Prediction of soil organic and inorganic carbon at different moisture contents with dry ground VNIR: a comparative study of different approaches

Visible and near infrared reflectance spectroscopy (VNIR) is a rapid and multi-sensing technology that can provide spatially dense soil data for many disciplines. Soil moisture is the main barrier to the use of this technology in the field. The objective of this study was to compare the effectiveness of five approaches, external parameter orthogonalization (EPO), direct standardization (DS), global moisture modelling (GMM), slope bias correction (SB) and selective wavelength modelling (SWM), to enable models based on dry ground VNIR spectra of soil to be applied directly to moist spectra to predict organic carbon (OC) and inorganic carbon (IC). A total of 352 archived soil samples were selected and scanned, and of these 185 samples were used to develop dry ground models. The remaining 167 samples were rewetted to eight different moisture contents for the development (100) and validation (67) of the five moisture-correction approaches. The results showed that EPO, DS and GMM account satisfactorily for the effect of moisture in soil spectra. They improved the prediction of OC substantially with an increase in R-2 from almost zero for no correction to over 0.5, and an RPIQ (ratio of performance to interquartile range) from 0.38 to over 1.7. Improvement was also achieved with EPO, DS and GMM for IC; RPIQ increased from 0.53 to over 1.2. External parameter orthogonalization and DS showed that their effectiveness in moisture correction depended on the moisture content of the sample (larger moisture contents decrease the effectiveness), whereas this dependence was not observed for the moisture-explicit DS and GMM methods. We concluded that EPO, DS and GMM are all viable approaches for moisture correction for VNIR-based proximal soil sensors.