Forty Minnesota soils were classified both by a non-numerical classification (7th Approximation) and by five numerical classifications, based on (i) all 22 properties which were measured, (ii) on properties measured in the field, (iii) on moisture-related properties, (iv) on nutrient-related properties, and (v) on horizon texture and thickness.
The numerical relationship between any two profiles was evaluated on the basis of similarity in properties and sequence of component horizons. Horizons were considered to be isotropic units, and each horizon in a given profile was compared to three horizons at a comparable depth in a second profile. Euclidean distance was used to measure horizon similarities. All primary soil properties, whether qualitative or quantitative, were made to contribute equally to distance. Properties which better described a primary property (e.g. mottles described by abundance, size, and contrast) affected the contribution of that primary property to distance. Clusters were separated from the matrices of distance coefficients by the weighted-pair group method of cluster analysis.
The resulting classifications were compared to one another. Dividing the classifications into many groups (lower levels in the hierarchies) resulted in better relationships than did reducing the number of groups. Large groups contained dissimilar individuals, so accuracy was reduced. The numerical groups were subjectively homogeneous. Because the numerical classifications did not heavily weight “diagnostic” horizons, the results of these classifications generally differed from those of the non-numerical classification. However, the non-numerical classification based on textural classes of the family and the numerical classification based on horizon texture and thickness, because they were based on nearly the same properties, did correspond.