Collaborative evaluation of near infrared reflectance analysis for the determination of protein, moisture and hardness in wheat

Publisher Website
Google Scholar
Abstract
Different methods for the calibration of near infrared reflectance (n.i.r.) instruments for determination of protein, moisture and kernel hardness in wheat by using 11 instruments of the same model series, were collaboratively assessed. The use of the same calibration constants (other than the bias or intercept constant which must be set separately for each instrument) resulted in significantly better reproducibility in the determination of protein in ground wheat than the use of constants determined separately for each instrument. In the case of accuracy against the reference Kjeldahl method the results using universal constants were not significantly different compared with those using constants individually determined by each laboratory for its own instrument. The use of different grinders of the same type resulted in between‐grinder biases in some cases but not in worse reproducibility or accuracy than using one grinder. Therefore, using a common calibration and grinder brings the different laboratories' n.i.r. protein results closer to each other but not necessarily closer to Kjeldahl results. In any case, the agreement of the protein results from one instrument to another or between n.i.r. and Kjeldahl should be within about ±0.50% if the instruments are correctly adjusted. Wheat moisture was measured using unground samples and the same constants (other than the bias) for each instrument when the reproducibility was 0.20% and average accuracy against oven drying was 0.39%. Individual calibrations were not attempted. Wheat kernel hardness, classified as ‘hard’ or ‘soft’, was assessed by eight laboratories on 20 samples using different models of grinder and different constants. Six laboratories achieved a near perfect classification with only three errors in all, but the other two had seven errors between them.