AI meets ecology – Identifying wetlands with GeoAI

Published on: January 16, 2026

In Sweden, Sweco carried out a mapping of Gotland’s alkaline fens (rikkärr) using GeoAI. Alkaline fens are a key wetland ecosystem on Gotland and biodiversity hotspots, hosting a specialised and species‑rich flora. These fens are mineral-rich (typically calcareous) with a relatively high pH (6-8). Drainage, intensive agriculture, lapsed active management and eutrophication have all contributed to a serious decline of alkaline fens in Europe. The County Administrative Board of Gotland sought to identify any alkaline fens missing from earlier inventories.

Sweco’s assignment consisted of mapping unknown potential alkaline fens and assessing their ecological status, using remote sensing in aerial photos and other available geospatial data. The analysis combined two AI methods: machine learning, which is a well-known but iterative and manual method, and deep learning, which requires more annotated training data and longer training times but can offer higher accuracy. The methods were tested separately and in combination. The combined approach, where outputs from the traditional method were used to train deep learning models, proved most effective.

Because alkaline fens are visually heterogeneous and hard to map directly, the team first focused on a more easily detected species: Great fen-sedge (Cladium mariscus). Earlier inventories indicate that the conservation status of alkaline fens has deteriorated due to overgrowth of Great fen-sedge. While dense stands of this reed-like grass are detrimental to biodiversity, they are relatively easy to detect using GeoAI.

“The project provided a landscape‑wide knowledge base for selecting future conservation measures in suitable alkaline fens.” – Annika Forsslund, client and biologist at the County Administrative, Board of Gotland.

Results:

The automated workflow identified alkaline fens and vegetation structures. Trained deep learning models learned patterns typical of alkaline fens and Great fen-sedge. The final output was a geodatabase with detailed mappings of Great fen-sedge, potential alkaline fens, water, shrubs and trees, providing the County Administrative Board of Gotland with map data to help prioritise conservation and restoration measures.

Collaboration and validation:

Close cooperation between Sweco and the County Administrative Board of Gotland ensured methods and models were adapted, tested and verified for reliability and accuracy. The project delivered an innovative, useful dataset, demonstrating both the potential and the practical challenges of using GeoAI to support biodiversity conservation and sustainable environmental management.

Practical outcome:

The geodatabase helps identify areas to conserve or restore and supports the planning of nature conservation actions on Gotland.