Discriminating between organic matter in soil from grass and forest by near-infrared spectroscopy

The absorbance of near-infrared (NIR) radiation by plants depends on the overtones of vibration, bending and stretching of the chemical bonds within their organic components. The wide variety of these bonds gives rise to characteristic NIR spectra, or 'fingerprints', of plant materials. Do these fingerprints remain in the soil when the plants decompose and allow discrimination between soils that have supported different types of vegetation? We have compared the NIR spectra of grassland soil with those of soil under forest in the Vosges mountains in France. Near-infrared reflectances, R, from 370 samples of soil were recorded as A = log(10)(1/R) at wavelengths between 1100 and 2500 nm and averaged over 10-nm intervals to give spectra each with 140 values. A canonical variate analysis (CVA) of the raw spectra discriminated well, but not perfectly, between grassland and forest soil, with Mahalanobis distance, D, of 4.87. Standardization along the spectra to remove effects of varied mineral composition achieved complete separation between the two sources with D = 9.81. Canonical variate analysis of first and second derivatives of the spectra distinguished the two groups even better, with D = 12.27 and D = 16.65, respectively. The results show much promise for inferring past vegetative cover from NIR spectra deriving from the organic matter in the soil. The next step will be to extend these studies to see how well we can distinguish other types of vegetation from such spectra.