Dataset description

Annual maps of emerged aquatic vegetation on intertidal mudflats in the German Bight, derived from Sentinel-2 MSI imagery at 10 m spatial resolution for the period 2020–2023. Per-pixel outlier-clipped maximum Normalized Difference Vegetation Index (NDVI) during July–September (JAS) serves as a proxy for emerged vegetation cover; within areas identified as seagrass by a machine learning classifier trained on phenological parameters from time-series of the Atmospherically Resistant Vegetation Index (ARVI) and relative emergence frequency, NDVI is translated into seagrass percent cover using an empirical relationship.

The dataset is split into two sections for the eastern (2020-2023_foccus_obs_oc_gbe_esm_l4-hr_P1Y.nc) and southern (2020-2023_foccus_obs_oc_gbs_esm_l4-hr_P1Y.nc) sections of the German Bight.

Limitations

The product quality is directly tied to the availability of cloud-free, low-tide Sentinel-2 observations. Spatial and temporal gaps due to cloud cover or tidal conditions may result in the JAS composite not capturing the true seasonal peak of vegetation, leading to underestimation of seagrass density and coverage. Effects due to spatial-temporal coverage are particularly pronounced in the Wadden Sea of Lower Saxony (German Bight South), where the frequency of suitable observations is substantially lower than in Schleswig-Holstein (German Bight East), resulting in incomplete spatial coverage in some years.

The seagrass percent cover estimates should be interpreted as the maximum density observed under the constraints of this method, not as an absolute measure of seagrass cover.

The XGBoost classifier is trained exclusively on field observations from Schleswig-Holstein and applied to Lower Saxony without retraining. Performance in Lower Saxony is therefore unvalidated, and classification errors are expected to be higher, particularly given the sparser and less occurrence of seagrass. Visual inspection indicates occasional misclassification of other intertidal vegetation types, in particular macroalgae on mussel reefs and microphytobenthos, which can exhibit NDVI and phenology patterns similar to seagrass. These confusions occur across both regions but are more frequent in Lower Saxony.

The Zoffoli et al. (2020) empirical model used to estimate seagrass cover is only reliable for cover values ≥20 % (NDVI ≥ 0.25). Percent cover estimates below this threshold should be interpreted with caution. Furthermore, the model was calibrated on Zostera noltii and meadows composed of mixed Zostera marina/Zostera noltii communities may exhibit larger discrepancies.  Field reflectance measurements used for algorithm calibration were collected over small 20 cm-diameter circles.  Scaling the algorithm to larger areas in the Wadden Sea, where small water ponds, scattered deposits of drifting macroalgae, and heterogeneous sediment conditions may occur within the meadow likely contributes to an systematic offset of approximately 26% observed between model predictions and field observations. In addition, in situ data collected along transects were available in six discrete classes, whereas the NDVI-based algorithm estimates seagrass percent cover along a continuous gradient.