Spatial Patterns in the Distribution of Tropical Tree Species

See allHide authors and affiliations

Science  26 May 2000:
Vol. 288, Issue 5470, pp. 1414-1418
DOI: 10.1126/science.288.5470.1414


Fully mapped tree census plots of large area, 25 to 52 hectares, have now been completed at six different sites in tropical forests, including dry deciduous to wet evergreen forest on two continents. One of the main goals of these plots has been to evaluate spatial patterns in tropical tree populations. Here the degree of aggregation in the distribution of 1768 tree species is examined based on the average density of conspecific trees in circular neighborhoods around each tree. When all individuals larger than 1 centimeter in stem diameter were included, nearly every species was more aggregated than a random distribution. Considering only larger trees (≥ 10 centimeters in diameter), the pattern persisted, with most species being more aggregated than random. Rare species were more aggregated than common species. All six forests were very similar in all the particulars of these results.

The spatial dispersion of individuals in a species is central in ecological theory (1, 2). Patchiness, or the degree to which individuals are aggregated or dispersed, is crucial to how a species uses resources, to how it is used as a resource, and to its reproductive biology. Spatial patterns have been a particularly important theme in tropical ecology, because high diversity in the tropics begets low densities. Since Wallace (3) noted how difficult it was to find two individuals of the same species, the hyperdispersion of tropical trees has focused much of theoretical tropical ecology.

In 1979, Hubbell (4) published a large study of dispersion of trees in a dry forest in Costa Rica. His results were contrary to Wallace's long-prevailing wisdom and the Janzen–Connell prediction (5, 6) that wide dispersion is a defense against predators. Most species were aggregated, so that near neighborhoods of a tree had a higher than average density of conspecifics. Since that study, though, contradictory results have appeared, particularly from Lieberman and Lieberman (7), who found that most species in a wet forest in Costa Rica, as well as from a literature survey, were not aggregated.

Over the past two decades, we have been assembling a long-term, large-scale, global research effort on spatial patterns and dynamics of tropical forests (8, 9). An international team has now fully censused six plots in five tropical countries, mapping and identifying every individual of ≥1 cm in stem diameter over 25 to 52 ha at each plot (Table 1). The large plot size is necessary to encompass substantial populations of most tree species in the community. Major goals of this effort have been to examine Janzen–Connell effects, density dependence, and the spacing pattern of individual species.

Table 1

Six tropical forest dynamics plots that have been fully censused at least once. Underlined census year is the one used in this report. Dry season months gives the number per year with mean rainfall < 100 mm. dbh, diameter at breast height.

View this table:

The six sites represent a wide variety of tropical forests (Table 1). At one extreme, the two plots in Malaysia are in tall, evergreen forest; have no regular dry season; and include over 800 tree and treelet species each. The Sinharaja forest is also very wet and evergreen, but its island setting reduces species diversity. The site in India is in dry forest with a fairly open canopy, grassy understory, and just 70 species; the Thai site is also dry and low in diversity. The single site in Central America is moist forest, structurally quite like the Malaysian sites, but intermediate in climate and diversity. The forests also cover a wide taxonomic range. Four of the Asian sites are dominated by the family Dipterocarpaceae, but few species are shared among them. The Indian and American sites are distinct taxonomically; they are not dominated by a single family and have few (India) or no (Panama) dipterocarps.

We evaluate spatial patterning by examining neighborhoods around individual trees. For each individual, we tallied the number of conspecifics between x and x + Δx meters for all x + Δxinside the plot. We also calculated the area inside the plot of each of these annuli. The number of neighborsNx and the areaAx in each annulus at distancex were then summed over all individuals of a given species.Dx = ΣNx Ax gives the density of neighboring conspecifics as a function of distance from the average individual. This is a biologically meaningful measure of clumping, because it evaluates the conspecific population density in the neighborhood of each tree. It is closely related to Ripley'sK statistic [called the correlation integral by astronomers when applied, in three dimensions, to the distribution of galaxies (10)], but K is a cumulative distribution, whereas our neighborhood density is a probability density function; that is, Kx refers to conspecifics of <x meters from the focal tree, andDx refers to an annulus betweenx and x + Δx meters. Although the K statistic is very popular (11), our approach has the advantage of isolating specific distance classes, whereas K confounds effects at larger distances with effects at shorter distances (12, 13).

