Co-ordinating Editor: Dr. Angelika Schwabe-Kratochwil.
Problem: The increasing availability of large vegetation databases holds great potential in ecological research and biodiversity informatics, However, inconsistent application of plant names compromises the usefulness of these databases. This problem has been acknowledged in recent years, and solutions have been proposed, such as the concept of “potential taxa” or “taxon views”. Unfortunately, awareness of the problem remains low among vegetation scientists.
Methods: We demonstrate how misleading interpretations caused by inconsistent use of plant names might occur through the course of vegetation analysis, from relevés upward through databases, and then to the final analyses. We discuss how these problems might be minimized.
Results: We highlight the importance of taxonomic reference lists for standardizing plant names and outline standards they should fulfill to be useful for vegetation databases. Additionally, we present the R package vegdata, which is designed to solve name-related problems that arise when analysing vegetation databases.
Conclusions: We conclude that by giving more consideration to the appropriate application of plant names, vegetation scientists might enhance the reliability of analyses obtained from large vegetation databases.
Vegetation databases that contain species co-occurrence data have been compiled all over the globe. It is estimated that in Europe alone nearly two million phytosociological relevés are stored electronically (Schaminée et al. 2009), and similar databases are emerging outside Europe (Ewald 2001). This legacy of plot data spanning 100 years is a powerful tool for vegetation science (Ewald 2003). These databases open new avenues for analysis, from classical synthetic classification to predictive mapping and tests of fundamental ecological hypotheses regarding functional traits, assembly rules and biodiversity patterns (Dengler et al. 2008). They potentially offer many options for analysing patterns and processes of global change caused by anthropogenic climate warming, land-use changes and biotic invasions (Schaminée et al. 2009).
By contrast, little acknowledgement has been given to problems associated with the use of plant names in vegetation databases. Actually, consistent application of “research object” names is an indispensable step in any analysis of biotic data. This is particularly true when combining plot data from different sources. In fact, the procedure of normalizing plant name usage can become one of the most time-consuming processes among all the methods. The problems we aim to address in this contribution arise from the fact that plant taxonomy is a dynamic discipline (Stuessy 2009). A consequence of this constant taxonomic flux and individual idiosyncrasy of taxonomic understanding is that the thousands of floras and checklists in use worldwide are seldom congruent in their taxonomy.
Focusing on vegetation databases, the aim of this paper is to: show how inconsistent application of plant names can lead to biased analyses; propose ways of minimizing such shortcomings; and demonstrate how our ideas can be put into practice.
When Inconsistent Use of Plant Names Cause Problems
Problems caused by inconsistent use of plant names arise in floristic or vegetation databases as a result of the accumulation of data from hundreds or thousands of researchers (with their regional and individual idiosyncrasies). For example, Schaminée et al. (2009) state that in order to establish a joint European vegetation database (SynBioSys Europe), 30 national species lists with 300 000 names (or rather: taxon views) have to be united. The effects on analyses are likely to be profound, but the methods and results in published vegetation studies will not allow the reader to detect any artifice, or make it possible to trace the problem back to its source in the vegetation database. Below, we highlight these difficulties using three arbitrarily chosen analyses of floristic and vegetation databases, in which effects of inconsistent use of plant names are apparent.
Using names that are interpreted differently in their extent by various authors is probably the most frequent source of problems in floristically well-studied regions. For instance, Benkert et al. (1996), referring to the taxonomic concept of Bäßler et al. (1996), presented distribution maps of the Festuca ovina aggregate and its micro-species in E Germany. According to these maps, F. ovina and F. brevipila would be equally frequent in lowland habitats. The latter species, however, is much more characteristic of this region than the former (e.g. Dengler 1994). The mapping error likely did not result from misidentifications but rather from different delimitations of “Festuca ovina” among the surveyors who contributed data to the atlas of Benkert et al. (1996). For example, Ascherson (1864) accepts only one species, F. ovina, with several infraspecific taxa (including F. ovina subsp. duriuscula sensu Ascherson =F. brevipila of modern floras), this taxon view corresponds to F. ovina agg. in most modern classifications. When compiling data from hundreds of field researchers, one has to account for the fact that at least these three different wide concepts of the name F. ovina are still regularly used in Germany (see Fig. 1). Thus, the assignment of all records with field name “F. ovina” to the F. ovina map in the atlas might not have been correct based on the taxon view adopted there.
