Central Atlantic, (7) North American Channel Islands and Hawai'i, and (8) South America (Falklands) (Supporting Information Fig. S1) were obtained from published articles and floras (Table S1). Island area (A), ‘altitude’ – altitude of the highest peak in the island (Alt) and absolute latitude (L) (considered to be positive for both Northern and Southern Hemisphere) were then obtained from websites of these islands or using Google Earth. Some small islands of the Bahamian archipelago (e.g. Rum Cay, Acklin's island, Mayaguana) were subsequently excluded from the analyses, as the orchid distribution records were not specific enough to compile species lists for individual islands (Correll & Correll, 1982) or their area and/or altitude were not available. In two regions (North American Channel Islands and Hawai'i, and the Mediterranean), the number of islands was not sufficient to enable a rigorous analysis and therefore these regions were excluded. Thus, four independent analyses were performed: (1) for combined global data of all orchid species (including data from all eight regions – 117 islands), for data on all orchid species in (2) the Caribbean (57 islands), (3) Western Pacific and Oceania (41 islands) and (4) Western Indian Ocean (six islands). We then performed separate linear regressions for the relationships between: (R1) loge(S) and loge(A) (i.e. the data were fitted by a power function), (R2) loge(S) and Alt and (R3) loge(S) and L. Coefficients of determination (R2) were always calculated and used as a best fit estimator of the models. where E(Y), is the expected value of Y; while Xβ is the linear function that can explicitly be non-normal (in this case Poisson), with a negative binomial link, the dependent variables are predicted from a linear combination of all the variables included in the model. All analyses were performed using ‘glm’ in R, Mass package (Venables & Ripley, 2002) and the function, glm.nb, where the dependent variable (number of species of orchids) is a function of the independent variables (altitude, absolute centroid of the latitude, landmass area, and oceanic region), from a negative binomial distribution, which is effective for data with overdispersion. We compared simple and multivariate models and report the Akaike second order information criterion (AICc) index, which is a measure of the fit of the model, penalizing for the complexity of the model. The differences between the model with the smallest AICc and other models (ΔAICc) are evaluated. Models with ΔAICc of < 2 are similar, the more parsimonious and a better fit, whereas those > 4 are considered substantially different (Arnold, 2010). The multiple regression analyses (Table 1) revealed that island area and altitude of its highest peak significantly correlated with the number of orchid species on an island in all four models studied, when the global data was considered. This was also true for the data from the Caribbean (except for area in Model 2) and in half of the models in Western Pacific and Oceania. It was not true for data from Western Indian Ocean, for which data from only six islands were available. However, the effect of latitude was significant, when the global dataset was considered, and in Western Pacific and Oceania – both datasets spanning a large range of latitudes. When archipelagos were spanning only several degrees, latitudinal effect was not significant. Model 1 explained most of the variability in all datasets explored and these relationships are depicted individually (Figs S2–S5). The results for the combined data on global orchid species diversity are shown in Fig. S2. Area was the best predictor of orchid species richness and explained 55% of the variability, followed by altitude, explaining 52% of the variability. However, when four island groups, which are either much more isolated from the mainland than the others (Madeira, Azores and Hawai'i) or are very far to the north (Iceland) are excluded, altitude becomes the best predictor of species richness with 68% of the variability explained. Latitude itself explained only 11% of the variability. Altitude was the best predictor of orchid species richness and explained 65% of the variability, followed by area, which explained 43% of the variability (Fig. S3). Latitude itself explained only 1.5% of the variation. The Lesser Antilles is the only archipelago in the Caribbean that runs north–south, which is where we expected to see the strongest latitudinal trend, and this is what we found until a dramatic drop in species richness occurs in the northernmost islands. The most species-rich islands in the Caribbean are the large islands of Cuba, Hispaniola and Jamaica, where the highest peaks in the archipelago can be found. Area was the best predictor of orchid species richness in the Western Pacific and Oceania and explained 71% of the variability, followed by altitude, explaining 62% of the variability (Fig. S4). Latitude itself explained < 10% of the variability. For Western Indian Ocean, we had data from only six islands, which makes the analysis difficult because of the low number of degrees of freedom. Thus, although the amount of variability explained looks impressive, the