Integration of agronomic information, vegetation indices (VIs), and meteorological data for phenological monitoring and yield estimation of rice (Oryza sativa L.)

Fernandez-Jibaja J.A.; Atalaya-Marin N.; Álvarez-Robledo Y.A.; Taboada-Mitma V.H.; Cruz-Luis J.; Tineo D.; Goñas M.; Gómez-Fernández D. · Smart Agricultural Technology (2025) · DOI: 10.1016/j.atech.2025.101203

Resumen

Rice (Oryza sativa L.) is a staple crop for sustaining global food security and is particularly important in tropical and subtropical regions. In this context, precision agriculture enables more efficient crop management to increase productivity and sustainability. This study proposes an integrated framework for monitoring the phenological development and estimating the yield of O. sativa by combining agronomic variables, vegetation indices (VIs), and meteorological data. Six rice varieties (Victoria, Esperanza, Bellavista, Puntilla, Capoteña, and Valor) were evaluated across six phenological stages using field data, 20 VIs and meteorological parameters. Field data revelated greater tillering of the Puntilla and Valor varieties (9–28 tillers), with Esperanza having the most stable chlorophyll values (21.5–38.7, σ = 10.46) during ripening. The temporal dynamics of the VIs consistently increased from the seedling to inflorescence emergence stage, followed by a decrease during flowering and ripening, which aligns with known physiological transitions in rice, however, significant differences in the NDVI index were detected during ripening (p > 0.05). For yield estimation, feature selection was performed using principal component analysis (PCA) and the least absolute shrinkage and selection operator (LASSO) to increase model efficiency and interpretability. Among the regression algorithms tested, support vector regression (SVR) demonstrated the highest predictive accuracy (R² = 0.81) for the Bellavista variety at the maximum tillering stage. Furthermore, the Valor variety presented the highest grain yield (13.70 t/ha). These results underscore the potential of integrating multisource data with machine learning techniques for high-resolution phenological monitoring and varietal performance assessment.