Formalized reproduction of an expert‐based phytosociological classification: A case study of subalpine tall‐forb vegetation

Abstract. Delimitation of vegetation units in phytosociology is traditionally based on expert knowledge. Applications of expert‐based classifications are often inconsistent because criteria for assigning relevés to vegetation units are seldom given explicitly. Still, there is, e.g. in nature conservation, an increasing need for a consistent application of vegetation classification using computer expert systems for unit identification. We propose a procedure for formalized reproduction of an expert‐based vegetation classification, which is applicable to large phytosociological data sets. This procedure combines Bruelheide's Cocktail method with a similarity‐based assignment of relevés to constancy columns of a vegetation table. As a test of this method we attempt to reproduce the expert‐based phytosociological classification of subalpine tall‐forb vegetation of the Czech Republic which has been made by combination of expert judgement and stepwise numerical classification of 718 relevés by TWINSPAN. Applying the Cocktail method to a geographically stratified data set of 21794 relevés of all Czech vegetation types, we defined groups of species with the statistical tendency of joint occurrences in vegetation. Combinations of 12 of these species groups by logical operators AND, OR and AND NOT yielded formal definitions of 14 of 16 associations which had been accepted in the expert‐based classification. Application of these formal definitions to the original data set of 718 relevés resulted in an assignment of 376 relevés to the associations. This assignment agreed well with the original expert‐based classification. Relevés that remained un‐assigned because they had not met the requirements of any of the formal definitions, were subsequently assigned to the associations by calculating similarity to relevé groups that had already been assigned to the associations. A new index, based on frequency and fidelity, was proposed for calculating similarity. The agreement with the expert‐based classification achieved by the formal definitions was still improved after applying the similarity‐based assignment. Results indicate that the expert‐based classification can be successfully formalized and converted into a computer expert system.     

Assignment of relevés to pre‐defined classes by supervised clustering of plant communities using a new composite index

Question: How does a newly designed method of supervised clustering perform in the assignment of relevé (species composition) data to a previously established classification. How do the results compare to the assignment by experts and to the assignment using a completely different numerical method? Material: Relevés analysed represent 4186 Czech grassland plots and 4990 plots from a wide variety of vegetation types (359 different associations or basal communities) in The Netherlands. For both data sets we had at our disposal an expert classification, and for the Czech data we also had available a numerical classification as well as a classification based on a neural network method (multi‐layer perceptron). Methods: Two distance indices, one qualitative and one quantitative, are combined into a single index by weighted multiplication. The composite index is a distance index for the dissimilarity between relevés and vegetation types. For both data sets the classifications by the new method were compared with the existing classifications. Results: For the Czech grasslands we correctly classified 81% of the plots to the classes of an expert classification at the alliance level and 71% to the classes of the numerical classification. Correct classification rates for the Dutch relevés were 64, 78 and 83 % for the lowest (subassociation or association), association, and alliance level, respectively. Conclusion: Our method performs well in assigning community composition records to previously established classes. Its performance is comparable to the performance of other methods of supervised clustering. Compared with a multi‐layer perceptron (a type of artificial neural network), fewer parameters have to be estimated. Our method does not need the original relevé data for the types, but uses synoptic tables. Another practical advantage is the provision of directly interpretable information on the contributions of separate species to the result.     

The management of vegetation classifications with fuzzy clustering

Questions: Does fuzzy clustering provide an appropriate numerical framework to manage vegetation classifications? What is the best fuzzy clustering method to achieve this? Material: We used 531 relevés from Catalonia (Spain), belonging to two syntaxonomic alliances of mesophytic and xerophytic montane pastures, and originally classified by experts into nine and 13 associations, respectively. Methods: We compared the performance of fuzzy C‐means (FCM), noise clustering (NC) and possibilistic C‐means (PCM) on four different management tasks: (1) assigning new relevé data to existing types; (2) updating types incorporating new data; (3) defining new types with unclassified relevés; and (4) reviewing traditional vegetation classifications. Results: As fuzzy classifiers, FCM fails to indicate when a given relevé does not belong to any of the existing types; NC might leave too many relevés unclassified; and PCM membership values cannot be compared. As unsupervised clustering methods, FCM is more sensitive than NC to transitional relevés and therefore produces fuzzier classifications. PCM looks for dense regions in the space of species composition, but these are scarce when vegetation data contain many transitional relevés. Conclusions: All three models have advantages and disadvantages, although the NC model may be a good compromise between the restricted FCM model and the robust but impractical PCM model. In our opinion, fuzzy clustering might provide a suitable framework to manage vegetation classifications using a consistent operational definition of vegetation type. Regardless of the framework chosen, national/regional vegetation classification panels should promote methodological standards for classification practices with numerical tools.     

