Chemical analysis is an important but often expensive and time-consuming step in the characterization of soils. Methods used for soil analysis ideally need to be rapid, accurate and relatively simple and infrared partial least squares (PLS) analysis is potentially one such method. Mid-infrared diffuse reflectance Fourier transform (DRIFT) spectra of powdered soils present the major mineralogical and organic components within each soil, relative to their concentrations. The theory indicates that experimentally derived soil properties may be correlated with the infrared spectra of some of these components, and the covariance between soil properties and spectra can then be modelled by PLS loadings and scores. Factors and scores can be derived independently for each Soil property using PLS-1, an extension of the more general PLS-2 method. This study evaluates the use of PLS-1 for the qualitative and quantitative study of soils, and in particular to classify the soil spectra and their associated major element chemistry by their PLS loadings and Scores. A subset of 100 soils, selected from a complete set of 298 samples from throughout eastern and southern Australia, was analysed by X-ray fluorescence (XRF) for major oxides as a calibration or training set to model the PLS loadings, scores and linear regression coefficients. Linear regressions resulted with R(2) values of 0 . 973-0 . 917 for XRF versus PLS predicted values for SiO2, Al2O3 and Fe2O3. Regressions for the other oxides, e.g. TiO2, MgO and CaO, were generally curved with a linear calibration giving severe underestimations at high concentrations. The PLS loadings and regression coefficients were then used to model the complete soil set to produce scores and concentration predictions for all the samples. The samples were plotted in bivariate score maps to give a visual representation of the spectral variability within the entire soil set. Samples were selected from the boundaries of the groups of soils in these maps for mineralogical characterization using X-ray diffraction (XRD) analysis. The XRD results confirmed the mineralogy obtained from the infrared spectra and PLS weight loadings. For this study, the depiction of the samples in the score maps was found to be of particular importance for demonstrating similarities in composition of the samples.
Comparing the effect of different sample conditions and spectral libraries on the prediction accuracy of soil properties from near- and mid-infrared spectra at the field-scale
