Rarity and Sparseness in Plant Communities: Impact of Minor Species Removal on Beta Diversity and Canonical Ordination

Abstract

Question

Among the “minor” species present in communities, we distinguish between true “rare” species, with infrequent occurrence (low occupancy) in a given regional data set, and “sparse” species, which may be present over most of the study area, but with low local abundance. Do rare and sparse species play a different role in the evaluation of beta diversity and in the constrained ordination of plant community data sets?

Methods

Based on their positions in the abundance-occupancy scatterplots of six contrasted vegetation data sets, we distinguished core, rural, urban, and satellite species. To disentangle the role of rarity and sparseness, we applied to each data set a progressive removal of either the least frequent or the least locally abundant species. We assessed impacts on beta diversity (q = 0, 1 and 2), and on model performance of RDA, without or after pretransformation of absolute cover values.

Results

Multiplicative beta diversity decreased with the number of removed rare species, with slightly higher values for q = 2, whereas it increased when removing sparse species, with much higher values for q = 0. With raw data or after binary or by-site transformation, the fraction of variation explained by RDA increased only slightly when removing rare species, with a more sensible increase of the relative contribution of the first canonical axis. By contrast, progressive elimination of sparse species, which mimics a lower sampling effort within each community, negatively affected model performance. Generally, the removal of rare species clearly improved the performance of RDA after double transformation (chi-square transformation), contrary to the removal of sparse species.

Conclusions

The frequently observed positive correlation between occupancy and abundance hides profound differences with critical impacts on vegetation analysis. Providing that meaningful transformations are applied, there is no need to remove rare species prior to RDA. Focusing only on abundant species during sampling is likely to limit the performance of ecological empirical models.