Organic Carbon and Total Iron Effect on Soil Vis-SWNIR Spectraand Quantification of Their Contents Using PLS R Models

Abstract

ABSTRACT Many factors could influence simultaneously soil spectra. We aimed to study the single effect of organic carbon and total iron in soil visible and short-wave near-infrared spectra and to quantify their contents. Two datasets of soil mixture samples were prepared by mixing, in various fractions, an organic carbon-rich material with a total iron-rich material and then with a total iron-poor material. For these two datasets, contents in organic carbon are quite similar but contents in total iron are significantly different. Results show that samples of the same dataset have the same overall spectral shape. Organic carbon has a decreasing effect that affects the whole spectral range without showing any specific absorption peaks. By contrast, total iron has specific absorption peaks. Spectra of the second dataset characterized by soil mixtures with higher total iron contents were more compact within the spectral bands 400–440 and 920–950 nm. Besides, continuum removal enables to exaggerate absorption peaks of wavelengths linked to total iron content. Partial Least Squares Regression (PLS R) models of both total organic carbon and total iron assign high coefficients to the wavelengths that are considered relevant and conversely low coefficients to those that are considered irrelevant. Both organic carbon content and total iron content were well predicted. For these models, coefficients of determination were superior to 0.9 and RMSE was closed to zero. The global models calibrated on all the samples demonstrated that PLS R was able to integrate sample heterogeneity.