AI identifies 10,000 previously unrecorded field islands in Estonia

Researchers at the University of Tartu have used artificial intelligence to identify 10,000 field islands in Estonia that were missing from existing records, ERR reported. Farmers can receive some support for preserving the features, which benefit nature.

Field islands are small wooded patches, often containing piles of stones, in the middle of fields. They form when stones are cleared from farmland and can become important habitats for plants, animals and birds.

Farmers currently have to mark field islands manually on the PRIA website when applying for European Union funding. The process takes time, and the resulting data quickly becomes outdated. Researchers trained an artificial neural network using images of nearly 15,000 known field islands, then used it to analyse orthophotos.

Farmers have recorded nearly 30,000 field islands, but the model found another 10,000. Researcher Holger Virro said this did not mean the existing records were wrong, but showed how difficult they were to keep up to date.

The Agricultural Registers and Information Board, known as PRIA, has received the map data and is considering whether to use it for automated monitoring. The researchers hope PRIA will eventually run the model itself. They also want to use similar methods to map hedges and stone walls, and plan to explore satellite data and expand the research across Europe.