Abstract
In the context of the climate crisis and increasing sustainability demands, the transition to precision agriculture is imperative to optimize input use and ensure global food security. This shift requires tools capable of capturing intra-plot heterogeneity through an integrated management approach. Interpreting the physical interaction between solar radiation and the crop canopy is of high strategic value, and leveraging the multispectral capabilities of the Sentinel-2 constellation can develop robust technological solutions.
Research demonstrates that analyzing spectral signatures, particularly in the Visible, Red Edge, and SWIR regions, enables superior physiological characterization of the crop compared to traditional vegetation indices. Under this premise, spectral reflectance serves as the cornerstone for modeling in decision-support tools. As a successful application, this study demonstrates its use in variable-rate nitrogen fertilization (VRT) and yield forecasting. Regarding variable fertilization, applying Random Forest algorithms to satellite data minimized the need for in situ monitoring, achieving a dosage precision of R2 = 0.97. Simultaneously, integrating multitemporal data with Machine Learning models (XGBoost and RF) enabled the anticipation of final production, achieving an R2 of 0.85. This predictive capacity captures spatial variability throughout the phenological cycle, facilitating the implementation of corrective actions to maximize yield.
These results underscore the potential of remote sensing as an advanced decision support system. By combining dynamic diagnostics with yield forecasting, this approach offers a technical solution that maximizes resource efficiency and ensures economic viability, transforming traditional agronomic management into a predictive, site-specific ecosystem.
