The application of statistical preprocessing on spectral data does not always guarantee the improvement of the predictive quality of multivariate models: Case of soil spectroscopy applied to Moroccan soils

Statistical preprocessing methods are widely used to improve the predictive performance of chemometric models especially on MIR and NIR spectral database. The main role of pretreatments in the case of soil spectroscopy is to remove/reduce the scattering impact, thus, highlight the part of the signal related to the chemical, physical or biological properties of interest. Nevertheless, given the complexity of the soil as a matrix, together the pure absorption data and the information hidden in the scattering are of key interest as they cooperatively describe the physicochemical state of the soil samples. Consequently, pretreatment methods that removes/ reduces the scattering material in the spectroscopic data may lower the predictive quality of the multivariate models. The purpose of this study was to explore the effect of preprocessing methods on FTIR spectra of soil samples and test the hypothesis that the use of scatter correction techniques as pretreatment (Viz. standard normal variate, Savitzky-Golay 1st and 2nd derivatives, and the multiplicative scatter correction) does not guarantee the improvement of the predictive performance of partial least squares regression models for the prediction of total carbon, organic carbon and total nitrogen in soil samples. The obtained results showed that among all the multivariate calibrations, the PLS models set-up on the unprocessed spectral data led to similar/better predictive qualities for the estimation of selected soil properties especially for total carbon with an R2 of 0.92 and RMSECV of 0.129, either because the information contained in the scattering background is important for the predictions or there is no scatter in the spectral data. Hence, FTIR spectroscopy as a simple, fast and nondestructive analytical method that in many cases will not require a supplementary effort when performing chemometric modeling by avoiding the preprocessing step could be highly recommended for soil health indicators prediction.