Analysis of factors that influence the variation in predictive ability of single and multi-environment prediction models.

The effects of the statistical model (GBLUP, RKHS-1 and RKHS-2), the trait (DTF, GY and PH), the cross-validation strategy (CV1 and CV2), drought stress environments (E1, E2 and E3) and different combinations of training environment and their interactions were evaluated.