Quantification of soil carbon from bulk soil samples to predict the aggregate-carbon fractions within using near- and mid-infrared spectroscopic techniques

There is a persistent general concern with carbon sequestration and modeling of soil carbon change affecting global issues, such as climate change and food security. To address these concerns requires the measurement of carbon everywhere and routinely, but the rate limiting step is the need to physically fraction the soil carbon to establish; where it is stored in soil, to model the formation of soil aggregates that physically protect soil carbon, and in-turn to populate soil carbon models. To remove the need for this fractionation pretreatment, commonly done by wet-sieving, this study scopes the notion of the efficacy of using near- (NIR) and mid- (MIR) infrared derived spectra taken of bulk soil samples to predict carbon in the separated aggregate fractions contained within. Forty five surface soil samples were collected from three bioregions of New South Wales providing for a range of soil types and associated soil carbon. The carbon content was measured of the bulk soil samples and their aggregate fractions of <63 mu m, 63-250 mu m, and >250 mu m subsequently separated by wet-sieving. The bulk soil samples were scanned in the spectral ranges 800-2500 nm (NIR region) and 2500-25,000 nm (MIR region). The Cubist regression tree model was used to predict the carbon content in the aggregate fractions scanned from the bulk soil samples. The cross-validation results reveal that the MIR demonstrated the strongest correlation between measured and predicted carbon of the aggregate fractions demonstrated by high R-2 (0.63-0.85) and ration of performance to inter-quintile distance (RPIQ 0.53-0.93). The wavelengths selected in the Cubist model coincide with wavelengths identified as characterizing adsorption due to chemistry of soil carbon in some recently published works in this area of research. (c) 2016 Elsevier B.V. All rights reserved.