Pseudoperonospora cannabina, the causal agent of downy mildew in Cannabis sativa, was first identified in New York in 2020. The pathogen causes angular brown lesions on foliar tissue and dark sporulation on the leaf underside, often leading to leaf curling, defoliation, and yield loss. To date, no sources of genetic resistance or effective chemical controls have been reported. In this study, 108 hemp entries were evaluated for resistance to downy mildew using three phenotyping methods: 1) high-throughput Blackbird robotic imaging coupled with convolutional neural network-based leaf disc analysis, 2) visual rating of the same leaf discs, and 3) detached leaf assays. Susceptibility ranged from 0 to 100%, with two entries, G 33532 and 'Santhica 27', exhibiting 0% severity across all three phenotyping methods. Detached leaf assays produced the lowest disease severity estimates, whereas Blackbird convolutional neural network ratings exhibited the lowest residual variance but consistently underestimated disease severity relative to visual assessments. Visual leaf disc ratings produced the highest severity estimates and had the greatest variability. These results suggest that while Blackbird convolutional neural network analysis holds promise for high-throughput phenotyping of the hemp downy mildew interaction, additional refinement is required to improve accuracy.
Herrmann et al. (Sun,) studied this question.