Content area
The cost of genotyping has decreased significantly in previous decades, providing plant breeders with vast amounts of genomic information. However, the cost and labor requirements for phenotyping have not declined at a comparable rate, creating a phenomic bottleneck in plant breeding. To maintain the pace of genetic gain in the future, it is essential to fully characterize the quantitative variation in important agronomic traits attributable to genetics. Achieving this requires next generation phenotyping methods that are objective and accessible for measuring complex traits, along with a robust framework to derive meaningful biological conclusions at phenomic, genomic, and temporal scales. This dissertation addresses these challenges through (1) an exploration of current and potential future methods for micro and macroscopic stress detection in plants, (2) the development of a deep learning pipeline for quantifying Fusarium head blight (FHB) in wheat, followed by (3) its application using mobile images to enhance genomic selection for resistant lines, and (4) a temporal analysis of the phenomics and genomics of maize canopy cover using unoccupied aerial vehicles (UAVs). For FHB in wheat, a high-throughput, deep learning-based image analysis pipeline was found to be more precise than visual ratings for disease scoring. The pipeline’s disease inferences remained valid across diverse environments, camera angles, and stages of disease progression. Moreover, the deep learning pipeline was successfully adapted for use with mobile phone images, offering a scalable solution for research groups studying this disease. Furthermore, genomic prediction models trained with image-based disease scores outperformed conventional training methods used by the University of Minnesota Wheat Breeding Program. For maize canopy cover, this study revealed that factors influencing canopy development varied throughout the growing season. While genetic constraints set upper and lower limits, environmental factors and genotype-by-environment (G×E) interactions influenced canopy dynamics along this spectrum. The rate of canopy accumulation emerged as a valuable dynamic indicator of plant growth and response to environmental conditions both within and between seasons. Additionally, by integrating multiple modeling approaches and temporal trait iterations, the study identified a diverse number of genomic regions associated with canopy traits, far exceeding what would have been detected using only terminal measurements. Together, these studies demonstrate how advanced phenotyping and modeling approaches can enhance plant breeding by improving trait measurement, increasing genomic prediction accuracy, and expanding our understanding of genotype-by-environment interactions.