{"slug": "ai-discovers-thousands-of-overlooked-field-islands-in-estonia", "title": "AI discovers thousands of overlooked field islands in Estonia", "summary": "Researchers at the University of Tartu used AI to automatically locate field islands across Estonia, identifying nearly 38,000—about 34 percent more than the official register lists. The deep-learning model, trained on over 15,000 labeled examples from aerial photographs, found around 10,000 additional field islands, which are small patches of trees and shrubs that provide critical habitat for farmland birds and beneficial insects. The findings highlight the difficulty of manual mapping and the potential of AI to assist in environmental monitoring.", "body_md": "# AI discovers thousands of overlooked field islands in Estonia\n\nResearchers at the University of Tartu have used artificial intelligence (AI) to automatically locate field islands across Estonia and discovered around 10,000 more of them than are listed in the official register.\n\nField islands are tiny patches of trees and shrubs in the middle of crop fields that may look insignificant, but they are among the most valuable habitats in agricultural landscapes.\n\n**Small features, big impact**\n\nTo a farmer driving a tractor, a field island might simply look like an inconvenient obstacle to navigate around. Yet these small field islands offer nesting sites for farmland birds while providing homes for wild bees and beneficial insects that pollinate crops and control agricultural pests. Most field islands cover only 0.01–0.5 hectares — roughly the size of a small garden up to about half a football field — yet they provide disproportionately important habitat for wildlife. Because of their ecological importance, farmers can receive support for preserving them under the European Union's Common Agricultural Policy.\n\n**Teaching AI to recognize the landscape**\n\nAs these features are small and scattered across millions of hectares of farmland, keeping track of them by hand is an almost impossible task.\n\nUntil now, drawing these islands on maps was slow, tedious work and farmers had to draw them by hand in the Agricultural Registers and Information Board (ARIB) digital portal when applying for European Union support, leading to maps that were incomplete or out of date.\n\n\"Our goal was to teach a computer to recognize these small landscape elements automatically from aerial photographs,\" explains PhD student Afif Fauzan.\n\nUniversity of Tartu researchers trained a deep-learning model using more than 15 000 labelled examples created from high-resolution aerial photographs and existing field island maps drawn by farmers. In a sense, the AI learned from thousands of maps that farmers had already drawn when applying for agricultural support. The model learned to recognize the unique shapes and colors of field islands and tell them apart from regular forests or simple rows of trees. Training the model required the University of Tartu's high-performance computing (HPC) cluster, but once trained, it could automatically analyze almost 2,000 aerial photographs covering the whole of Estonia and locate field islands across the country. The researchers have also made both the dataset and the computer code openly available so that similar methods can be applied elsewhere.\n\n**AI finds an extra 10,000 field islands**\n\nOnce trained, the AI surveyed the entire Estonian countryside. It identified almost 38,000 field islands, around 34 percent more than are listed in the official ARIB registry. The highest numbers were found in central Estonia where large agricultural fields are common. In most places, the AI found more field islands than official ARIB records showed, highlighting how difficult it is to keep these small landscape features up to date through manual mapping alone. The only region where the model consistently found fewer field islands was northeastern Estonia where many field islands consist mainly of rocks and shrubs rather than trees.\n\nOne of the biggest surprises was that the AI became more accurate when shown slightly lower-resolution images, as it could view larger chunks of the landscape at once. Seeing a wider area helped it understand the landscape context, making it easier to distinguish isolated field islands from hedgerows or forest edges while also speeding up the analysis.\n\n\"Our findings do not mean that the official maps are inaccurate. Rather, they show just how difficult it is to keep track of thousands of small landscape features across an entire country using manual mapping alone. AI can rapidly scan aerial images of the whole of Estonia and highlight locations where experts should take a closer look. It still occasionally mistakes hedgerows for field islands or misses rocky field islands without trees, but it provides an efficient first pass for identifying places that deserve expert review,\" says researcher Holger Virro.\n\n**Seeing the bigger picture**\n\nField islands may occupy only tiny patches of farmland, but together they form a network of habitats stretching across the country. By making these overlooked features visible, AI is giving researchers, farmers and decision-makers a clearer picture of Estonia's agricultural landscapes and new opportunities to protect them for future generations.\n\nAfif Fauzan, Holger Virro and Evelyn Uuemaa published their findings in the scientific journal [Geocarto International](https://www.tandfonline.com/doi/full/10.1080/10106049.2026.2691353).\n\n--\n\nEditor: Marcus Turovski", "url": "https://wpnews.pro/news/ai-discovers-thousands-of-overlooked-field-islands-in-estonia", "canonical_source": "https://news.err.ee/1610127139/ai-discovers-thousands-of-overlooked-field-islands-in-estonia", "published_at": "2026-09-03 14:05:00+00:00", "updated_at": "2026-09-03 14:22:53.274730+00:00", "lang": "en", "topics": ["artificial-intelligence", "computer-vision", "machine-learning"], "entities": ["University of Tartu", "Agricultural Registers and Information Board", "European Union", "Afif Fauzan"], "alternates": {"html": "https://wpnews.pro/news/ai-discovers-thousands-of-overlooked-field-islands-in-estonia", "markdown": "https://wpnews.pro/news/ai-discovers-thousands-of-overlooked-field-islands-in-estonia.md", "text": "https://wpnews.pro/news/ai-discovers-thousands-of-overlooked-field-islands-in-estonia.txt", "jsonld": "https://wpnews.pro/news/ai-discovers-thousands-of-overlooked-field-islands-in-estonia.jsonld"}}