4 Classification of Plant Communities
Plant community classification, at its core, is the grouping of similar assemblages of plant species into classes for the purpose of communication and further study (Whittaker 1973). The goals of the classification were to identify recurring assemblages of plant species and describe the floristic composition of these assemblages. The 472 releves inventoried in the summer of 1997 were used to develop a plant community classification for DPG. The classification of DPG communities was a four-step process.
1. Placement of samples (releves) into one of four physiognomic classes (Dwarf Woodland, Woodland, Shrubland, Herbaceous). These physiognomic classes were derived from the standardized national vegetation classification system (SNVCS) and preliminary field data collected in 1996 (The Nature Conservancy 1994b).
2. Arrangement of the samples into four sample-by-species abundance matrices reflecting the initial physiognomic classification in step 1.
3. Classification of samples in each physiognomic matrix into community associations through multivariate analysis and association table work.
4. Naming of the community associations based on the framework of the SNVCS and development of association descriptions.
Raw vegetation data, provided by DPG, were arranged into a single sample-by-species abundance matrix using the cover class midpoints found in Table 1 as data, with each releve representing a sample (Daubenmire 1959; Gauch 1982; Bonham 1989). The large number of releves (samples) resulted in a large, uninterpretable data matrix. As a result, investigators reduced the data by assigning each sample to one of four physiognomic class categories based upon the SNVCS (The Nature Conservancy 1994b) and preliminary data collected in 1996. This initial physiognomic assignment served to improve subsequent multivariate classification by reducing the statistical "noise" commonly associated with large ecological data sets (Gauch 1982; Krebs 1989). Each sample was assigned to a physiognomic class using the following key:
1. Juniperus osteosperma over above 2 meters 2
No J. osteosperma cover above 2 meters 3
2. Crowns widely spaced, cover 10 to 25% Dwarf Sparse Woodland
Crowns not touching, cover 25 to 60% Dwarf Woodland
3. Woody shrub cover > 10%, under two meters Shrubland
Woody shrub cover < 10%, under two meters Herbaceous
After the initial physiognomic class assignments were made, each matrix was examined. Both woodland classes were combined for the subsequent multivariate classification because of their obvious similarities. As a result, three matrices (combined Woodland, Shrubland, and Herbaceous) were analyzed using multivariate methods.
Gauch (1982) recommends the use of nonhierarchical clustering (NHCL) techniques when working with large and/or unfamiliar ecological data for classifying vegetation data. However, NHCL requires the number of clusters be supplied by the investigator. To designate an ecologically realistic number of clusters for the NHCL procedure, two additional multivariate techniques, hierarchical clustering (HCL) and multidimensional scaling (MDS), were used to determine the likely number of clusters within each data matrix. Each of the three physiognomic class sample-by-species data matrices underwent the multivariate analysis separately (Figure 3). Multivariate analysis was performed by Syntax 5.0 (a computer program for multivariate data analysis, Podani 1993).
Hierarchical Clustering and Multidimensional Scaling
Sample dissimilarity, using the percentage difference algorithm, was calculated for each physiognomic matrix. The samples were then hierarchically clustered by the unweighted pair-group method using arithmetic averages (van Tongeren 1987). This method of HCL is considered a sound method for the identification of plant communities (Gauch and Whittaker 1981; Gauch 1982; Krebs 1989). Dendrograms of the cluster analysis were generated and interpreted following the suggestions of Faith (1991).
Multidimensional scaling has been found to be a robust method for detecting patterns in community ecology (Minchin 1987; Austin 1991). Multidimensional scaling was also used to further investigate the number of possible clusters in each of the physiognomic data matrices. The percentage difference algorithm was used to calculate sample dissimilarity. The results of the MDS were plotted in a two-dimensional ordination space and the number of obvious groups noted.
Nonhierarchical Clustering
Nonhierarchical clustering was performed on each physiognomic matrix, with the number of clusters for each based on the results of the HCL and MDS. The percentage difference algorithm was used to calculate species dissimilarity. The NHCL results were summarized in separate Braun-Blanquet association tables, ostensibly representing individual plant associations, and were examined (Mueller-Dombois and Ellenberg 1974; Gauch 1982).
Ecological Sense
Further refinement of the association tables was deemed necessary in some instances because the cluster did not make ecological sense. For instance, some tables showed combinations of species that had not previously been reported and whose recognized distributions did not overlap. The individual association tables requiring further refinement underwent the same multivariate analysis procedures (Figure 3). The results were interpreted and additional association tables created, reflecting the further refinement. The association tables were reexamined and limited association table work was used to finalize each association table. Each final association table represented a single floristic association at DPG.
Data Summarizations
Associations. Once the association tables had been finalized, mean vegetative cover was calculated for each species in the stratum or strata in which it occurred. Species constancy was also calculated for all species in the association tables. Species constancy was defined as number of samples a species occurred in divided by the number of samples in the association (Kent and Coker 1992). A species was considered dominant if it had a constancy of 1.0 and high vegetative cover relative to the other species within the association. The association names were derived from the dominant species in each stratum.
After the floristic composition of each association was determined, MDS was used to examine the similarity relationships between each association (Figure 4. Associations are identified using the first two letters of the genus and species names of the dominant species.). In addition, the frequency of each association (number of plots in each association divided by the total number of plots) was calculated.
Figure 4. Similarity relationship of the identified associations at DPG calculated by MDS.
Alliances and Formations. The alliances and formations were derived from the final associations. Since the SNVCS is a hierarchical system, it is possible to derive less detailed levels of the classification system from more detailed levels. Alliances were identified by the genera of the dominant species in the upper most stratum. Formations were named based on environmental and physiognomic characteristics of the floristic associations.
Environmental Relationships. The relationship between the identified community associations and the environment was examined. There was no attempt to directly correlate the distribution of the associations with environmental variables, primarily because detailed edaphic data was not collected.
