A Multidimensional Data Fusion Method Incorporating a Temporal Feature for Estimating Tomato Aboveground Biomass

Aboveground biomass (AGB) serves as a key indicator of tomato seedling growth status and provides a critical basis for guiding precision management. However, existing AGB estimation methods are typically confined to single-time-point assessments and fail to account for phenotypic variations across different growth ages. To address this limitation, this study proposes a growth-stage-aware estimation method for tomato seedling AGB utilizing cross-sectional data collected across multiple growth stages. The proposed method constructs multidimensional input variables from stem phenotypic traits, RGB vegetation indices, and image texture features, and incorporates plant age (days after sowing) as a temporal feature to capture stage-dependent variations during growth. In this study, Bayesian Ridge Regression (BR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were integrated into an ensemble learning framework to leverage the complementary strengths of these models. The proposed stacking ensemble approach yields improved estimation performance, showing potential to mitigate overfitting effects and prediction bias compared with single base models under this evaluation setting. The method achieved an R² of 0.8559 and an RMSE of 0.0363 for tomato AGB estimation. This study provides a novel technical strategy for accurate and multidimensional growth status estimation of greenhouse tomato seedlings.