Intertidal areas, which emerge during low tide, form a vital link between terrestrial and marine environments. Seagrasses, a well-studied intertidal habitat, provide a multitude of different ecosystem goods and services. However, owing to their relatively high exposure to anthropogenic impacts, seagrasss meadows and other intertidal habitats have seen extensive declines. Remote sensing methods that can capture the spatial and temporal variation of marine habitats are essential to best assess the trajectories of seagrass ecosystems. An advanced machine learning method has been developed to map intertidal vegetation from satellite-derived surface reflectance at a 12-band multispectral resolution and distinguish between similarly pigmented intertidal macrophytes, such as seagrass and green algae. The Intertidal Classification of Europe: Categorising Reflectance of Emerged Areas of Marine vegetation with Sentinel-2 (ICE CREAMS v1.0), a neural network model trained on over 300,000 Sentinel-2 pixels to identify different intertidal habitats, was applied to the open-access long term archive of systematically collected Sentinel-2 imagery to provide 7 years (2017–2023) worth of intertidal seagrass dynamics in 6 sites across Western Europe (471 Sentinel-2 Images). A combination of independently collected visually inspected Uncrewed Aerial Vehicle imagery and in situ quadrat images were used to validate ICE CREAMS. Having achieved a high seagrass classification accuracy (0.82 over 12,000 pixels) and consistent conversion into cover (19% RMSD), the ICE CREAMS model outputs provided evidence of site specific variation in trajectories of seagrass extent, when appropriate consideration of intra-annual variation has been considered. Inter-annual dynamics of sites showed some instances of consistent change, some indicated stability, while others indicated instability over time, characterised by increases and decreases across the time-series in seagrass coverage. This methological pipeline has helped to create up-to-date monitoring data that, with the planned continuation of the Sentinel missions, will allow almost real-time monitoring of these habitats into the future. This process, and the data it provides, could aid management practitioners from regional to international levels, with the ability to monitor intertidal seagrass meadows at both high spatial and temporal resolution, over continental scales. The implementation of Earth Observation for high-resolution monitoring of intertidal seagrasses could therefore allow for gap-filling seagrass datasets, and sustain specific and rapid management measures.

中文翻译：

---

Sentinel-2 时间序列的潮间带海草范围显示西欧的独特轨迹


潮间带在低潮时出现，形成陆地和海洋环境之间的重要联系。海草是一种经过充分研究的潮间带栖息地，提供多种不同的生态系统产品和服务。然而，由于受到人为影响的程度相对较高，海草草甸和其他潮间带栖息地已大幅减少。能够捕捉海洋栖息地时空变化的遥感方法对于最好地评估海草生态系统的轨迹至关重要。一种先进的机器学习方法已经开发出来，可以根据卫星表面反射率以 12 波段多光谱分辨率绘制潮间带植被图，并区分类似色素的潮间带大型植物，例如海草和绿藻。欧洲潮间带分类：使用 Sentinel-2 对海洋植被新兴区域的反射率进行分类 (ICE CREAMS v1.0)，这是一种神经网络模型，经过超过 300,000 个 Sentinel-2 像素的训练来识别不同的潮间带栖息地，该模型被应用于开放式访问系统收集的 Sentinel-2 图像的长期档案，以提供西欧 6 个地点 7 年（2017-2023）的潮间带海草动态（471 个 Sentinel-2 图像）。结合独立收集的目视检查的无人驾驶飞行器图像和原位样方图像来验证冰淇淋。 ICE CREAMS 模型实现了较高的海草分类精度（超过 12,000 像素为 0.82）和一致的覆盖转换（19% RMSD），在适当考虑年内变化的情况下，ICE CREAMS 模型输出提供了海草范围轨迹中特定地点变化的证据。已被考虑。 站点的年际动态显示出一些一致变化的情况，一些表明稳定，而另一些则表明随着时间的推移不稳定，其特点是海草覆盖范围随时间序列的增加和减少。该方法管道有助于创建最新的监测数据，随着哨兵任务计划的继续，这些数据将允许在未来对这些栖息地进行几乎实时的监测。这一过程及其提供的数据可以帮助区域到国际层面的管理从业者，能够在大陆尺度上以高空间和时间分辨率监测潮间带海草草甸。因此，实施地球观测对潮间带海草进行高分辨率监测可以填补海草数据集的空白，并维持具体和快速的管理措施。