Prediction accuracy in structured (A) and non-structured (B) traits using the core-collection as training population.

<p>Mean prediction accuracy was calculated on all 10 replicates of the simulation using four methods (cof, RR, BLR, cofRR). Models were built on the core-collection (Call) and applied to the four breeding sub-populations separately (dWW, dWE, dTE and Mixed, each composed of 200 individuals) and on the whole breeding meta-population (dall, 800 individuals). The two figures on the left side represent accuracies observed on structured traits and the other two figures accuracies on non-structured traits.