cated a reasonable normality of distribution for all 

 variables. 



Initial ordinations were performed using data for all 

 horizons and all 89 pedons. These ordinations produced 

 groupings, based on the presence or absence of data for a 

 single horizon, within a larger sequence of horizons. The 

 number of pedons having data for a fifth and sixth hori- 

 zon was too few to allow meaningful analysis with those 

 horizons included in the data set. Analysis was then 

 reduced to using the physical characteristics of the first 

 four horizons only. Ordination groups created from this 

 reduced data set still contained very dissimilar soils ex- 

 cept for the presence or absence of a thin A horizon or the 

 presence or absence of a fourth horizon. The fourth hori- 

 zon, when present, contained genetic horizon data that 

 described highly dissimilar B, C, or R type characteristics. 

 Although stratification of the data by parent material was 

 considered to have future utility, further ordination 

 analysis, based on inclusion of the fourth horizon data, 

 was deemed meaningless. 



Ordinations were next performed using data from the 

 upper three horizons and only those samples having an A 

 horizon present. A second set of ordinations was then 

 conducted on this same set of samples using only data 

 from the second and third horizons. Comparison of re- 

 sults of these ordinations indicated that very little infor- 

 mation was lost due to removal of the A horizon charac- 

 teristics. All further analyses use only data from the 

 second and third horizons. Because the data consist of 

 the same 11 variables found in two consecutive horizons, 

 a numerical suffix was added to the name of each of the 

 22 variables to identify the horizon of origin. Even 

 though an A horizon (that is, the first horizon) did not 

 occur in all pedons analyzed, for consistency the suffixes 

 used were 2 and 3. 



Widely differing parent materials produce significantly 

 different textural and structural qualities, coarse frag- 

 ment contents, and pH values, but often do not create 

 differences in color or depth. Data were stratified into 

 coarse-textured vs. fine-textured parent material groups 

 in an attempt to eliminate these confounding factors. 

 Basalt was grouped separately due to its basic properties, 

 as opposed to the acidic nature of the other parent materi- 

 als. Three groups were created: 



Coarse-textured 



n=55 



Alluvium - coarse 

 Glacial drift 

 Gneiss 

 Granite 



Mixed sedimentary 

 Quartzite 

 Quartz monzonite 

 Sandstone 

 Schist - coarse 



Fine-textured 



n=31 



Alluvium - fine 



Argillite 



Loess 



Mica schist 

 Phyllite 

 Schist - fine 

 Siltite 

 Silts tone 



Basalt 



n=3 



Basalt 



In all cases, volcanic ash, where present, is an overlying 

 amendment to the parent materials. 



Pattern Analysis 



If a soil-survey-oriented taxonomy can be developed 

 based on a combination of quantifiable and categorical 



horizon variables, then numerical taxonomic analysis of 

 these variables should assign the same samples to clus- 

 ters of closely equivalent taxonomic units. One problem 

 created by the monothetic design of the soil taxonomy 

 (USDA SCS 1975) is the emphasis placed on single vari- 

 ables in the delineation of taxonomic units. Two soil se- 

 qua similar in all respects except color of the epipedon can 

 vary taxonomically in Order, Suborder, and/or Great 

 Group. The emphasis in this study was not to mimic the 

 currently accepted soil taxonomy, but rather to investi- 

 gate the classification of polypedons based on multivariate 

 statistical analysis of physical attributes. Because of this 

 approach, the data from individual horizons were not 

 combined into a control section format as used in soil 

 taxonomy (USDA SCS 1975), nor was emphasis in the 

 form of weighting placed on any single variable or set of 

 variables. 



As soils are extremely variable and multivariate in 

 character, ordination was selected as the means to sum- 

 marize and reduce dimensionality of the data (Gauch 

 1982). Using the four ordination techniques and the 11 

 variables for each of two horizons as outlined above, no 

 identifiable relationships were discerned between numeri- 

 cally generated soil groupings, soil taxonomic units (using 

 all hierarchical units from Order to Family), and habitat 

 types within the full data set. Further stratification of 

 the data set to reduce internal variation appeared neces- 

 sary. The coarse-textured parent material group of 55 

 samples was selected for all further analyses. 



Analyses of this reduced data set by three of the ordina- 

 tion techniques ranked samples in similar positions 

 within their respective ordinations (Neiman 1986). Even 

 though the rankings of each technique concurred in a 

 general way, a large amount of variation occurred among 

 the soils. Low eigenvalues of the principal component 

 analysis indicated that only 19 percent of the total vari- 

 ation was explained by the first axis, 62 percent by the 

 first five axes, and 86 percent by the first 10 axes. The so- 

 called "cloud" of sample points in multidimensional space 

 in this case truly lived up to its name. This large amount 

 of unexplained variation in the data indicated that either 

 the selected variables were not suitable for numerical 

 grouping or that identifying soil groups numerically at 

 this level of stratification has no statistical or ecological 

 interpretive power. Yet, the ability to develop consistent 

 rankings of samples by the various analytical techniques 

 indicated a potential to define soil groups. The problem in 

 doing so appears to be the small data set and high vari- 

 ation inherent in soils. Variation could be further reduced 

 by stratifying the coarse-textured parent material group 

 to create a subset containing samples from only granite, 

 quartz monzonite, quartzite, and gneiss. This was not 

 performed due to sample size restrictions. 



Soil- Vegetation Relationships 



The second objective was to investigate relationships 

 between soil characteristics and forest habitat types. A 

 lack of correlation between the two taxonomic units can 

 be seen in appendix A. If the work of Jenny (1941, 1958) 

 and Major (1951) is correct, then some relatively discrete 

 relationship between the functional factors for soil and 



6 



