Drought tolerance classification using unmanned aerial systems based on RGB and multispectral data.

Area of Science:

• Agricultural Science
• Plant Breeding
• Remote Sensing

Background:

• Drought is a major threat to global food security, necessitating efficient crop breeding strategies.
• Current methods for selecting drought-tolerant cultivars are time-consuming and costly.
• Developing new cultivars requires advanced techniques for trait selection.

Purpose of the Study:

• To evaluate the use of spectral data from RGB and multispectral sensors for classifying drought tolerance.
• To assess the performance of various machine-learning models in practical breeding program scenarios.
• To determine the feasibility of spectral data for automating drought tolerance assessments.

Main Methods:

• Genotypes were evaluated under optimal and drought-stress conditions over two years.
• Field traits and spectral data (RGB and multispectral vegetation indices) were collected.
• Machine learning models were trained using spectral data, with drought trials providing the best training data.

Main Results:

• Multispectral sensors generally outperformed RGB sensors across evaluation metrics.
• AdaBoost and linear discriminant analysis models showed strong consistency in predictions.
• Models achieved high accuracy (0.71), specificity (0.56), and F1-Score (0.77), with NIR band being influential.

Conclusions:

• Spectral information, particularly vegetation indices, is valuable for complementing field-based drought tolerance evaluations.
• This data-driven approach enables automation and speeds up genetic gains in breeding pipelines.
• Utilizing spectral data improves resource utilization efficiency in crop development programs.