Should we estimate plant cover in percent or on ordinal scales?

Prepared by Jürgen Dengler & Iwona Dembicz

How much does species x cover in the plot? Two vegetation scientists estimating plant covers in a steppe of Southern Ukraine during the 15th EDGG Field Workshop. (Photo Credit: Jürgen Dengler).

Vegetation-plot records (relevés) containing a full list of species with cover information on each of them are the main source of information in vegetation science. In the Braun-Blanquet school, and also other schools of vegetation science, cover estimation on ordinal scales has a long tradition, and in various countries around the world is still the preferred method over direct estimates of % cover. In our simulation study, we analysed whether sticking to the tradition of ordinal scales is a wise decision or not.

Evidently, every cover estimation of a plant species comes with an estimation error, which, depending on the vegetation’s complexity and the researcher’s experience, might be smaller or bigger. If using direct percent estimates, the initial estimation error is identical to the error that enters the final calculations. If using an ordinal scale, there are two translation steps in between, first from the visual estimate to a cover class and then back from the cover class to a numerical value (typically the class midpoint), which is required because one cannot apply the usual numerical operations to categories, such as “r”, “+” or “2b”. Both additional steps will inevitably increase the final error. We found that under a realistic scenario, the wide-spread 7-step Braun-Blanquet scale led to an “error inflation” by a factor of 2, i.e. given the same accuracy in the field, the values entering the analyses were two times less precise. Evidently, the finer an ordinal scale is, the less information loss it will introduce. However, it is not only about the number of steps of an ordinal scale, but it also depends on the range of each category. Unfortunately, the probably most widespread ordinal scale, the 7-step Braun-Blanquet scale, performs particularly poorly – because its cover class “2” comprises a five-fold cover range (5–25%). In this respect, even the 5-step Hult-Sernander-Du Rietz scale is better.

Visualisation of the different steps involved in observing a real plant in a vegetation plot and using its estimated cover in a wide range of numerical analyses. In both cases, the same estimation error of the observer is involved (in the case depicted, the botanist estimates 10% while the plants of this species only have 5% cover). In the case of direct percent estimation (blue path), this estimation error is the only source of error prior to statistical analysis, but when using an ordinal scale (red path), there are two additional transformation steps involved, each of which adds to the overall error. (Credit: Iwona Dembicz, from the paper).

However, the question remains: why use an ordinal scale at all when it necessarily comes with an information loss? From our experience, having first done thousands of plots with the Braun-Blanquet scale and since a couple of years ago another thousands with direct percent estimates, direct % estimates do not need more time. Instead, on the contrary: if you are confident that a plant covers approximately half of your plot (i.e. something between 45 and 55%), with direct % estimation, you could easily note “50%”, while using the Braun-Blanquet scale you would need to think carefully whether it is rather 49% (which would be interpreted as 37.5% later on) or 51% (which would be interpreted as 62.5% later on). After this initial study on the general effects of % vs. ordinal scales, we are already planning two follow-up simulations on the consequences of these two cover estimation methods (a) on diversity indices (such as Shannon diversity and Shannon evenness) and (b) on results of resurvey studies. We would welcome ideas to complement these simulation studies with experiments quantifying the estimation errors vs. time needed for different researchers working with different methods.

This is a plain language summary for the paper of Dengler & Dembicz published in Vegetation Classification and Survey (