Habitats are increasingly used to assess the status of biodiversity. Mapping the distribution of habitats is vital for successful conservation, management and monitoring of biodiversity. The habitat classification of Delarze, Gonseth et al. (2015) is the most widely used in Switzerland. While there has been some regional modelling of this classification, there is currently no spatially explicit map of these habitats across Switzerland.
This project takes advantage of advances in remote sensing technologies to develop a semi-automated methodology to map the current extent of habitat types in Switzerland, taking into account impacts of human disturbances and interventions. We employ an extensive suite of earth-observation and mapping data as inputs in an approach involving habitat distribution modelling, image segmentation and classification. High-resolution 3D information from digital aerial photogrammetry allows differentiating shrubs and trees and identifying buildings. Phenological dynamics are determined from seasonal variation in vegetation indices (e.g. NDVI), derived from high temporal resolution Sentinel-2 satellite imagery. Data describing climate, topography, soil and land use (from the topographic landscape model TLM) provide additional covariates. Habitat distribution models are developed via machine learning approaches trained with field data available from large scale Swiss vegetation and biodiversity monitoring programmes.
Within the software eCognition, airborne orthoimagery (1m resolution) is segmented into ‘image primitives’ on the basis of reflectance in the RGB and NIR bands, and values of the metrics NDVI and NDWI. In a rule-based approach, habitat types can be assigned to segments based on the input data and distribution models, resulting in a high spatial resolution Swiss-wide map of habitat types. This semi-automated approach can be re-applied with updates of the base data at specified time intervals, enabling use for monitoring purposes.
The current version of the Habitat Map of Switzerland in available for download on EnviDat
2019 - 2021