Recently, the Innovation Team of Plant Environmental Engineering of Protected Agriculture at the Institute of Environment and Sustainable Development in Agriculture, Chinese Academy of Agricultural Sciences, in collaboration with Wageningen University & Research (WUR), made progress in greenhouse tomato growth modeling. The related findings have been published in Computers and Electronics in Agriculture.

Knowledge-and-data-driven SATCN-Informer modeling framework

Accurate estimation of crop growth is critical for intelligent greenhouse control. However, traditional process-based models are often difficult to parameterize and limited in generalization, while purely data-driven models may produce distorted predictions due to a lack of physiological constraints. Therefore, integrating temporal environmental data with crop growth patterns to achieve high-precision and physiologically plausible growth predictions remains a key challenge.

The research team proposed a knowledge-and-data-driven scale-adaptive temporal convolutional hybrid model (SATCN-Informer) for tomato growth prediction. This model dynamically adjusts the scale of temporal feature extraction across different tomato growth stages. In conjunction with pre-training on synthetic data from a process-based model (TOMSIM) and physiological threshold-based constraints, it achieves accurate prediction of dry mass and leaf area index (LAI) in different organs of tomato. The model enables non-destructive estimation of plant growth based solely on regular greenhouse environmental data, holding promising practical value for advancing intelligent environmental control and yield prediction in greenhouses.

This work was supported by the Central Public‑interest Scientific Institution Basal Research Fund.

Linkage: https://doi.org/10.1016/j.compag.2026.112281