No abstract

Abstract:The potential of near-infrared reflectance spectroscopy (NIRS) for determining the chemical composition of heterogeneous, botanically complex semi-natural grassland herbage was assessed. Samples were collected over four consecutive years at different maturation stages and were analysed by chemical and NIRS procedures for crude protein, neutral detergent fibre (NDF), acid detergent fibre (ADF), acid detergent lignin and cellulose. A subset of samples was selected on the basis of spectral features in order to establish calibration equations, while the other samples were used for validation. The wavelengths selected by using multiple regression were similar to others previously reported. The ranges of correlation coefficients and standard errors of prediction, respectively, for the various components when NIRS and chemical procedures were

No abstract

: Near infra-red reýectance spectroscopy (NIRS) was used to develop calibration equations to measure nitrogen and ash contents in grassland samples. Four populations of samples were collected or prepared : total herbage, with a heterogeneous and complex botanical composition, and its botanical components (grasses, legumes and forbs). A set of samples from each population was selected to develop the speciüc calibration equations using three mathematical data treatments (log 1/R, ürst and second derivative). Six and seven wavelengths were selected by multiple regression to predict nitrogen and ash contents, respectively. Calibration equations were evaluated by comparing the values obtained by reference methods (Kjeldahl test and dry ashing measurements) with those predicted by NIRS. The three data treatments generally provided similar results as regards estimations of nitrogen contents. The correlation coefficients varied from 0.94 to 0.98, and the standard errors of prediction ranged from 1.10 to 1.49 g kg-1 (total herbage), from 0.73 to 0.98 g kg-1 (grasses), from 0.99 to 1.30 g kg-1 (legumes) and from 0.76 to 0.80 g kg-1 (forbs). The most suitable treatments to predict ash contents were log 1/R and the ürst derivative. The best performance was obtained for legumes, using log 1/R, with a correlation coefficient of 0.95 and a standard error of prediction of 3.54 g kg-1. The calibration equations became more accurate as the components of the sets of samples became botanically simpler. Prediction accuracy was greater when the speciüc calibration equations for each population of samples were used.

No abstract