Effects of soil composition and preparation on the prediction of particle size distribution using mid-infrared spectroscopy and partial least-squares regression
Soil composition and preparation can affect prediction accuracy using diffuse reflectance mid-infrared Fourier transform spectroscopy (DRIFTS). In the present study, we evaluated the effect of soil composition, preparation and carbonate content on the accuracy of particle size distribution (PSD) predictions in four contrasting sets of soils, including calcareous soils, using partial least-squares regression (PLSR). The soils were scanned as <2- and <0.1-mm fine-ground samples. Regression calibrations were derived for individual soil sets, as well as a composite of the four sets. Predictions for clay and sand for the <2-mm composite calibration resulted in good accuracy (coefficient of determination R2=0.78; ratio of the standard deviation of reference values to the prediction error (RPD)=2.2), but were less accurate for clay in the calcareous soils (R2=0.70–0.78; RPD=1.8–1.1) and similarly accurate for sand (R2=0.68–0.80; RPD=1.7–2.2). Predictions for silt were poor. Accuracies improved by fine grinding (R2=0.88, RPD=2.9 for clay; R2=0.84, RPD=2.9 for sand). It was concluded that single, large and highly variable sets rather than site-specific calibrations could be used for the PSD predictions of specific soil sets. Changes in the PLSR loading weights, resulting from grinding, could be linked to an improved access of the infrared beam to the soil matrix by removal or dilution of surface coatings, resulting in a reduction of inter- and intraparticulate heterogeneity.