The objective of this research was to examine three techniques for prediction of the chemical composition by NIR spectroscopy (1100 – 2500 nm) from semi-natural grassland: modified partial least squares (MPLS) regression; partial least square (PLS) regression and principal component regression (PCR). A spectral data for a total of 150 samples originated from seminatural grassland were used. Standard errors of calibration (SEC) for crude proteins (CP) were 6.52, 4.87 and 6.94 for MPLS, PLS and PCR, while standard errors of cross validation (SECV) were 8.16, 6.13 and 7.56 respectively. SEC for organic matter (OM) were 7.69, 7.61 and 7.37 for MPLS, PLS and PCR, while SECV were 8.08, 8.27 and 7.57 respectively. Higher SEC and SECV were reported for neutral detergent fibre (NDF) and acid detergent fibre (ADF) content than reported for CP and OM content. Hyperspectral analysis by PLS resulted in the highest accuracy for the estimation of crude protein, organic matter and neutral detergent fibre and acid detergent fibre, while MPLS was the best in predicting acid detergent fibre. The greatest accuracy in this research was achieved for CP, then NDF, OM, and finally ADF content. Prediction for NDF, OM, and especially ADF content should be improved in the future by involving specific semi-natural grassland samples.

seminatural grassland, chemical composition, NIR spectroscopy, PLS, MPLS, PCR

Published by University of Zagreb, Faculty of Agriculture, Zagreb, CROATIA
Sponsored by Ministry Science and Education of the Republic of Croatia