STUDYING GXE INTERACTION UNDER DIFFERENT MANAGEMENT SYSTEM AND YIELD LEVEL USING LINEAR-BILINEAR MODELS: THE CASE OF CIMMYT INTERMEDIATE TO LATE HYBRIDS TRAILS IN EASTERN AFRICA

Abstract

CIMMYT, with national programs, conducts selection of stress-tolerant genotypes under managed stress conditions; this investigation is expected to add information to the existing knowledge. Data sets used in this study comes from Intermediate to Late Hybrid Trails (ILHT) conducted in five Eastern and Central Africa (ECA) countries from 2008 to 2011. Trials, ranging from 18 in 2009 to 29 in 2010 were used. Trials are categorized into four management systems and two yield levels. Variance Components, broad sense heritability (H), Site Regression (SREG), Genotypic Regression (GREG), Completely Multiplicative Model (COMM) and Factor Analytic (FA) models were fitted. Results are discussed and compared with those stated in literature. We argue that it is preferable to first fit the fixed effect models before proceeding to the mixed effect model, as the former shows the level of complexity of the GE component and number of Axis required to explain it. The fixed effect model, SREG2, is preferable for trails targeting to compare hybrids with checks. From the GGE biplots it was noted that the first two PC did not account for sufficient percentage of variation for all years which witnessed complexity in the GE component for this data. Nevertheless, since PC1 accounted for large percentage of variation than PC2, the plot still gives some idea of which hybrids are favored and where. Most importantly, location of genotypes along PC1 can serve for judging yielding potential of the genotypes to guide in selection decision. Equivalence between Finlay – Wilkinson and GREG was established. The few environmental covariables obtained for 2009 was used to fit Partial Least Square (PLS) regression. The result indicated complexity in the GE component, as PLS latent factors accounted for small percentage of variation. It was recommended to use information from SREG2, GREG2 and FA(1) models in order to identify stable genotype.