Effects of Phenotypic Adjustment Using Genetic and Spatial Models on Genomic Prediction and Top-N Recall in Early-Stage Maize Hybrid Breeding

Single-replicate field trials are common in early-stage maize hybrid screening, but raw plot observations contain field noise that can weaken genomic prediction. We evaluated whether pedigree and spatial adjustment improve the predictability of phenotypic targets and Top-N recovery, whether joint pedigree-spatial adjustment provides complementary gains, and whether target construction has a larger influence than baseline model choice. Across four sites, five agronomic traits, four phenotypic inputs, and six genomic prediction models, pedigree-corrected phenotypes (Ped) increased mean predictive ability from 0.465 ± 0.157 for raw observations (Raw) to 0.573 ± 0.131 and increased upper-tail Top-20 recall from 0.558 ± 0.146 to 0.656 ± 0.146 under a 20% predicted retention window. Spatial correction alone (Spa) was less predictable than Raw, whereas joint pedigree-spatial correction (Ped+Spa) improved over Raw but did not exceed Ped. Phenotypic input explained substantially more variation in predictive ability than model choice, whose main effect was not significant (F5428 = 1.59, p = 0.161). These results support pedigree correction as a useful target-construction strategy in the evaluated single-replicate trials while showing that spatial adjustment is conditional on model adequacy and trial structure. Top-N recall directly represents favorable direction recovery for grain yield; for grain moisture, plant height, ear height, and silking date, the archived upper-tail metric is interpreted as rank consistency because the favorable direction or ideotype is breeding-program-dependent.