Predictive modeling for prediction of technological maturity and microterroir zones of grapevines in subtropical region in Brazil

Abstract

Determining grape quality through conventional methods is a destructive and costly laboratory process that limits the understanding of spatial and temporal variability within vineyards. This study proposes to validate non-destructive predictive approaches, grounded in precision viticulture and machine learning, for monitoring technological maturity and delineating microterroir zones. The research focuses on the double-pruning regime (winter wine production) in subtropical regions of Brazil, investigated under two complementary perspectives of scale and data acquisition. In the first study, the focus was on interseasonal macroprediction (2024–2025) and microterroir zoning in a subtropical area. The integration of Growing Degree-Days (GDD) and Vegetation Indices (VIs) revealed the Random Forest algorithm as the most robust for the temporal prediction of Soluble Solids (SS, R² = 0.73) and Titratable Acidity (TA, R² = 0.71), although pH modeling exhibited performance degradation on independent datasets. Concurrently, K-Means unsupervised clustering integrated multiyear canopy vigor (via Google Earth Engine), topography, and soil proxies to delineate microterroir zones, achieving high efficacy in the stratification of macrophysiological traits (Variance Reduction of 33.62% for chlorophyll b). In the second study, a high-spatialresolution predictive framework was developed for the same management system. The fusion of agrometeorological data with multispectral imagery acquired by an Unmanned Aerial Vehicle (UAV) significantly outperformed the isolated use of VIs. The XGBoost algorithm achieved the highest accuracy (R² = 0.89 for SS; R² = 0.77 for TA), demonstrating that GDD and cultivar determine the primary maturation dynamics, while VIs capture the refinement of spatial variability in vigor at a sub-metric scale. In summary, it is concluded that integrating climatic data with remote sensing platforms (orbital and aerial) provides a robust analytical framework for precision viticulture in double-pruning areas, optimizing the assessment of technological maturity, the management of microterroir zones, and harvest decisions.