Fruit cracking in grape production remains a concern for growers across different climate zones. Weather conditions, irrigation practices, and vineyard management can all influence how berries develop and respond to stress during the season.
Within the CrackSense project, pilot activities in Greece and Israel are helping researchers better understand how environmental conditions affect grape production and cracking incidence. The work combines field observations, UAV imaging, meteorological monitoring, and machine learning models.
Why does this matter for growers? Because understanding stress signals earlier may support better vineyard decisions before losses become visible in the field.
Greece Pilot Activities In Table Grape Vineyards
Monitoring Vineyards Across Multiple Growing Seasons
The Greek experimental and pilot activities, led by Agricultural University of Athens and Pegasus Agrifood Coop, were conducted during three consecutive growing seasons between 2023 and 2025 in vineyards near Kiato in the northeastern Peloponnese, a region with a long viticulture tradition.
During the 2023 and 2024 seasons, researchers monitored 32 vines under controlled experimental conditions. In 2025, the activities expanded to three commercial vineyards, with 12 vines selected in each field. This expansion allowed the team to test the developed models under different pedoclimatic conditions and vineyard management practices.
The Greek pilot aimed to collect detailed field and remote sensing data to better understand how environmental conditions, plant physiological status, and management practices influence fruit cracking and yield variability in vineyards. To achieve this, researchers combined ground measurements, UAV-based remote sensing, and meteorological observations to capture vine performance throughout multiple growing seasons.
Figures 1–3: Field data collection and monitoring activities
Testing Irrigation, Growth Regulators, And Fertilisation
The experimental setup included three separate sub-trials focused on irrigation regimes, plant growth regulators (PGRs), and fertilisation practices on vine physiology and productivity.
The irrigation trial compared two watering strategies between March and August. The first treatment consisted of a high-frequency irrigation regime, where water was applied weekly for five hours, while the second followed a lower frequency schedule with irrigation every two weeks for the same duration and rate.
Researchers also evaluated how plant growth regulators influenced berry development. Applications included GA₃* treatments aimed at promoting cell elongation and berry enlargement, alongside CPPU* applications designed to stimulate cell division. A separate CPPU treatment was applied in repeated applications during the early fruit development phase to help maintain berry firmness. Untreated vines were maintained as controls.
A third trial focused on fertilisation practices. The team compared balanced NPK* applications applied at the recommended commercial rate, intensive fertilisation regime, and reduced fertilisation treatment used as controls.
In 2025, the focus shifted from applying new treatments to validating the developed models under commercial vineyard conditions, providing additional environmental variability useful for model validation. This step helped researchers evaluate how well the models performed outside controlled experimental settings.
Combining UAV Imaging With Field Measurements
The monitoring framework combined UAV based remote sensing with physiological and environmental measurements collected directly in the vineyard.
Flights were conducted using the DJI Mavic 3 Multispectral and DJI Mavic Thermal platforms under stable weather conditions to ensure consistent image quality. The flights took place around midday with sufficient solar elevation to minimise shadow effects.
At the same time, researchers collected tree level measurements during four important phenological stages, from post fruit set to final ripening. Measurements included stem water potential, trunk diameter, berry weight, soil moisture, and final yield. Meteorological data such as air temperature, humidity, precipitation, wind speed, and vapour pressure deficit were collected through the TOMMY monitoring system.
Figure 4: TOMMY, a robotic system equipped with thermal and optical cameras to detect early stress indicators that may lead to cracking
The team also extracted vegetation indices from UAV imagery, including NDVI*, GNDVI*, MSAVI*, NDRE*, Clred*, and Clgreen*. These indices were spatially aligned with individual vine data to support machine learning analysis.
Yield Prediction Results And Future Outlook
One important observation from the Greek pilot was that no fruit cracking events were recorded during the monitored seasons. Prolonged heatwaves and dry climatic conditions likely reduced the occurrence of cracking in the studied vineyards.
As a result, researchers focused primarily on yield prediction and vine water status rather than cracking estimation.
The machine learning model used meteorological, multispectral, and phenological variables collected during the 2023 and 2024 seasons. The training dataset produced strong performance results, while prediction accuracy decreased when tested on independent datasets.
These results indicate that while the model successfully captured relationships within the training dataset, its ability to generalize to unseen data was limited, potentially due to the small sample size and seasonal variability.
