What question did this study set out to answer?

The study aims to enhance yield prediction and decision-making in Mediterranean vineyards by integrating diverse data sources with machine learning techniques.

March 21, 2026Open Access

Improving Yield Prediction and Decision-Making in Mediterranean Vineyards Using Machine Learning on Multisource Remote Sensing and Soil Data

Key Points

The study aims to enhance yield prediction and decision-making in Mediterranean vineyards by integrating diverse data sources with machine learning techniques.
Used multi-temporal satellite data, UAV imagery, and soil measurements.
Applied machine learning algorithms like KNN, SVM, DT, and RF for analysis.
Evaluated models using metrics such as Pearson’s correlation coefficient and RMSE.
KNN achieved the highest prediction accuracy with PlanetScope imagery (r = .9988, RMSE = 0.0105).
Statistically significant differences were observed in vegetation indices based on sensor type (p < .05).
The methodology allows for better resource management and sustainable farming practices.

Abstract

Despite advances in remote sensing, vineyard yield prediction often suffers from limited multi-source data integration, low spatial resolution, and a lack of reproducibility in Mediterranean environments. This study addresses these gaps by presenting an innovative, standardized methodology that integrates multi-temporal satellite data (Sentinel-2, PlanetScope, and VHR), high-resolution UAV imagery (1–3 cm 2 ), and soil property measurements with machine learning (ML) algorithms to improve yield prediction and crop management. The research was conducted in a 9.2 ha experimental vineyard in Villamena, Granada (Spain). A statistically significant difference was detected between vegetation indices obtained from different sensors ( p < .05), highlighting that sensor type strongly affects the measured index values. Vegetation indices and soil variables were analyzed using advanced ML algorithms (KNN, SVM, DT, and RF) to identify key patterns influencing crop vigor and performance. Model evaluation metrics included Pearson’s correlation coefficient, Mean Absolute Percentage Error (MAPE), Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE). Among the tested models, KNN achieved the highest accuracy, particularly when using PlanetScope imagery ( r = .9988, RMSE = 0.0105). The findings demonstrate the practical value of data integration: UAVs provide precise, site-specific monitoring during key growth phases, while satellite imagery supports broader temporal and spatial coverage. By applying this methodology, farmers and agronomists can make informed decisions regarding fertilization, irrigation, and harvest timing, optimizing resource use and enhancing sustainability. Furthermore, the developed models are transferable to other Mediterranean vineyard regions. This work provides a concise framework for implementing precision viticulture strategies, bridging the gap between advanced remote sensing analytics and actionable agricultural management.

Improving Yield Prediction and Decision-Making in Mediterranean Vineyards Using Machine Learning on Multisource Remote Sensing and Soil Data

Key Points

Abstract

Cite This Study

Also Consider

Also Consider