Machine Learning Techniques for Predicting Crop Photosynthetic Capacity from Leaf Reflectance Spectra

Area of Science:

• Plant physiology
• Spectroscopy
• Machine learning in agriculture

Background:

• Increasing crop yield requires harnessing natural variation in photosynthetic capacity.
• Current physiological phenotyping methods are too laborious for large-scale genetic screens.
• Leaf reflectance spectroscopy presents a potential high-throughput alternative.

Purpose of the Study:

• To evaluate leaf reflectance spectroscopy for predicting photosynthetic capacity in C3 (Brassica oleracea) and C4 (Zea mays) crops.
• To identify optimal machine learning algorithms and assess model transferability across species.
• To simulate the application of this method in selective breeding for improved crop photosynthetic capacity.

Main Methods:

• Systematic evaluation of reflectance spectra properties across species.
• Machine learning methods, including recursive feature elimination and partial least squares regression, were assessed.
• Model transferability was tested using a Brassica relative, Moricandia.

Main Results:

• Reflectance spectra properties were found to be similar across a wide range of species.
• Recursive feature elimination followed by partial least squares regression demonstrated the highest predictive power.
• Intra-species models achieved high accuracy in predicting crop photosynthetic capacity, with cross-species performance not predictable by phylogenetic proximity.

Conclusions:

• Leaf reflectance phenotyping is an efficient method for improving crop photosynthetic capacity.
• The developed models have the potential to significantly enhance breeding success through high-throughput photosynthetic phenotyping.
• This approach offers a viable solution to the limitations of traditional phenotyping methods in large-scale genetic screens.