Yield prediction in a peanut breeding program using remote sensing data and machine learning algorithms.

Area of Science:

• Agricultural Science
• Plant Breeding
• Remote Sensing

Background:

• Peanut is a vital global food crop, necessitating advancements in breeding for increased genetic gain.
• Direct yield estimation in peanuts via remote sensing is challenging, requiring indirect methods using above-ground traits.
• High-throughput phenotyping is crucial for accelerating crop improvement.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting peanut yield using UAV-derived phenotypic data.
• To assess the effectiveness of random forest and eXtreme Gradient Boosting (XGBoost) algorithms in peanut yield estimation.
• To demonstrate the utility of these models in improving the efficiency of peanut breeding programs.

Main Methods:

• Unmanned aerial vehicles (UAVs) were used for high-throughput phenotyping of peanut surface traits.
• Multitemporal growth curves (canopy cover, height) were constructed from UAV imagery.
• Latent phenotypes from growth curves informed random forest and XGBoost models for yield prediction.

Main Results:

• The random forest model achieved high predictive accuracy for peanut yield (R² = 0.93).
• The eXtreme Gradient Boosting (XGBoost) model also demonstrated effective yield prediction (R² = 0.88).
• Both models proved valuable for classifying genotypes, aiding in the selection process within breeding pipelines.

Conclusions:

• Machine learning models, particularly random forests and XGBoost, show significant potential for predicting peanut yield.
• UAV-based phenotyping combined with machine learning can substantially improve the efficiency of peanut breeding programs.
• These methods facilitate the identification of superior genotypes and the filtering of underperforming ones.