Finlay-Wilkinson random regression to predict yield and yield stability in cereals

Authors: SANDRO, PABLO - University Of Wisconsin; BLANCON, JUSTIN - University Of Clermont Auvergne, Clermont-Ferrand, France; Neyhart, Jeffrey; DAWSON, JULIE - University Of Wisconsin; GUTIERREZ, LUCIA - University Of Wisconsin

Submitted to: The Plant Genome
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 6/1/2026
Publication Date: N/A
Citation: N/A

Interpretive Summary: Environmental variability significantly impacts agriculture by reducing crop yields, necessitating the selection of plant varieties that can endure these changes by offering greater stability in yield and other traits. Traditional plant breeding methods often face challenges due to the sparse nature of multi-environment datasets, where not all varieties are evaluated in all environments; however, these datasets are essential for evaluating grain yield stability. This study compared a new type of predictive model for stability to the standard predictive model as alternate approaches for selecting plants based on grain yield and yield stability across complete and incomplete data sets from barley, wheat, and oat field evaluations. The new predictive model was more accurate and with less data than the traditional approach, as long as the data originated from environments of yield extremes. In many cases, plant breeders are aware of those environments that tend to lead to yield extremes, so this research offers a practical approach for breeders to select for greater yield stability early in the breeding pipeline. Breeders and seed developers will use these predictive models to select and release crop varieties with stable yields, and farmers will benefit by planting those varieties that perform reliably under year-to-year conditions.

Technical Abstract: Year to year climate variability poses a challenge for agriculture by compromising crop yields; hence, there is a need to select genotypes that can withstand climatic variability. Through plant breeding, we can select genotypes with yield stability capable to withstand variability. There are several indices available to evaluate grain yield stability, most stability indices require complete datasets spanning multiple environments. A downside is that breeding programs typically generate sparse multi-environment datasets. Methods such as Finlay-Wilkinson random regression (RFW) allow us to use sparse datasets and incorporate genomic data to leverage phenotypic information from related genotypes to predict yield stability. Our objective was to integrate Finlay-Wilkinson random regression (RFW) as a genomic prediction tool for characterizing yield stability, compare it to the traditional gBLUP as reference models. As well as examine how the number of environments used and variance among those environments affect stability predictions. We used three different training populations, two completely balanced training populations with a different number of environments in barley and wheat, and one highly unbalanced training population for oats. We fit FW and RFW models in various scenarios where we estimated stability and predicted grain yield. We found that the estimated stability values obtained were similar using whole datasets for FW or RFW. RFW achieved high predictive ability for stability in both balanced and unbalanced datasets under CV1 cross-validation (new genotypes). When we look at the number of environments, selecting the right set of environments is as important than adding more environments.