Fig. 4.

Accuracies of the predictions of the TS genotypes in the rice germplasm. The calibration sets were defined by maximizing CDmean; minimizing PEVmean; maximizing CDmean within cluster; stratified proportional sampling and random sampling. Four different population sizes (25, 50, 100, 150 and 175) were used for the optimization algorithm in four different traits (a florets per panicle, b flowering time, c plant height, d protein content). Standard error is indicated for each point over the 50 runs. Optimization of CDmean, PEVmean and StratCDmean was made with the heritability measured for each trait in each germplasm (color figure online)