Figure 4

Cross-validated accuracies of prediction for genomic selection with regard to location and year. Cross-validation was performed using RR-BLUP across genotype within location [CV_{G (L)}] or across genotype and location (CV_G×L), and across genotype within year [CV_{G (Y)}] or across genotype and year (CV_G×Y). ES in the scheme on top refers to the estimation set, TS represents the test set. In each case, the dataset was divided into five subsets (S1-S5), of which S1 to S4 were assigned to the ES and S5 to the TS. The mean accuracy of prediction r_p values for genomic selection based on data from population A (Pop-A) and population B (Pop-B) for traits grain yield and plant height are shown below. The mean r_p across genotype based on the eight location-year combinations for which data from two years could be included (BEK10, BEK11, PET10, PET11, WAL10, WAL11, WOH10, WOH11; excluding HOH10) serves as reference (CV_G; white). The mean r_p across genotype within location [CV_{G (L)}] or across genotype and location (CV_G×L) was determined by selecting data from two locations in the years 2010 and 2011 (e.g. BEK10, BEK11, PET10, PET11) for estimation, the mean r_p across genotype within year [CV_{G (Y)}], or across genotype and year (CV_G×Y) from estimation sets based on data from four locations in one year (e.g. BEK10, PET10, WAL10, WOH10). Test sets were derived either from the same location-year combinations as the estimation sets [CV_{G (L)}; light grey; CV_{G (Y)}; dark grey] or from the location-year combinations not used for estimation (CV_G×L; intermediate grey; CV_G×Y; black).