Spectroscopic based partial least-squares models to estimate soil features

Spectroscopic data of Dystrudepts and Haplodux soils with variable selection are scarcely found but they are currently investigated as alternative tools to minimize the deleterious impacts of classical analytical techniques on environment. In this study, Fourier transform mid-infrared spectroscopy, energy dispersive X-ray fluorescence and photoacoustic spectroscopy were applied to the analysis of these soils and data were used to build models by interval partial least-squares regression and partial least-squares discriminant analysis to estimate soil properties (type, geographic origin, management and sand, silt and clay contents). The combination of spectroscopy techniques and statistical tools allows the classification of soil by region and all samples (whole soil and their fractions, topsoil, and subsoil) were appropriate to this aim. The most accurate technique for the distinction of soil types and classification of fractions by content was energy dispersive X-ray fluorescence-based interval partial least-squares-discriminant analysis model for whole soil and fractions, the model exhibited above 97.5% (for validation) of correct classification of samples. On the other hand, to differentiate soil management, Fourier transform mid-infrared spectroscopy-based interval partial least-squares-discriminant (iPLS-DA) analysis model using whole soil presented the best results, with 100% of correct classification samples. The soil management iPLS-DA models of energy dispersive X-ray fluorescence and photoacoustic spectroscopy did not have adequate prediction capability and are not useful to differentiate soil management. The most representative discriminant variables were discussed. Energy dispersive X-ray fluorescence (silt and sand content, RPDCV > 2.0), Fourier transform mid-infrared spectroscopy (clay and sand content, RPDCV > 2.0) and photoacoustic spectroscopy (RPDCV > 1.4) combined with interval partial least-squares-regression and cross validation can be a useful approach to estimate the texture in Dystrudepts and Haplodux soils, and the one of the three textual classification calculated for difference.

Automated summary

This study applied spectroscopic techniques and statistical tools to analyze Dystrudepts and Haplodux soils, and built models for soil classification and estimation of soil properties. The results showed that energy dispersive X-ray fluorescence and Fourier transform mid-infrared spectroscopy were the most accurate techniques for soil classification, while photoacoustic spectroscopy was suitable for estimating soil texture.