Studies in numerical taxonomy of soils

Wickramagamage, Piyasena


Wickramagamage, Piyasena (1982) Studies in numerical taxonomy of soils.

Our Full Text Deposits

Full text access: Open

10098477.pdf - 43.29 MB


A series of established numerical taxonomic strategies was applied to soil data from three sources: USDA (1975), De Alwis (1971) and the Soil Survey of England and Wales. The first two sources provided data for 41 soil profiles, which were classified without reference to their geographical location. The data obtained from the Soil Survey of England and Wales related to a particular geographical area (West Sussex Coastal Plain) and the geographical relationship between soil individuals was also examined.

Two methods of soil characterization (soil profile models) were compared with respect to their effect on the results produced by two hierarchical agglomerative strategies based on two measures of inter-individual similarity. Comparison of results, obtained from the agglomerative strategies for the two soil profile models, was made. The nature of inter-attribute correlation for depth levels modelled as arrays of independent attributes was examined, and all attributes were classified on the basis of inter-attribute correlation.

Seven hierarchical agglomerative strategies were examined with respect to their goodness-of-fit in the original space and also the relationship between goodness-of-fit and clarity of clusters was examined. From these comparisons, two agglomerative strategies were chosen to represent two classes of strategy: (a) strategies with minimum of distortion, (b) strategies with a greater distortion but clear clusters. The average linkage method from the first category and the Ward's error sum of squares (ESS) method from the second category were selected.

These two strategies were applied to the data sets described above using two measures of similarity namely (a) squared Euclidean distance and (b) Mahalanobis D2, and a divisive strategy, REMUL, was also applied to classify the soil populations. The classifications obtained from these strategies were compared by Wilk's Criterion A and the classification which had the lowest A was treated as the best initial partition. The best two partitions of the two populations obtained from the agglomerative strategy, Ward's ESS method, were further analysed. The optimum number of groups (G) in each population was decided by the relationship between LambdaG2 and G. The soil profile groups produced by these methods were further examined and improved by a reallocation strategy based on the Mahalanobis distance between individuals and the group centroids. Reallocation was done using 30 attributes from the uppermost soil horizons.

Canonical analysis was performed on the populations both before and after the classification. Canonical plots were produced and a comparison was made with the dendrograms obtained for the best partitions.

The classifications obtained were examined in relation to parent material classes. The spatial relationship of the soil groups of the West Sussex Coastal Plain was also investigated.

As shown by this study, it is possible to produce a better classification of soils by numerical taxonomic methods compared with traditional methods. For this end, it is not necessary to use all attributes of soils, but a sufficiently large number of properties, which can be empirically determined, is adequate for the purpose of producing a natural classification. The soil groups produced by numerical methods showed a closer association with parent materials.

Information about this Version

This is a Accepted version
This version's date is: 1982
This item is not peer reviewed

Link to this Version

Item TypeThesis (Doctoral)
TitleStudies in numerical taxonomy of soils
AuthorsWickramagamage, Piyasena
Uncontrolled KeywordsGeography; Social Sciences; Numerical; Soils; Studies; Soils; Taxonomy
DepartmentsDepartment of Geography



Deposited by () on 01-Feb-2017 in Royal Holloway Research Online.Last modified on 01-Feb-2017


Digitised in partnership with ProQuest, 2015-2016. Institution: University of London, Bedford College (United Kingdom).