How can Sentinel‐2 contribute to seagrass mapping in shallow, turbid Baltic Sea waters?. Issue 3 (7th December 2021)

• Record Type: Journal Article
• Authors: Kuhwald, Katja; Schneider von Deimling, Jens; Schubert, Philipp; Oppelt, Natascha
• Editors: Scales, Kylie; Lecours, Vincent
• Is Part Of: Remote sensing in ecology and conservation. Volume 8:Issue 3(2022)
• Page Start: 328
• Page End: 346
• Publication Date: 2021-12-07
• Subjects: Baltic sea -- benthic habitat mapping -- eelgrass -- random forest -- Sentinel‐2 -- submerged aquatic vegetation; Remote sensing -- Periodicals; Ecology -- Research -- Periodicals; Ecology -- Methodology -- Periodicals; Ecology -- Remote sensing -- Periodicals; Nature conservation -- Methodology -- Periodicals; 577.0723
• DOI: 10.1002/rse2.246
• Languages: English
• ISSNs: 2056-3485

Abstract

Abstract: Seagrass meadows are one of the most important benthic habitats in the Baltic Sea. Nevertheless, spatially continuous mapping data of Zostera marina, the predominant seagrass species in the Baltic Sea, are lacking in the shallow coastal waters. Sentinel‐2 turned out to be valuable for mapping coastal benthic habitats in clear waters, whereas knowledge in turbid waters is rare. Here, we transfer a clear water mapping approach to turbid waters to assess how Sentinel‐2 can contribute to seagrass mapping in the Western Baltic Sea. Sentinel‐2 data were atmospherically corrected using ACOLITE and subsequently corrected for water column effects. To generate a data basis for training and validating random forest classification models, we developed an upscaling approach using video transect data and aerial imagery. We were able to map five coastal benthic habitats: bare sand (25 km²), sand dominated (16 km²), seagrass dominated (7 km²), dense seagrass (25 km²) and mixed substrates with red/ brown algae (3.5 km²) in a study area along the northern German coastline. Validation with independent data pointed out that water column correction does not significantly improve classification results compared to solely atmospherically corrected data (balanced overall accuracies ~0.92). Within optically shallow waters (0–4 m), per class and overall balanced accuracies (>0.82) differed marginally depending on the water depth. Overall balanced accuracy became worse (<0.8) approaching the border to optically deep water (~ 5 m). The spatial resolution of Sentinel‐2 (10–20 m) allowed delineating detailed spatial patterns of seagrass habitats, which may serve as a basis to retrieve spatially continuous data for ecologically relevant metrics such as patchiness. Thus, Sentinel‐2 can contribute unprecedented information for seagrass mapping between 0 and around 5 m water depths in the Western Baltic Sea. Abstract : Sentinel‐2 provides unprecedented information on seagrass distribution in the Western Baltic Sea between 0 and 5‐m water depths. Classification results and accuracies are similar using only atmospherically or additionally water‐column corrected reflectance data. The turbid waters limit analyses to water depths less than 5 m. In these shallow waters, the new maps provide spatially continuous data on seagrass distribution patches, for example patches or large, connected meadows, which is highly interesting for seagrass ecology (e.g. restoration projects and management).