Regression Analysis of Genotype‐Environment Interaction in Smooth Bromegrass1

Publisher Website
Google Scholar
Abstract
Twenty‐one single crosses and seven selected parental clones of smooth bromegrass (Bromus inermis Leyss.) were grown in replicated trials at four locations in Alberta (Lethbridge, Edmonton, Kinsella, and Beaverlodge). Lethbridge and Edmonton were higher yielding environments than either kinsella or Beaverlodge. Plants at Beaverlodge had the highest yield per unit area and tiller dry weight.The analyses of variance revealed that genotypes interacted significantly with environments for all characters. The genotype ✕ environment interaction source was further partitioned into the heterogeneity among regression and the residual. Environments were assessed in two ways: (i) the mean expression of all genotypes and (ii) the average performance of the seven parental genotypes. When environments were measured by the mean of all genotypes, a significant part of the interaction was ascribed to the heterogeneity among regression lines for all characters. The residual components were also significant in most cases indicating the presence of some unpredictable variation. The test of the heterogeneity of regressions against residuals revealed that the linear model would retain considerable predictive value for characters such as tiller density, yield per area, and fall yield since most of the interactions were accounted for by the heterogeneity of regressions.When environment was assessed independently by the mean of the seven parents of the single crosses, the proportion of the interaction attributed to heterogeneity of regression differed considerably from that assessed dependently by overall genotypes. Height was the only character which failed to fit the linear model. Despite the differences obtained from the test of significance for heterogeneity of regressions, the rankings of the genotypes on the basis of their linear regression coefficients were similar. The results showed that the conclusions drawn from regression data were unaffected by the choice of measure of environment and that predictability of genotypes varied with characters measured.

This publication has 3 references indexed in Scilit: