Simulation‐based optimization of mega‐environmental designs for sparse multi‐environment testing

Abstract

Plant breeding programs aim to identify superior varieties by evaluating potential varieties in a network of locations, and these evaluations are often called multi‐environment trials. It is important to optimize multi‐environment trials in terms of resource allocation to ensure accurate and efficient experiments. It has been shown that mega‐environmental designs (MEDs), which are sparse designs for multi‐environment trials that explicitly account for the mega‐environment structure within a target population of environments, can substantially increase the response to selection compared to other common designs. The objective of this simulation study was to further optimize MEDs in three areas to provide general guidelines for their use: the amounts of replication and sparsity to use in the designs, the choice of model for analysis, and the ideal structure for the target population of environments. Optimized MEDs were compared to other commonly used designs in simulated multi‐environment trials using real yield data and the real genotype‐by‐environment interaction structure of 223 barley ( Hordeum vulgare L.) genotypes evaluated in 30 location‐year environments in the United States and Canada. MEDs always had the highest response to selection when heterogeneous genetic variances across mega‐environments were modeled, and sparser MEDs were generally best. However, interactions among trait heritability, strategies in modeling the relationships between genotypes, and the alignment of the environments with historical mega‐environment structure make choosing the most appropriate experimental design case‐specific.