Soil differentiation using fingerprint Fourier transform infrared spectroscopy, chemometrics and genetic algorithm-based feature selection

Abstract

Soil is a complex environmental medium, particularly with respect to its biological and organic characteristics. Analytical approaches based on specific aspects of biological or organic variation may fail to detect broader differences in function or composition. Fourier transform infrared spectroscopy (FT-IR) spectra provide a more comprehensive description of soil characteristics in the form of complex multivariate data sets. Soils at different stages of recovery from degradation following opencast mining and from undisturbed land were investigated to evaluate the use of powerful chemometric approaches to differentiate their FT-IR spectra. FT-IR data sets were first subjected to principal component analysis (PCA), then discriminant function analysis (PC-DFA), with genetic algorithms (GAs) subsequently used to determine important discriminatory variables (wavenumbers) in the PC-DFA models. Whilst PCA of spectra was partially successful in grouping soils, PC-DFA discriminated undisturbed from reclaimed soils and differentiated between reclaimed soils of differing age. GA-DFA analyses identified wave numbers related to stretching of aromatic and aliphatic C–H bonds, and C–O or C–OH groups in glycopeptides or complex carbohydrates, as important discriminatory variables. GA-multiple linear regression (GA-MLR) analyses indicated that the recovery of disturbed soils may not be complete after 50 years. Chemometric analyses of FT-IR spectra offer a potentially efficient approach to screen for complex changes in soils and to indicate their nature. Additional work across a range of soils, comparing outputs from this approach with those from other widely used analyses, would further evaluate its effectiveness and aid in the interpretation of findings.