Prediction of genotype performance for untested years based on additive main effects and multiplicative interaction and linear mixed models: An illustration using dolichos bean (Lablab purpureus (L.) Sweet) multiyear data

Abstract

AbstractCarrying out multi‐environment trials (MET) is a regular and mandatory procedure for identifying and recommending superior genotypes as cultivars of crops with no exception of dolichos bean. The accuracy of a crop MET can be increased using more efficient statistical tools such as Additive Main effects and Multiplicative Interaction (AMMI) and mixed linear models via best linear unbiased prediction (BLUP) procedure. AMMI is not a single model, but rather, a family of models. Considering genotypes, environments or both as random variables, three types of BLUPs, namely BLUPg, BLUPe and BLUPge, respectively are possible. Diagnosis and use of the best AMMI model family member and type of BLUP is the key to identify the best genotype(s) for use as cultivars with a hypothesis that they will perform well in farmers' fields in future years. We diagnosed the best AMMI model family member and type of BLUP based on between‐year predictive accuracy using a 5‐year dataset in dolichos bean. Replication‐wise mean fresh pod yield of different combinations of 4‐years' was used as prediction datasets to build AMMI and BLUP models. The observed mean fresh pod yield of genotypes evaluated in the year, which is not used in modelling, was used as a validation dataset. Predictive accuracy was measured as root mean squared differences between AMMI and BLUP model‐predicted and observed mean fresh pod yield of genotypes. Our results showed that parsimonious AMMI1 model was far better than any type of BLUP in predicting the genotype performance for untested years.