Digital transformation in space management: AI solutions for Vodnjan and Šolta with the support of EDIH Adria

Challenges

The town of Vodnjan and the Municipality of Šolta, as local self-government units in Adriatic Croatia, face similar challenges in managing their territory. Preserving space, preventing illegal construction, identifying illegally disposed waste (wild dumps) and monitoring the contracted use of agricultural land leased for years are key tasks. Traditional methods of monitoring, which rely on field surveys of municipal monitors and manual verification of documentation, are often slow, resource-intensive and insufficiently effective to cover the entire area of local self-government units, especially given the limited human capacities. There is therefore a clear need to modernise and introduce digital tools to improve the monitoring and efficiency of utilities.

Solution

Through the “Test Before Invest” (TBI) programme and with the expert support of the EDIH Adria consortium, the City of Vodnjan and the Municipality of Šolta tested the potential of the application of artificial intelligence (AI) and analysis of satellite and aero-photo images to solve these challenges. The aim was to examine whether AI models can automate the detection of changes in space and provide timely information to the relevant services.

As part of the TBI activity, various machine learning approaches and AI models were tested:

• Detection of objects (buildings and cultivated surfaces): Models such as YOLO (You Only Look Once) architecture and OWL-ViT (Open World Vision Transformer) were tested. The aim was to automatically identify buildings and determine whether there is a building permit for them, and to distinguish processed from uncultivated agricultural areas by comparing the images over time. Prototypes, such as the “SatelliteGuard” system developed for Vodnjan, demonstrated a high-precision detection capability based on locally relevant data.

• Detection of Wild Landfills: To identify irregular patterns that point to wild dumps, more advanced models such as Google Gemini 2.0 Flash, capable of analyzing complex visual patterns and classifying areas as suspicious of illegal waste disposal, have been tested, even with a limited number of examples for training specific detection models.

Solutions were tested using publicly available images (e.g. Geoportal, Regulated Land) and potentially commercial satellite images, and a prototype solution was developed to visualize detection results through simple user interfaces (e.g. based on Streamlit UI).

Results and Benefits

Testing conducted in the City of Vodnjan and the Municipality of Šolta has shown significant potential of AI technologies for improving space management, enabling:

• Increased efficiency: Automated image analysis can drastically reduce the time it takes to identify potential space irregularities compared to manual fieldwork.
• Improved accuracy and consistency: AI models provide objective and consistent analysis, reducing the possibility of human error or detection failure.
• Better resource allocation: Municipal guards can focus their time and resources on checking locations that the AI system has flagged as highly suspicious, rather than random or extensive field inspections.
• Timely action: Faster detection allows for a faster reaction of the competent services, preventing further devastation of space or consolidation of illegal objects.
• Support to spatial planning and environmental protection: The system provides valuable data for monitoring changes in land use, planning the development and implementation of environmental protection measures.
• High application potential: Tested YOLO models (e.g. in the City of Vodnjan) achieved high metrics (mAP > 80%), while models for waste detection showed high accuracy in binary classification (>80%), even without a specific fine-tuning for local data.

Lessons learned

During the TBI phase, key insights for successful implementation were identified:

• Data quality and relevance: For optimal results, it is crucial to train the model on high-quality satellite/aero images that reflect specific local characteristics (architecture, vegetation, soil types).
• Need to adapt the model: Generic, pre-trained models are often not enough. Fine tuning or custom model development (e.g. YOLOv10 based on local data) is required to achieve the desired precision.
• Combined approach: Using different AI models (e.g. YOLO for objects, Gemini/OWL-ViT for complex samples) can provide a more comprehensive solution.
• Integration with GIS systems: It is essential to ensure the integration of AI solutions with existing GIS (GeoInformation System) platforms used by city/municipal services to facilitate the use and visualisation of data.
• User interface: It is necessary to develop an intuitive interface that allows easy use of the system and interpretation of results by staff without advanced technical knowledge.
• Acquisition strategy of recordings: It is necessary to define the optimal strategy for the acquisition of satellite images (frequency, resolution, price) through the analysis of available commercial providers.

Estimated effects of implementation

Based on testing prototype solutions and analyzing existing processes, it is estimated that the full implementation of AI systems could bring significant improvements in space management. The time needed for the preliminary identification of potentially suspicious sites is expected to be drastically reduced compared to lengthy manual inspections or fieldwork. This acceleration would directly lead to the optimization of the field work of communal monitors, as it would reduce the need for indiscriminate surveys, allowing them to focus their resources on checking locations that the AI system has labeled as high-risk, resulting in a more efficient use of their working hours. Furthermore, systematic and continuous aerial surveillance increases the likelihood of early detection of illegal activities and wild dumps that might otherwise go unnoticed. The implementation would also enable more regular and systematic monitoring of changes in the use of agricultural land and other important elements in space.

Conclusion

“Test Before Invest” projects implemented in the City of Vodnjan and the Municipality of Šolta have successfully demonstrated that the application of artificial intelligence and analysis of satellite images is a feasible and very potent solution for modernizing the supervision and management of space in local self-government units. Acquired insights and developed prototype models provide a solid basis for making informed decisions about procurement and implementation of complete commercial solutions. EDIH Adria continues to support the City of Vodnjan, Šolta Municipality and other local self-government units as well as small and medium-sized enterprises (SMEs) in their digital transformation efforts, helping them to take advantage of advanced technologies for more efficient, transparent and sustainable management of their resources