Fruit cracking in pomegranate is one of the biggest concerns for pomegranate growers. A single cracking event close to harvest can reduce market quality, affect storage potential, and lead to financial losses. But what if growers could identify cracking risks earlier in the season? And how can monitoring technologies support decisions in the orchard before visible damage appears?
This blog presents recent progress from the CrackSense pilot activities focused on pomegranate production in Greece and Israel. The work combines field observations, environmental monitoring, UAV imagery, and Artificial Intelligence models to better understand how fruit cracking in pomegranate develops under different growing conditions.
Pilot activities in Greece
The pomegranate pilot activities in Greece, led by Agricultural University of Athens, were carried out over three cultivation seasons between 2023 and 2025 in the Dalamanara area near Argos, Peloponnese. This region is widely known for intensive pomegranate cultivation and provided the basis for long term monitoring activities.
During the 2023 and 2024 seasons, monitoring focused on a commercial orchard planted with the ‘Wonderful’ cultivar of Punica granatum. A stable group of nine trees was monitored each year.
In 2025, the pilot activities expanded further. Three additional pilot fields were included, increasing the monitoring network to 36 trees under different environmental and soil conditions. This wider coverage supported broader validation of the predictive models across different environments.
Irrigation Management and Cracking Susceptibility
A major part of the experimental work examined how irrigation management influences fruit cracking and yield formation. Three irrigation strategies were applied between March and October during the 2023 and 2024 seasons.
The first treatment followed the local standard irrigation protocol. The second reduced irrigation volume by 30%, while the third increased water application by 20%. These treatments allowed researchers to study how water availability affects plant physiology, fruit growth, and cracking susceptibility.
Would reduced irrigation increase stress related cracking, or could excess water create additional pressure on fruit development? These were among the questions guiding the pilot activities.
Combining UAV Monitoring With Field Observations
To support the analysis, UAV campaigns were conducted across fifteen monitoring dates during the three seasons. The flights targeted important fruit development stages before harvest.
Two UAV platforms were used for image acquisition. The DJI Mavic 3 Multispectral captured RGB* and multispectral imagery, while the DJI Mavic Thermal collected canopy temperature observations.
Flights were carried out during midday hours to ensure consistent light conditions and minimise shadow effects. Images were captured using calibrated sensors to maintain comparable image quality across all monitoring campaigns.
Remote sensing observations were combined with physiological field measurements, soil observations, and meteorological data collected through the TOMMY monitoring platform. This integrated dataset enabled detailed analysis of plant physiological responses and environmental conditions at the tree level.
Machine Learning Models For Yield And Cracking Prediction
The modelling incorporated four main categories of input variables:
- Vegetation indices derived from multispectral UAV imagery, including sixteen spectral and derived indicators such as NDVI*, SIPI*, MSAVI*, OSAVI*, and several CW* wavelength bands
- Meteorological variables, including hourly environmental parameters such as air temperature, dew point, vapour pressure deficit (VPD)*, relative humidity, precipitation, and wind speed from a nearby meteorological station
- Physiological measurements, including stem water potential, stomatal conductance, trunk diameter, fruit diameter, soil moisture, and soil texture at depths of 30 cm and 60 cm
- Phenological observations, recording the BBCH* growth stage for each sampling event
All datasets were quality checked by outlier detection, removal of inconsistent measurements, and temporal alignment across different data sources. UAV imagery was processed to extract vegetation indices, which were then linked to individual tree locations using spatial data. This allowed remote sensing information to be directly combined with field measurements for machine learning analysis.
The collected datasets were used to train Random Forest models for predicting yield and fruit cracking in pomegranate. The model showed strong performance during training, however, performance decreased on independent testing datasets. This suggests that additional data and further model refinement are still needed to improve generalisation across orchards and seasons.
Variable importance analysis showed that multispectral vegetation indices were the most important predictors of yield. In particular, MSAVI, IPVI*, and NDVI had the highest importance, as they reflect canopy vigour and photosynthetic activity. Other indices such as OSAVI, GNDI*, and SAVI* also contributed to the model, while NDRE* had a smaller effect. Overall, these results suggest that canopy structure and physiological characteristics are strongly linked to yield variability in pomegranate orchards.
A Random Forest model was also developed to predict fruit cracking using physiological variables (fruit diameter, trunk diameter, stomatal conductance), meteorological data (rainfall, air pressure), and vegetation indices. The model performed well on the training data, but again showed lower accuracy on the test data, indicating limited ability to generalise to new conditions.
Feature importance analysis showed that fruit diameter and trunk diameter were the strongest predictors of cracking. This highlights the role of fruit growth and tree vigour in cracking susceptibility. Vegetation indices such as NDVI, Clred*, IPVI, and MSAVI also contributed by describing canopy health and physiological status. Meteorological factors such as rainfall and air pressure further supported the model, suggesting that environmental conditions interact with plant physiological responses to influence cracking occurrence.
Pomegranate Pilot Activities In Israel
Developing Predictive Models For Fruit Cracking
The pilot activities in Israel, led by Volcani Institute, focused on developing a prediction model fruit cracking in pomegranate using physiological measurements, meteorological data, and UAV based remote sensing observations.
