A robust-weighted AMMI modeling approach with generalized weighting schemes

Abstract

The additive main effects and multiplicative interaction (AMMI) model and its variations are widely used to identify genotypes with specific adaptability and stability under environmental conditions in crop improvement breeding programs. However, atypical data points, arising from measurement errors, genotype characteristics, diseases, or climate phenomena, can significantly impact the model’s performance, by contributing to the violation of its underlying assumptions. To address this challenge, we propose a hybrid modeling framework called robust-weighted AMMI (RW-AMMI), which combines robust and weighted algorithms to effectively model genotype-by-environment interaction (GEI) in the presence of data contamination and heteroscedasticity. We also introduce a comprehensive set of nine weighting schemes for the weighted (W-AMMI), robust (R-AMMI), and RW-AMMI models. Our extensive Monte Carlo simulations, which encompass both contaminated and uncontaminated data with and without heterogeneous error variance, demonstrate that several models within the W-AMMI, R-AMMI, and RW-AMMI classes perform competitively relative to the conventional AMMI model. Furthermore, we validate the effectiveness of the proposed approach using real crop data, where we leverage ensemble strategies to enhance genotype recommendations, providing practical evidence of its applicability. This work provides a hybrid framework for genotype selection under diverse environmental conditions, offering breeders a reliable tool for improving stability and adaptability.