Application of unsupervised classification in NIR spectroscopy of soil
Abstract
Assessment of soil quality requires expensive and time-consuming measurements in the laboratory
and the field. Many repetitions of the measurements are required to deal with high soil
variability. As a result, scientists have been unable to measure and monitor soil quality and soil
degradation over large areas. Sensing Soil Quality is a technological approach for rapid assessment
and large area surveillance of soil condition. The technology is based on rapid screening
of soil quality using a portable reflectance spectrometer. The resulting data set requires reliable
statistical methods that can aid to fast extraction of information. Several clustering algorthim
procedures has been used to find the optimal classes in the NIR spectroscopy data set for
global soils. However, there were no significant soil classes obtained for the 697 topsoil samples.
Therefore, other methods with potential of handling large dimensional data need to be adopted.
For, example performing direct calibration of the data set or using both physical and chemical
soil properties data. We are also recommending where possible to integrate clustering results
with advanced graphical representation for example by using minimum spanning tree- MST.
Citation
Master of Science (Biometry)Publisher
University of Nairobi School of Mathematics