Mapping perennial brown macroalgae in the Baltic sea using multitemporal Sentinel-2 imagery

Satellite observations can extend benthic biodiversity monitoring beyond sparse point surveys, but their value depends on robust field validation and transparent uncertainty. We used high-quality Sentinel-2 MSI (S2) imagery from 2016–2023 and ground-truth observations to map the spatial extent of ecologically significant perennial brown macroalgae habitats across three Baltic Sea sites: Vilsandi (Estonia), Gotland and Blekinge (Sweden). Classification performed best in Vilsandi, where shallow waters, high water transparency and extensive reference data provided favourable conditions. The high revisit frequency of S2 enabled the generation of pixel-wise occurrence frequency (OF) products – the proportion of times a pixel is classified as brown macroalgae (BROWN) class across multiple S2 images – to supress sporadic errors in individual classification. Unlike conventional single-scene classifications, the OF framework interprets repeated classification agreement across time as an indicator of habitat persistence and classification confidence. No single OF threshold transferred perfectly across sites, but threshold close to 50% generally balanced between over- and underestimation of BROWN class. Across all sites, OF-based estimates of BROWN class extent were lower than mean single-image estimates, indicating that individual images tended to inflate habitat area through transient misclassifications. Multi-temporal S2 products therefore provide a scalable, conservative basis for mapping stable brown macroalgal habitats, while highlighting the continuing need for high-quality in situ data.