Evaluating site-specific management of varieties and nitrogen fertilization across management zones in wheat

Wheat (Triticum aestivum L.) remains a cornerstone of global agriculture; however, its production continues to face substantial spatial and temporal variability driven by differences in soil properties, topography, and weather patterns. This heterogeneity complicates the implementation of uniform management practices and results in significant yield variation even within individual fields. This study, conducted in the Río Negro Department, Uruguay, aimed to quantify the effect of assigning wheat varieties and nitrogen fertilization strategies differentially across zones defined by their current productivity, soil properties, and terrain attributes (management zones, MZs). To achieve this objective, innovative precision agriculture tools and Bayesian statistical inference were applied, combining geostatistical methods and hierarchical spatial modeling (INLA)—an advanced methodology that surpasses traditional statistical approaches—enabling capture of real within-field variability in a large-scale experiment established under commercial farming conditions (10 ha). The experiment consisted of a factorial combination of two wheat varieties ("Don Alberto" and "Nogal") and three nitrogen fertilization strategies across four management zones. Results demonstrated that the effectiveness of site-specific management strongly depended on the interaction between wheat variety and management zone. Significant responses were observed in the most restrictive zone for wheat growth, characterized by a greater risk of excess-water limitations. Under these conditions, the variety "Nogal" achieved the highest performance when combined with incremental nitrogen fertilization. The optimized site-specific management strategy produced an average yield gain ranging from 0.17 to 0.51 Mg ha⁻¹ compared with conventional uniform management, corresponding to a total yield increase of 1.7 to 5.1 Mg across the entire 10-ha study area. The study concludes that explicit spatial modeling is essential for capturing treatment effects in commercial-scale experiments, as models incorporating spatial structures consistently outperformed simpler alternatives. The Bayesian framework implemented through R-INLA provides a robust decision-support tool, enabling probabilistic quantification of production scenarios and associated risks. Although the experimental evidence is limited to a single growing season, which constrains direct extrapolation to different climatic scenarios, this study establishes a reproducible and transferable methodology for precision agriculture, with future potential to incorporate multi-year data and informative priors derived from historical agronomic data, which would substantially improve predictive accuracy and the reliability of site-specific management recommendations for commercial producers.