Predicting the yield of varieties of maize in a target environment using regression analysis

Previous analyses on variety yield have usually focused on regression coefficients as an indicator to measure the stability and adaptation of a specific variety under experimental conditions. Due to the huge differences between experimental plots and farm fields, the model results from experimental plots can hardly be applied to farm fields. In this study, a regression analysis was conducted between the variety yield and an on-trial environment index (the mean yield of all varieties in the same test site). Then, using the average proportional coefficient between the on-trial environment index and the on-farm environment index (the statistical maize production yield of the growing county containing the test site) as a bridge, the on-farm environment index was converted to the corresponding on-trial environment index, which was then applied to the regression model generated from the on-trial plot-scale data. This procedure ensured the homogeneity of the model parameters and successfully predicted the yield of maize varieties under a target environment. The procedure also produced the 95% confidence interval predicted yield, making the results more practically significant. By introducing the proportional coefficient and confidence interval, the new approach provides a feasible solution for studying the performance of varieties under on-farm conditions. Finally, we used the maize variety NH1101 as an example to illustrate the modeling procedures. The results indicated that the model produced promising results. The new method provides direct support for variety recommendation, and facilitates the identification of better-adapted varieties.