Guest blog post by Jürgen Dengler, Iwona Dembicz & Łukasz Kozub
This post refers to the article Translation of the Braun-Blanquet scale to percent in TURBOVEG can bias diversity metrics in Vegetation Classification and Survey (https://doi.org/10.3897/VCS.198373)
The most widespread approach to vegetation sampling is the use of vegetation plots, i.e., defined areas in which all species present are recorded together with their cover. Cover refers to the percentage of the plot area occupied by the superficial parts of a given species. Still, many vegetation ecologists prefer recording cover on ordinal scales, such as variants of the Braun-Blanquet scale, rather than recording it directly as a percentage. Estimating cover always comes with errors, but using an ordinal scale increases this error as it involves a double transformation, first in the field from the fractional cover to a category of the ordinal scale, and then back from the ordinal scale to a metric scale for all kinds of analyses (Dengler and Dembicz 2023; Dembicz and Dengler 2025). Typically, the arithmetic or geometric midpoint of the class boundaries of the respective ordinal scale is recommended when back-translating the data. However, the most widely used software for storing and handling vegetation-plot databases, TURBOVEG (Hennekens and Schaminée 2001), has an inbuilt back-translation deviating from this principle (see Figure 1). We thus asked whether and how the mid-point translation and the TURBOVEG translation might bias a set of widely used biodiversity metrics, namely Shannon diversity, Shannon evenness and Simpson diversity.
In our case study using three different datasets covering a wide array of vegetation types, we found that the two mid-point translations had partly significant, partly non-significant, but in all cases small effects on the biodiversity metrics. By contrast, the inbuilt translation of TURBOVEG had severe distorting effects on all three metrics (see Figure 2). The average increases were about 0.7 units for Shannon diversity and 0.2 units for Shannon evenness, which corresponds to an approximately 30% difference in both cases. For Simpson diversity, the increase on average was smaller, with about 0.1 unit, corresponding to roughly 10%, which is due to the fact that the true Simpson diversity values in our datasets were above 0.8 and thus already close to the theoretical maximum of 1.0.
These differences due to the TURBOVEG settings were way higher than differences in biodiversity metrics often reported as statistically significant and ecologically meaningful. This means that using the TURBOVEG default translations can lead to wrong conclusions in studies using datasets that partly have been recorded on the Braun-Blanquet scale and partly directly in percent. For example, in a temporal comparison where a larger fraction of the older plots has been recorded on the Braun-Blanquet scale than in the newer plots, this methodological artifact could erroneously suggest a biodiversity decline. Likewise, spatial biodiversity patterns can be biased when the fraction of plots recorded with variants of the Braun-Blanquet scale varies between different geographic entities (e.g., countries). It is self-evident that the TURBOVEG default translations will also bias any other response variable that relies on species cover, namely other cover-based biodiversity metrics or community-weighted means of functional traits, albeit we did not quantify the effect sizes in these cases.
We thus recommend that researchers using TURBOVEG should not use the default back-translations of this program but set their own back-translations. This is particularly important when receiving data from the two largest vegetation-plot databases in the world, EVA in Europe (Chytrý et al. 2016) and sPlot globally (Bruelheide et al. 2019), as these databases are run under TURBOVEG 3. However, with an adequate export of the requested data, users can overwrite the TURBOVEG default percent values with more proper values that cause less distortion. It is to be hoped that in a future release of TURBOVEG these mistakes rooted in the early days of the program will be corrected.
