Figure 5.

Box plot of predictive ability for mean dry matter yield from all seasonal harvests (DMY All Cuts) in TP1 using single-trait (ST-CV1) and multi-trait (MT-CV1, MT-CV2, and MT-CV3) genomic prediction models. In ST-CV1 model, DMY All Cuts was used to train and predict without any secondary trait in the model. In MT-CV1, DMY All Cuts and one of five correlated secondary traits (GS All, GS Spring, GS Late Spring, GS Summer and GS autumn) were used to train the model and predict DMY All Cuts in the test set. Under the MT-CV2 scheme, phenotypic data of a secondary trait was used in both the training and test set to predict DMY All Cuts. The MT-CV3 approach is similar to MT-CV2, except that phenotypes of the secondary trait were excluded from the training set, leaving secondary trait phenotypes in the test set only, for prediction of the primary trait.