Mahecha & Schmidtlein (2008) analysed species distribution patterns, based on floristic grid mapping data from Germany (FlorKart), with multidimensional ordination methods. The authors state that a significant proportion of the patterns they found can be attributed to biases caused by different sampling intensities, as well as to regional differences in taxonomic concepts. They highlight the distribution of “Tripleurospermum perforatum” versus “T. maritimum” in the FlorKart database, which is highly correlated to biogeographic patterns. The first species seemingly occurs in nearly all grid cells of W Germany, Berlin and Saxony, but is absent from the rest of the former German Democratic Republic (GDR). The second seems to be highly frequent in Thuringia, scattered in the rest of the former GDR, and nearly absent from W Germany. Actually, T. maritimum (sensuWisskirchen & Haeupler 1998) is a coastal species that does not occur in Thuringia at all, while T. perforatum (sensuWisskirchen & Haeupler 1998) is very common in all German federal states (Jäger & Werner 2005). Thus, we witness two problems. Similar to Festuca ovina, the name T. maritimum is frequently applied in a sense deviating from FlorKart, i.e. including T. perforatum (sensuWisskirchen & Haeupler 1998) as T. maritimum subsp. inodorum. Further, the near complete “absence” of both taxa from NE Germany is caused merely by the fact that they have been mapped at the aggregate level in that area (Benkert et al. 1996).
In our third example, Wamelink et al. (2005) modelled species response curves against pH for 556 taxa. In several cases, both infraspecific taxa and the corresponding species have been modelled separately, with deviating results for the mean response (e.g. Erodium cicutarium subsp. cicutarium: pH=7.4; Erodium cicutarium: pH=7.0). Modelling species in addition to subspecies can be reasonable when more than one subspecies is involved, but would require including all the data of the subordinate taxa at the higher level. Further, we can assume that in their study area each of the species occurred with only one subspecies (i.e. they were “regionally monotypic”). Thus, while being two valid taxa of different rank, the biological “content” of the species and the subspecies was identical. The authors seemingly modelled the “names” in the database, neglecting their hierarchical relationships.
To exemplify the extent of the named problems, we provide a (non-exhaustive) list of critical names from the German flora in Appendix S1.
Steps of Taxonomic Interpretation
Three steps of taxonomic interpretation are essential when moving from the real-world object to the results of scientific research (Fig. 2). In the following, we discuss the difficulties and necessary methods in each step.
Field survey and species determination
The first step of taxonomic interpretation occurs in the field, when recording relevé data (Fig. 2). Many studies have demonstrated that even experienced surveyors can typically overlook or misidentify a fraction of the plants present (e.g. Tüxen 1972; Scott & Hallam 2003; Archaux et al. 2006; Vittoz & Guisan 2007). The problems arising at this point are not within the focus of this article, but shortfalls at this first step can hardly be re-addressed at a later stage of database entry or preparation for analysis. In this context, it appears that students in vegetation ecology are increasingly less trained in handling taxonomy-related issues and making proper use of floras or checklists. The taxonomic education at universities should thus be strengthened to match the demands for high quality primary data.
Surveyors should document their references for determination and naming for every single taxon, i.e. it is important to disclose the author of the taxonomic concept (who is not identical to the author of the taxon name) when the determination or naming deviates from the identification guide that is otherwise followed in the study. In many cases (particularly in botanically less well studied regions of the Earth), it is essential to collect voucher specimens of critical plant taxa and to quote the herbarium where these have been deposited. Additionally, we recommend further explanation of the uncertainty of determination by specifying at which taxonomic level the uncertainty is located (cf. Festuca ovina, F. cf. ovina agg. or F. cf. ovina) or which is the next higher taxon known with certainty. Further, it would be useful to indicate the degree of uncertainty on a rough ordinal scale (“certain”, “with some uncertainty”, “uncertain, but best guess”).
The second step is the data entry into an electronic list or database (Fig. 2). Widely used vegetation database programs such as TURBOVEG (Hennekens & Schaminée 2001) do not provide a field for the entry of information on the preciseness of plant identification as default. Accordingly, information is often lost, be it by replacing “Festuca cf. ovina” with F. ovina, by assigning it to some superior taxon, or by omitting it altogether. Instead, we suggest retaining information on lack of certainty and leaving decisions on how to deal with such entries to the step involving preparation for analysis (see Appendix S2).
More troublesome are problems caused by different meanings of the same name. The problem cannot be avoided by publishing the taxon names together with their authority. Festuca ovina, for example, bears the same correct author citation “L.” (for C. Linnaeus) irrespective of whether a narrow or a wide taxon view is applied (see Fig. 1). Accordingly, the taxonomic concept must be disclosed, whether by citing the flora used or by describing the taxonomic concept. However, there are only a few systems, e.g. Vegbank (http://www.vegbank.org/vegdocs/design/planttaxaoverview.html) and VegetWeb (Ewald et al. 2006, http://www.planto.de/OekoArt/ModellLog.php), that allow the referencing of each plant name to a source, but even these databases actually currently work with only one preferred taxonomic view. BIOTABase (http://www.biota-africa.de/biotabase_ba.php) is another example of a database system designed to map the relationships between field names and accepted names in a very flexible way, together with links to voucher specimens. This software has been designed to account for the absence of checklists useful for standardizing the use of names of African plants, but its approach is useful for other regions as well.
