Prediction of soil organic carbon for different levels of soil moisture using Vis-NIR spectroscopy

Visible and near infrared diffuse reflectance spectroscopy has produced promising results to infer soil organic carbon (SOC) content in the laboratory. However, using soil spectra measured directly in the field or with airborne imaging spectrometers remains challenging due to uncontrolled variations in surface soil conditions, like vegetation cover, soil moisture and roughness. In particular, soil moisture may dramatically degrade predictions of SOC content when using an empirical approach. This study aims to quantify the effect of soil moisture on the accuracy of SOC predictions, and propose a method to determine SOC content for moist samples with unknown moisture content. More than 100 soil samples were collected along a transect, in the Grand-Duchy of Luxembourg. The soil samples were air-dried, moistened in steps of 0.05g water g soil−1 until saturation, and scanned in the laboratory with a visible and near infrared diffuse reflectance spectrometer. We computed the normalized soil moisture index (NSMI) to estimate the soil moisture content of the samples (R2=0.74), and used it to spectrally classify the samples according to their moisture content. SOC content was predicted using separate partial least square regressions developed on groups of samples with similar NSMI values. The root mean square error of prediction (RMSE) after validation was always below 5g C kg−1, with a ratio of prediction to deviation (RPD) greater than 2. The SOC content prediction models with a-priori knowledge of soil moisture gave similar RMSE as the ones after the NSMI classification. Hence, the NSMI might be used as a proxy of moisture content to improve SOC content prediction for spectral data acquired outside the laboratory since the method is simple and does not need other data than a simple band ratio of the spectra.