The complex syntaxonomy of Balkan dry grasslands addressed with an innovative classification approach

The post provided by Kiril Vassilev & Jürgen Dengler

Study area in the Central and Eastern Balkan Peninsula with the distribution of the analysed plots of three widespread and one rare alliance. From the article.

This post refers to the article Classification of the high‐rank syntaxa of the Central and Eastern Balkan dry grasslands with a new hierarchical expert system approach by Vassilev et al. published in Applied Vegetation Science (

The Balkan Peninsula is one of the European hotspots of biodiversity and at the same time home to extensive stands of semi-natural grasslands, many of which still qualify as High Nature Value grasslands. At the same time, numerous small countries with different languages led to idiosyncratic classifications of plants and plant communities at regional to national scales, impeding broad-scale overviews. This motivated us more than 10 years back to initiate a collaborative project to overcome this situation for dry grasslands as one of the main vegetation types occurring in the region. The first step was to set up a comprehensive regional vegetation-plot database of dry grasslands, the Balkan Dry Grassland Database (BDGD; Vassilev et al. 2012). An inspiring moment was also the third international EDGG Field Workshop in NW Bulgaria in 2011, which resulted in a first attempt to resolve the syntaxonomic problems with a European perspective (Pedashenko et al. 2013) and an in-depth analysis of the diversity patterns (Dembicz et al. 2021).

However, the task selected by us turned out to be far more demanding and time-consuming than we anticipated – otherwise we might have not started it at all. The challenges arose from solving numerous nomenclatural issues in both plant taxonomy and syntaxonomy, but also from our intention to provide something really meaningful by advancing several methodological issues. Moreover, we had no funding for the project, except some small seed money that allowed several research stays of Kiril Vassilev in the group of Jürgen Dengler. Further delay came when Jürgen moved to a new position in Switzerland where international research is not in the focus and he first had to focus on getting established. It was a long dry spell and we asked ourselves more than one time whether the paper would ever be finished. But in summer 2023, with a last big effort, we finalised and submitted the manuscript with its extensive three printed and 30 electronic appendices. We are glad and relieved that we now can celebrate the publication.

Simplified flowchart of the ICO-HES approach (iterative cluster optimisation for hierarchical expert systems). From the article.

In a way, this paper is two publications in one – as a major aspect was the development and implementation of a complex new classification approach. It builds on two widely used elements of modern classification approaches: (1) electronic expert systems (ES’s) for supervised classification using species composition and (2) the determination of diagnostic species using statistical measures, such as phi-values. We were dissatisfied by the situation that in most studies that employed both elements thus far these are disconnected, i.e. the species of the ES do not match the diagnostic species. We, therefore, developed a solution that unifies the two approaches: We derived the species of the ES from the diagnostic species and then again determined the diagnostic species from the units derived from the application of the EC. This was done iteratively until both species lists converged. This had previously been done by García-Mijangos et al. (2021) in a similar manner, albeit without iteration. Moreover, unlike the large majority of published ES’s to date, ours is hierarchical, i.e. reflecting the hierarchies of phytosociological classification systems (Dengler et al. 2008). Finally, unlike in the original proposal by Chytrý et al. (2002), we compared the phi values not between the target vegetation unit and all other vegetation units combined, but with the next similar unit of the same rank as previously suggested by Tsiripidis et al. (2009). We consider this comparison as the far more relevant one in practice, e.g. for the recognition of units.

With this combination of established and new methodological elements, called iterative cluster optimisation for hierarchical expert systems (ICO-HES), we were able to classify the heterogeneous dataset successfully at the class, order and alliance level. ICO-HES left only about 5% of all plots unclassified, which is a very low rate compared to other recent ES’s. According to our results, the dry grasslands of the study region belong to four classes, eight orders and 12 alliances: (1) Tuberarietea guttatae (Romuleion); (2) Stipo‐Brachypodietea distachyi (Clinopodio alpini‐Thymion striati); (3) Festuco‐Brometea with Brachypodietalia pinnati (Chrysopogono‐Danthonion calycinae and Cirsio‐Brachypodion pinnati), Festucetalia valesiacae (Festucion valesiacae), an unnamed order of rocky steppes (with Pimpinello‐Thymion zygioidis) and Koelerietalia splendentis (Centaureo‐Bromion fibrosi, Saturejion montanae and Diantho haematocalycis‐Festucion hirtovaginatae); (4) Koelerio‐Corynephoretea with Sedo acris‐Festucetalia (Festucion vaginatae) and Trifolio arvensis‐Festucetalia ovinae (Armerio rumelicae‐Potentillion and Minuartio montanae‐Poion molinerii).