Synanthropic vegetation of the <Emphasis Type="Italic">Eragrostion cilianensi-minoris</Emphasis> alliance in the Czech Republic

A synthesis of the alliance Eragrostion cilianensi-minoris in the Czech Republic is presented on the basis of 82 relevés including new unpublished data. A TWINSPAN classification and detrended correspondence analysis were used to identify the main vegetation types included in the alliance Eragrostion cilianensi-minoris. A syntaxonomic revision of the data set revealed five associations of the alliance: Digitario sanguinalis-Eragrostietum minoris, Portulacetum oleraceae, Eragrostio poaeoidis-Panicetum capillaris, Cynodontetum dactyli, and Hibisco trioni-Eragrostietum poaeoidis. The latter was recently found in several arable fields in Southern Moravia (Czech Republic) and was newly characterized.     

Picea abies andAbies alba forests of the austrian alps: Numerical classification and ordination

A TWINSPAN classification of a representative set of 3026 relevés of spruce and fir forests from the Eastern Alps (Austria) is presented. Ecological features of relevé clusters and species groups are described by means of Ellenberg indicator values, site factors and stand characteristics. The most important floristic discontinuity in the data set separates acidophilous communities on mostly silicate substrates from basiphilous communities on mostly carbonate substrates. Further divisions reflect a combined gradient of temperature, nutrient regime and shading. This is supported by the correlation of average Ellenberg values of sample plots with DCA axes. A qualitative comparison between TWINSPAN clusters and a syntaxonomic system widely used in the Austrian Alps is drawn. The two ordersPiceetalia excelsae andAthyrio-Piceetalia largely coincide with the clusters of the first level of divisions. Alliances are partly reproduced by TWINSPAN. Clusters on the fifth and fourth level of division mostly correspond to associations. However, a considerable portion of the lower level clusters is of a transitional type. Out of thirteen fir and spruce associations described for the Austrian Alps, five associations are not reproduced by TWINSPAN, i.e.Bazzanio-Piceetum, Veronico-Piceetum, Adenostylo alliariae-Abietetum, Asplenio-Piceetum andCarici-Piceetum. Three associations are split on the second level of division, i.e.Larici-Piceetum, Luzulo nemorosae-Piceetum andCalamagrostio variae-Piceetum.     

JUICE,software for vegetation classification

Abstract. The program JUICE was designed as a Microsoft® WINDOWS® application for editing, classification and analysis of large phytosociological tables and databases. This software, with a current maximum capacity of 30 000 relevés in one table, includes many functions for easy manipulation of table and header data. Various options include classification using COCKTAIL and TWINSPAN methods, calculation of interspecific associations, fidelity measures, average Ellenberg indicator values, preparation of synoptic tables, automatic sorting of relevé tables, and export of table data into other applications (word processors, spreadsheet programs or mapping packages). JUICE is optimized for use in association with TURBOVEG which is the most widespread database program for storing phytosociological data in Europe.     

Local ranges of phytosociological associations: are they reflected in numerical classification?

In the tradition of European phytosociology, delimitations of vegetation units such as associations are mostly based on data from small areas where more detailed vegetation sampling has been carried out. Such locally delimited vegetation units are often accepted in large-scale synthetic classifications, e.g. national vegetation monographs, and tentatively assigned to a small geographical range, forming groups of similar (vicarious) vegetation units in different small areas. These vicarious units, however, often overlap in species composition and are difficult to recognize from each other. We demonstrate this issue using an example of the classification of dry grasslands (Festuco-Brometea) in the Czech Republic. The standard vegetation classification of the Czech Republic supposes that the majority of accepted associations (66 out of 68) have a restricted distribution in one of the two major regions, Bohemia or Moravia. We compared the classification into traditional associations with the numerical classification of 1440 phytosociological relevés from the Czech Republic, in order to test whether the traditionally recognized associations with small geographical ranges are reflected in numerical classification. In various comparisons, the groups of relevés identified by numerical analysis occupied larger areas than the traditional associations. This suggests that with consistent use of total species composition as the vegetation classification criterion, the resulting classification will usually include more vegetation units with larger geographical ranges, while many of the traditional local associations will disappear.     

