What genotype do offspring have?
Genotype-environment interaction was analysed 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 whenabundant
external information is available on genotypes and/or environments.
Experiments were conducted at 5 sites in France during 1991-92. The
approach, focused on environmental characterization, was performed
with two kinds of covariates: (1) deviations of yield components
measured on 4 probe genotypes; and (2) usual indicators of
yield-limiting factors. The first step was based on analysis of a
crop diagnosis measured on 4 probe genotypes. Difference of grain
number to a threshold number (DKN)and reduction of 1000-grain
weight from a potential value (RTKW) were used to characterize
grain-number formation and grain-filling periods, respectively.
Grain yield was analysed according to a biadditive factorial
regression model using 8 environmental covariates (DKN and RTKW
measured on each of 4 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 wasperformed on the analysis of DKN and
RTKW for the 4 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, sumof 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.
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 theinteraction 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.
