Comparison of Ecological Communities: The Problem of Sample Representativeness

Obtaining an adequate, representative sample of ecological communities to make taxon richness (TR) or compositional comparisons among sites is a continuing challenge. Although randomization in the collection of sample units is often used to assure that sampling is representative, randomization does not convey the concept of how well samples represent the community or site from which they are drawn. In ecological surveys, how well a sample represents a community or site literally means the similarity in taxon composition and relative abundance between a sample and the community from which it is drawn. Using both field and simulated data, we show that the proportion of the total taxon richness at a site (%TTR) achieved with a fixed sample size varies across sites, which in turn causes changes in site-to-site differences in observed TR with sample size. This means that equal-sized samples may differentially represent the communities from which they are drawn. However, the similarity of a sample to the community from which it is drawn cannot be measured directly because the taxon composition and relative abundance of the community is usually unknown. We propose to estimate it by measuring the average similarity among replicate samples randomly drawn from a community, i.e., autosimilarity, which is measured with Jaccard Coefficient in this study. Using the same data sets, we found that: (1) samples of equal size from different sites or communities achieved different levels of autosimilarity, with lower levels achieved in taxon richer sites, indicating variation in how well samples of equal size represent their respective communities; (2) %TTR was positively and almost linearly correlated with autosimilarity, indicating that autosimilarity might be a good predictor of TTR; and (3) when samples were compared at the same level of autosimilarity, similar %TTRs across different sites were achieved (i.e., the relative differences in taxon richness among sites became independent of sample size). We conclude that standardization on autosimilarity, rather than on sample size, can improve the accuracy of taxon richness comparisons.