the 18 environmental variables recorded (Table 1; note - 

 radiation index was used in these analyses rather than aspect) . 

 Both TWINSPAN and DCA are based on the same mathematical strategy 

 (i.e., reciprocal averaging; Hill 1979a, b) and thus offer direct 

 comparisons between the results of ordination and classification. 



All default options in the TWINSPAN algorithm v;ere used 

 except that pseudospecies cut levels were set at , 2, 5, 20, and 

 50 percent cover. Also, all default options were used in running 

 the ordinations except that rare species were downv;eighted. 

 First, the entire data matrix of 125 stands and 230 species was 

 analyzed. To reduce the amount of variation being considered, 

 which is substantial in the v.'hole matrix, the data set was also 

 subdivided into forest, shrubland, and grassland groups which 

 were analyzed separately. 



In some instances, a particular TWINSPAN class included a 

 plot or plots that, based on field experience and ordination 

 patterns, appeared to be better placed in a different existing 

 TWINSPAN class. These plots were repositioned in the classifica- 

 tion as appropriate. 



In addition to helping refine the classification, the ordi- 

 nations assisted in describing and interpreting general patterns 

 of vegetation communities and environment. For example, DCA 

 extracts the dominant compositional gradients from the species 

 data matrix, irrespective of site variables, whereas DCCA 

 extracts the dominant gradients given the constraint that they 

 must be orthogonal linear comibinations of the supplied environ- 

 mental variables (Ter Braak 1988) . 



Finally, species richness and diversity measures (Hill 1973) 

 were calculated for each plot using the AID computer program 

 (Overton et al. 1987). For those communities represented by more 

 than one plot, means and standard errors of means for each 

 measure were also calculated. 



Taxonomic Considerations 



Nomenclature follows Kartesz and Kartesz (1985) . Scientific 

 names of all species in this study, their code names, and their 

 synonyms (from GPFA 1986) are listed in Appendix A. 



Stipa spartea was observed on two of the study's 125 plots 

 and was a dominant element at both of these sites. However, 

 close examination of initial TWINSPAN and DCA patterns suggested 

 close ecological similarity between S. spartea and S. comata 

 within the study area. Because of this similarity, the two 

 species v;ere grouped under S. comata for all analyses reported 

 here. 



12 