Phytosociological data give biased estimates of species richness

Abstract. Large phytosociological data sets of three types of grassland and three types of forest vegetation from the Czech Republic were analysed with a focus on plot size used in phytosociological sampling and on the species‐area relationship. The data sets included 12975 relevés, sampled by different authors in different parts of the country between 1922 and 1999. It was shown that in the grassland data sets, the relevés sampled before the 1960s tended to have a larger plot size than the relevés made later on. No temporal variation in plot sizes used was detected in forest relevés. Species‐area curves fitted to the data showed unnatural shapes, with levelling‐off or even decrease in plot sizes higher than average. This distortion is explained by the subjective, preferential method of field sampling used in phytosociology. When making relevés in species‐poor vegetation, researchers probably tend to use larger plots in order to include more species. The reason for this may be that a higher number of species gives a higher probability of including presumed diagnostic species, so that the relevé can be more easily classified in the Braun‐Blanquet classification system. This attitude of phytosociologists has at least two consequences: (1) in phytosociological data bases species‐poor vegetation types are underrepresented or relevés are artificially biased towards higher species richness; (2) the suitability of phytosociological data for species richness estimation is severely limited.     

Plot sizes used for phytosociological sampling of European vegetation

Abstract. In European phytosociology, variable plot sizes are traditionally used for sampling different vegetation types. This practice may generate problems in current vegetation or habitat survey projects based on large data sets, which include relevés made by many authors at different times. In order to determine the extent of variation in plot sizes used in European phytosociology, we collected a data set of 41 174 relevés with an indication of plot size, published in six major European journals focusing on phytosociology from 1970 to 2000. As an additional data set, we took 27 365 relevés from the Czech National Phytosociological Database. From each data set, we calculated basic statistical figures for plot sizes used to sample vegetation of various phytosociological classes. The results show that in Europe the traditionally used size of vegetation plots is roughly proportional to vegetation height; however, there is a large variation in plot size, both within and among vegetation classes. The effect of variable plot sizes on vegetation analysis and classification is not sufficiently known, but use of standardized plot sizes would be desirable in future projects of vegetation or habitat survey. Based on our analysis, we suggest four plot sizes as possible standards. They are 4 m² for sampling aquatic vegetation and low‐grown herbaceous vegetation, 16 m² for most grassland, heathland and other herbaceous or low‐scrub vegetation types, 50 m² for scrub, and 200 m² for woodlands. It has been pointed out that in some situations, sampling in either small or large plots may result in assignment of relevés to different phytosociological classes or habitat types. Therefore defining vegetation and habitat types as scale‐dependent concepts is needed.     

Two-step vegetation analysis based on very large data sets

A two-step method for the classification of very large phytosociological data sets is demonstrated. Stratification of the set is suggested either by area in the case of a large and geographically heterogeneous region, or by vegetation type in the case of a set covering all the plant communities of an area. First, cluster analysis is performed on each subset. The resulting basic clusters are summarized by calculating a ‘synoptic coverabundance value’ for each species in each cluster. All basic clusters are then subjected to the same procedure. Second order clusters are interpreted as community types. The synoptic value proposed reflects both frequency and average cover-abundance. It is emphasized that a species should have a high frequency to be used as a diagnostic species. The method is demonstrated with a set of 1138 relevés and 250 species of coastal sand dune vegetation in Yucatan treated with the programs TWINSPAN and TABORD. Some problems and perspectives of the approach are discussed in the light of hierarchy theory and classification theory.     