Whileour broad-scale analysis with ICO-HES confirmed previous viewpoints (Mucina et al. 2016; Willner et al. 2019) in the case of the orders Brachypodietalia pinnati nom. cons. propos. (see Dengler and Willner 2023) and Festucetalia valesiacae, it led to suggested changes in the other major syntaxa: Among the Mediterranean and sub-Mediterranean grasslands where Mucina et al. (2016) proposed several classes with more than half a dozen alliances for the study region, we could distinguish only one alliance from the acidophytic class Tuberarietea guttatae and one from the basiphytic class Stipo-Brachypodietea distachyae. Within the class Festuco-Brometea, we suggest combining all rocky dry grasslands from the Balkan mountains, whether on limestone or serpentine, into the order Koelerietalia splendentis nom. cons. propos. (including the order Halacsyetalia sendtneri)  as both groups share many joint species. The Koelerietalia splendentis are a southern vicariant of the Stipo pulcherrimae-Festucetalia pallentis. By contrast, the rocky grasslands in the steppe zone near the Black Sea coast with plenty of submediterranean elements (alliance Pimpinello‐Thymion zygioidis) turned out to differ from the orders Festucetalia valesiacae, Stipo pulcherrimae-Festucetalia pallentis and Koelerietalia splendentis so much that we suggest that they are separate at ordinal level, while refraining from a formal description for the time being. Finally, the class Koelerio-Corynephoretea is represented by two quite different orders in the region: The Sedo acris-Festucetalia (subcontinental and continental sand steppes) contain the sole alliance Festucion vaginatae, occurring both on inland sands along the Danube and in grey dunes of the Black Sea coast. To the Trifolio arvensis-Festucetalia ovinae (meso-xeric grasslands on sandy and siliceous soils), we assigned the Armerio rumelicae‐Potentillion (originally placed in the order Astragalo-Potentilletalia, Festuco-Brometea) as previously done by Pedashenko et al. (2013) and the Minuartio montanae‐Poion molinerii all. nov. (= Thymion jankae nom. inval., originally placed in the order Halacsyetalia sendtneri, Festuco-Brometea) as previously indicated by Kuzmanović et al. (2016), based on floristic and ecological reasons. All proposed units are supported by the hierarchical classification system and fully reproducible with the accompanying electronic ES. Our classification system considerably reduces the often criticized inflation of high rank syntaxa by synonymising several previously accepted orders and alliances that were not sufficiently separated floristically.

Stand of the rocky grassland alliance Saturejion montanae (class Festuco-Brometea) in Chepun mountain, Western Bulgaria. Photo credit: Kiril Vassilev

A particular strength of the paper lies in its figures, tables, three print appendices and 30 online supplements that together provide a very comprehensive documentation that should facilitate the implementation and transfer of the results:

  • The hierarchical expert system allows the reproducible classification of any dry grassland plot from the study region with the program JUICE.
  • We have applied this expert system to any type relevé of dry grassland associations from the region to allow the unanimous determination of the correct names for all alliances and orders.
  • To facilitate the uptake of the proposed syntaxonomic and nomenclatural novelties, we have already prepared the respective applications to the Committee for Change and Conservation of Names (CCCN; Willner et al. 2021) and the European Vegetation Classification Committee (EVCC; Biurrun and Willner 2020) and will submit them in the near future.
  • All alliances are characterised by distribution maps, boxplots of their species richness, structure and environmental conditions (via mean EIVE values; Dengler et al. 2023) and a standardised description with diagnostic, frequent and dominant species.

To conclude, on the one hand our paper provides a modern data-based classification of the higher-rank syntaxa of four vegetation classes in a highly diverse part of Europe that makes a major contribution to a consistent classification system of Europe, on the other hand, we advanced current classification approaches to a unified methodology that could inspire future classifications of complex syntaxa.

Cryptogam-rich stand of the alliance Armerio rumelicae-Potentillion from Bulgaria, which according to the study should be included in the class Koelerio-Corynephoretea. Photo credit: Jürgen Dengler