While we strongly support the application of the “taxon view” approach in vegetation databases, i.e. storing both the original plant name and a link to a reference that defines his content, we are sceptical as to whether this alone would solve the “nomenclatural confusion” for such databases. In many published sources of vegetation relevés, no explicit references to floras or checklists are made, or the general declaration is not reliable for all species. Further, names are often used that do not exist in the source (“Taraxacum officinale” is probably one of the most frequently occurring names in vegetation relevés worldwide, despite the fact that many recent floras reject such a taxon), or names are used in an aberrant mode from the source (e.g. Festuca ovina in the sense of F. ovina agg.).
Thus, we suggest that a taxonomic interpretation of names, together with documentation of the decision, should take place instead of an automatic assignment of names. While often the assignment to a higher taxonomic level is required, in other cases, a-posteriori application of a subordinate name might be sensible. For example, Silene latifolia has four accepted subspecies in Europe (Tutin et al. 1993), of which only subsp. alba occurs in Germany (Wisskirchen & Haeupler 1998). Thus, researchers in Germany often do not note the subspecies because they consider this information superfluous. Thus, in the data entry step, it is both reasonable and useful to re-assign any German field record of S. latifolia to S. latifolia subsp. alba. When combining databases at a supranational level, the infraspecific classification becomes relevant, and the German data are already correctly assigned to subspecies. These examples show that it is important to have the digitized data “verified” by an experienced botanist before they are added to a larger database.
Expert knowledge is also required when combining datasets based on different taxonomic reference lists, as is typical for international collaborations. The correct assignment of names and concepts can be a very time-consuming process, but is essential for credible results. In the software package vegdata, the function tv.compRefl is included, which compares taxon numbers and/or taxon names of different TURBOVEG reference lists as a starting point (see Appendix S3).
Preparation for analyses
For the third step, the stored data have to be prepared for the needs of the scientific analyses (Fig. 2). If the interpretations suggested for steps 1 and 2 remain unresolved, they have to be dealt with now – which obviously will be more complicated and error-prone because the necessary information about taxon views is usually not available. In any case, the outcome should be the correct fusion of synonyms and homonyms into accepted names of the chosen taxonomic reference. If alternative taxon views are available in the vegetation database, this would be the point to select one. To ensure sound and consistent results, four further decisions are necessary prior to the analyses, depending on the type of study.
1A choice has to be made as to the use of uncertain determinations (often indicated with the abbreviation “cf.” in field names), with two options: acceptance as a confident determination or aggregation at next highest taxonomic level (then see 4).
2Nested taxa may make sense in some studies and not in others. For example, one could model separate response curves along a gradient for a species and its subspecies, with the ecological amplitude of the latter probably being narrower. By contrast, counting a species in addition to its subordinate taxa is unacceptable when species richness is analysed. In both cases, however, all data of subordinate taxa have to be included in the analysis of the higher-rank taxon.
3Also, for the taxonomic level of the analyses, there are two options: one may choose (i) the “terminal” taxa (i.e. the lowest-ranked accepted taxa with available data, which in one case may be a species and in another a subspecies or variety), or (ii) a uniform taxonomic rank (e.g. species).
4Further, one has to decide what to do with taxon records available only at coarser taxonomic levels than desired (e.g. genus level): either combining the more precise data to the higher level (see 2) or excluding the coarse-scale data from the analyses.
For none of these four decisions is there a general right or wrong method. Instead, we face a trade-off between different solutions, each of which necessarily introduces additional uncertainty to the data. If the introduced uncertainty is too big, the exclusion of the whole relevé from the analyses might be the better choice.
In order to apply the principles presented in our paper to real datasets and to automate the process of taxonomic interpretation, the first author has written scripts for the statistical environment R (R Development Core Team, Vienna, Austria). These are capable of solving problems occurring at the last step, but also assists with checking results of the second (and first) step. The scripts are included in the package vegdata (version 0.2, http://cran.r-project.org/package=vegdata), whose present version supports data access from TURBOVEG. In Appendix S2 we present an example session with a commentary.
Taxonomic Reference Lists – A Necessary Prerequisite
As outlined above, reference to taxonomic lists is needed during all three interpretative steps (Fig. 2). These lists could be (i) floras, which define the meaning of plant names by use of keys and descriptions, (ii) checklists (providing reference to species belonging to the current flora of a region) or (iii) specialized reference lists for survey data. They may simply present a preferred taxon view, potentially accompanied by synonyms (e.g. Ehrendorfer 1973), but in a more useful variant, they additionally outline the links between the preferred and other taxon views (see Wisskirchen & Haeupler 1998; Koperski et al. 2000). Such lists are the “backbones” of all databases containing information related to plant taxa (e.g. museum collections, distribution, taxonomy, vegetation, traits) and for the connection of different types of such databases. However, floras and checklists are typically developed according to other goals, and so they seldom fully meet the needs of survey databases.
