Selecting the number of multiplicative terms in AMMI and GGE models

• Applicant: Professor Dr. Hans-Peter Piepho
• Subject Area: Plant Breeding and Plant Pathology
• Term: from 2014 to 2019
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 255643789

Series of variety and plant breeding trials conducted at multiple environments, so-called multi-environment trials (MET), are the basis for the development and dissemination of new crop varieties. A salient feature of MET is the presence of genotype-environment interaction, and this complicates the selection of the best varieties. Careful modelling and study of the interaction is therefore of utmost importance for the success of breeding programs. Models with multiplicative terms for genotype-environment interaction such as the Additive Main effects and Multiplicative Interaction (AMMI) model and the Genotype and Genotype-Environment (GGE) model are commonly used for analysing MET data. Their popularity stems from the multiple uses of the model fits, such as the delineation of mega-environments for breeding locally adapted varieties and the facility to obtain visual displays such as the biplot which allow studying the genotype-environment interaction pattern.A key problem in the use of these models is the choice of the number of multiplicative terms to be fitted. If too few terms are fitted, the resulting estimates of genotype-environment means are biased, whereas fitting too many terms leads to inefficient estimates suffering from inflated variance. Several procedures have been proposed so far for the model selection problem, including significance tests and cross-validation procedures. But empirical experience has not yet identified a single best strategy to determine the number of multiplicative terms, and each of the procedures proposed so far has some drawbacks.The purpose of the proposed project is to develop new methods to select the number of multiplicative terms in AMMI and GGE models and to evaluate and compare these empirically to contending approaches, using MET data from German plant breeding and variety testing programs as well as Monte Carlo simulation.

• DFG Programme: Research Grants