Assessing GeoAI methods' accuracy for wetland detection in the region of Stockholm
Project title: Assessing GeoAI methods' accuracy for wetland detection in the region of Stockholm
Project leader: Jenny Lindblad, SoM, KTH
Participants: Andrea Nascetti, KTH, Lina Suleiman, KTH, Ioannis Ioannidis, KTH
Funder: Region Stockholm, 2025
Project period: 2025
For decades, there has been growing momentum to protect and restore wetlands for the vital ecosystem services they provide, including habitat for endangered species, water purification, flood prevention, and carbon storage. However, wetlands continue to be lost, primarily due to urban development. Sweden’s national environmental goal, supported by a dedicated budget, aims to restore and protect wetlands, but recent data shows that losses from urbanization and infrastructure outweigh restoration efforts. Thus, integrating urban development with wetland protection is critical.
A major challenge is the fragmented and limited knowledge of wetlands in urban areas, as highlighted in a 2022 pilot study performed by a project team member. Traditional wetland mapping methods, like field surveys and aerial mapping, are resource-intensive, which led the Swedish national wetland inventory (VMI) to exclude wetlands smaller than 10 hectares in their surveys. This leaves out small wetlands in urban environments, despite their significant ecosystem services and their high presence in the Stockholm region. Moreover, existing local inventories are fragmented and not digitalized, hindering the proactive inclusion of wetlands in urban planning. Recent advances in remote sensing technologies and Geospatial artificial intelligence (GeoAI) offer possibilities for accurate ecosystem detection and new means for improving the consideration of wetlands in urban practice and policy.
The purpose of the project is to contribute to the improvement of the integration of wetland protection efforts into urban planning processes. The goal of the project is to assess the accuracy and flexibility of existing GeoAI methods based on Geospatial Foundation Models (GFMs) for wetland detection. To perform accuracy assessments, the project will collect reference data by compiling fragmented documentation of field-surveyed wetlands figuring in environmental impact assessments and nature inventories dispersed across planning documents. The project fulfills the goal by addressing the following research questions:
The project will result in new knowledge about how well GFMs for wetland detection perform in the complex and fragmented wetland ecosystems typical of urban environments like the region of Stockholm. The results will provide critical insights into the suitability of GFMs as planning support for wetland protection and restoration efforts, laying the groundwork for integrating these innovative tools into practical regional and municipal land use policy and practice.
