A bi-stage data-driven process-based model for sorghum breeding and yield prediction: coupling explainable artificial...

Area of Science:

• Agricultural Science
• Computational Biology
• Plant Breeding

Background:

• Global population growth necessitates advanced breeding methods for increased food supply.
• Sorghum is a vital cereal crop with diverse types (Grain, Forage, Dual Purpose, Photoperiod-Sensitive), requiring tailored breeding strategies.
• Genotype x Environment (GxE) interactions significantly impact crop performance, demanding sophisticated modeling approaches.

Purpose of the Study:

• To develop a bi-stage, data-driven and process-based crop model for sorghum breeding.
• To provide breeding recommendations by analyzing Genotype x Environment (GxE) effects.
• To enhance the interpretability and flexibility of crop models through explainable AI methods.

Main Methods:

• Integrated a process-based crop model with explainable data-driven techniques.
• Utilized seven years of hourly weather data, soil factors, management practices, and parental information from 651 males and 131 females.
• Predicted hourly dry weight (leaves, stems, grain) and final yield, incorporating management practices.

Main Results:

• Achieved a combined Relative Root Mean Squared Error of 16%-19% across various environmental conditions, demonstrating robust predictive accuracy.
• Successfully identified elite sorghum hybrids across four distinct types, reducing the need for extensive field trials.
• Revealed significant variability in GxE interactions, underscoring the importance of environment-specific breeding strategies.

Conclusions:

• The explainable hybrid model framework significantly improves crop modeling and plant breeding.
• This approach enhances agricultural efficiency and sustainability by optimizing breeding recommendations.
• Tailored breeding strategies based on GxE analysis are crucial for maximizing crop performance.