Errors in Forage-Quality Data Predicted by Near Infrared Reflectance Spectroscopy
1991
Buxton, D. R. | Mertens, D. R.
Near infrared reflectance spectroscopy (NIRS) is rapidly being accepted for routine prediction of forage quality and chemical constituents. Accurate predictions require calibration with samples analyzed by reference methods. This study was conducted to determine whether deviations between values predicted by NIRS and those determined by conventional techniques are random. In all, 108 samples from the basal 150 mm of immature and mature stems of replicated, field-grown cultivars of alfalfa (Medicago sativa L.), birdsfoot trefoil (Lotus corniculatus L.), red clover (Trifolium pratense L.), smooth bromegrass (Bromus inermis Leyss.), and orchardgrass (Dactylis glomerata L.) were studied. Residuals were obtained from calibration equations developed on all 108 samples. The overall accuracy with which 12 forage-quality and chemical traits were predicted by NIRS was similar to that reported in the literature. When residuals were tested by analyses of variance, significant differences were identified. Significant sources of variation were associated with field replicate, year, entry ✕ year interaction, and cultivars within species. Three traits with the most significant deviations were in vitro true digestibility, acid-detergent fiber, and cellulose, whereas the traits with the least significant deviations were neutral-detergent fiber, hemicellulose, glucose from the cell wall, and vanillin from nitrobenzene oxidation of core lignin. Unless NIRS results are checked for bias in subsets and verified by conventional analyses, erroneous conclusions can be drawn when NIRS is used to predict for management, physiology, and breeding studies.
显示更多 [+] 显示较少 [-]