Oat kernel size uniformity

Abstract

Optimal conditions for dehulling oats with an impact dehuller are affected by kernel size. Oat mills frequently separate oats according to kernel size before dehulling to optimize milling yields. Therefore, oat kernel size uniformity is of concern to oat milling operations. We have evaluated oat kernel size distributions by sequential sieving with slotted sieves and with digital image analysis. The use of slotted sieves is faster, less expensive, and less technically challenging. However, digital image analysis provides more detailed information. Graphical analyses of oat kernel size distributions indicate that they do not resemble normal Gaussian distributions, but rather appear bimodal or multi-modal. A statistical model was developed using likelihood analysis to determine whether a mixture of two subpopulations better describes oat kernel size distributions. The analysis estimates means, variances and numerical proportions of each putative subpopulation and calculates a bimodal coefficient which increases with increasing bimodality. Analysis of ten oat cultivars grown in five locations over three years indicated that all samples were better described by a bimodal model than by a normal distribution. The source of the bimodality is likely to be the architecture of the oat spikelet. Most oat spikelets contain two kernels, where the primary kernel is significantly larger than the secondary. Thus, the larger kernel subpopulation appears to be derived from primary kernels of double kernel spikelets and the smaller kernel subpopulation appears derived from the secondary kernels. The presence of single kernel spikelets and triple kernel spikelets tend to make the distributions less distinctly bimodal. Because of the non-uniform nature of the oat spikelet, it appears unlikely that a totally uniformly sized oat can easily be developed. However, selection based on the methods described here potentially could lead to more uniform kernel size distributions. We have found that the application of bimodal analysis to digital image analysis data can reliably estimate the mean sizes of primary and secondary kernels from double kernel spikelets. Significant variation in the bimodal coefficient was also found among the ten cultivars tested thus far, indicating the existence of selectable genetic variation for this trait. Greater kernel size uniformity should therefore be achievable selecting for low size differences between the primary and secondary kernels.