somewhat later and they were clipped in late August. These exclosures con- 

 tained the Andropogon scoparius , Juniperus horizontal is/Andropogon scoparius - 

 Agropyron spicatum , Andropogon scoparius - Agropyrc n spicatum , Bouteloua 

 gracil is , Bouteloua gracil is/Agropyron smithii-Bouteloua gracilis , and 

 Artemisia cana / Agropyron smithi i - Bouteloua gracil is community types. 

 Whitman (1975) described the average time of peak production for some species 

 found in the study area. He reported that thirty percent of the seasonal 

 aerial production was achieved by May 20, eighty percent by June 1, ninety- 

 three percent by July 31, and about ninety-seven percent by August 31. 

 Appendix C mentions a few of the shortcomings of this method of measuring 

 grassland productivity. 



Grasses were clipped to ground level. Plant matter was sorted by spe- 

 cies for the dominant species, and by life form for other species. Litter 

 was also collected. The material was oven-dried and weighed at Montana State 

 University. Statistical summaries were provided by the DNRC. 



CLASSIFICATION 



Cluster Analysis 



The DNRC is aware of the controversy surrounding the classification of 

 vegetation (See Appendix C), but it seems that classification is best suited 

 to meeting the objectives of the study. The Department of State Lands re- 

 quires that vegetation types be defined on the basis of dominance, and so a 

 number of criteria for classification such as this, using fidelity or con- 

 stancy, for example, did not have to be considered. (Whittaker (1962) 

 thoroughly discusses classification schemes.) However, all floristic data 

 from the samples were used, which should better express the relationship of 

 communities to one another and to environment than using only dominants. 



The sample plot data were agglomeratively clustered by the pair group 

 method and average linkage using Sorenson's (1948) index of similarity. The 

 technique is briefly explained below. 



Each sample from the reconnaissance data was compared to ewery other 

 sample using Sorensen's index of similarity. The midpoint values of the 

 cover classes were used in the index. Sorensen's index offers the ad- 

 vantage (over Jaccard's (1928) or Gleason's (1926) for example) of equating 

 two plots with the same recorded coverages. 



The most similar samples were then grouped, their coverages averaged, 

 and the resulting data compared with all other samples. The process was re- 

 peated until all samples joined all others. 



Species were not weighted prior to clustering because weighting adds 

 an unnecessary element of subjectivity into the analysis and can obscure 

 relationships inherent in the data. This is not to say, however, that 

 weighting is inherently bad. Since species have neither identical ecolog- 

 ical amplitudes nor ranges, weighting could be valuable, but unfortunately, 

 the autecologies of virtually all species under study are not sufficiently 

 understood to allow reliable a priori weighting. 



13 



