Phenomic prediction of club wheat milling yields

Abstract

Near‐infrared reflectance spectroscopy (NIRS) provides a nondestructive method for estimating physical and chemical grain properties and is widely used in breeding programs to phenotype traits such as texture, color, moisture, protein, and oil content. Club wheat ( Triticum aestivum subsp. compactum ), a specialty subclass of soft white wheat grown primarily in the Pacific Northwest, is valued for its high break flour yield and suitability for confectionery products. Current milling evaluations are destructive and require more than 200 g of grain per sample, limiting the ability to identify and cull poor‐performing lines in early generations. Using 2850 grain samples collected across five growing seasons, we evaluated the accuracy of NIRS models for predicting total flour yield and break flour yield in a set of club wheat breeding materials. Both traits were predicted with comparable accuracy using local partial least squares regression, with cross‐validation correlations of r  = 0.81–0.84. Incorporating samples from a growing season as additional training observations increased validation accuracy for individual years by 13%–16% based on Pearson's correlation coefficients. Spectral calibrations for soft wheat milling yields can effectively identify genotypes with poor milling characteristics, enabling nondestructive screening in early breeding stages and improve the efficiency of enhancing wheat end‐use quality.