Soil Sci Plant Nutr. 40-3, 403-414 (1994)
Key words: numerical classification, paddy soils, rice production, soil mapping
In order to determine how a conventional soil map on a
local scale can be reproduced from soil sample information, a set of numerical
procedures was applied to the soil chemical data from 159 soil samples
collected in paddy fields of the Aizu Basin, in the northern part of Japan.
The digital elevation data were also included as an attribute.
Various steps were involved. In the first step, the sparse
data were converted into a closely spaced data set, This first step was
previously described in Part 1 of this paper. In the second step a multivariate
ordination (principal component analysis) was applied to the sampling data
in order to avoid redundancy of information. Large contributions to the
first component were derived from the values of CEC and exchangeable Ca
and Mg. In the third step the sampling sites were numerically classified
using the results of four or five components obtained in the second step.
A hierarchical clustering method was applied to proved the initial set
of clusters for non-hierarchical clustering. In the fourth step the closely
spaced data collected it the first step were allocated to an appropriate
class obtained by using the non-hierarchical clustering method applied
in the third step. for the allocation method, the geographic location of
the sampling sites was included as an additional characteristic of the
data.
In regional partition map based on the numerical procedures,
each soil group corresponded to a soil series i the original conventional
soil map which depend predominantly on detailed morphological and topographical
differences, In addition, the regional partition map was found to provide
useful information for the identification of factors closely related to
soil fertility, by statistically examining the differences in rice yield
among the soil groups.