Maize (Zea mays L.) is one of the world's most productive crops, benefiting from advancements in agronomic practices and breeding techniques that harness hybrid vigor. Over the past century, maize yields have experienced an eightfold increase, driven by these innovations. As global challenges related to food security intensify with a rapidly growing human population and increased protein consumption, higher breeding goals for maize have become imperative. The maize canopy profoundly impacts plant growth and yield. By using high throughput phenotyping and manual measurements of various leaf and canopy parameters, we found late-season canopy traits significantly impact yield components. The integration of these leaf canopy traits led to the development of a predictive model for yield, achieving an R2 value of 0.483. Innovative leaf angle measurements and a simulation method for leaf curvatures were also introduced. Integrating yield analysis with canopy traits offers critical insights for maize breeding and cultivating high-yield varieties, advancing productivity by leveraging a deeper understanding of canopy dynamics. Through the analysis of multi-year high-throughput phenotyping data, we established a regression method to align Normalized Difference Vegetation Index (NDVI) data with Growing Degree Days (GDD) across diverse environments. Retaining "stay green" ability emerged as a critical factor impacting yield, quantified through our NDVI-GDD curve, exhibiting strong linear correlations with yield at both plot and environmental levels. Leveraging this, we developed physiological and genomic prediction models for yield, demonstrating promising predictive capabilities across practical breeding scenarios. NDVI, as a high-level trait, displayed correlations with several manually measured traits. Our Genome-Wide Association Study (GWAS) using NDVI as an input trait identified significant signals for these correlated traits, suggesting the potential to replace current manual measurements in breeding programs. Though genomic selection has become commonplace to expedite breeding cycles, accurately forecasting the performance of new genotypes remains a significant challenge. A study evaluating cross-subpopulation genomic prediction within a small tested cross NAM population revealed that the accuracy of cross-subpopulation prediction fell below that of randomly sampled genetics pools. Additionally, prediction accuracy varied considerably among traits and within prediction subpopulations. We delved into these differences, identifying potential explanations. To enhance cross-subpopulation prediction accuracy, we explored the impact of dominant relationship matrices, Gaussian kernel-based relationships, and LD-adjusted methods, which provided limited improvement. The findings highlight the complexity of genomic prediction in diverse breeding scenarios and the need for further research to enhance accuracy and applicability.
Read
