Five additional pedon summarization or site-specific 

 variables were included in the analysis of soil horizon 

 data: Total depth of organic litter layers; total depth of 

 sequum to C horizon; total effective depth, calculated as 

 the summation of each horizon depth times [(100 - per- 

 cent coarse fragment)/]. 00] down to but not including the 

 C horizon; and total available water capacity, a summa- 

 tion of all horizon AWC's. Soil temperature, moisture 

 regime, or chemical composition data, such as base satu- 

 ration or cation exchange capacity, were not available for 

 analysis. A complete set of these data and definitions for 

 variables are presented in Neiman (1986). 



Data Matrix Design — Since root systems are not gen- 

 erally affected by the minor differences that are signifi- 

 cant to soil horizon classification, horizon data was ana- 

 lyzed in a simple sequential order, based on the depth 

 rather than genetic horizon (that is, first, second, third 

 horizon vs. Al, A2, AB, B2, . . .). This design was also 

 dictated by the similarity-dissimilarity index analysis and 

 ordination techniques available, wherein the presence or 

 absence of data for a group of variables is weighted more 

 heavily than are the individual quantitative values. Con- 

 sider, for example, two pedons identical in all respects 

 except for the presence of a 1-cm-deep A horizon in one of 

 the sequa. Based on the presence-absence relationships 

 in the first set of A horizon variables, ordination tech- 

 niques would place these two pedons in highly dissimilar 

 positions, whereas the presence of such a shallow A hori- 

 zon should be subordinate to similarities for variables in 

 the rest of the horizons. 



Because categorical names are simply a summarization 

 of horizon characteristics (such as color, texture, . . .), the 

 quantitative data for these characteristics should contain 

 equivalent if not more definitive information. A major 

 problem arises when sequential horizonation rather than 

 genetic horizonation is used for analysis. The problem 

 occurs when one soil description begins with an A horizon 

 and another sample begins with a B horizon. By not us- 

 ing categorical names in the analysis, the ability to differ- 

 entiate A from B is lost. Forest soils of northern Idaho 

 often do not develop an A horizon, yet when present, it 

 was considered to be potentially significant in analysis of 

 soil -vegetation relationships. Therefore, the first set of 27 

 horizon characteristics was allotted to only A horizon 

 data, allowing for simplified analysis of presence-absence 

 or quantitative data within only A horizons. For samples 

 having more than one A horizon, a weighted-by-thickness 

 average for all characteristics was used as the single set 

 of A horizon data. The second and subsequent sequential 

 horizon data sets record all other horizonation, and thus 

 are restricted to AB, E, B, C, and R type illuvial and par- 

 ent material horizons. 



Data Analysis — Analysis was divided into three sepa- 

 rate processes: The first investigated noise and redun- 

 dancy of variables in the data set of 27 characteristics per 

 horizon; the second attempted to delineate naturally oc- 

 curring patterns of soil physical characteristics and assess 

 their relationship to the vegetation types that they sup- 

 port; and the third developed discriminant functions 

 based on soils data that are predictive for habitat type 



classification. Due to a disparity in both size and units of 

 measure, all variables were standardized to a mean of 1 

 and a standard deviation of 0.1 (SAS 1982b). All data, 

 raw and standardized, were analyzed for normal, skewed, 

 or bimodal distribution (SAS 1982a) across the entire data 

 set and within sets stratified by habitat type. 



Noise was considered as variation in one characteristic 

 being not coordinated with variation in another (Gauch 

 1982). Noise analysis was restricted to use of means and 

 range data, with only those variables which were constant 

 across the data (and therefore contain no useful informa- 

 tion) being removed from further analysis. Correlation 

 analysis of all possible pairs (SAS 1982a) and principal 

 components analysis (Gauch 1977) were used to evaluate 

 redundancy within and relationships between variables 

 across the entire data set and for data stratified by either 

 habitat types or parent material groups. The objective of 

 these analyses was to create a reduced data set of as few 

 independent variables as possible without sacrificing 

 meaningful information. 



Pattern analysis was conducted using a series of ordina- 

 tion techniques: polar ordination (Bray and Curtis 1957); 

 principal components analysis (Gauch 1977); two-way 

 indicator species analysis (Hill 1979b); and detrended 

 correspondence analysis (Hill 1979a). All of these tech- 

 niques are described as dimensionality reduction tech- 

 niques, but each approaches the problem from a slightly 

 different perspective. All four techniques allow for ordi- 

 nation of both variables and samples in the same analy- 

 sis, which makes them useful for exploring variable re- 

 duction within samples, pattern analysis between 

 samples, and delineation of variables related to patterns 

 of samples. 



Vegetation-soil relationships were analyzed using a 

 subset of samples stratified by parent material and fur- 

 ther stratified by habitat type. Techniques used to iden- 

 tify significant discriminators were: factor analysis (SAS 

 1982b); stepwise discriminant analysis (Dixon 1981); and 

 canonical discriminant analysis (SAS 1982b). Using the 

 set of significant variables identified by these programs, 

 classification models based on discriminant functions 

 were developed using discriminant analysis (SAS 1982b). 



RESULTS AND DISCUSSION 

 Data Reduction 



Criteria for retaining a variable in the data were as 

 follows: continuous or a class of continuous values; not 

 related to short-term vegetational changes or person- 

 caused disturbance; suited to accurate assessment in the 

 field; requires minimal subjective interpretation; and not 

 influenced by other characteristics. Based on these crite- 

 ria, a subset of 11 variables per horizon was selected for 

 use in all further analyses. These were: depth; moist color 

 value; moist color chroma; structural size and shape; 

 percentages of clay, silt, gravel, cobble, and stone; and pH. 

 All variables selected are quantified in terms of continu- 

 ous or classes of continuous units, except for structural 

 shape, which was quantified into categories whose in- 

 creasing values denote increasing development through 

 illuviation of fine soil material. Univariate analysis indi- 



5 



