Error structures in biplot análisis of multienvironment trials

BALZARINI M; CASANOVES F; MACCHIAVELLI R

Montreal-Canada

Conferencia; International Biometric Conference; 2006

Multi-Environment  yield trials (MET) are crucial tools to recommend cultivars in plant breeding. Because of the genotype-by-environment interaction (GEI), METs are conducted in various environments using designs which involve several replications (plots) for each genotype at each environment.  Biplot analysis from additive main effects and multiplicative interaction (AMMI) models as well as site regression (SReg) models are commonly used for visualizing GEI. Frequently, biplots parameters are estimated under simplistic error structures.  Beside the model mean structure, mixed models have been suggested to taking into account within-trial spatial correlation among plots as well as between-trial residual variance heterogeneity. This work describes the analysis of  GEI in a serie of METs regarding yields of peanut-crops. For each database, we analyzed GEI by means of AMMI and SReg models with and without modeling spatial plot correlations and residual variance heterogeneity. The MET-specific best fit  and the corresponding biplot for exploring GEI is interpreted. The heterogeneity of residual variances across trials is dominant. The consideration of correlations and variance heterogeneity results in a better model fit in most cases and adjust  the interpretation of GE patterns.