Genomic and phenomic prediction models improve selection accuracy for complex traits in alfalfa

Abstract

Predictive breeding has been proposed as an effective approach to accelerate genetic gain for complex traits. Genomic prediction (GP) models have been developed in alfalfa ( Medicago sativa L.) for key traits in the last decade. More recently, phenomic prediction (PP) models have been proposed as a low‐cost, high‐throughput alternative to GP in several crops. Although PP holds potential for accelerating genetic gains in plant breeding, environmental factors and data acquisition timing substantially influence predictive ability. To assess the viability of phenomics for perennial forage improvement, we compared PP and GP of dry matter yield (DMY) and persistence in alfalfa across multiple environmental conditions, including different harvest schedules and fertilizer management regimes. GP models were developed using the genomic best linear unbiased prediction (GBLUP) approach. For PP models, a similar approach was used by creating a relationship matrix from reflectance data collected with an unmanned aerial vehicle (phenomic best linear unbiased prediction). Different cross‐validations scenarios were implemented to test the performance of prediction models. For DMY, PP models resulted in greater predictive ability than GP for predictions within harvest and within management. For predictions across harvests or management practice, the predictive ability for PP models decreased and was similar or lower than GP. For stand persistence, GP and PP models achieved moderate predictive ability, ranging from 0.34 to 0.4 and 0.25 to 0.52, respectively. For PP optimization, training models built with ground truth data collected from adjacent experimental units had similar predictive ability than models built with random sampling across the whole experiment, which is operationally more effective. These results suggest that GP and PP can be implemented in alfalfa, and other perennial forages, to breed improved cultivars for complex traits.