Improved spatial prediction of soil properties and soil types combining semi-automated landform classification, geostatistics and mixed effect modelling
Date
2011Author
Paron, Paolo
Omuto, Thine Christian
Type
ArticleLanguage
enMetadata
Show full item recordAbstract
Landforms classifications using a range of techniques
from pure automatic classification to hybrid semiautomated
model / expert opinion was tested as a base for a
robust modelling of soil characteristics spatial distribution.
The current approaches for producing soil maps use a single
model which either blocks/controls the grouping effects or
do not statistically recognize the natural landscape groupings.
This study tested mixed-effects modelling technique for
ingenious recognition of soil groupings and consequent improvement
of the accuracy of the resultant soil maps. It further
tested the various landscape classification for soil mapping.
Mixed-effects modelling is a form of regression analysis
for simultaneous modelling of the average landscape characteristics
and individual units within the landscape. Hence, it
can model a family of curves and potentially remove inadequacies
inherent in the current models for soil mapping. Its
potential in regression kriging of continuous and categorical
soil attributes has been shown in this paper. Compared to
the current application of a single model in regression kriging,
mixed-effects modelling produced about five times improvement
of the mapping accuracy. It is anticipated that its
adoption will contribute to improved soil mapping.