Wheat MET trial across 8 locations: AMMI surfaces 2 mega-environments
12 advanced wheat lines across 8 sites in north and central India. Reading the AMMI biplot to cluster sites and pick varieties for each cluster.
Anita is a wheat breeder in a state agricultural university. Her target programme covers eight test sites across north and central India. She has 12 advanced lines (G1 to G12) and one season of yield data per site. The breeding committee meets in three weeks and wants to know which line to recommend for which zone.
Question
Are the eight sites really one mega-environment for which one variety is enough, or are they two or more clusters that warrant separate recommendations?
Data, in StatVeda format
genotype,E1,E2,E3,E4,E5,E6,E7,E8 G1,4.21,3.88,5.62,4.05,4.12,4.31,5.45,4.18 G2,4.55,4.12,5.31,4.62,4.42,4.55,5.21,4.61 G3,3.95,3.71,5.84,3.88,4.02,4.18,5.71,3.95 G4,5.12,4.66,5.05,5.21,4.95,5.08,5.02,5.18 G5,4.78,4.34,5.42,4.71,4.51,4.65,5.32,4.71 G6,5.31,4.88,4.92,5.45,5.18,5.31,4.95,5.42 G7,4.02,3.65,5.71,3.92,3.85,4.01,5.62,3.91 G8,4.62,4.21,5.18,4.55,4.41,4.55,5.05,4.52 G9,4.91,4.48,5.05,5.02,4.78,4.91,5.01,4.99 G10,4.12,3.78,5.61,4.01,3.95,4.12,5.51,4.05 G11,5.21,4.71,5.01,5.32,5.05,5.18,4.92,5.28 G12,4.32,3.95,5.55,4.21,4.11,4.28,5.42,4.18
What she does in StatVeda
Open Biplots, GxE category. Pick AMMI biplot. Paste the long form CSV (or paste the wide form above and let StatVeda pivot). Run.
IPCA1 explains 71 percent of the GxE sum of squares, IPCA2 18 percent. Together 89 percent of the interaction is captured in two axes.
Two clear environment clusters emerge along IPCA1. E3 and E7 sit together, far from the origin, on the positive IPCA1 side. E1, E2, E4, E5, E6, E8 cluster on the negative IPCA1 side, closer to the origin.
On the genotype side, G3, G7, G10, G12 sit on the same side as E3 and E7. G4, G6, G9, G11 sit on the opposite side, near the E1-cluster.
What it means
E3 and E7 are a separate mega-environment. Looking at the original data, those are the two sites where every genotype yielded above 4.9 t/ha and the responsive lines G3, G7, G10 went above 5.5. The other six sites form a second mega-environment where yields are lower and the broadly-adapted lines G4, G6, G9, G11 dominate.
Recommend G3 (top adjusted yield, far from origin on IPCA1) for the E3 plus E7 zone.
Recommend G6 (top adjusted yield, opposite cluster) for the other six sites.
Drop G1, G2, G5, G8, G12 from advancement: nothing distinctive in either cluster.
Anita writes up the AMMI section of her report with the biplot screenshot, the partitioning table, and the two-zone recommendation. The breeding committee accepts the proposal to release G3 in the E3 plus E7 zone and G6 in the wider zone.
Why two zones, not one
If she had averaged across all eight sites and ranked on grand mean, G4 (mean 5.10) and G6 (5.18) would have come out joint top, and the recommendation would have been a single broadly-adapted line. That would leave 25 percent of her target zone (E3 plus E7) under-served: the lines that suit those two sites are the ones that look mediocre on average. The biplot makes the trade-off visible in a way the ranked-mean table cannot.
What she will do next season
Repeat the trial with G3, G4, G6, G9, G11 only across the same eight sites, plus add two new test lines from the next breeding cohort. With fewer entries she gets more replications per site, tighter standard errors and a sharper picture of the two-zone recommendation. The AMMI biplot from year two will either reinforce the year-one picture or flag year-on-year instability, which is itself useful information.