The advantages of unequal sampling effort for measuring species richness
The intuition that data collected in the same way are therefore comparable is not valid when estimating species richness; sites with more species require relatively more sampling. Instead, samples can be standardized by coverage, an unbiased, information-theoretic estimator of the proportion of individuals in the sampled community belonging to species already collected. Coverage enables researchers to compare richness estimates from sites even though across sites, species total and relative abundances vary. Here we test coverage, the most quantitatively rigorous standard by which to estimate relative richness, against metrics that are commonly used in ecological studies, all calculated based on equal sampling effort across sites: abundance, observed richness, and asymptotic richness estimated using the Chao1 estimator. We also test coverage against observed and asymptotic richness estimated from data sets rarefied to an equal number of specimens, which is an alternative method of sample standardization. For these comparisons, we collected data on diverse communities of bees at 50 sites in New Jersey, USA.
We collected a total of 6281 individuals of 150 bee species, with a mean of 30.2 species per site. None of the commonly-used metrics predicted coverage-based richness estimates well. The best metric was observed richness (R2=0.36, p < 0.001). In contrast, abundance and, surprisingly, Chao1-estimated asymptotic richness both performed poorly (R2<0.1, slopes not significantly different from 0). Similarly, neither rarefied richness nor rarefied Chao1 asymptotic richness predicted coverage-based richness (R2<0.1, slopes not significantly different from 0). In general, two factors determine a richness estimate: sampling effort and the species-abundance distribution of the community. The commonly used metrics of richness consider only sampling effort and do not fully account for variation in species-abundance distributions, which decouples sampling effort from the proportion of a community represented in a sample. We found that commonly used metrics are nearly uncorrelated with the more informed, coverage-based estimate. Nevertheless, species richness estimates remain contingent upon sample coverage, underlying species abundance distribution, and scale, and not solely on the number of species in a sampled community.