PREDICTION OF BOTANICAL COMPOSITION IN GRASSLAND HERBAGE SAMPLES BY NEAR-INFRARED REFLECTANCE SPECTROSCOPY

Near infrared reflectance spectroscopy (NIRS) was evaluated as a methodto predict the botanical composition of seminatural grassland in`dehesa' systems. Samples of herbaceous biomass were harvested overfour consecutive years, determining in each-by manual separation-theproportion by weight of the following taxonomic groups: grasses,legumes and the rest of the families in a single block ('others'). After reconstructing the natural samples they were analysed by NIRS. One set of samples (calibration set) was selected for the development ofthe equations, assaying different mathematical treatments (log 1/R,first derivative and second derivative). The ranges of coefficients ofmultiple determination and standard errors of calibration, respectively,for the various components were: grasses, 0.86 to 0.92 and 6.66 to9.14; legumes, 0.77 to 0.81 and 6.82 to 7.43; and `others', 0.85 to 0.88and 8.17 to 9.54. The remaining samples not included in the developmentof the NIRS equations (prediction set) were used for the purposes ofvalidating the best equations. Standard errors of performance were: grasses, 6.12; legumes, 7.56 and `others', 7.70.