The experimental work was conducted with pomegranate (Punica granatum L.) cv. ‘Wonderful’ during the 2023 and 2024 seasons in a commercial orchard located at Kibbutz Tsor’a in central Israel. Researchers monitored stem water potential, stomatal conductance, leaf area index, trunk and fruit growth rates, and cracking incidence at harvest.
UAV-based remote sensing observations included multispectral, thermal, and LiDAR* approaches for estimating plant water status at both tree and plot scales. The collected data were integrated into machine learning models designed to predict fruit cracking under varying environmental and orchard management conditions.
The Role Of Canopy Development
One of the strongest findings from the Israeli pilot activities was the relationship between plant area and fruit cracking incidence.
Analysis of September data, approximately one month before harvest, showed that trees with smaller canopy areas had considerably higher cracking incidence, while larger and more developed canopies were associated with lower cracking levels. The Random Forest prediction model confirmed this trend and showed good agreement between measured and predicted cracking rates.
Could canopy structure become one of the most practical indicators for growers monitoring cracking risks in commercial orchards? The pilot results suggest that this may be possible.
Testing The Predictive Model In Commercial Orchards
In 2025, the activities progressed to a dedicated pilot experiment designed to test the predictive model under commercial orchard conditions.
Three orchard plots were selected for the pilot observations. One plot was located in Kibbutz Tsor’a, while two additional plots were established in Kibbutz Mishmar Hanegev in the Negev region.
The orchards differed in climate conditions and fruit quality characteristics, allowing researchers to test the model under varying production environments. QGIS* and R software were used to identify variability zones within each orchard and optimise tree selection across the plots.
For every orchard, twelve trees were selected from varied areas. Physiological, climatic and remote sensing data collection was carried out during three fruit development stages based on skin colour progression, from early green fruit to harvest stage. The number of total fruits and cracked fruits was recorded for each monitored tree during harvest.
The pilot observations again confirmed that plant area remained one of the most influential predictors of cracking incidence.
Practical Considerations And Next Steps
The results also highlighted several practical considerations for growers managing fruit cracking in pomegranate.
- First, the canopy area showed a strong relationship with both yield and fruit cracking. Yield generally increased with larger canopy development when mid-season data were used, showing that canopy development during this period is a strong indicator of final yield. In contrast, fruit cracking decreased as plant area increased in late-season data, but the relationship was not consistent across all tree sizes, meaning the effect varied rather than following a simple pattern.
- Second, the methodology is considered transferable to any pomegranate orchard because the required tools and parameters are already accessible to growers. At the same time, canopy or plant area may depend on tree-shaping protocols that are specific to local traditions. Naturally, the pomegranate plant has a bushy architecture, but we apply trellising to the branches to allow the full potential of canopy expansion. Thus, the predictive model may only be applicable to specific pomegranate plant architectures.
- Looking ahead, the Israeli team is now preparing for a second year of piloting observations. The goal is to further verify the factors that determine pomegranate fruit cracking and to validate the power of the predictive model.
Towards More Informed Orchard Management
Together, the pilot activities in Greece and Israel show how integrated monitoring and data analysis can support earlier detection of fruit cracking in pomegranate and contribute to more informed orchard management practices.
As more data are collected across seasons and regions, the predictive models can become more robust and scalable, allowing their integration into the CrackSense Spatial Decision Support System. This could ultimately support growers across different production areas in reducing yield losses, improving fruit quality, and enhancing the sustainability of pomegranate production systems.
Further updates from our other pilots will explore how these methods translate to different fruit types and regional conditions. You can follow progress through our Newsroom or on LinkedIn.
Abbreviations
* RGB – Red, Green and Blue imagery used in standard colour imaging.
* NDVI – Normalized Difference Vegetation Index, a spectral index used to assess vegetation health and greenness.
* SIPI – Structure Insensitive Pigment Index, a vegetation index related to plant pigment content and stress.
* MSAVI – Modified Soil Adjusted Vegetation Index, a vegetation index designed to reduce soil background influence.
* OSAVI – Optimized Soil Adjusted Vegetation Index, a spectral index developed to minimise soil effects in vegetation analysis.
* CW – Continuous Wave wavelength bands used in spectral analysis.
* VPD – Vapour Pressure Deficit, an indicator of atmospheric moisture demand affecting plant water stress.
* BBCH – Biologische Bundesanstalt, Bundessortenamt and Chemical industry scale, a standard system for describing plant growth stages.
* IPVI – Infrared Percentage Vegetation Index, a spectral index related to vegetation density and health.
* GNDI – Green Normalized Difference Index, a vegetation index associated with plant vigour and nitrogen status.
* SAVI – Soil Adjusted Vegetation Index, a vegetation index developed to reduce soil brightness effects.
* 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.
* LiDAR – Light Detection and Ranging, a remote sensing method using laser pulses to measure canopy structure and terrain.
* QGIS – Quantum Geographic Information System, open source software used for spatial data analysis and mapping.