EGU24-18776, updated on 11 Mar 2024
EGU General Assembly 2024
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

A human-in-the-loop approach to monitor blue carbon ecosystems at scale with Copernicus Sentinel-2 imagery

Gyde Krüger, Silvia Huber, Lisbeth T. Nielsen, Paul Daniel, Charalampos Malathounis, and Lars B. Hansen
Gyde Krüger et al.
  • DHI, Denmark (

A human-in-the-loop approach to monitor blue carbon ecosystems at scale with Copernicus Sentinel-2 imagery

Blue carbon ecosystems, such as seagrass meadows, mangroves or coral reefs play an essential role for food provision, erosion control, disaster resilience, biodiversity and as habitats, and in addition they serve as important natural sinks for carbon. With increasing climate pressures and human impacts related to eutrophication, overfishing and habitat fragmentation, the coverage and health of these coastal habitats have declined globally. To address this, it requires first and foremost an accurate quantification of the distribution and quality of these ecosystems. This poses challenges from multiple angles (i.e., lack of manpower, fiscal limitations, etc.). By exploiting the full capacity of Copernicus Sentinel-2 imagery and AI technology, we have developed a cloud-based interactive platform, MCSAV – short for Mangrove, Coral and Submerged Aquatic Vegetation (Figure 1) – to map and improve the planning, management and monitoring of blue carbon ecosystems worldwide. The platform is designed with a focus on making coastal mapping as easy as possible for users with greater local knowledge of blue carbon ecosystems but without expert knowledge in satellite image processing or machine learning. The entire mapping process, from the selection of suitable satellite imagery to the final classification, can be executed in just a few clicks with the platform. The backbone of the classification model that is integrated into the backend of the platform, is a pretrained convolutional neural network (DeepResUNet). A Human-in-the-Loop component allows fine-tuning of the pre-trained classification model with additional training data.

Our approach has already been applied to high latitude regions with success [1] and is currently applied to Semporna in Sabah, Malaysia, as part of the United Nations Development Programme’s (UNDP) third cohort of Ocean Innovations on marine protected areas, area-based management, and blue economy.

We will give an introduction into the project, present the methods implemented in our interactive mapping approach, and demonstrate the coastal mapping tool. We will conclude with some lessons learnt and an outlook.

Figure 1 MCSAV interface showing Copernicus Sentinel-2 image of 13 March 2023 after pre-processing module has been run (top) and an example habitat classification (preliminary) and deepwater areas for Mapul region (bottom).



[1]          S. Huber et al., “Novel approach to large-scale monitoring of submerged aquatic vegetation: A nationwide example from Sweden,” Integr Environ Assess Manag, vol. 18, no. 4, pp. 909–920, 2022, doi:


How to cite: Krüger, G., Huber, S., Nielsen, L. T., Daniel, P., Malathounis, C., and Hansen, L. B.: A human-in-the-loop approach to monitor blue carbon ecosystems at scale with Copernicus Sentinel-2 imagery, EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-18776,, 2024.