the bilinear and simple linear models were performed by calculating
the root mean square error (RMSE) described in Potter and
Williams (1994).
Different attributes recorded on a plot basis were analyzed to
identify any possible association between the occurrence of the
breakpoint and a generalized observation across genotypes, densities
and years. For each plot (a particular combination of genotype,
density, replicate and year), we selected two values of cumulative
IPAR—one immediately above and the other immediately below the
breakpoint estimated by the bilinear model (Eqs. (6) and (7)). We
chose the closer value and identified the sampling date. A database
was built recording the corresponding LAI and F values for every
selected IPAR value. We also recorded the developmental stage
and number of leaves of five plants harvested from each plot in
the same date. All developmental data corresponding to the same
genotype were pooled across densities and years. Then, we built
cumulative frequency distributions for developmental stage and
leaf number for each genotype (n = 70), and for F and LAI across
treatments and years (n = 56). For each attribute, we calculated the
20th, 50th and 80th percentiles.