Fig. 3

Performance of single-trait and multi-trait genomic prediction for wheat yield. A 8 methods for predicting Grain Yields of 1,092 bread wheat lines. Genetic value prediction accuracy was estimated by cross-validation. Complete data (yield, marker genotypes, and 620 hyperspectral wavelength reflectances) was available for all lines, but 50% of the yield values were masked during model training. Genetic value prediction accuracy was estimated as \(\rho _{g} = \text {cor}_{g}(\hat {\mathbf {u}},\mathbf {y})\sqrt {h^{2}(\hat {\mathbf {u}})}\) because hyperspectral data and actual yields were collected on the same plots [24]. Bars show average estimates (± standard error) over 20 replicate cross-validation runs for each method. Details of each model are presented in Additional File 1: Supplemental Methods. Briefly, the three single-trait methods only used yield and genotype data. The five multi-trait methods additionally used hyperspectral data measured on all 1,092 lines. B Phenotypic correlation (black lines), and estimates of genetic correlation (red lines) between each hyperspectral wavelength measured on each of the 10 flight dates with final grain yield. Genetic correlations were estimated with the MegaLMM GBLUP method using complete data. Ribbons show the 95% highest posterior density (HPD) intervals