Abstract:
Soil parent material exerts a fundamental control on many environmental processes.
Nevertheless, resulting from the separate mapping programmes of the geological and
soil surveys, parent material is currently poorly mapped in the United Kingdom. This
research develops and tests four methods of predicting soil parent material using three
study areas in England. The qualities of desirable parent material maps were stated, and
then used to create new map value metrics to assess the success of the four
methodologies.
Firstly, translations of surface and bedrock geology maps to parent material maps were
tested, using international and national parent material classifications. Secondly,
qualitative expert knowledge of parent material, captured from published literature, was
formalised into inputs for a corrected probability model. Parent material likelihood was
predicted using three map evidence layers: geology, slope and soil. Thirdly, extensive
data mining was used to create fully quantitative inputs for the same probability model,
and the results were compared. The final method provided a quantitative framework for
the expert knowledge model inputs by the incorporation of sparse data sampling.
The expert knowledge method created parent material maps of higher value than those
created by the translation of geological maps. However, the inputs derived from
qualitative expert knowledge were demonstrated to benefit from the addition of
quantitative sample data. The resulting maps achieved overall accuracies between 60%
and 90% and contained numerous detailed classes with explicit probabilities of
prediction. Extensive parent materials were shown to be predicted well, and physically
and chemically distinctive parent materials could be effectively predicted irrespective of
their extent. Parent material class confusion arose between units where the evidence
datasets were unable to provide the sufficient geographic or descriptive detail necessary
for differentiation. In such cases, class amalgamation was used to overcome consistent
misclassification. Recommendations are provided for the application of this research.
