Digital soil pattern recognition in the lower Namoi valley using numerical clustering of gamma-ray spectrometry data

Abstract

Digital soil mapping involves the use of ancillary data (e.g. proximal and remotely sensed) as surrogates for soil information to produce maps of soil type or classes. Two of the most popularly used sources of ancillary data are the proximally sensing electromagnetic (EM) induction instruments and remotely sensed digital elevation model data (DEM). However, these data have limitations at the catchment level and are of limited use on predominantly flat alluvial landscapes, respectively. Another option is the use of remotely sensed gamma-ray (γ-ray) spectrometry data which has successfully been used to map the regolith. In this paper, the use of γ-ray spectrometry data (i.e. potassium—K, uranium—U, thorium—Th and total count—TC), coupled with a numerical clustering algorithm (i.e. fuzzy k-means (FKM) algorithm) is explored in order to identify geological and geomorphological units at the district and soil mapping units at a broader catchment level. We do this by using the Euclidean distance and the measures of fuzziness performance index (FPI) and normalized classification entropy (NCE) to identify k=11 classes and a fuzziness exponent (ϕ)=2.0 for interpretation. The k=11 classes produce contiguous classes, which are consistent with geological and geomorphological interpretations of an eroded landscape, alluvial lands and dust-mantled alluvial lands at the district level. At the sub-catchment level the k=11 classes also match broad soil mapping units and also elucidate subtle differences of the alluvial lands which characterise the agriculturally significant parts of the lower Namoi valley. Fuzzy canonical analysis shows that K and TC contribute most to the discrimination of the classes associated with the trachyte rich parts of the eroded landscape, the alluvial lands and most of the dust-mantled alluvial lands, whilst U and Th discriminate the basaltic outliers and other geomorphological units of the eroded lands. We conclude that the approach allows soil management and landscape units to be identified with the information used as a first approximation to determine where soil sample locations need to be collected to validate the map units identified. In order to better incorporate and characterise subsoil properties the inclusion of EM signal data (e.g. EM38 or DUALEM-1) may also be appropriate.