An assessment of Sentinel-2 multispectral satellite reflectance data for ecological mapping of temperate raised peatlands

Raised bogs are ombrotrophic peatlands defined by dome-like topography that is elevated above the water table. Once widespread in temperate northern Europe, their exploitation has necessitated special conservation status, which requires assessment of raised bog conditions at the scale of tens of thousands of hectares. Satellite remote sensing offers a solution to this challenge, but upscaling relationships between ground-based ecological surveys at temperate raised bogs and remotely sensed data are unclear. 

We provide a first statistical analysis of the relationship between ecological survey data obtained at temperate raised bogs at c. 80 m² scale to multispectral remote sensing data recorded by the Sentinel-2 satellite at 100-400 m² scale.  We analyse data from 34 images obtained in summer and winter of 2020-2025 in conjunction with 2827 ecological survey points recorded in 2023-2024 from six Irish raised bogs that cover an area of 3.91 km². We test for correlations between ecological communities as established at each survey point and the reflectance spectra and amplitudes in the corresponding Sentinel-2 pixels. 

Our results show statistically significant differences between reflectance of pixels associated with the end-member ecological communities as well as in the broader classes of ‘active’ and ‘inactive’ raised bog habitat. Differences are most pronounced in red-edge and near infrared bands as well as indices composed of these bands. A general reduction of reflectance values in winter, likely related to phenology, moisture and solar radiance impacts, does not greatly diminish the statistical strength of differentiation between ecotopes. Winter images can be prone to frost/ice, however, which produces anomalous reflectance distributions that can be detected (and thus filtered) by principal component analysis. These findings help to underpin the use of multispectral data for large scale automated mapping of these vulnerable habitats via e.g. machine learning approaches.