Additive Main Effect and Multiplicative Interaction Analysis of National Turfgrass Performance Trials: II. Cultivar Recommendations

A parsimonious additive main effect and multiplicative interaction (AMMI) model has often been shown to be more accurate than the cell means from which it is computed in predicting future performance, especially where complex genotype × environment (GE) interactions exist, as is typical of National Turfgrass Evaluation Program (NTEP) variety trials. NTEP relies on ANOVA procedures for estimates (means averaged over replicates) to predict turfgrass quality performance. The objectives of this research were (i) to compare the predictive accuracy and statistical efficiency of AMMI models relative to the cell means model (means averaged over replicates or AMMI‐F full model) and (ii) using mega‐environment analysis of AMMI, identify subregions within the cool‐season turfgrass growing region having similar GE interaction patterns and cultivar recommendations. The 1990 Kentucky bluegrass (Poa pratensis L.) and perennial ryegrass (Lolium perenne L.) variety trials were analyzed. The GE interactions in Kentucky bluegrass (KBG) and perennial ryegrass (PRG) data sets contain 33.3 and 40.6% noise, respectively. Noise complicates cultivar recommendations and AMMI was shown to recover noise selectively in a discarded residual. Based on validation studies, AMMI reduced models (AMMI‐2 for PRG and AMMI‐7 for KBG) were shown to be more predictively accurate with a statistical efficiency relative to AMMI‐F of 2.05 (KBG) and 5.6 (PRG). Mega‐environment analysis identified several subregions, each representing several NTEP locations. Mega‐environments follow a cultural intensity gradient that can be altered by mowing height and nitrogen fertilization. Locations from the same subregion had better predictive value for other locations from the same subregion. This may allow the same genotypes to be targeted to all locations from the same subregion, simplifying cultivar recommendations. The data suggest that if the goal is to recommend the most reliable genotype, then priority should be given to AMMI adjusted means.