Premium
The numerical classification of soils: a case study using data from West Sussex, England
Author(s) -
WICKRAMAGAMAGE P.,
FISHER G. C.
Publication year - 1988
Publication title -
journal of soil science
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.244
H-Index - 111
eISSN - 1365-2389
pISSN - 0022-4588
DOI - 10.1111/j.1365-2389.1988.tb01201.x
Subject(s) - mahalanobis distance , weighting , cluster analysis , mathematics , hierarchical clustering , statistics , homogeneous , set (abstract data type) , data set , data mining , artificial intelligence , pattern recognition (psychology) , computer science , combinatorics , radiology , programming language , medicine
SUMMARY A set of numerical clustering strategies was applied to soil data for 65 profile descriptions from profiles in West Sussex. The numerical classifications were compared to an original (1967) traditional classification, and at the higher categorial levels the composition of groups was found generally to be different. Such differences were ascribed to attribute weighting, missing information and restrictions on the types of data which could be employed by certain methods. Wilks’ Criterion, A, was used to find the classification producing the most homogeneous groups and, on the basis of tests using a restricted number of quantitative variables, numerical methods were found to give better groupings. An agglomerative, heirarchical procedure based on similarities calculated by Mahalanobis D 2 and clustering by minimizing error sums of squares was found most satisfactory, however, the difficulty of defining a best classification is discussed.