Robust greenhouse tomato growth prediction under severe data resolution mismatch using a feedback-guided diffusion approach

Accurate crop growth forecasting is fundamental to precision agriculture, yet its application in real-world greenhouses faces challenges including mismatches in data acquisition frequencies. Even though environmental sensors generate millions of data points, growth measurements capture less than 1% of that information, revealing a major data resolution mismatch in greenhouse monitoring. Data resolution mismatches often lead to a "lab-to-field gap", where models fail to generalize from benchmark datasets to operational settings. To address this challenge, this study introduces a feedback-guided temporal diffusion framework that generates biologically realistic and contextually consistent synthetic growth data. The framework integrates a temporal diffusion model to learn from sparse growth patterns, an environment-growth feedback network to enforce causal relationships, and a guided reverse diffusion process to ensure realistic outcomes. The framework was trained on a large-scale public dataset, and then validated through cross-domain testing on an independent operational tomato greenhouse dataset over the growing season with an even more severe mismatch in data temporal resolution. Ablation studies revealed that standard diffusion improved predictions for continuous variables, but it failed on discrete variables like leaf count, increasing the Mean Absolute Error (MAE) by 37.3%. In contrast, our feedback-guided approach mitigated this and achieved improvements across all metrics. MAE for leaf count decreased by 90% from the baseline, reaching the lowest Normalized Root-Mean-Square Error (NRMSE) of 0.019. Among forecasting models, XGBoost demonstrated optimal stability and accuracy when trained on the augmented dataset. Our proposed framework provides a robust solution to agricultural data resolution mismatch and represents significant progress toward bridging the “lab-to-field gap” for practical deployment of AI-driven crop management systems.