Possibilities of NIRS for the prediction of organic carbon in soils

For the determination of organic carbon in soils, the availability of a fast, low-cost analysis method is required. The aim of this study was to evaluate the possibilities of near infrared reflectance spectroscopy (NIRS) to build a spectral database and to develop calibrations for the prediction of the organic carbon content in grassland soils. NIRS spectra of 1485 soil samples from different origins were collected between 1100 and 2500 nm. NIRS calibrations were developed with modified partial least square regression and tested with independent validation samples. The best equations were obtained with the first derivative of the spectra without scatter corrections. For the global calibration containing the samples of all origins the standard errors of calibration (SEC) and of prediction (SEP) were respectively 2.60 (r(2) = 0.91) and 2.77 g kg(-1) (r(2) = 0.89). The ratio of the standard deviation of the reference validation data to the SEP (SD/SEP), indicating the performance of the calibration, was 3.0. Dividing the samples in groups according to their origin improved the SEP on average with 18%.