Report

Beta-Diversity in Tropical Forest Trees

See allHide authors and affiliations

Science  25 Jan 2002:
Vol. 295, Issue 5555, pp. 666-669
DOI: 10.1126/science.1066854

Abstract

The high alpha-diversity of tropical forests has been amply documented, but beta-diversity—how species composition changes with distance—has seldom been studied. We present quantitative estimates of beta-diversity for tropical trees by comparing species composition of plots in lowland terra firme forest in Panama, Ecuador, and Peru. We compare observations with predictions derived from a neutral model in which habitat is uniform and only dispersal and speciation influence species turnover. We find that beta-diversity is higher in Panama than in western Amazonia and that patterns in both areas are inconsistent with the neutral model. In Panama, habitat variation appears to increase species turnover relative to Amazonia, where unexpectedly low turnover over great distances suggests that population densities of some species are bounded by as yet unidentified processes. At intermediate scales in both regions, observations can be matched by theory, suggesting that dispersal limitation, with speciation, influences species turnover.

Beta-diversity is central to concepts about what controls diversity in ecological communities. Species turnover can reflect deterministic processes, such as species' adaptations to differences in climate or substrate, or it can result from limited dispersal coupled with speciation, delayed response to climatic change, or other historical effects. Perhaps more important, beta-diversity is as important as alpha-diversity for conservation, because species turnover influences diversity at large scales. Recently, Hubbell (1) and Harte et al.(2, 3) have derived theories relating species turnover with distance to species-area relations and total species richness. In very rich forests of the neotropics, these theories may allow us to interpolate species turnover and estimate species distributions and diversity at scales relevant to conservation even with the sparse data from forest plots that are currently available.

To measure beta-diversity and test factors influencing it, we identified all trees in 34 plots near the Panama Canal, 16 plots in Ecuador's Yasunı́ National Park, and 14 plots in Peru's Manu Biosphere Reserve (4–7). All plots were in terra firme, or unflooded, forests. Over 50,000 trees ≥10-cm stem diameter were tagged, measured, and sorted to morphospecies. The similarity between two plots was measured three different ways: Sørensen's and Jaccard's measures of the fraction of species shared and the probability F that two trees chosen randomly, one from each plot, are the same species (8). The Sørensen and Jaccard indices weight all species equally; F is influenced primarily by common species. We used the overall decay of similarity in species composition with distance as a measure of beta-diversity (9).

In all three regions, the similarity between two 1-ha forest plots declined with increasing distance between them (Fig. 1). Adjacent hectares shared 70% of their species in Panama and 55% in Amazonia (10). No pair of hectares separated by over 2 km shared this high a fraction. Similarity declined rapidly with distances up to 3 to 5 km in all three regions. In Panama, this rapid decline persisted to 50 km, at which distance two plots typically shared only 1 to 15% of their species (Fig. 1). In South America, however, similarity hardly changed from 5 to 100 km, with plots at those distances consistently sharing 30 to 40% of their species (Fig. 1).

Figure 1

Sørensen similarity index between pairs of 1-ha plots as a function of distance between the plots. Only the four corner hectares of the BCI 50-ha plot were used to avoid undue influence of the single site. In Ecuador, the 25-ha plot was not included here, because species names have not yet been matched with the single hectares. Solid lines connect average values in various distance categories: red for Ecuador, black for Peru, and blue for Panama. Individual points for Peru were omitted to reduce clutter.

Panamanian plots shared few species with plots in Amazonia (averaging 8% with single plots in Peru and 5% with Ecuador). Ecuadorian and Peruvian hectares 1400 km apart shared, on average, 20% of their species—more than hectares only 50 km apart in Panama. How do these measures of beta-diversity compare with other forests? Over 9000 km of lowland boreal spruce forest (11), the natural logarithm of the Jaccard index between plots declined by 0.19 per 1000 km of distance. Between Peru and Ecuador, the same decline was 0.55 per 1000 km, whereas from Panama to Ecuador, it was 1.85. In these tropical regions, species turnover is higher than in boreal forest.

We presume that varied climate and geology accelerate species turnover in Panama. Annual rainfall is <2000 mm near the Pacific and >3000 mm near the Caribbean, and many different geological formations underlie the plots (4). Habitat type influences species distribution: For example, tree species common in dry areas reappear on rapidly draining soils in wet areas (4). In contrast, the plots in Peru and Ecuador have relatively similar soils (12), and climate varies little within either region. Unlike Panama, species turnover in western Amazonia should reflect mainly dispersal limitation: Seeds seldom travel far (13), so distant sites are less likely to share species.

To assess the influence of limited dispersal on beta-diversity, we consider a model for how similarity should change with distance in a community where only dispersal and speciation affect species distributions. This theory provides a null hypothesis by which we can measure the impact of influences that the model ignores; without it, we were unable to assess the role dispersal limitation might play in beta-diversity. To generate quantitative predictions, the model makes the simplifying assumptions of Hubbell's neutral theory (1)—all species are identical, trees mature instantly, and new species arise from single individuals. Despite these simplifications, a dispersal model of beta-diversity is warranted, given the ample discussion on how dispersal affects forest communities at both local and continental scales (13).

