Formalized reproduction of an expert-based phytosociological classification

604

Kočí, M. et al.

associations. Therefore in some Cocktail definitions of

associations we used dominance of single species in

combination with occurrence of species groups. A thresh-

old cover value of either 25% or 50% was used as

dominance criterium.

Similarity-based assignment of relevés to vegetation

units

Some Cocktail definitions of associations created in

this study marginally overlapped, i.e. a few relevés

could be assigned to more than one association. These

relevés and the relevés unassigned to any of the associ-

ations were subsequently assigned to one of the associ-

ations by calculating their similarity to the constancy

columns of a synoptic table created from the relevés

which were unequivocally assigned to the associations

by Cocktail definitions.

For the similarity-based assignment of relevés to

vegetation units we designed a new index, called Fre-

quency-Fidelity Index (FFI), which is a combination of

Frequency Index (FQI) and Fidelity Index (FDI). The

Frequency Index is derived from the measure of

‘compositional satisfaction’ proposed by Hill (1989),

modified so as to deal with percentage frequencies of

species occurrence rather than with constancy classes:

FQI = 

Σ

 

i

∈R

 FQ

i

 

/

 

Σ

 

i

∈C

 FQ

i

(1)

FQ

i

 is the frequency (constancy) of species i in a con-

stancy column of a synoptic table. Species present in the

relevé are indicated as i 

∈R and species present in the

constancy column as i 

∈C. In the numerator, frequen-

cies are summed over all species of the constancy col-

umn that are also present in the relevé considered, while

in the denominator the sum is calculated over all species

of the constancy column.

The Frequency Index satisfactorily measures the

similarity of relevés to constancy columns in terms of

species composition; however, it gives the same weight

to diagnostic (specialist) species, i.e. those with a dis-

tinct concentration of occurrence in a certain vegetation

unit, and generalist species, i.e. those occurring in most

vegetation units of a synoptic table. Thus, for a con-

stancy column of a vegetation unit consisting of an

equal proportion of generalist and diagnostic species, a

relevé containing none of the diagnostic species but

sharing all the generalist species will have the same FQI

value as a relevé sharing all the diagnostic species but

lacking the generalists. As this feature may be disadvan-

tageous in phytosociological applications, we introduce

an alternative index, called Fidelity Index (FDI), which

is based on fidelity, i.e. concentration of species occur-

rence in a given vegetation unit:

FDI = 

Σ

 

i

∈R

 FD

i

 

/

 

Σ

 

i

∈C

 FD

i

(2)

This index is calculated in an identical way as the

Frequency Index but uses a fidelity measure (FD) in-

stead of frequency (Chytrý et al. 2002). In the calcula-

tions, we considered only positive fidelity values, which

clearly indicate one particular vegetation unit. By con-

trast, negative fidelity values of a particular species are

often very similar for several vegetation units. E.g., a

specialist species occurring in a single vegetation unit

will have nearly identical values of negative fidelity in

all other vegetation units. The fidelity measure (FD)

used in this paper was the phi coefficient (Sokal &

Rohlf 1995; Chytrý et al. 2002).

A major disadvantage of the Fidelity Index is that it

poorly discriminates between the relevés composed

exclusively of generalist species shared with the con-

stancy column on one hand, and the relevés composed

of totally different species than the constancy column

on the other hand. Both of these two types of relevés

yield a FDI value close or equal to zero. Therefore we

combined the Frequency Index and Fidelity Index into

a single Frequency-Fidelity Index (FFI), which retains

the advantages and lacks the disadvantages of both:

FFI = (FQI + FDI) / 2

(3)

Comparison of expert-based and formalized classi-

fication

To compare the expert-based and formalized classi-

fication, we applied the Cocktail definitions of asso-

ciations to the data set of 718 relevés originally used

for the expert-based classification by Kočí (2001). We

assigned the relevés that met these definitions to the

associations. Subsequently we calculated the Fre-

quency-Fidelity Index for the relevés that had been

assigned to more than one association by the Cocktail

method and assigned them to that of the candidate

associations for which the highest FFI  value was

yielded. Then we applied the same procedure to the

relevés that had not been assigned to any of the asso-

ciations by the Cocktail definitions.

The assignment of relevés to the associations in the

expert-based classification was first compared with their

assignment by the Cocktail definitions only, and sec-

ond, with the combined assignment by the Cocktail

definitions and similarity calculations. All analyses in

this paper were performed with the JUICE program

(Tichý 2002; www.sci.muni.cz/botany/juice.htm).

---