Technical Comments

Comment on “Disentangling the Drivers of β Diversity Along Latitudinal and Elevational Gradients”

See allHide authors and affiliations

Science  30 Mar 2012:
Vol. 335, Issue 6076, pp. 1573
DOI: 10.1126/science.1216393


Kraft et al. (Report, 23 September 2011, p. 1755) argued that the latitudinal trend in β diversity is spurious and just reflects a trend in γ diversity. Their results depend on the idiosyncrasies of their data, especially the latitudinally varying degree of undersampling and a local sampling setup that is not suitable for analyzing drivers of β diversity.

Kraft et al. (1) started with the premise that trends in β diversity are indicative of species responses to environmental gradients and claimed that the spatial scale of their data set “is appropriate to capture responses to fine-grained environmental heterogeneity.” They then cautioned that “It is widely recognized that β diversity is a simple function of α and γ diversity...and, therefore, is not independent of variation in either α or γ diversity.” To determine whether β diversity within the transects of their data set just reflects variation in γ diversity, they randomly shuffled the species identities of the tree stems in each transect and subtracted the resulting mean β diversity from that actually observed. This β-deviation value showed no latitudinal trend, so Kraft et al. claimed latitudinal trends in β diversity to lack ecological relevance. There are three major problems with these assertions.

First, the value of β diversity in a data set can be indicative of species responses to environmental gradients only if the sampling setup is appropriate. In particular, environmental variability within sampling units should be small relative to the environmental variability among sampling units for any species-environment relationships to be detectable in the data. If latitudinal comparisons are made, the degree of environmental variability should also be sufficiently constant among the samples representing different latitudes. Kraft et al. used a data set of 197 transects, each of which was established by Alwyn Gentry to quantify local species richness in a relatively uniform forest site. Gentry divided the transects into ten 50-m subunits because his delimitation tool was a 50-m tape measure, not in order to analyze within-transect heterogeneity. Indeed, there is no documentation of either the degree of environmental variability in the transects or on the spatial configuration of the subunits in relation to environmental heterogeneity or even to each other, so the data cannot be used to assess whether the ecological processes determining species composition differ among latitudes. In effect, each subunit provides a haphazard sample of the local species pool, so it is not surprising that Kraft et al. found the results using Gentry's original data to be very similar to results obtained with simulated random sampling.

Second, the assertion of Kraft et al. that variation in β diversity is dependent on variation in γ diversity because β is a simple function of α and γ is mathematically invalid. They defined α diversity as the average number of species observed in the subunits of a transect and γ diversity as the total number of species observed in the entire transect. They used these values to calculate proportional species turnover (2) βP = 1 – α/γ, which they called β diversity. An increase in γ does not need to have any effect on βP [or the other definitions of β used in the supporting online material of (1)] as long as α is free to vary: If α increases by the same factor that γ does, β remains unchanged (24). When all subunits are compositionally identical, βP equals zero because then α equals γ no matter how many species are involved.

This leads to the third problem: Kraft et al. observed a correlation between βP and γ and concluded that βP is causally dependent on γ. They touched on a real and relevant issue but failed to notice that the correlation between βP and γ is as spurious as the correlation between βP and latitude. Both correlations are due to sampling constraints inherent in Gentry's data set, and these cause a latitudinal gradient in the degree to which α is prevented from becoming as large as γ. Both α and γ depend on the relation between the observed number of individuals and the number of species in the local species pool (Fig. 1). It has been known for a long time that compositional differences among communities are overestimated if the communities are undersampled—that is, not all the species that are present in the community are included in the sample representing the community (57). If the local tree species pool is small (a handful of species, as in high-latitude areas), a few dozen tree stems are sufficient for all of the relevant species to be present in a subunit. However, in tropical forests, a single site can contain more than 1000 tree species (8), so each subunit has to contain thousands of stems to be representative of the local community (Fig. 1) (9). Gentry's subunits had an average of just 34 stems, so the degree of undersampling in the data is strongly correlated with latitude, which forces observed β diversity to correlate with latitude as well.

Fig. 1

The dependency of α and γ diversity (A), β diversity (B) and proportional species turnover (C) on species pool size and within-subunit number of individuals in 10-subunit data sets. Each color corresponds to a different species pool size: red, 10 species; blue, 50 species; orange, 250 species; black, 1250 species. Each line shows the mean of 1000 replicates where 10 subunits of either 10, 50, 250, 1250, or 6250 individuals were randomly drawn from a log-normal species-abundance distribution. In (A), the lower line for a given species pool size shows α diversity and the upper line, γ diversity. With a sufficiently large number of individuals, α and γ converge; all subunits were drawn from the same species pool, so any compositional differences among them are caused by undersampling. In (B), the factor by which γ exceeds α (βM = γ/α) decreases with the increasing number of individuals sampled. In (C), the proportion of species in the entire set of 10 subunits that does not fit within a single subunit (βP = 1 – α/γ) decreases with increasing number of individuals sampled.

Such results on latitudinal trends are dependent on the properties of the data set used and should not be interpreted as if they were valid for β diversity in general (10). Of course, latitudinal trends in undersampling are not only a problem of local-scale studies. Macroecological studies are also affected, because the availability of field observations on species occurrences is in general poorer in tropical areas than at higher latitudes, especially if related to the size of the regional species pool (1013).

Although Kraft et al. titled their Report “Disentangling the drivers of β diversity along latitudinal and elevational gradients,” they used an example data set where the only important driver of β diversity is the degree to which sampling units fail to provide adequate representation of the local species pool. As a consequence, their results shed no light on whether there are latitudinal gradients in community assembly processes or ecological mechanisms driving β diversity. These questions remain unanswered until they are addressed with data sets that provide an adequate, documented sampling of within-site environmental variability in addition to solving the undersampling problem.


View Abstract

Navigate This Article