Accounting for Spatial Variability in Breeding Trials: A Simulation Study

Several techniques to control for spatial heterogeneity in breeding trials were compared through the use of simulated data for a field site with 256 genotypes (i.e., treatments). Various experimental designs, error structures, and polynomial functions were modeled. The error structures studied included first-order autoregressive with and without measurement error (or nugget) and independent errors. Also, several nearest neighbor methods (Papadakis [PAP] and moving average [MA]) were used. The results indicated that, of models with independent errors, row-column designs gave the best correlation between the predicted and true treatment effects (CORR). Once the autoregressive error structure, with or without nugget, was incorporated, CORR values were even higher. Also, failing to incorporate the nugget produced bias in the correlation parameters of the error structure. Nearest neighbor technique were also among the best options, where some variants of the Papadakis method were almost as good as models that incorporated the error structure.