Back to search

Agricultural University of Athens

Investigating the application of precision agriculture and big data analytics in vineyards

Abstract

dc:description

The use of vegetation indices, specifically the normalized difference vegetation index (NDVI), is common in the production of wine grapes as a way to differentiate management units (MU) and explain variability in quantity and quality. This study compares NDVI data from four different sources – two terrestrial crop reflectance sensors and UAV and Sentinel-2 imagery – over two seasons to determine the similarity of the data and its ability to explain grape yield and quality. CropCircle, SpectroSense+GPS, Parrot Sequoia+ multispectral camera equipped UAV, and Sentinel-2 imagery were used to collect high-resolution multispectral multi-temporal data over two seasons (2019 and 2020). The study aims to investigate the differences between vegetation index (VI) maps derived from various sensing systems in a production context and to determine the extent to which NDVI values obtained from these different systems are correlated and how well they can explain variations in yield and quality. Data pre-processing techniques, including data quality assessment, data interpolation onto a grid (20 m x 10 m), and data normalization, were applied to the data. Methods, such as exploratory correlation analysis, principal component analysis, and analysis of variance and the creation of management unit maps were used to compare the data and determine how similar the sources were and identify relationships between NDVI values and grape yield and quality. This study also explores the use of machine learning techniques and NDVI data collected at different growth stages to predict wine grape quality characteristics non-destructively. The transformed data was then used to train and test statistical and machine learning algorithms, including linear and nonlinear regression models and ensemble methods based on decision trees. Finally, this study investigates the use of automated machine learning (AutoML) techniques in combination with NDVI data to predict grape quality attributes in vineyards. Ordinary Least Square (OLS), Theil-Sen and Huber regression models, and tree-based methods were used in the analysis. Support Vector Machines (SVMs) and Automatic Relevance Determination (ARD) were also included. Results show strong similarities between NDVI values obtained from similar sensors, but differences between proximal and remote sensing observations. NDVI data had high correlations with grape yield and total soluble solids, but no relationship was found with must titratable acidity or pH. Proximal sensors were found to better explain variability in yield and quality in the early season and later growth stages, while satellite data was found to be less consistent due to its coarse spatial resolution. NDVI values obtained from different sensors, especially satellite platforms, generated differing information for decision making and were not equally useful when converted to production decisions. Proximal sensors were more accurate in predicting wine grape quality parameters in the early season, while remote sensors were more accurate during later growth stages. The UAV and Spectrosense+GPS data were found to be particularly accurate in predicting sugar content, especially during mid-late season with full canopy growth. Regression models using AutoML (R2 = 0.65) provided slightly improved results compared to those using manually fine-tuned machine learning (R2 = 0.61). When combining data from multiple sensors and growth stages within a single year, the coefficient of determination (R2) increased to an average of 0.66 for the best-fitting regressions. When considering combinations of sensors and growth stages across both cropping seasons, UAV and Spectrosense+ GPS, as well as Véraison and Flowering, each had the highest average R2 values. These results suggest that AutoML has potential for long-term performance and may be a useful tool in improving the efficiency of crop quality prediction. Overall, the results suggest that proximal sensors are more accurate in predicting wine grape quality characteristics in the early season, while remote sensors are more accurate during later growth stages. Combining data from multiple sensors and growth stages can improve the accuracy of predictions. The use of NDVI data in combination with machine learning techniques may provide a more efficient and cost-effective method for predicting wine grape quality characteristics.

Degree

thesis:*
Grantor dc:publisher
Agricultural University of Athens
Year dc:date
2023

Author and committee

dc:creator, dc:contributor.*
Authors dc:creator
  • Kasimati, Aikaterini
  • Κασιμάτη, Αικατερίνη

Subjects

dc:subject × 21

Rights

Language dc:language
eng

Identifiers

dc:identifier.*
Identifier
10.12681/eadd/53898

Chain of custody

source
Harvested from
Greek National Archive of PhD Theses
Base URL
phdtheses.ekt.gr/eadd_oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
citation

Kasimati, Aikaterini; Κασιμάτη, Αικατερίνη. Investigating the application of precision agriculture and big data analytics in vineyards. Agricultural University of Athens, 2023. http://hdl.handle.net/10442/hedi/53898