To derive the theory, we borrow population genetic methods for analyzing how allelic similarity changes with distance (14,15). With these methods, we calculate the probabilityF(r) that two randomly selected treesr km apart are conspecific. Let all trees in the forest have the same prospects of death, reproduction, and dispersal. When a tree dies, let a seed-parent chosen at random from the dead tree's neighbors provide an instantly maturing replacement. Let this replacement have probability ν of being an entirely new species. Define the dispersal function P(r) as the probability that a tree at a particular location r km away is the parent of the replacement and let P(r) be a radially symmetric Gaussian density, centered on the replacement. Assume that speciation is in complete balance with extinction, so thatF(r) does not change with time (a balance that may take 2/ν generations to attain). Then the probabilityF(r) that two trees r km apart are conspecific isEmbedded Image(1)when r > σ, andEmbedded Image(2)when r < σ.K 0 is the modified Bessel function, 2σ2 is the mean square dispersal distance from parent to surviving offspring, ρ is tree density, and ν is speciation rate. For large r, Eq. 1 also holds at least approximately for any dispersal kernel with a finite third moment. Analogous approximations can be derived for the “fat-tailed” Cauchy kernel (16). These derivations are sketched in the supplemental material (7).

The theory suggests that similarity decays monotonically with distance and that, over a wide range of distances, the decline is linear with log-distance. This aspect of the theory resembles data from Panama and Western Amazonia. In addition, values for the dispersal parameter close to those measured in the 50-ha plot in Panama—a mean of 39 m for 65 species (17)—produce theoretical similarity curves resembling those observed (Fig. 2). For example, with σ = 55 m in Ecuador, the theoretical curve matches data from r = 0.2 km tor = 50 km (Fig. 2). Higher beta-diversity in Panama can be fit with a lower dispersal parameter (σ = 40 m; Fig. 2).

Figure 2

The probability F that randomly selected pairs of trees are the same species, as a function of distancer, on a semilogarithmic scale, in Panama (top) and Ecuador (bottom), and a best fit of the dispersal model to the data for r > 100 m. Within the large plots (▵), F was calculated in 5-m distance categories from all pairs of individual trees (8). Because there is some habitat variation in species composition in the 50-ha BCI plot (27), we only used pairs when both trees were on the lower part of the flat plateau in the center of the plot, excluding the slopes around it, a swamp, and the higher part of the plateau. Likewise, in the Yasunı́ 25-ha plot, we only consider pairs when both trees were on ridges, excluding a flat, wet valley. For 1-ha plots (•),F was found by summing the product of relative abundances from each plot (8), and results are presented as the mean and bootstrap-estimated confidence limits for F within several distance categories. The prediction from Eqs. 1 and 2 was found by setting ρ = one tree per 23.3 m2 in Panama and one tree per 15.3 m2 in Ecuador, the observed tree densities. Then distance r must be measured in “tree units,” so a unit of distance is 4.8 m in Panama and 3.9 m in Ecuador; in the figure, distances were reconverted to kilometers. The parameters σ and ν were fit to the data (including Fin large plots from distance bins where the sample size exceeded 1000 but with r > 100 m and all 1-ha plots) by minimizing the sum of squared deviations with a Nelder-Mead search. The lines show the resulting fit: Panama, σ = 40.2, ν = 4.8 × 10−8; Ecuador, σ = 54.8, ν = 3.6 × 10−11; and Peru (not shown), σ = 73.0, ν = 1.7 × 10−14 . The solid line is from Eq. 1, and the dashed line is from Eq. 2.

Closer comparison of the observed and predicted beta-diversity suggests, however, that habitat variation is the cause of at least some species turnover in Panama. Variance in similarity at a given distance is three times higher in Panama than in Amazonia (18), but according to the theory, variance can be due only to sampling error, which should be identical in both regions. Furthermore, there are instances where Panamanian plots on distinct substrate differ more in vegetation than plots on the same substrate (4,19). Is species turnover steepened by habitat variation in Panama but governed chiefly by dispersal limitation in western Amazonia?

It seems not. Even in Amazonia, dispersal theory alone is insufficient: It cannot simultaneously accommodate the very steep decay in similarity observed in Ecuador from 0 to 100 m, the more gradual decline seen at both sites in Amazonia between 0.5 and 50 km, and the very slight decline between 50 km and 1400 km (Fig. 2; the steep decline within 100 m was also observed in Panama). The dispersal parameter σ must be set to 16 m to fit the data from 0 to 100 m in the 25-ha plot in Ecuador, 55 m to fit the data from 0.2 to 50 km in Ecuador, and 81 m to fit the similarity between Ecuador and Peru. This suggests that different factors influence beta-diversity at different scales.

The rapid decline of similarity at short distance suggests that species are more aggregated than dispersal theory predicts. This may reflect old light gaps that only a few species happened to colonize or high variation in adult reproductive output; both can produce dense aggregations of conspecifics (20). The high similarity between Ecuador and Peru arises because many tree species are common at both sites (6), suggesting a factor favoring similarity that partially overrides dispersal limitation (21). For example, the palm Iriartea deltoidea is the most common species in most plots in Ecuador and Peru (6), as well as at one wet site in Panama. Our dispersal theory cannot account for such an abundant, widespread species. High similarity over long distances could reflect equilibrating processes that control density of species over wide areas, such as differences in life history or pest resistance. Once a species reaches a site, its population tends toward a “preferred” density, overcoming the influence of dispersal limitation.

We have shown striking differences in beta-diversity in forests of Central Panama versus western Amazonia and have argued that the patterns cannot be explained by limited dispersal and speciation alone. Although our null model fits species turnover for plots separated by 0.2 to 50 km, discrepancies at other scales suggest that additional factors must be important. The role of habitat heterogeneity at local scales and the impact of widespread species would not have been evident without a quantitative null model for beta-diversity. A full understanding of turnover in tree species composition at all scales will require reckoning not only with speciation and limited dispersal but with habitat structure and species differences.

REFERENCES AND NOTES

View Abstract

Navigate This Article