Despite these challenges, the analysis identified that dew point, NDRE, and OSAVI* were among the most influential predictors of yield variation, followed by GNDI*, Clred, and maximum air temperature. Growth stage and sampling date also contributed to the model, highlighting the importance of seasonal crop development dynamics.
Overall, the results suggest that grapevine yield variability is strongly influenced by a combination of environmental conditions, canopy spectral characteristics, and phenological factors.
The Greek pilot activities demonstrated how combining UAV imaging, field measurements, and environmental monitoring can support precision viticulture approaches. Over time, expanding the dataset across additional vineyards and seasons may improve model reliability and strengthen decision support tools for growers.
Figure 5: Grapes in the monitored vineyard
Israel Pilot Activities In Commercial Vineyards
Monitoring Cracking Incidence In Diverse Vineyard Conditions
The Israeli pilot activities focused directly on predicting fruit cracking incidence and the related occurrence of sour rot in vineyards.
The experiments were conducted during 2025 in four vineyard plots near Lachish village (Figure 6). In each plot, researchers monitored 12 vines selected to represent variability within the vineyard.
Figure 6: Location of the pilot plots and the vines within each plot: A-the Lachish area with pilot plots 1 and 2 in the entrance to the village and pilots 3 and 4 in a valley south of the village. B- location of the vines in pilot plots 1 and 2 on a stratified map of the vineyards. C – location of the vines in pilot plots 3 and 4 on a stratified map of the vineyards.
One important finding was the strong variability observed both between vineyards and within the same vineyard. Growth conditions differed considerably even among nearby vines. Researchers observed large differences in fruit load within the same plot (Figure 7 and 8), highlighting how variable vineyard performance can be across relatively small areas.
Figure 7-8: Examples of high and low yield in nearby vines in pilot 4.
Cracking incidence also varied significantly between pilot plots. In pilot 1, there were hardly any cracked berries, while pilot 4 experienced damage affecting more than 40% of the clusters.
Figure 9: A cracked berry
These observations reinforced the importance of monitoring vineyards at a detailed spatial level rather than relying only on broader field averages.
Integrating Field Data With Machine Learning Models
The Israeli pilot combined UAV based multispectral imaging with physiological and environmental measurements collected directly in the field.
The objective was to develop machine learning models capable of predicting cracking incidence using the collected datasets.
At present, the research team is integrating the field observations with the modelling framework to evaluate how different environmental and physiological parameters influence cracking development.
The pilot activities also included the use of the TOMMY monitoring system during field measurements (Figure 10). The collected information will contribute to further model training and validation during future seasons.
Figure 10: The research team using the TOMMY system in pilot plot 2
What happens next? The team is already preparing for the 2026 pilot measurements, which are expected to begin shortly.
The continuation of the monitoring activities will help researchers expand the dataset, evaluate seasonal differences, and improve the predictive capacity of the developed models.
Looking Ahead
The grape pilot activities in Greece and Israel show how different environmental conditions can influence vineyard performance, stress responses, and cracking incidence.
As the CrackSense project moves into the next phase of pilot activities, the collected datasets and field experience will continue supporting the development of practical tools for vineyard management and cracking risk assessment.
Read more pilot insights in our Newsroom and follow us on LinkedIn for the latest project updates and field activities.
Abbreviations
* GA₃ – Gibberellic acid, a plant growth regulator used to promote cell elongation and berry enlargement.
* CPPU – Forchlorfenuron, a synthetic cytokinin used to stimulate cell division and improve berry firmness and development.
* NDVI – Normalized Difference Vegetation Index, a spectral index used to assess vegetation health and greenness.
* GNDVI – Green Normalized Difference Vegetation Index, a spectral index used to assess plant vigour, chlorophyll activity, and canopy condition.
* MSAVI – Modified Soil Adjusted Vegetation Index, a vegetation index designed to reduce soil background influence.
* NDRE – Normalized Difference Red Edge Index, a vegetation index sensitive to chlorophyll content and crop stress.
* Clred – Chlorophyll Red Edge Index, a vegetation index associated with chlorophyll content and plant physiological status.
* Clgreen – Green Chlorophyll Index, a vegetation index used to estimate chlorophyll content and monitor plant physiological status.
* OSAVI – Optimized Soil Adjusted Vegetation Index, a spectral index developed to minimise soil effects in vegetation analysis.
* GNDI – Green Normalized Difference Index, a vegetation index associated with plant vigour and nitrogen status.