Contents - Index

Mega-environment classification based on a covariate-effect biplot


Previous topic: Interpreting the trait-effect by environment biplot

You might have noticed that the environments in the above biplots fall into two distinct clusters, although these biplots are not ideal for classifying environments. For this purpose, the singular values should be partitioned entirely to the environment scores, i.e., SVP =2 should be used so that the biplot looks like:



Clearly the 25 environments more clearly fall into two groups. The group on the right contains all environments from Ontario (ON), Prince Edward Island (PE), Manitoba (MB), and one of the two Quebec (QC) environments. Since the environments in this group are located mostly in the eastern part of Canada, it may be regarded as the Eastern mega-environment, although it also contains two environments from Saskatchewan (SK), and one environment from Montana (MO, should be shortened as MT).

The group on the left contains all environments from Alberta (AB), most environments from SK, and environments from some USA states, including Washington (WA), North Dakota (ND), and Alaska (AK). One exception is that it includes a QC environment. Therefore, this group may be called the Western mega-environment.

Interestingly, this mega-environment classification is very similar to that based on the GGE biplot of yield, indicating that the GGE of yield can be effectively explained by the covariate effect by environment interactions and therefore exploited by indirect selection for these explanatory traits.


Continue to read: Exploiting GE through selection for explanatory traits