Soil organic carbon and its fractions estimated by visible-near infrared transfer functions

The capture and storage of soil organic carbon (OC) should improve the soil's quality and function and help to offset the emissions of greenhouse gases. However, to measure, model or monitor changes in OC caused by changes in land use, land management or climate, we need cheaper and more practical methods to measure it and its composition. Conventional methods are complex and prohibitively expensive. Spectroscopy in the visible and near infrared (vis-NIR) is a practical and affordable alternative. We used samples from Australia's Soil Carbon Research Program (SCaRP) to create a vis-NIR database with accompanying data on soil OC and its composition, expressed as the particulate, humic and resistant organic carbon fractions, POC, HOC and ROC, respectively. Using this database, we derived vis-NIR transfer functions with a decision-tree algorithm to predict the total soil OC and carbon fractions, which we modelled in units that describe their concentrations and stocks (or densities). Predictions of both carbon concentrations and stocks were reliable and unbiased with imprecision being the main contributor to the models' errors. We could predict the stocks because of the correlation between OC and bulk density. Generally, the uncertainty in the estimates of the carbon concentrations was smaller than, but not significantly different to, that of the stocks. Approximately half of the discriminating wavelengths were in the visible region, and those in the near infrared could be attributed to functional groups that occur in each of the different fractions. Visible-NIR spectroscopy with decision-tree modelling can fairly accurately, and with small to moderate uncertainty, predict soil OC, POC, HOC and ROC. The consistency between the decision tree's use of wavelengths that characterize absorptions due to the chemistry of soil OC and the different fractions provides confidence that the approach is feasible. Measurement in the vis-NIR range needs little sample preparation and so is rapid, practical and cheap. A further advantage is that the technique can be used directly in the field.