Table 8 Estimated prediction accuracy (correlation between predicted and adjusted grain yield, averaged over 50 TRN-TST partitions): cycle 2012−2013 (W3), CV1
Models/EnvironmentsaCorrelationChange%bNumberc
Single environment
 0iFN0.561(0.035)
 2iBN0.445(0.051)
 5iBH0.628(0.037)
 5iBN0.360(0.046)
 5iFN0.312(0.055)
Pairs of environmentsInteraction ModelAcross-Env Model
 0iFNCor = 0.170.559(0.036)0.411(0.049)−0.3%; 36.1%18; 50
 2iBN0.445(0.051)0.317(0.060)0%; 40.4%29; 49
 0iFNCor = 0.300.563(0.036)0.466(0.042)0.2%; 20.7%34; 50
 5iBH0.626(0.037)0.552(0.039)−0.3%; 13.4%9; 50
 0iFNCor=-0.100.559(0.036)0.356(0.048)−0.5%; 57.1%13; 50
 5iBN0.360(0.046)0.154(0.051)−0.1%; 134.2%27; 50
 0iFNCor=-0.010.556(0.036)0.400(0.045)−0.9%; 39%8; 50
 5iFN0.311(0.054)0.136(0.057)−0.5%; 129.5%20; 50
 2iBNCor=-0.030.441(0.052)0.236(0.049)−0.7%; 87.2%11; 50
 5iBH0.625(0.037)0.439(0.052)−0.4%; 42.3%9; 50
 2iBNCor = 0.120.444(0.051)0.370(0.049)−0.1%; 20.2%27; 49
 5iBN0.361(0.046)0.300(0.047)0.2%; 20.2%31; 48
 2iBNCor = 0.040.446(0.050)0.345(0.052)0.3%; 29.4%31; 50
 5iFN0.313(0.054)0.180(0.056)0.1%; 73.7%24; 50
 5iBHCor=-0.090.627(0.038)0.451(0.051)−0.2%; 38.9%14; 50
 5iBN0.358(0.046)0.130(0.054)−0.6%; 174.4%20; 50
 5iBHCor = 0.020.625(0.038)0.485(0.046)−0.4%; 28.8%10; 50
 5iFN0.305(0.053)0.125(0.060)−2.2%; 144.1%7; 50
 5iBNCor = 0.550.352(0.046)0.316(0.052)−2.2%; 11.5%15; 46
 5iFN0.309(0.053)0.286(0.054)−1.2%; 7.9%19; 42
All environments
 0iFN0.560(0.037)0.311(0.049)−0.2%; 80.2%21; 50
 2iBN0.443(0.051)0.209(0.059)−0.4%; 112%21; 50
 5iBH0.620(0.039)0.333(0.059)−1.3%; 86.1%0; 50
 5iBN0.357(0.046)0.156(0.048)−0.7%; 129.4%16; 50
 5iFN0.309(0.052)0.111(0.064)−1.2%; 179.2%15; 50
  • TRN-TST, training-testing.

  • a Environments are described by a sequence of codes: 0i, 2i, and 5i denote the number of irrigation cycles; B/F denotes whether the planting system was “bed” (B) or “flat” (F); N/H denotes whether planting date was normal (N) or late (H, simulating heat); Z indicates no tillage.

  • b Change in prediction accuracy of the M×E model relative to the prediction accuracy of the single-environment (before semicolon) and relative to the prediction accuracy of the across-environment model (after semicolon).

  • c Number of partitions (out of 50) for which the M×E model gave greater accuracy than the single-environment (before semicolon) and the across-environment model (after semicolon).