Classification of soil samples based on Raman spectroscopy and X-ray fluorescence spectrometry combined with chemometric methods and variable selection

Abstract

Soil classification is crucial for its cultivation preparation in countries that export several agricultural commodities. The soil classification system adopted in Brazil is based on chemical parameters and physical and morphological changes. This system possesses disadvantages because many analyses are time-consuming, especially during the sample preparation stage. Raman spectroscopy is a non-destructive technique that enables rapid soil sample characterisation. In this study, Raman spectroscopy was used to discriminate different soil samples. Although the Raman spectra of a substance can be used as a phase fingerprint due to its specificity, this technique is not adequate for sample discrimination and suffers from matrix interferences, especially during the analyses of soil samples. However, a synergic effect with satisfactory results regarding prediction and classification problems occurs when this method is coupled with chemometric tools. In this research, a robust classification method for analysing soil samples using Raman spectroscopy combined with a support vector machines (SVM-C) method and genetic algorithm (GA) for variable selection was developed. The results obtained from the combination of the proposed GA–SVM-C based on the figures of merit were sensitivity (1.000), specificity (1.000), and misclassification error (0.0%) in the validation step. This soil discrimination methodology was validated using X-ray fluorescence spectrometry. These tools can be used in routine analyses, reducing laboratory costs with good efficiency.