To compare species with various population densities, we standardized Dx by dividing it by the mean density of a given species across the whole plot. We call this standardized index the relative neighborhood density, or Ω. In a perfectly random distribution, Ωx = 1 for all distances x. Aggregation is indicated when Ωx > 1 at short distances, whereas Ωx < 1 at short distances indicates spacing at some scale, or hyperdispersion. A great advantage of this standardized statistic is that it is sample-size independent, which allowed us to directly compare species and stem diameter classes and offered a bootstrap method for estimating confidence limits (14). Rare species, with N < 50 individuals, had to be dealt with carefully, and most of our statistics are based only on species with at least one individual per hectare (15).

Nearly every species was aggregated when all diameter classes of ≥1 cm were included. In the six plots, 1768 species had at least one individual per hectare: 1753 were aggregated at 0 to 10 m (Ω0–10 > 1), 1490 significantly so (95% confidence limits around Ω0–10 did not include 1); 1759 were aggregated at 10 to 20 m, and 1730 were aggregated at 20 to 30 m (16). No plot had fewer than 96% of its species aggregated at 0 to 10 m, and every one of 772 species at Lambir was aggregated. Relative neighborhood density almost invariably declined with distance (Fig. 1): Ω10–20 < Ω0–10 in 1714 of 1768 species, and Ω20–30 < Ω10–20 in 1581 species. Nevertheless, Ω values in nearby distance classes were highly correlated with one another. Thus, we can use Ω0–10—the mean conspecific density within 10 m of a tree (relative to the species' overall density)—as a simple measure of the intensity of aggregation of a species.

Figure 1

Relative neighborhood density (Ωx as defined in the text) as a function of distance x in sample species. Vertical bars give 95% confidence limits (14), sometimes too small to see.

There was an enormous range in Ω0–10. Most species had values of <10, but there were a few species with much higher aggregation. The highest of all among the species with ≥1 individual per hectare, Ω0–10 = 906, was inLagerstroemia sp. (Lythraceae) in the HKK plot; 51 of the 59 individuals of this species occurred in a single clump in an area of 400 m2. Nineteen species had aggregation indices over 100, eight at Lambir and four at HKK. These results are dominated by saplings, because the vast majority of trees ≥ 1 cm diameter are <10 cm diameter.

Results hold for larger trees, though. Nearly all species were aggregated when only diameter classes of ≥10 cm were considered. There were 543 species with more than 1 individual per hectare at this size in all six plots, and 488 had Ω0–10 > 1; for 257 species, the aggregation was significant (16). But aggregation intensity weakened at greater sizes in most species: 321 of 543 had a higher Ω0–10 for all trees than for larger trees and of those with a significant difference, 84 of 102 were more aggregated at the smaller than at the larger size. At BCI, Pasoh, HKK, and Lambir, about two-thirds of the species were more aggregated at the smaller diameter class, but the pattern reversed at Sinharaja and Mudumalai and most species became more aggregated at the larger size. We repeated the analysis for trees of ≥30 cm diameter and again found most species aggregated (16).

Rare species were substantially more aggregated than common species at all but the Mudumalai site (17). Median Ω0–10 was 4 to 10 times higher in the rare abundance class than in the commonest classes (Table 2). In the most abundant species, Ω0–10 < 5, and at BCI and Pasoh mostly Ω0–10 < 2, whereas in species with fewer than 500 individuals, Ω0–10 typically ranged from 5 to 30 or higher (Fig. 2). This trend held for species as rare as 10 individuals per 50 ha, even though many species had Ω0–10 = 0 when N ≅ 10 (15). For N < 10 individuals, median scores were zero, but arithmetic means were very high. Thus, the rarest species were the most aggregated of all, and across the entire range of abundances (four orders of magnitude), the degree of clumping correlated negatively with species density. These results match Hubbell's conclusion from dry forest in Costa Rica (4) and two previous analyses of the Pasoh plot (18, 19). All results from the neighborhood density statistic Ω were confirmed by nearest-neighbor analysis (16).

