Species abundance distributions should underpin ordinal cover-abundance transformations

Abstract

Questions: The cover and abundance of individual plant species have been recorded on ordinal scales for millions of plots world-wide. Ordinal cover data often need to be transformed to a quantitative form (0%-100%), especially when scrutinising summed cover of multiple species. Traditional approaches to transforming ordinal data often assume that data are symmetrically distributed. However, skewed abundance patterns are ubiquitous in plant community ecology. The questions this paper addresses are (a) how can we estimate transformation values for ordinal data that account for the underlying right-skewed distribution of plant cover; (b) do different plant groups require different transformations; and (c) how do our transformations compare to other commonly used transformations within the context of exploring the aggregate properties of vegetation?

Location: Global.

Methods: We assigned Braun-Blanquet cover-abundance ordinal values to continuous cover observations. We fitted a Bayesian hierarchical beta regression to estimate the predicted mean (PM) cover of each of six plant growth forms within six ordinal classes. We illustrate our method using a case study (2,809 plots containing 95,812 observations), compare the model-derived estimates to other commonly used transformations and validate our model using an independent dataset (2,227 plots containing 51,497 observations) accessed through the VegBank database.

Results: Our model found that PM estimates differed by growth form and that previous methods overestimated cover, especially of smaller growth forms such as forbs and grasses. Our approach reduced the cumulative compounding of errors and was robust when validated against an independent dataset.

Conclusions: By accounting for the right-skewed distribution of cover data, our alternate approach for estimating transformation values can be extended to other ordinal scales. A more robust approach to transforming floristic data and aggregating cover estimates can strengthen ecological analyses to support biodiversity conservation and management.