Geographical and ecological differentiation in Greek Fagus forest vegetation

Question: Which are the gradients of floristic differentiation in Greek beech (Fagus sylvatica) forests? Which is the role of geographical and ecological factors in this differentiation? Location: Beech forests of the plant geographical regions Northeast, North Central and East Central Greece. Methods: A total of 1404 published and unpublished phytoso‐ciological relevés were used in the analyses. TWINSPAN and DCA were applied to classify and ordinate the relevés. Altitude, Indicator Values of relevés and their X and Y coordinates were used in a posteriori interpretation of the ordination axes. Kendall's correlation coefficients were calculated between DCA relevé scores and explanatory variables. Multiple linear regression was used to partition the variation explained by the first two DCA axes, between the geographical and the ecological variables. Results: Classification resulted in 14 vegetation units defined by species composition. Two types of gradients, ecological and geographical, were revealed by the DCA of all releves. The partition of the variation accounted for by the first and second DCA axis was attributed mainly to ecological and geographical variables, respectively. Conclusions: Beech forests of northeast and Central Greece show phytogeographical differences, while ecologically similar vegetation units occur in both regions. A west‐east gradient is revealed in Greek beech forest vegetation. The extent of the study area, its position along regional gradients and the comprehensiveness of the data set that is analysed determine the types of the gradients which can be revealed in a vegetation study.     

A phytosociological classification of Swiss mire vegetation

The mapping and monitoring of Swiss mires has so far relied on a classification system based on expert judgement, which was not supported by a quantitative vegetation analysis and which did not include all wetland vegetation types described in the country. Based on a spatially representative sample of 17,608 relevés from 112 Swiss mires, we address the following questions: (1) How abundant are wetland vegetation types (phytosociological alliances) in Swiss mires? (2) How are they distributed across the country––is there a regional pattern? (3) How clearly are they separated from each other? (4) How clear and reliable is their ecological interpretation? Using published wetland vegetation relevés and lists of diagnostic species for phytosociological units (associations and alliances) established by experts, we developed a numerical method for assigning relevés to units through the calculation of similarity indices. We applied this method to our sample of 17,608 relevés and estimated the total area covered by each vegetation type in Switzerland. We found that vegetation types not included in previous mapping were either rare in Switzerland (partly due to mire drainage) or poorly distinguished from other vegetation units. In an ordination, the Swiss mire vegetation formed a triangular gradient system with the Sphagnion medii, the Caricion davallianae and the Phragmition australis as extreme types. Phytosociological alliances were clearly separated in a subset of 2,265 relevés, which had a strong similarity to one particular association, but poorly separated across all relevés, of which many could not be unequivocally assigned to one association. However, ecological gradients were reflected equally well by the vegetation types in either case. Overall, phytosociological alliances distinguished until now proved suitable schemes to describe and interpret vegetation gradients. Nevertheless, we see the urgent need to establish a data base of Swiss wetland relevés for a more reliable definition of some vegetation units.     

A new measure of fidelity and its application to defining species groups

Abstract. The first objective of this paper is to define a new measure of fidelity of a species to a vegetation unit, called u. The value of u is derived from the approximation of the binomial or the hypergeometric distribution by the normal distribution. It is shown that the properties of u meet the requirements for a fidelity measure in vegetation science, i.e. (1) to reflect differences of a species’relative frequency inside a certain vegetation unit and its relative frequency in the remainder of the data set; (2) to increase with increasing size of the data set. Additionally (3), u has the property to be dependent on the proportion of the vegetation unit's size to the size of the whole data set. The second objective is to present a method of how to use the value of u for finding species groups in large data bases and for defining vegetation units. A species group is defined by possession of species that show the highest value of u among all species in the data set with regard to the vegetation unit defined by this species group. The vegetation unit is defined as comprising all relevés that include a minimum number of the species in the species group. This minimum number is derived statistically in such a way that fewer relevés always belong to a species group than would be expected if the differential species were distributed randomly among the relevés. An iterative algorithm is described for detecting species groups in data bases. Starting with an initial species group, species composition of this group and the vegetation unit defined by this group are mutually optimized. With this algorithm species groups are formed in a data set independently of each other. Subsequently, these species groups can be combined in such a way that they are suited to define commonly known syntaxa a posteriori.     

Vegetation structure along the altitudinal gradient at the treeline of Mount Paektu,North Korea

Vegetation structure of the treeline was studied on the southeastern slope of Mount Paektu (North Korea). Fifteen temporary plots (20m×20m) were situated along the altitudinal transect. Seven square samples (phytosociological relevés 5 m×5 m) were analyzed in each plot. Although the character of the vegetation was rather continuous altitudinally, five similar groups of vegetation samples were distinguished by the TWINSPAN procedure. The distribution of samples in the classification showed not only the effect of altitude but also the effect of competition between the tree and the shrub layers and other factors. In the canonical correspondence analysis (CCA) ordination several environmental variables were used (such as altitude of each plot, distance of each plot from the forest edge and the cover percentage of tree, shrub, herb and moss/lichen layers). The variable ‘cover of tree layer’ had the strongest relation to the species data. Continuous changes of species richness along the altitude were confirmed. Species richness decreased significantly with altitude.     

