ARPA, a Malta-based government agency, needed an AI solution to predict land use in Malta, Gozo, and Comino from satellite images. Smart Studios developed a custom AI model tailored to the unique vegetation and field sizes of the Maltese Islands.
The vegetation of the Maltese Islands was distinct compared to other regions, which meant that off-the-shelf AI models, trained on generalized datasets, performed poorly in predicting local land use. ARPA needed a custom AI solution that could recognize these specific characteristics.
The agricultural fields in Malta, Gozo, and Comino were much smaller than those seen in other parts of Europe, further complicating land use detection. Existing AI models struggled to accurately interpret such small-scale fields, necessitating a specialized approach.
A critical task for ARPA was to detect any false land use claims made by farmers in their EU grant applications. The AI system needed to compare the model’s predicted land use against the registered land use in order to identify discrepancies, including false registrations.
A custom AI model had to be developed and trained on local data, including satellite imagery specific to the Maltese Islands. Gathering and preparing this training data to reflect the landscape and field sizes in Malta was essential.
Manually verifying land use claims is a time-intensive process, taking months to complete. ARPA needed an automated solution that could significantly reduce the time required to detect discrepancies.
The output of the AI model had to integrate seamlessly with ARPA’s existing data formats, particularly in the form of GeoPackage (GPKG) files, preserving geometry and enabling easy querying of the attribute tables to detect changes.
Smart Studios partnered with ARPA to create a custom-built AI model designed to meet the specific challenges of land use prediction and verification in Malta. The project was rolled out in two main phases:
Smart Studios initially conducted a POC to test whether an off-the-shelf AI model could be used or if a custom-built AI model was needed. The POC revealed that the unique vegetation and small field sizes in Malta required a custom solution.
The team developed a custom AI model, trained on local satellite data, that could accurately predict land use across the islands, providing much better results than generic models.
Following the successful POC, ARPA extended the project to fully integrate the AI model into its grant management workflow.
The AI system automatically compared the land use declared by farmers with the predicted land use and flagged any discrepancies, allowing ARPA staff to quickly detect false registrations.
The output was provided in the form of GPKG files, which preserved the geometries of the fields and applied necessary SQL queries to the attribute tables. This allowed ARPA’s team to seamlessly view and analyze discrepancies between registered and actual land use.
Smart Studios developed a specialized training dataset using satellite imagery and local geospatial data from Malta, Gozo, and Comino. This ensured that the AI model was tailored to recognize the distinct characteristics of the Maltese landscape and the small size of the fields.
The AI model was designed to automatically compare predicted land use with the registered land use in farmers’ EU grant applications. When discrepancies were detected, the system flagged these instances, enabling ARPA staff to investigate potentially false land use claims efficiently.
To ensure compatibility with ARPA’s existing systems, the AI model’s output was formatted as GPKG files. These files maintained field geometries and applied SQL queries to attribute tables, allowing ARPA to easily detect changes and identify discrepancies without requiring extensive manual processing.
Manually verifying land use claims could take several months. By automating the process, the AI model reduced this timeframe to a matter of days or weeks. This significant time saving allowed ARPA to process grant applications more quickly and efficiently, improving overall operational efficiency.
The AI model’s ability to automatically compare predicted and registered land use significantly improved ARPA’s ability to detect false land use registrations, ensuring that EU grants were awarded based on accurate and verified data.
The automation of the land use verification process saved ARPA several months of manual labor. What once took months to verify could now be done in a few days, allowing ARPA to allocate resources more effectively.
The AI model’s output in GPKG format ensured a smooth integration with ARPA’s existing systems. The preserved geometry and applied SQL queries allowed ARPA staff to efficiently detect discrepancies between registered and predicted land use, improving the accuracy and efficiency of the grant approval process.
The custom-built AI model was specifically designed for Malta’s unique landscape, ensuring that ARPA received accurate predictions that generic off-the-shelf models could not provide. This local data-driven approach led to much higher accuracy in land use predictions.
The AI system was designed to be scalable, allowing ARPA to extend the use of the model to future projects or different geospatial analysis tasks, enhancing its overall value to the agency.
The partnership between ARPA and Smart Studios resulted in the development of a custom AI solution that not only improved the accuracy of land use predictions but also streamlined ARPA’s grant management process. The AI model successfully reduced the time and effort needed to detect false land use registrations, allowing ARPA to process grant applications more efficiently. The seamless integration of AI-generated data into GPKG files, along with the model’s ability to compare predicted and registered land use, made the system a vital tool for ensuring the integrity of EU grant distribution in Malta.
This project demonstrated Smart Studios’ expertise in developing tailored AI solutions that address unique local challenges while delivering measurable improvements in efficiency, accuracy, and business process automation.
