RELIABLE ESTIMATES OF BETA DIVERSITY WITH INCOMPLETE SAMPLING
Using various recent and fossil datasets of benthic marine communities, we explore how the mean pairwise proportional dissimilarity changes when varying amounts of information are included in its calculation. We compute the mean proportional dissimilarity using only the n most abundant taxa from each sample. The values converge rapidly with increasing n, and the standard error of the mean dissimilarity of the complete matrix, based on abundance counts of all taxa, is reached between n=2 and n=10.
We also test the correlation between the dissimilarity matrix derived from the complete community composition matrix and the one derived from a degraded matrix, in which only the n most abundant taxa are included. Correlation coefficients are > 0.9 at low values of n (3 to 11) and correlations are all highly significant, even with only the single most abundant species considered in each sample.
To test which factors affect the accuracy of beta diversity estimates, we use empirical as well as simulated data to compare the correlation values with different ecological indices. The results lead us to assume that – when the data are reduced to the most abundant taxa – the community composition is better characterized when the number of taxa shared between samples increases.
Including only the five most abundant taxa for each sample depicts the dissimilarity patterns very well for most datasets. Beta diversity estimates are usually statistically indistinguishable from the estimate of the full community. When using datasets with a very high mean dissimilarity (≥ 0.9), we recommend considering the ten most abundant taxa.