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

Background : The majority of vegetation plots has been and still is being recorded on ordinal scales, meaning that the values have to be back-transformed to percent for many typical analyses. Little is known about how different back-transformation approaches affect this procedure. While generally the midpoints of the class borders are recommended for back transformation, one of the most widely used back-transformation for the 7-step Braun-Blanquet scale (as the most widely used ordinal scale) is the default back transformation in the software TURBOVEG. Thus, we asked whether this approach possibly biases results. Case study : We selected three datasets of different vegetation types recorded with percent and calculated how the use of the 7-step Braun-Blanquet scale with three possible back-transformations (arithmetic and geometric mid-point, TURBOVEG) would have influenced Shannon diversity, Shannon evenness and Simpson diversity. We found small and inconsistent changes for the first two but always substantial increases in case of the TURBOVEG back-transformation. Conclusions : The default TURBOVEG translation of the 7-step Braun-Blanquet scale can strongly bias analyses of vegetation plots stored in regional and collaborative databases in the case of cover-based biodiversity metrics, but likely also for any other analysis that involves species cover. It should thus be avoided.