Numerical reproduction of traditional classifications and automatic vegetation identification

Questions: Is it possible to develop an expert system to provide reliable automatic identifications of plant communities at the precision level of phytosociological associations? How can unreliable expert‐based knowledge be discarded before applying supervised classification methods? Material: We used 3677 relevés from Catalonia (Spain), belonging to eight orders of terrestrial vegetation. These relevés were classified by experts into 222 low‐level units (associations or sub‐associations). Methods: We reproduced low‐level, expert‐defined vegetation units as independent fuzzy clusters using the Possibilistic C‐means algorithm. Those relevés detected as transitional between vegetation types were excluded in order to maximize the number of units numerically reproduced. Cluster centroids were then considered static and used to perform supervised classifications of vegetation data. Finally, we evaluated the classifier's ability to correctly identify the unit of both typical (i.e. training) and transitional relevés. Results: Only 166 out of 222 (75%) of the original units could be numerically reproduced. Almost all the unrecognized units were sub‐associations. Among the original relevés, 61% were deemed transitional or untypical. Typical relevés were correctly identified 95% of the time, while the efficiency of the classifier for transitional data was only 64%. However, if the second classifier's choice was also considered, the rate of correct classification for transitional relevés was 80%. Conclusions: Our approach stresses the transitional nature of relevé data obtained from vegetation databases. Relevé selection is justified in order to adequately represent the vegetation concepts associated with expert‐defined units.     

Determination of diagnostic species with statistical fidelity measures

Abstract. Statistical measures of fidelity, i.e. the concentration of species occurrences in vegetation units, are reviewed and compared. The focus is on measures suitable for categorical data which are based on observed species frequencies within a vegetation unit compared with the frequencies expected under random distribution. Particular attention is paid to Bruelheide's u value. It is shown that its original form, based on binomial distribution, is an asymmetric measure of fidelity of a species to a vegetation unit which tends to assign comparatively high fidelity values to rare species. Here, a hypergeometric form of u is introduced which is a symmetric measure of the joint fidelity of species to a vegetation unit and vice versa. It is also shown that another form of the binomial u value may be defined which measures the asymmetric fidelity of a vegetation unit to a species. These u values are compared with phi coefficient, chi‐square, G statistic and Fisher's exact test. Contrary to the other measures, phi coefficient is independent of the number of relevés in the data set, and like the hypergeometric form of u and the chi‐square it is little affected by the relative size of the vegetation unit. It is therefore particularly useful when comparing species fidelity values among differently sized data sets and vegetation units. However, unlike the other measures it does not measure any statistical significance and may produce unreliable results for small vegetation units and small data sets. The above measures, all based on the comparison of observed/expected frequencies, are compared with the categorical form of the Dufrêne‐Legendre Indicator Value Index, an index strongly underweighting the fidelity of rare species. These fidelity measures are applied to a data set of 15 989 relevés of Czech herbaceous vegetation. In a small subset of this data set which simulates a phytosociological table, we demonstrate that traditional table analysis fails to determine diagnostic species of general validity in different habitats and large areas. On the other hand, we show that fidelity calculations used in conjunction with large data sets can replace expert knowledge in the determination of generally valid diagnostic species. Averaging positive fidelity values for all species within a vegetation unit is a useful approach to measure quality of delimination of the vegetation unit. We propose a new way of ordering species in synoptic species‐by‐relevé tables, using fidelity calculations.     

