Machine-learning-assisted comparative analysis of rice growth and yield formation in field and plant factory systems

Plant factories provide a controlled platform for rice cultivation and rapid breeding, yet the effects of controlled environments on rice growth and yield formation remain poorly understood relative to field conditions.

This study systematically compared growth duration, plant architecture, root and leaf traits, biomass accumulation, yield components, and environmental dynamics of three representative rice cultivars grown under field and plant factory conditions. Logistic model fitting was used to characterize plant height growth dynamics, and five machine learning models were further applied to predict plant height and assess the relative importance of growth- and environment-related variables.

The results showed that a plant factory shortened the average growth duration from 138 to 95 days compared with field cultivation and promoted early vegetative development, including stronger tillering, larger total leaf area, longer roots, and greater fresh biomass accumulation. It also increased the total panicles per unit area and the grain number per panicle, whereas seed-setting rate, 1000-grain weight, and final grain yield were not significantly increased. The machine learning models achieved high predictive accuracy for plant height (R²>0.9), with growth duration as the dominant predictor in both systems, followed by CO₂ concentration in the plant factory and cultivar in the field.

These findings reveal that controlled-environment cultivation reshapes rice developmental rhythm and vegetative-reproductive allocation, providing a physiological basis for understanding rice plasticity and optimizing plant factory-based rice production.