Spatial Statistics Enhance Genotype Selection in Wheat Breeding Trials

In wheat breeding, the selection of superior genotypes in late stages for cultivar recommendation involves high-dimensional experimental fields. Traditional analyses using ordinary least square methods do not effectively account for the structured nature of this data, particularly in spatial trends. Mixed model-based approaches offer an alternative, managing data unbalancing and modeling residual effects. Thus, the aim of this study was to evaluate the suitability of spatial statistics for modeling the residual effects in single-trial analysis on a F7 wheat lines trial and to select ten top-performing genotypes for grain yield. The data were subjected to linear mixed model analysis for estimation of the genetic parameters through the restricted maximum likelihood approach and predicting the genotypic values via the best linear unbiased predictor. Four models were fitted, accounting for different residual effects. The first model (NSP) did not account for correlations among rows and columns, the second (SP1) accounted for spatial correlations among rows, the third (SP2) accounted for spatial correlations among columns and the fourth model (SP3) accounted for correlations among rows and columns. Model comparisons were made using the Akaike Information Criteria. The significance of the genotypic effect was tested via likelihood ratio test, and the broad sense heritability and the accuracy were also estimated for each model. The likelihood ratio test showed significant genotype effect (p<0.01) for all models considered, indicating variability between the genotypes for grain yield and a favorable environment for selection. Heritability and accuracy ranged from 0.65 to 0.71 and 0.79 to 0.84, respectively, with the best model accounting for correlations in both rows and columns. The AIC decreased progressively from 4219.95 (NSP) to 4195.8 (SP3), indicating that the best-fitted model is the one which accounted for correlations among rows and columns. An 80% coincidence index between the most and least accurate models revealed two genotypes that differed in selection. These findings highlight the importance of considering heterogeneity patterns in adjacent plots and demonstrate the increased accuracy in predicting genotypic effects when accounting for spatial dependencies. The ten top-performing genotypes selected through the SP3 model for grain yield comprised four commercial cultivars: ORS Guardião, ORS Senna, ORS Feroz, and BRS 264, as well as six lines developed by the UFV Wheat Breeding Program, each derived from distinct biparental crosses.