Removing the Effects of Iron Oxides from Vis‐NIR Spectra for Soil Organic Matter Prediction

Abstract

Core Ideas Iron oxides decreased the prediction accuracy of soil organic matter (SOM). External parameter orthogonalization (EPO) was used initially to alleviate iron oxides effects. Prediction accuracy of SOM was greatly improved from original models. Recently, visible and near‐infrared (vis‐NIR) spectroscopy has been widely used to predict soil organic matter (SOM) contents because it is highly efficient and inexpensive. However, iron oxides interfere with the spectral characteristics of SOM, thereby decreasing the SOM prediction accuracy. In this study, the spectra of 267 soil samples from South China were used to investigate the effects of iron oxides on spectral reflectance, and external parameter orthogonalization (EPO) was implemented to remove the effects of the iron oxides. To examine the effectiveness of EPO, two commonly used modeling methods—partial least squares regression (PLSR) and least‐squares support vector machines regression (LS‐SVMR)—were applied to establish prediction models using spectra with and without EPO correction. The EPO algorithm effectively reduced the effects of iron oxides. Compared with the results calculated from the original spectra, the prediction accuracies derived from the EPO‐processed spectra using the two modeling methods were greatly improved. Additionally, the LS‐SVMR method performed better than the PLSR method, with a coefficient of determination of the prediction (R2p) of 0.91, a root mean square error of the prediction (RMSEp) of 4.31 g kg–1, a ratio of the standard deviation to the root mean square error of the prediction (RPD) of 3.31 and a ratio of performance to interquartile range (RPIQ) of 4.45. The results demonstrated that the EPO method improved the accuracy of SOM prediction in soils rich in iron oxides. This method has the potential to improve the accuracy of SOM prediction in other soils rich in iron oxides worldwide.