Determining environmental covariates which explain genotype environment interaction in winter wheat through probe genotypes and biadditive factorial regression

Genotype-environment interaction has been analyzed in a winter-wheat breeding network using biadditive factorial regression models. This family of models generalizes both factorial regression and biadditive (Or AMMI) models; it fits especially well when abundant external information is available on genotypes and/or environments. Our approach, focused on environmental characterization, was performed with two kinds of covariates: (1) deviations of yield components measured on four probe genotypes and (2) usual indicators of yield-limiting factors. The first step was based on the analysis of a crop diagnosis on four probe genotypes. Difference of kernel number to a threshold number (DKN) and reduction of thousand-kernel weight from a potential value (RTKW) were used to characterize the grain-number formation and the grain-filling periods, respectively. Grain yield was analyzed according to a biadditive factorial regression model using eight environmental covariates (DKN and RTKW measured on each of four probe genotypes). In the second step, the usual indicators of yield-limiting factors were too numerous for the analysis of grain yield. Thus a selection of a subset of environmental covariates was performed on the analysis of DKN and RTKW for the four probe genotypes. Biadditive factorial regression models involved environmental covariates related to each deviation and included environmental main effect, sum of water deficits, an indicator of nitrogen stress, sum of daily radiation, high temperature, pressure of powdery mildew and lodging. The correlations of each environmental covariate to the synthetic variates helped to discard those poorly involved in interaction (with \correlation\ <0.3). The grain yield of 12 genotypes was interpreted with the retained covariates using biadditive factorial regression. The models explained about 75% of the interaction sums of squares. In addition, the biadditive factorial regression biplot gave relevant information about the interaction of the genotypes (interaction pattern and sensitivities to environmental covariates) with respect to the environmental covariates and proved to be interesting for such an approach.