Energy dispersive X-ray fluorescence and scattering assessment of soil quality via partial least squares and artificial neural networks analytical modeling approaches.

Soil quality assessment (SQA) calls for rapid, simple and affordable but accurate analysis of soil quality indicators (SQIs). Routine methods of soil analysis are tedious and expensive. Energy dispersive X-ray fluorescence and scattering (EDXRFS) spectrometry in conjunction with chemometrics is a potentially powerful method for rapid SQA. In this study, a 25 m Ci (109)Cd isotope source XRF spectrometer was used to realize EDXRFS spectrometry of soils. Glycerol (a simulate of "organic" soil solution) and kaolin (a model clay soil) doped with soil micro (Fe, Cu, Zn) and macro (NO(3)(-), SO(4)(2-), H(2)PO(4)(-)) nutrients were used to train multivariate chemometric calibration models for direct (non-invasive) analysis of SQIs based on partial least squares (PLS) and artificial neural networks (ANN). The techniques were compared for each SQI with respect to speed, robustness, correction ability for matrix effects, and resolution of spectral overlap. The method was then applied to perform direct rapid analysis of SQIs in field soils. A one-way ANOVA test showed no statistical difference at 95% confidence interval between PLS and ANN results compared to reference soil nutrients. PLS was more accurate analyzing C, N, Na, P and Zn (R(2)>0.9) and low SEP of (0.05%, 0.01%, 0.01%, and 1.98 μg g(-1)respectively), while ANN was better suited for analysis of Mg, Cu and Fe (R(2)>0.9 and SEP of 0.08%, 4.02 μg g(-1), and 0.88 μg g(-1) respectively).