Phenotyping the hidden half: combining UAV phenotyping and machine learning to predict barley root traits in the field.

Improving crop root systems for enhanced adaptation and productivity remains challenging due to limitations in scalable non-destructive phenotyping approaches, inconsistent translation of root phenotypes from controlled environments to the field, and a lack of understanding of genetic controls. This study serves as a proof of concept, evaluating a panel of Australian barley breeding lines and cultivars in field experiments conducted across two contrasting environments. A diverse subset of 20 genotypes was subjected to ground-based root and shoot phenotyping at key growth stages, and this dataset was used in combination with unmanned aerial vehicle (UAV)-captured vegetation indices (VIs) to train machine learning models to predict root distribution and above-ground biomass for the untested panel comprising 544 genotypes across the two seasons. Unlike previous root studies that have focused on above-ground traits or indirect proxies, this approach predicts root traits in the field using machine learning. Haplotype-based mapping using predicted root and shoot traits in the broader panel revealed key genomic regions. These include novel regions, previously reported root quantitative trait loci, and EGT2-a recently cloned gene that regulates root gravitropism in barley. This scalable phenotyping approach offers opportunities to advance root research across crops and support the development of future varieties adapted to changing climates.