Figure 2

Aggregation index (Ω0–10, the relative density of conspecifics within 10 m of focal trees) for all species with ≥100 individuals at three plots, as a function of the abundance of each species, on a log-log scale.

Table 2

Median aggregation index, Ω0–10, across species in various abundance categories in the six large plots. Number of species within each abundance category is listed under spp. Last row gives overall median Ω0–10 for all species with at least 50 individuals at each plot.

View this table:

The six forests were remarkably similar in community-wide patterns of aggregation. The range of values for Ω0–10 was similar at all sites, and the relationship between this aggregation index and abundance was essentially identical across sites. However, there were significant differences (20). Pasoh and BCI had low aggregation indices at a given abundance, whereas Sinharaja and Lambir had high indices (Table 2). These results correspond with habitat variation within the plots in that Pasoh and BCI are topographically uniform and Lambir and Sinharaja are relatively rugged (Table 1). Lambir also has a sharp soil gradient (21). Mudumalai has steep topography, but there is little indication that species respond to it, whereas at Lambir and Sinharaja many species' distributions follow topographic features (22). There are clear examples of habitat-related patchiness at most of the plots, especially Sinharaja (Fig. 3). Species in which larger trees were more densely aggregated than juveniles are also suggestive of habitat-related patchiness, because adults might be collected in sites most favorable for the species, whereas juveniles are widely dispersed.

Figure 3

Distribution maps for species also used in Fig. 1. Small circles, trees of 1 to 9.9 cm diameter; open circles, trees of ≥10 cm diameter. Grid squares = 1 ha. Vatica clumps follow ridges at Lambir. Rinorea clumps at BCI do not correlate with any known canopy, topographic, or soil feature, and the patches are probably due to limited seed dispersal (seeds disperse from exploding capsules). Shorea follows ridge tops at Sinharaja, and Eugenia is very rare at Sinharaja, but most individuals are close to several conspecifics. Additional maps published elsewhere (32, 33) illustrate many cases of habitat and dispersal limited patchiness.

On the other hand, there are many species in all the plots whose aggregated distributions indicate dispersal limitation. These species occur in circular clumps that do not correspond with topography (Fig. 3), and they had the highest values of Ω0–10. Their neighborhood density functions Ωx declined abruptly with distance (Fig. 1). We tested the importance of dispersal limitation by comparing aggregation intensity in poorly versus well-dispersed trees. Species whose seeds are dispersed by animals were assumed to be better dispersed than wind- or explosively dispersed species, and canopy trees were assumed to have well-dispersed seeds relative to understory treelets (23). We also considered the Dipterocarpaceae, a family of trees with poorly dispersed, winged seeds that dominates Southeast Asian forests.

The prediction that better dispersal reduces aggregation was partially borne out. There was no significant difference in aggregation intensity between canopy and understory species at Pasoh, but at BCI there was (24). At BCI, there was no significant difference in aggregation for animal versus nonanimal dispersed species, but the difference was fairly pronounced in the predicted direction (25). Finally, at the two Malayasian plots, dipterocarps were strikingly more aggregated than nondipterocarps (26).

Finally, the observation that aggregation is weaker in larger diameter classes supports the notion that herbivores and plant diseases play a role in reducing aggregation. Other evidence from the BCI and Pasoh plots indicates that pests have already substantially weakened aggregation intensity by the time trees enter the census at 1 cm diameter (27, 28). After 1 cm diameter, the pest effect is not particularly dramatic, and we saw no indication for further loosening of aggregations between 10- and 30-cm diameters.

  • * To whom correspondence should be addressed. E-mail: ctfs{at}


View Abstract

Stay Connected to Science

Navigate This Article