Optimization of target population of environments for breeding programs

Adequately representing the target population of environments (TPE) is a central challenge in breeding programs, as it directly affects the characterization of genotype-by-environment interaction (GEI) and supports recommendation decisions. This study proposes a data-driven approach that integrates phenotypic and environmental information to precisely define the TPE based on the response of genotypes to environmental variation. Environmental coefficients were estimated via partial least squares (PLS) regression on phenotypic and environmental data from 4,791 maize genotypes evaluated across 29 locations of the Genome to Fields initiative (2014–2023), capturing the main environmental gradients associated with GEI. An environmental similarity matrix, built with a Gaussian kernel, revealed consistent environmental groupings. Representative environments (pivots) were then selected using a genetic algorithm based on the PEVMEAN criterion, identifying five distinct environmental groups with similar genotypic response patterns. The approach efficiently captured environmental heterogeneity and revealed redundancies within the experimental network, allowing highly similar environments to be removed with minimal loss of representativeness. These findings demonstrate the methodology's potential as a decision-support tool for optimizing multi-environment testing networks, reducing operational costs and increasing the efficiency of breeding programs.