A non‐destructive method avoiding the utilization of toxic and corrosive reagents is presented as an alternative for the determination of organic matter (OM) in soils. This method is based on a multivariate calibration procedure using partial least squares (PLS) regression to establish the relationship between the organic matter content in soils determined by conventional chemical measurements and by diffuse reflectance spectra in the mid‐infrared region. The spectra are presented as reflectance (R) or log(1/R) and in Kubelka–Munk (K/S) units. Several data pretreatments such as multiplicative scatter correction (MSC), smoothing, derivation and normalization of the spectral data were employed to improve the performance of the method. The PLS analysis on the data expressed as R and log(1/R) after smoothing, differentiation and normalization showed better results, with RMSEPs equal to 0·63% (for R) and 0·69% (for log(1/R)) and linear correlation coefficients between reference and predicted OM values equal to 0·981 (for R) and 0·972 (for log(1/R)). Copyright © 1999 John Wiley & Sons, Ltd.

SUMMARYA non-destructive method avoiding the utilization of toxic and corrosive reagents is presented as an alternative for the determination of organic matter (OM) in soils. This method is based on a multivariate calibration procedure using partial least squares (PLS) regression to establish the relationship between the organic matter content in soils determined by conventional chemical measurements and by diffuse reflectance spectra in the mid-infrared region. The spectra are presented as reflectance (R) or log(1/R) and in Kubelka-Munk (K/S) units. Several data pretreatments such as multiplicative scatter correction (MSC), smoothing, derivation and normalization of the spectral data were employed to improve the performance of the method. The PLS analysis on the data expressed as R and log(1/R) after smoothing, differentiation and normalization showed better results, with RMSEPs equal to 0⋅63% (for R) and 0⋅69% (for log(1/R)) and linear correlation coefficients between reference and predicted OM values equal to 0⋅981 (for R) and 0⋅972 (for log(1/R)).

No abstract

No abstract