Discrimination of soils managed with different sources of fertilization and plant species in organic and conventional farming through near-infrared spectroscopy and chemometrics

Abstract

Background: The combination of near-infrared (NIR) spectroscopy and chemometrics can be used to group or discriminate soils based on spectral response. In this study, we conducted discrimination and classification analyses on soils managed with different sources of fertilization and plant species grown in organic and conventional farming systems.

Results: Principal component analysis explained 96% (PC1) and 3% (PC2) of the data variability and separated the soil samples of organic and conventional management systems. The wavenumbers that contributed most to the separation of the management systems were in the range of 3600 and 7300 cm^-1 , especially the absorption peaks of 3700 and 4600 cm^-1 (characteristic of CH and NH combinations), and 5200 and 7000 cm^-1 (typical of OH combinations). Machine learning analysis using k-nearest neighbor and random forest algorithms was efficient in classifying soil samples according to management system with an accuracy of 97.8% and can therefore be used for future classification studies.

Conclusion: Based on the results, we strongly recommend the use of NIR spectroscopy associated with chemometrics for discriminating soils grown with Malus domestica, Musa spp., Oryza sativa and Solanum tuberosum L. under organic and conventional management systems through spectral response. © 2021 Society of Chemical Industry.

Keywords: Malus domestica; Multivariate analysis; Musa spp; Oryza sativa; R software; Solanum tuberosum L.