Evaluating UAV-based phenotyping strategies for Megathyrsus maximus

Area of Science:

• Agricultural Science
• Plant Breeding
• Remote Sensing

Background:

• High-throughput phenotyping is vital for plant breeding, but optimal UAV imaging parameters for forage crops are undefined.
• Understanding the impact of ground sampling distance (GSD), environment, and harvest date on UAV-derived traits is crucial.

Purpose of the Study:

• To optimize UAV-based phenotyping for Megathyrsus maximus.
• To evaluate the accuracy of RGB-derived digital traits in predicting yield and canopy height.
• To assess the influence of GSD, environment, and harvest date on trait prediction.

Main Methods:

• Applied machine learning algorithms and mixed model analyses to predict yield and canopy height.
• Examined correlations between digital traits (pixel count, entropy, vegetation indices) and ground-truth measurements.
• Investigated the effect of different GSD resolutions (0.27-1.5 cm) on prediction accuracy.

Main Results:

• Pixel count and Haralick's entropy strongly correlated with yield, especially in Environment 2.
• Machine learning significantly improved prediction of green and dry matter yield (r > 0.80) and canopy height (r = 0.71).
• Moderate GSD (0.5-1.0 cm) yielded the best results; high heritability (0.7-0.87) observed for yield traits.

Conclusions:

• Optimized UAV phenotyping framework using advanced digital traits and machine learning accurately predicts key agronomic traits in M. maximus.
• This approach enhances genotype selection efficiency in forage breeding programs.
• Moderate GSD resolutions are recommended for effective UAV-based phenotyping in this species.