Submitted to: Applied Engineering in Agriculture
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 8/30/2006
Publication Date: 11/1/2006
Citation: Bramble, T., Dowell, F.E., Herrman, T.J. 2006. Single-kernel near-infrared protein prediction and the role of kernel weight in hard red winter wheat. Applied Engineering in Agriculture. Vol. 22(6):945-949.
Interpretive Summary: Protein content is an important indicator of bread quality, with loaf volume generally increasing as protein content increases. However, protein content varies within wheat fields, plots, rows, plants, heads, spikelets, and position within spikelets. Measuring protein content in bulk samples will not give an accurate indication of the protein content variation since protein content varies from kernel to kernel. Thus, technology and calibrations are needed to automatically measure single kernel protein content. This technology and information can then be used in automated sorting instrumentation to help breeders develop varieties with more uniform kernel-to-kernel protein content. This research documented the kernel variability that occurs in the growing environment, and resulted in protein calibrations that can be used in automated sorting instrumentation. This research also documented the relationship between kernel weight and protein content, and revealed the importance of including kernel weight in prediction models. These results can be used by breeders as they seek to develop cultivars with specific, uniform quality traits for specific end-use markets, and to more effectively compete in world markets.
Technical Abstract: A near-infrared single-kernel protein calibration for hard red winter wheat (Triticum aestivum L.) was developed to support research mapping the variance structure of single kernel protein in commercial wheat fields. The hierarchical sampling design used to map the variance structure included fields, plots, rows, plants, heads, spikelet, and kernels from 47 fields containing the cultivars Jagger, 2137, Ike, or TAM 107. Each kernel was evaluated for protein content using an automated single kernel NIR system. Five hundred kernels were selected as the model development set and reference protein content was determined using a combustion nitrogen analyzer. The resulting 11 factor PLS model had a standard error of prediction based on a cross validation (SECV) of 1.21% and r2 = 0.84. Application of a kernel weight correction improved model performance statistics (SECV 0.40%, r2 = 0.89). A moderate negative correlation was observed (r = -0.55) between kernel weight and protein content. Previous research exploring single kernel protein had not documented this relationship. The PLS model containing a kernel weight adjustment was most accurate with Jagger kernels (SECV 0.32%, and r2 = 0.92) and least accurate with TAM 107 kernels (SECV 0.51%, r2 = 0.82). The application of the weight correction factor resulted in a lower SECV than previous research. Currently, single kernel protein analysis instruments have not included a kernel weight apparatus, which represents a constraint in accurately predicting single kernel protein using NIR technology.