In order to be useful for taxonomic interpretation in the framework of vegetation databases, a taxonomic reference list for survey data should fulfill the following requirements:
1Use of taxon views to link the present taxonomic interpretation to the views of major floras, other reference works (e.g. Red Lists) and previous editions of the reference list.
2Unambiguous recording of different taxon views. All widely applied names used in the past should find a direct counterpart in a new reference list, instead of referring to several names pro parte or only partially referring to one higher-rank taxon. For example, if Tripleurospermum maritimum in the sense of earlier floras is split into T. maritimum and T. perforatum in the reference list, then T. maritimum of the older sources must be included in the new reference list (e.g. as T. maritimum agg.) and all records must be mapped to it.
3Inclusion of hybrids, non-naturalized neophytes and frequently cultivated plants. Most checklists purposefully exclude these three groups from listing, since in statistical evaluations of floras, such taxa normally should not be counted. Nevertheless, taxonomic reference lists for survey data should provide reliable nomenclatural data for all taxa occurring in the vegetation.
4Hierarchical levels. All taxa should be assigned to hierarchical levels unambiguously to make automatic data checks and taxonomic refinement possible and to allow analyses at different taxonomic resolutions.
5Information on regionally monotypic taxa. When a species is present in a region with only one infraspecific taxon, it is necessary to list which infraspecific taxon is present, even if it is the nominal taxon (e.g. Silene latifolia subsp. alba).
While points 1-5 apply to printed lists, additional considerations become relevant when implementing reference lists as an electronic tool within the framework of vegetation databases (e.g. GermanSL, see Jansen & Dengler 2008):
6Joint list for all “plant” taxa. The use of vegetation databases requires that reference lists of different “plant” taxa (vascular plants, bryophytes, lichens, macroscopic algae, macroscopic cyanobacteria), which are usually covered in separate printed lists, are combined in one electronic list.
7Continuous update and versioning. Unlike print versions, the electronic format allows frequent updates to accommodate newly recorded plants and new taxonomic views. This is necessary and highly useful, but sequential versioning is necessary to unambiguously link to previously used taxonomies and nomenclature.
8Comprehensive documentation. The electronic reference list should either completely match the printed version, or all deviations must be disclosed unambiguously.
9Unique and unambiguous names with clear reference to the represented taxon view. To be able to convey the information from the original source with an unambiguous name, we urge the use of abbreviations such as “s.str.” or “s.l.” for differently wide species concepts and “auct. non” for misapplications in those, but only in those cases. Of course, the source of the applied taxon view (“secundum”) has to be referenced in additional fields of the list.
10Warning about critical taxa: It would be useful if database programs provide warnings during data entry or merging of different databases when “critical” taxon names are involved that can be misapplied (i.e. when there are variants of the same taxonym with and without “s.l.”, “s.str.”, “agg.” or “auct. non” or if monotypic species occur), and require a specific decision/confirmation of the researcher for such assignments.
Within the rapidly evolving field of biodiversity informatics (see Canhos et al. 2004; Jones et al. 2006), vegetation databases can contribute as a comprehensive and informative data source, as they contain not only many millions of records of individual species but also combine these with information on species composition, vegetation structure and environmental conditions in a spatially and temporally explicit manner. Tapping the full potential of vegetation databases currently remains seriously hindered by unappreciated nomenclatural problems that arise when combining data from different sources. In solving these problems, the “taxon view” approach developed for organismic taxonomy (e.g. Berendsohn 1995; Zhong et al. 1996; Koperski et al. 2000) is a welcome contribution. However, vegetation scientists should be aware that the situation for their data is even more complicated because field records normally are not connected to voucher specimens that would allow a later correction of the identification. While reference lists are a central tool for the correct use of plant names in vegetation databases, their utility is often limited because their layout does not cover the requirements of the users. In conclusion, vegetation scientists have to accept that their data will always contain some uncertainty, but with a taxonomic reference list layout, as proposed in this contribution, and appropriate tools like the R package vegdata shown in Appendices S2 and S3, they should be able to avoid flawed results that would ensue from inconsistent use of plant names. Finally, also editors and reviewers of journals should put more emphasis on taxonomic accuracy in ecological and biodiversity articles and allow for – even long if need be – taxon lists in research articles.
Acknowledgements. We thank Erwin Bergmeier, Curtis Björk, Zoltán Botta-Dukát, Manfred Finckh, Michael Manthey and Gerhard Muche for useful comments on former versions of the manuscript. Curtis Björk also polished our English usage.