Near-infrared reflectance spectroscopy (NIRS) for the mandatory labelling of compound feedingstuffs: chemical composition and open-declaration

Abstract

Near-infrared (NIR) calibrations were developed for the instantaneous and simultaneous prediction of the chemical and ingredient composition of compound feedingstuffs for different types of animals. Different sample presentation modes were also compared (ground versus unground) in order to demonstrate that, although the ground presentation is common in NIR analysis, the analysis of unground samples can also be viable, obviating the need for tedious milling. Calibration equations for the prediction of chemical composition in compound feedingstuffs showed similar accuracy for the different modes of analysis assayed. The equations obtained for unground compound feedingstuffs ( N = 433) are reliable enough for quality control at the feed mill. The coefficient of determination ( R 2) and the standard error of cross-validation (SECV) values ranged from good to very good, depending on the parameter: moisture (0.87 and 4.9 g kg −1), crude protein (0.97 and 5.5 g kg −1), crude fat (0.94 and 6.1 g kg −1), crude fibre (0.98 and 5.2 g kg −1) and ash (0.85 and 6.9 g kg −1). Equations for predicting ingredient composition (g kg −1) also showed similar accuracy for the analysis of ground and unground compound feedingstuffs. The equations showed an excellent ability ( R 2 ≥ 0.9; RPD ≥ 3) to predict the amount of sunflower meal, gluten feed, lucerne, beet pulp, palm meal, meat and bone meal (MBM), poultry meal, total meat meal (MBM + poultry meal), animal fat and mineral–vitamin supplement. For other ingredients, including corn, barley, lupin, manioc, soybean meal, wheat bran, molasses, calcium carbonate, dicalcium phosphate and methionine, the precision and accuracy of the equations obtained could be considered good ( R 2 ≥ 0.7). Calibrations for wheat, fish meal, lysine and salt displayed limited predictive ability (0.5 ≤ R 2 ≤ 0.7). Moreover, a partial least squares (PLS) discriminant model was developed, which classified correctly all samples as free or as containing MBM.