Abstract
Genomic prediction models that fit multiple environments globally are valuable tools for assessing cultivar performance across diverse and variable growing conditions. We analyzed 2,064 strawberry (Fragaria × ananassa) accessions genotyped with 12,591 SNP markers. Soluble solids content (SSC) was measured in multi-year trials conducted at seven locations spanning the U.S., Europe, and Australia. Population structure analysis grouped accessions into two major clusters corresponding to subtropical and temperate origins, which was confirmed by significant differences in allele frequency distributions. To improve prediction accuracy across environments, we developed factor analytic models focusing on genotype-by-environment interactions rather than covariance between sub-populations. We compared three genomic prediction approaches: (i) a standard GBLUP model (Gfa), (ii) a GBLUP model incorporating principal component analysis eigenvalues and re-parameterization (Pfa), and (iii) a multi-population GBLUP model that fits sub-population genomic relationship matrices (Wfa). The Pfa and Wfa models achieved the highest prediction accuracy (r = 0.8) for SSC, outperforming individual environment models and the standard GBLUP. These findings demonstrate that accounting for population structure and genotype-by-environment interactions enhances multi-environment genomic prediction and supports practical implementation of genomic selection in global strawberry improvement programs.