Machine Learning Approach for Prescriptive Plant Breeding.

Area of Science:

• Plant breeding
• Agricultural science
• Machine learning applications

Background:

• Accurate in-season seed yield (SY) prediction is crucial for effective crop management and cultivar development.
• Integrating high-dimensional phenomic data with advanced analytical methods can enhance predictive capabilities.

Purpose of the Study:

• To explore the fusion of phenomic data and machine learning (ML) for in-season SY prediction.
• To develop tools for prescriptive cultivar development and targeted agro-management practices.
• To identify key phenotypic traits influencing SY prediction across different management systems.

Main Methods:

• Phenotyping 32 SoyNAM parent genotypes across contrasting agro-management treatments (row spacing, seeding density).
• Collecting phenotypic trait data (canopy temperature, chlorophyll content, hyperspectral reflectance, leaf area index, light interception) at three growth stages.
• Utilizing Random Forest (RF) ML models to train for SY prediction using phenotypic traits, optimizing variable combinations for accuracy.

Main Results:

• Developed RF models for SY prediction using phenomic data, achieving optimal accuracy through specific temporal variable combinations.
• Identified key phenotypic traits for SY prediction, noting some traits' importance varied conditionally with agro-management.
• Demonstrated the capability of integrating phenomics and ML for in-season SY prediction.

Conclusions:

• The integration of phenomic data and ML provides powerful tools for plant breeders.
• This framework enables both in-season seed yield prediction and prescriptive cultivar development.
• Optimized agro-management strategies can be informed by predictive phenomic insights.