Refining grapevine vegetation status by utilizing grassland and remote sensing data.

Area of Science:

• Agricultural Science
• Remote Sensing
• Ecology

Background:

• Vegetation indices (VI) are crucial for monitoring plant health and environmental stress.
• Normalized Difference Vegetation Index (NDVI) is widely used, but its accuracy can be limited by factors like soil type and topography.
• Integrating data from diverse sources, such as grasslands and satellite imagery, can enhance vegetation monitoring capabilities.

Purpose of the Study:

• To investigate specific vegetation indices (VI) in grassland and vineyard ecosystems.
• To evaluate the potential of grassland remote sensing (RS) data to refine NDVI values in vineyards.
• To compare the performance of different machine learning models in predicting grapevine VI.

Main Methods:

• Field monitoring of NDVI, Photochemical Reflectance Index (PRI), and Photosynthetically Active Radiation (PAR) at grassland and vineyard sites.
• Acquisition and analysis of Sentinel-2 (S2) spectral data.
• Application of machine learning techniques, including linear regression (LR), random forest (RF), and XGBoost, to refine NDVI measurements.

Main Results:

• Significant differences in VI were observed between sites, correlating with soil chemistry.
• NDVI indicated overall canopy vigor, while PRI showed higher sensitivity to short-term physiological changes and stress.
• Ground-based and RS NDVI showed good correlation (r=0.68).
• The RF model, trained with day of year and grassland data, achieved the highest accuracy (r=0.787).
• All machine learning models showed improved grapevine VI prediction when grassland NDVI was included in training.

Conclusions:

• Vegetation indices vary significantly based on ecosystem type and soil properties.
• Grassland remote sensing data can enhance the accuracy of vineyard vegetation monitoring.
• Machine learning models, particularly Random Forest, show promise for refining NDVI measurements and assessing grapevine health.