Genomic prediction of agronomic traits in a switchgrass (Panicum virgatum L.) half-sib progeny panel evaluated across multiple environments.

Switchgrass (Panicum virgatum L.) improvement requires selection methods that remain effective across environments. Biomass yield is strongly influenced by genotype-by-environment (G × E) interaction. We evaluated genomic prediction models for biomass yield, spring emergence (SE), and flowering time (FT) in half-sibs at three southeastern US locations (Watkinsville, GA; Tifton, GA; and Knoxville, TN). Predictive performance was assessed within each site-year using five-fold cross-validation, comparing a parental general combining ability (GCA) baseline with Bayesian models, genomic BLUP (GBLUP), and a dominance model (GBLUPD). We further quantified the sensitivity of yield prediction to single-nucleotide polymorphism (SNP) density. Predictive abilities as Pearson correlation coefficient (PCC) increased as the SNP number rose from the minimum of 100 to 2500-5000 SNPs, with only minimal gains observed up to 10,000 SNPs and beyond. Genomic models consistently outperformed parental GCA baseline for yield, with the highest PCC of 0.45-0.55 in 2022 for Georgia, followed by declines (PCC = 0.25-0.35) in 2024. This indicated stronger G × E (where E represents combined year-location) impacts on mature switchgrass stands. FT showed higher predictive ability than yield, with PCC > 0.50 in Georgia and PCC > 0.28-0.33 in Tennessee. SE exhibited intermediate-to-high PCC of 0.65-0.70 in Georgia during 2022-2023 and a PCC of 0.45-0.47 in Tennessee. The GBLUPD provided small repeatable gains for yield and SE in comparison to GBLUP. FT exhibited genetic correlations exceeding 0.90 at Knoxville, which showed it as a transferable predictor for multi-environment selection. Collectively, these results indicate that modest SNP sets can be sufficient for yield prediction and highlight environment-dependent model performance.