Bayesian classification of vegetation types with Gaussian mixture density fitting to indicator values

Question: Is it possible to mathematically classify relevés into vegetation types on the basis of their average indicator values, including the uncertainty of the classification? Location: The Netherlands. Method: A large relevé database was used to develop a method for predicting vegetation types based on indicator values. First, each relevé was classified into a phytosociological association on the basis of its species composition. Additionally, mean indicator values for moisture, nutrients and acidity were computed for each relevé. Thus, the position of each classified relevé was obtained in a three‐dimensional space of indicator values. Fitting the data to so called Gaussian Mixture Models yielded densities of associations as a function of indicator values. Finally, these density functions were used to predict the Bayesian occurrence probabilities of associations for known indicator values. Validation of predictions was performed by using a randomly chosen half of the database for the calibration of densities and the other half for the validation of predicted associations. Results and Conclusions: With indicator values, most reléves were classified correctly into vegetation types at the association level. This was shown using confusion matrices that relate (1) the number of relevés classified into associations based on species composition to (2) those based on indicator values. Misclassified relevés belonged to ecologically similar associations. The method seems very suitable for predictive vegetation models.     

Formal definitions of Slovakian mire plant associations and their application in regional research

We applied the Cocktail method to a large data set of 4 117 relevés of all Slovak vegetation types with the aim to create formalised definitions of all Slovakian mire plant associations. We defined 21 groups of species with the statistical tendency of joint occurrences in vegetation. These groups differed substantially in their position along the pH/calcium gradient. We further defined 24 plant associations according to presence and/or absence of certain groups and/or strong dominance of some species. Only six traditional plant associations were not possible to be reproduced this way. We applied our formalised definitions to the regional data set of mires from the surrounding of the Vysoké Tatry Mts. Combined with frequency-positive fidelity index this method has led to the classification of the majority of vegetation plots into ten associations. When the vegetation types obtained from Cocktail-based classification and from cluster analysis were compared with respect to measured pH and conductivity in the study region, 82% of pairs differed significantly either in pH or in water conductivity in the former classification and 69% in the latter one.     

Diversity,variability and habitat characteristics of the communities dominated by Sesleria species (Poaceae) in the Western Carpathians

This study provides a view of vegetation types dominated by Sesleria species in the Western Carpathians (Sesleria caerulea, S. heufleriana, S. tatrae, and S. uliginosa). We also took into account characteristics/traits such as plant life forms, autochthonous status, endemism, and ploidy level occurring within each relevé in our data set. Altitude and Ellenberg indicator values derived for each relevé were considered as well. Eight vegetation types/formations/habitats were recognized in a data set of 942 phytosociological relevés: (1) synanthropic vegetation, (2) rock fissures and screes, (3) alpine grasslands and heaths, (4) mesic grasslands and pastures, (5) springs and fens, (6) forests, (7) xerophilous shrubland, and (8) high-mountain scrubs (krummholz). Results corroborated and clearly emphasized that Sesleria caerulea has the widest ecological amplitude of all studied species because the species occurred among all studied vegetation types. Sesleria tatrae was present only in several vegetation types occurring from montane to alpine vegetation belts in the highest mountains. Sesleria heufleriana and S. uliginosa were recorded only in low-altitude areas. The difference between them lies mainly in the dampness of each locality. Sesleria heufleriana was frequently found in xerophilous communities, whereas S. uliginosa preferred humid habitats of springs and fens.     

Re-assessing the ecology of rare flood-meadow violets (<Emphasis Type="Italic">Viola elatior,V. pumila</Emphasis> and<Emphasis Type="Italic">V. persicifolia</Emphasis>) with large phytosociological data sets

This study demonstrates how conventional ecological knowledge on species together with models resulting from functional traits can be tested and refined by tapping large data sources that have been made available through recent electronic compilations. The study is based on the comparison of three rare, closely related flood-meadow violet species, which have been supposed to have similar ecological behaviour and niche occupation. In contrast, the analysis of 335 Central European relevés using different methods of numerical ordination, classification and calibration revealed distinct differences in habitat preferences between the three species. Detrended correspondence analysis ordination, and multiple-response permutation procedures and TWINSPAN classification displayed the separation ofViola persicifolia fromV. elatior andV. pumila along a moisture and base-richness gradient, while the latter two differed mostly in terms of mowing compatibility. Although the three violets are considered to be weak competitors they may be found under nutrient-poor as well as nutrient-rich site conditions. The distribution of C-S-R strategy types in relevés supported the hypothesis that at more fertile sites the violets crucially depend on disturbance by management or flooding events that create gaps and weaken strong competitors, while at less productive sites they may persist for a long time even under fallow conditions. The S/R strategy is shared by all three violets. Problems and perspectives arising with the use of phytosociological databanks as a source of ecological information are discussed.