AMMI vs regression based methods in yield prediction
Pro Research Analysisby 
Searched over 200M research papers
AMMI vs Regression-Based Methods in Yield Prediction
Yield Prediction Models: AMMI and Regression-Based Approaches
The Additive Main Effects and Multiplicative Interaction (AMMI) model and regression-based methods, such as the Finlay-Wilkinson and Eberhart-Russell regression models, are widely used for analyzing yield data from multi-environment trials. Both approaches aim to understand genotype-by-environment (G×E) interactions, but they differ in methodology and effectiveness.
Comparative Effectiveness in Capturing G×E Interactions
AMMI combines analysis of variance (ANOVA) for additive effects with principal components analysis (PCA) for interaction effects, allowing it to partition variation into genotype, environment, and G×E interaction components. This structure enables AMMI to capture more complex patterns in the data compared to linear regression models, which typically focus on linear responses of genotypes to environmental indices Yau1995Gauch2006Gauch1988.
Empirical studies have shown that AMMI consistently explains a higher percentage of the interaction sum of squares (SS) than regression-based methods. For example, in wheat yield trials, AMMI accounted for an average of 37% of the interaction SS, while joint regression analysis (JRA) explained only about 11%, regardless of data transformation or environmental diversity. This suggests that AMMI is more effective at modeling the complexity of G×E interactions, especially in large and diverse trials .
Predictive Accuracy and Practical Implications
AMMI has demonstrated superior predictive accuracy over regression-based methods and even over the use of simple treatment means. This is attributed to AMMI’s ability to use information from the entire trial, rather than relying solely on individual genotype-environment combinations. The model’s predictive power increases with trial size and data noisiness, making it particularly valuable for large, multi-location trials with significant G×E interaction Gauch1990Gauch1988Gauch1988.
In direct comparisons, AMMI has outperformed both regression-based models and Best Linear Unbiased Prediction (BLUP) in terms of root mean square error prediction, indicating more accurate yield estimates for specific genotype-environment combinations . Additionally, AMMI’s structure allows for more precise grouping of genotypes and environments, aiding in the identification of stable, high-yielding varieties .
Model Selection, Stability, and Usability
While regression-based models like Eberhart-Russell and Finlay-Wilkinson provide useful information for selecting stable varieties, AMMI offers improved estimation precision by isolating error and interaction effects more effectively. This makes AMMI a relatively ideal method for evaluating variety stability and yield prediction in multi-environment trials Falkenhagen1996Li2006.
AMMI also facilitates better visualization and interpretation of data, as it separates main effects and interactions, which is beneficial for agricultural decision-making. However, the choice of the best AMMI model for a given dataset should be guided by model diagnosis to maximize predictive accuracy .
Conclusion
In summary, AMMI models generally outperform regression-based methods in yield prediction for multi-environment trials. They provide higher predictive accuracy, better capture of G×E interactions, and more precise identification of stable, high-yielding genotypes. These advantages make AMMI the preferred choice for detailed studies and practical applications in plant breeding and agronomy, especially when dealing with complex and noisy yield data Yau1995Sa’diyah2016Li2006+5 MORE.
Sources and full results
Most relevant research papers on this topic