Machine Learning-Driven Construction of High-Yielding Cucumber Plant Architectures in Greenhouse Environments

Area of Science:

• Agricultural Science
• Plant Breeding
• Computational Biology

Background:

• Declining arable land necessitates efficient resource utilization for food security.
• Developing optimized plant architectures is crucial for sustainable agriculture.

Purpose of the Study:

• To identify high-yielding cucumber plant architectures for greenhouse cultivation.
• To predict cucumber yield based on aboveground and root traits using machine learning.

Main Methods:

• Collected yield and trait data from 263 cucumber varieties.
• Utilized Gradient Boosting Decision Tree (GBDT) and Support Vector Machine (SVM) algorithms for prediction.
• Performed scenario simulations on 157,464 phenotypic combinations.

Main Results:

• Cucumber yield can be predicted using traits like flower node position, leaf width, stem diameter, and root angle (R²=0.6155).
• Identified antagonistic and synergistic interactions between aboveground and root structures.
• Phenotypes with compact, robust aboveground structures and shallow, large-diameter root systems showed up to 20% higher yields.

Conclusions:

• Machine learning models effectively predict cucumber yield based on plant architecture.
• Optimized plant structures, balancing aboveground and root traits, enhance greenhouse cultivation efficiency.
• Provides a theoretical basis for designing high-yielding cucumber varieties.