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