Fig. 1
The different scenarios of calibration and validation of the GP models used to predict the phenotype. Among the fertile plants extracted from the PCT27 population, 384 were used to train the model (PCT27A), while another set of 334 (PCT27B) was considered for validation of the model. The different scenarios of calibration and validation of the GP models used to predict the phenotype of the PCT27B at the S0:4 generation in Santa Rosa (SRO). The red area represents the validation set (VS), the green and blue represent the training set (TS), from Santa Rosa and Palmira (PAL), respectively. The percentage in the colored areas represents the fraction of the population used to calibrate or validate the model. The x% of S0:3 families phenotyped in SRO included in the TS in Multi2 scenario varied from 25, 50 and 75%. Scenarios can be summarized as follow: Uni1: cross-validation to estimate the predictive ability of a model calibrated with the information of PCT27B in a single location (SRO); Uni2 and Uni3: families from PCT27A at generation S0:2 or S0:3, respectively, were used as a TS to estimate the genomic breeding values of all the families of PCT27B at generation S0:4. Only one environment (SRO) was included in these scenarios. Multi1: data from two locations (PAL and SRO) from a single generation (S0:4) were used, TS was composed of 100% and 70% of the PCT27B families phenotyped at PAL and SRO, respectively, and the VS was composed of the remaining 30% of the PCT27B families phenotyped in SRO. Multi2: TS consisted of 100% of the families phenotyped in PAL at the S0:2 generation and 25, 50 or 75% of the families measured at SRO at the S0:3 generation