High‐throughput phenotyping of stay‐green in a sorghum breeding program using unmanned aerial vehicles and machine learning

Abstract

Abstract As climate change continues to influence global weather patterns, the frequency and severity of drought conditions are expected to increase, posing a significant challenge to crop production. In sorghum ( Sorghum bicolor L. Moench), a key cereal crop, the stay‐green trait is of particular importance as a measure of how well a genotype can tolerate post‐anthesis drought conditions, which are critical for harvestable yield. Despite its importance, there is a pressing need for a more efficient, accurate, and precise method to phenotype stay‐green in sorghum to enhance breeding efforts. To address this need, this study explores the application of random forest and XGBoost machine learning models for phenotyping the stay‐green trait in sorghum. These models provide quantitative measurements that have the potential to enhance genomic studies and offer additional benefits. Although correlations with vegetation indices were occasionally high, they were not sufficiently reliable to be used exclusively. The machine learning models, in contrast, showed high percentages of genetic variation explained and had high repeatability. The values generated by these algorithms enable plant breeders to efficiently make selections in their stay‐green breeding programs. Further research is needed to assess the robustness of these models across different environments and genetic material. Additionally, comparing these models with other machine learning approaches will help determine if decision tree‐based models are the most effective for this application. Overall, the models presented in this study serve as a promising foundation for improving the efficiency of stay‐green breeding programs in sorghum, but they require further validation and comparison with alternative approaches.

Code and data availability

The paper's data availability statement explicitly releases the raw tabular stay-green phenotyping data and the authors' Python machine learning scripts in a public GitHub repository, directly supporting this paper's phenotyping measurements and analysis.