Default cover translation in TURBOVEG can inflate diversity metrics

…

…

The practical significance of these findings becomes clearer when one considers how deeply embedded TURBOVEG is in the infrastructure of vegetation science. Developed originally in the Netherlands as a tool for managing relevé data, the software has become the de facto standard for storing vegetation-plot records, and the two largest international archives of such data—the European Vegetation Archive and its global counterpart sPlot—depend on it, directly or indirectly, for much of their ingested material. Because the software ships with a default translation table that converts the ordinal categories of the Braun-Blanquet scale into percent cover values, and because most users neither inspect nor modify these defaults, the specific numbers embedded in that table have propagated silently into thousands of published analyses. The authors of the current study demonstrate that this is not a trivial technical detail: the default values deviate substantially from both the arithmetic and geometric midpoints of the class intervals, and these deviations are concentrated precisely in the categories that matter most for diversity calculations.

The direction of the bias deserves particular attention. In the TURBOVEG table, the two lowest categories, r and +, are assigned cover values of 1 and 2 percent respectively, whereas the arithmetic midpoints would be 0.05 and 0.55 percent and the geometric midpoints 0.01 and 0.32 percent. This means that rare species—those present with only a few small individuals or a trace of cover—are effectively inflated by an order of magnitude or more when the default translation is applied. Since Shannon diversity, Shannon evenness, and Simpson diversity all reward the presence of additional species, and since evenness in particular is sensitive to how evenly abundance is distributed among species, systematically inflating the contribution of rare species raises all three metrics. The case study quantified this effect: Shannon diversity increased by roughly 0.7 units on average, Shannon evenness by about 0.2 units, corresponding to an increase of approximately thirty percent. These are not marginal discrepancies; they are large enough to alter ecological conclusions, for example when comparing diversity across sites, habitats, or regions, or when testing hypotheses about the drivers of diversity patterns.

It is worth emphasizing why the choice of midpoint matters so much for the low cover classes. Cover values in vegetation plots follow a strongly skewed distribution: most species in most plots are rare, occupying only a tiny fraction of the plot area, while a few species dominate. The lower categories of any ordinal scale therefore contain a disproportionate share of all species records, and any distortion in how those categories are translated into numbers affects a large fraction of the data. The Braun-Blanquet category r, for instance, is meant to capture individuals that occupy essentially negligible area—perhaps a single seedling in a plot of ten or one hundred square metres. Assigning such a record a cover value of one percent, as the TURBOVEG default does, overstates its contribution by a factor of ten to a hundred depending on the true value. When hundreds of such records accumulate in a plot-level diversity calculation, the cumulative effect is substantial.

The study also highlights a more fundamental problem with the Braun-Blanquet scale itself: the absence of a universally agreed definition of its categories. Although the scale is attributed to Braun-Blanquet and is used worldwide, the original 1964 textbook proposed variants with five or six steps, whereas the seven-step and nine-step versions that dominate actual practice were largely developed by later authors. Moreover, different published versions of the scale define the category boundaries differently, and some of the most widely used variants are combined cover-abundance scales in which the lower categories are defined by the number of individuals rather than by cover. For such combined scales, a proper back-translation to percent cover is not merely difficult but logically impossible, because the original information needed to reconstruct a cover value was never recorded. This means that even a perfect translation table cannot fully compensate for the loss of information inherent in ordinal recording; the double translation—from estimated percent to an ordinal category and back to an inferred percent value—introduces avoidable error at every step.

Given these considerations, the recommendation that vegetation scientists record cover directly in percent wherever feasible follows naturally. Direct percent estimates eliminate the intermediate step entirely, preserving the full resolution of the observer’s judgment and avoiding the arbitrary choices embedded in any translation table. The authors note that direct percent recording has become more prominent in recent decades, and the statistics from the European Vegetation Archive reflect this shift: among plots recorded in the twenty-first century, direct percent estimates account for thirty percent of records, compared with thirty-four percent for Braun-Blanquet variants and twenty-nine percent for presence-absence only. Nevertheless, the enormous legacy of ordinally recorded plots—more than half of all records in the archive use some variant of the Braun-Blanquet scale—means that back-transformation will remain a necessary step for the foreseeable future, and the choice of transformation will continue to shape the results of large-scale comparative analyses.

For researchers working with existing databases, the practical implications are straightforward. First, the default TURBOVEG translation should be avoided in any analysis involving species cover, whether the focus is biodiversity metrics or other cover-dependent quantities such as biomass estimation, functional trait weighting, or community similarity measures. Second, when a choice must be made between arithmetic and geometric midpoints, the evidence suggests that neither is systematically superior; both introduce some bias relative to direct percent estimates, but the biases are small and inconsistent compared with the large and consistent inflation produced by the TURBOVEG defaults. Third, authors of future studies should document explicitly which transformation they applied, since the choice can shift diversity values by amounts comparable to the ecological differences being tested. Database managers, meanwhile, might consider whether the default translation tables in widely used software should be revised or at least accompanied by prominent warnings, given that the silent propagation of these defaults has already affected a substantial body of published research.

Beyond the immediate practical recommendations, the study contributes to a broader methodological conversation about measurement error in ecology. Vegetation science has long relied on visual estimation of cover, a technique that is fast and inexpensive but subject to observer bias and, as this work shows, to additional distortion when ordinal categories are imposed on top of the estimates. The finding that the choice of back-transformation can alter diversity metrics by thirty percent underscores that seemingly minor methodological conventions—rarely reported in the methods sections of papers and almost never tested—can have consequences as large as the ecological patterns under investigation. This argues for greater transparency in reporting how ordinal data were converted to numeric values, and for sensitivity analyses in which key results are checked against alternative transformations. It also strengthens the case for community-wide standards that favour direct percent recording in new surveys, so that future generations of researchers inherit datasets that require fewer assumptions to analyse.

In summary, the evidence assembled here demonstrates that a widely used, largely unnoticed default in vegetation database software can substantially and consistently inflate cover-based diversity metrics. Because the affected databases underpin much of modern comparative vegetation ecology, the finding has implications well beyond the three case-study datasets examined. Researchers, database curators, and software developers all have a role to play in ensuring that the numerical foundations of vegetation analyses are as sound as the ecological questions they are meant to answer.

Subject of Research: …

Article Title: Translation of the Braun-Blanquet scale to percent in TURBOVEG can bias diversity metrics

Article References: Dengler, J., Dembicz, I., & Kozub, Ł. (2026). Translation of the Braun-Blanquet scale to percent in TURBOVEG can bias diversity metrics. Vegetation Classification and Survey, 7, 95-101. https://doi.org/10.3897/vcs.198373

Image Credits: AI Generated

DOI: 10.3897/VCS.198373

Keywords: Translation, Braun-Blanquet, scale, percent, TURBOVEG, bias, diversity, metrics, scientific research