results for this region, shown in Fig. S5, must be treated with caution. Area was the best predictor of orchid species richness here and explained 92% of the variability, followed by altitude, explaining 66% of the variability. Latitude itself explained only 9% of the variability. Of all the models evaluated two appeared to be better models. The best model (lowest AICc) included range of altitude, area and oceanic region, which differed only slightly from the second best model which included also the latitude at the centroid (see Table 2). Altitude was part of the eight best models, irrelevant of the inclusion or exclusion of the other variables. The variable geographical distribution shown as ‘x’ in Table 2 was present in the best model. Orchid diversity is not uniformly distributed across the globe, thus this is not a surprising result. No other model clearly is as effective at including a large portion of the variance as the first two models. For orchids, species–area relationships are exceptionally strong worldwide (Schödelbauerová et al., 2009), and among Caribbean islands they are even stronger when using altitude of the highest point on the island as a predictor (Ackerman et al., 2007). While archipelagos worldwide have unique geological, climatological and biological histories, we sought to determine whether the general patterns seen in the Caribbean, a complex region comprising three archipelagos (Bahamian Archipelago, Greater Antilles, Lesser Antilles), are applicable globally. The strongest predictor of the number of orchid species on an island was altitude of its highest peak, both in the combined data for the entire world and for the Caribbean, but not for data from Western Pacific and Oceania and those from Western Indian Ocean. This is in accordance with Ackerman et al. (2007) and Schödelbauerová et al. (2009), but in contradiction with Kreft et al. (2008). The effect of latitude exists, but is weak compared to other factors, such as island size and altitude of its highest peak. For relatively small archipelagos, this effect can therefore be ignored. While the general patterns are clear, the processes that lead to them may vary considerably, even within a particular archipelago. For small islands, especially low-lying ones, there is a considerable random element to species diversity (Ackerman et al., 2007). Furthermore, island age, distance from mainland source areas, dispersal tracks, latitude, niche conservatism, rates of diversification and extinction are some of the many factors that may contribute or generate variation to the strong patterns associated with area and altitudinal amplitude (e.g. Wallace, 1863; Wilson, 1988; Baldwin & Sanderson, 1998; Whittaker et al., 2007; Kozak & Wiens, 2010; Kisel et al., 2011). Here we have shown through multivariate analyses that a general worldwide pattern exists for orchid diversity on islands and may be used as an initial means of identifying and prioritizing areas of conservation concern. Specifically, orchid species richness is strongly associated with habitat diversity as expressed by island area and altitude of its highest peak, while latitudinal extent is important for some archipelagos, but overall it offered only minor refinement. However, what is presented here are generalizations, that should be modified according to local situations. To maximize returns on conservation efforts (Efimov, 2011; Feldman & Prat, 2011; Tsiftsis et al., 2011) on specific islands or archipelagos, factors other than those considered here may be as important if not more so. Indeed, oceanic region did have a significant effect (Table 2), so it follows that the generalizations should be carefully applied for each region separately. This research was supported by the grant no.14-36098G of the GA CR to P.K. and by the MSMT within the National Sustainability Program I (NPU I), grant no. LO1415 to I.T. and Z.Š. D.L.R. was supported in part by the Guy Harvais Studentship for Orchid Research at the University of Aberdeen. I.T. and Z.Š. collected the data, I.T. and R.L.T. did the analyses, J.D.A., R.L.T., D.L.R. and P.K. wrote the text. Please note: Wiley Blackwell are not responsible for the content or functionality of any Supporting Information supplied by the authors. Any queries (other than missing material) should be directed to the New Phytologist Central Office. Fig. S1 Eight regions in the world analyzed for orchid species diversity. Fig. S2 Relationship between number of orchid species and island area, altitude of its highest peak and latitude for all islands in the world, for which data were available. Fig. S3 Relationship between number of orchid species and island area, altitude of the highest peak and latitude for data from the Caribbean. Fig. S4 Relationship between number of orchid species and island area, altitude of its highest peak and latitude for data from Western Pacific and Oceania. Fig. S5 Relationship between number of orchid species and island area, altitude of its highest peak and latitude for data from Western Indian Ocean. Table S1 List of references to orchid flora in each of the studied regions Please note: The publisher is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.