Development of a Chemometric Energy Dispersive X-Ray Fluorescence and Scattering Spectroscopy (EDXRFS) Method for Rapid Soil Quality Assessment
Abstract
Sustainable land use and agricultural productivity especially in precision farming depends on
soil quality management and thus necessitates soil quality assessment (SQA). This calls for
simple, affordable and rapid but accurate analysis of soil nutrients (herein called soil quality
indicators, SQIs). This study presents results of the systematic experimental study on the
applicability of chemornetrics-assisted energy dispersive X-ray fluorescence and scattering
(EDXRFS) spectroscopy for rapid and non-destructive characterization of soils for SQA.
The utility of I09Cd radioisotope-excited EDXRF technique has been extended by further
exploiting the scatter profiles obtained from soil samples to (i) correct for matrix effects
observed in the spectrum deconvolution of both fluorescence and scatter peaks, to the
concentration of selected SQIs, and (ii) to develop multivariate modeling and calibration
strategies utilizing principal component analysis (PCA), partial least squares (PLS) and
artificial neural networks (ANNs) for quantitative analysis of the SQIs. PCA was used for
spectral data compression and pattern recognition, while PLS and ANNs were used to design
and test the calibration strategies based on kaolin as a model soil matrix spiked with
simulated composition of micro (Fe, Cu, Zn) and macro (N03-, so,', H2P04-) nutrients. The
developed method was applied to determine the concentrations of micro (Fe, Cu, Zn) and
macro (OC, N, Na, Mg, P) nutrients in real field soils.
PLS and ANNs modeling resulted in varying quantitative prediction capability for both micro
and macronutrients with respect to calibration accuracy, matrix effects correction, resolution
of spectral overlaps, and signal-to-noise ratios (SNR) of the spectral signals. PLS
performance was optimal for the linear models i.e. those for Mg and Zn, while ANNs was
optimum for the non-linear models i.e. those for Fe, Cu and the macronutrients. The PLS and
ANNs predicted SQIs compared with the reference values showed no statistical difference at
95 % confidence interval using a one-way ANOV A test. The developed method furnished
bio-available SQls rapidly (t = 200 s - 750 s) and simultaneously with good dynamic range in
the trace (ug/g) level regime for micronutrients and percent levels for macronutrients even at
low SNR. Therefore, chemometrics-assisted EDXRFS spectroscopy allows for rapid, direct
and reliable predictions of SQls in real soils, making the approach useful for rapid SQA
Citation
Master of Science (M.Sc.) in Nuclear ScienceSponsorhip
University of NairobiPublisher
Institute ofNucJear Science & Technology University of Nairobi