Breeding for low deoxynivalenol (DON) mycotoxin content in wheat is challenging due to the complexity of the trait and phenotyping limitations. Recent advances in prediction-based breeding using phenomic and genomic data have improved breeding strategies for such traits. Since phenomic prediction relies on non-additive effects and genomic prediction on additive effects, their comparison and complementation are necessary. Five Bayesian generalized linear regression models and two machine learning models were trained for phenomic and genomic predictions using advanced breeding lines evaluated in 2021 and 2022 independently, and the calculated BLUEs across years to account for year variation. Across all training sets and models, phenomic predictions using wavebands in the visible light spectrum (400-700 nm) had higher predictive ability than genomic predictions or phenomic predictions using the full waveband range (400-1000 nm). Forward prediction and model averaging were conducted on two sets of F_4:5 selection candidates evaluated independently in 2022 and 2023. The phenotypic and genetic correlations, as well as indirect selection accuracies, of the model averages of phenomic predictions and combined phenomic and genomic predictions, were higher than those of genomic predictions alone, but depended on the combination of training set and selection candidates. Unsupervised K-Means clustering-based ensembling of predicted values indicated distinct groupings of candidates with varying DON content, depending on the models, predictors, and training sets used. This study demonstrates the potential of hyperspectral imaging-based phenomic prediction to complement genomic prediction, highlighting considerations for prediction-based selection of low DON in soft winter wheat.