Artificial Intelligence from Space: Tracking the Pulse of the Mediterranean’s Seagrass Meadows

• The study, published in Ecological Indicators, presents a deep learning framework capable of accurately mapping Posidonia oceanica meadows across the Mediterranean using satellite imagery.
• The research is the result of a collaboration between the Institute for Cross-Disciplinary Physics and Complex Systems (IFISC, CSIC-UIB) and the Centre d’Estudis Avançats de Blanes (CEAB-CSIC).

Seagrass meadows are among the most valuable yet threatened ecosystems on the planet. In the Mediterranean Sea, Posidonia oceanica meadows play a crucial role in protecting coastlines, supporting marine biodiversity, and storing carbon. However, these underwater forests are disappearing at an alarming rate due to coastal development, pollution, and climate change. Monitoring their health and distribution is therefore vital for conservation, but traditional field surveys are costly and time-consuming.

A team of researchers from the Institute for Cross-Disciplinary Physics and Complex Systems (IFISC, CSIC-UIB) and the Centre d’Estudis Avançats de Blanes (CEAB-CSIC) has developed a new artificial intelligence framework that could revolutionize how we observe and protect these essential habitats. The model, named CAMELE (Consensus for Automated Marine Ecosystem Labelling and Evaluation), uses deep learning techniques applied to high-resolution satellite imagery to automatically detect and map Posidonia oceanica and other benthic habitats with unprecedented accuracy.

“Our approach moves beyond local proof-of-concept studies to provide a robust, generalizable tool for large-scale habitat mapping”, explains author Manuel A. Matías (IFISC, UIB-CSIC). “By combining machine learning with extensive ecological data, we can now monitor the Mediterranean seafloor faster, more reliably, and at a fraction of the cost of traditional methods”.

The framework relies on convolutional neural networks trained on high-resolution multispectral images from PlanetScope, a constellation of Earth-observing satellites that provides daily global coverage, combined with 19 years of detailed habitat data provided by the Government of the Balearic Islands. The resulting dataset covers around 2,500 square kilometres of coastline, including Mallorca, Menorca, Ibiza, Formentera, and Cabrera. To test how well the model could generalize, the researchers trained it using data from only one island and validated its predictions on others. Remarkably, the model successfully identified seagrass meadows even in ecologically distinct environments, demonstrating a high capacity to adapt to new regions.

In benchmark tests, CAMELE achieved an average accuracy above 90% in identifying and outlining seagrass meadows, as measured by the Intersection-over-Union score, a standard metric that quantifies how closely the model’s predictions match real mapped areas. The study also introduced new evaluation criteria specifically designed for image segmentation tasks, providing a more realistic assessment of model precision. “What makes CAMELE stand out is its robustness”, says IFISC researcher Àlex Giménez. “Even when exposed to unfamiliar environmental conditions, the system produced reliable maps of seagrass coverage that are consistent with field observations”.

Beyond its scientific innovation, the open-access nature of CAMELE makes it a valuable tool for marine conservation. The trained models and an online visualization platform are freely available, enabling researchers, environmental agencies, and policymakers to adapt the framework across the Mediterranean. This accessibility supports coordinated conservation actions and consistent long-term monitoring.

According to Manuel A. Matías, “AI-driven monitoring tools like CAMELE can help detect early signs of seagrass loss or fragmentation, informing faster and more effective management responses”. Looking ahead, the team aims to expand the framework to track additional indicators of ecosystem health and extend its application beyond the Balearic Islands. By combining artificial intelligence with remote sensing, the study opens the way to a new generation of ecological monitoring systems capable of keeping pace with rapid environmental change.

Figure: Example of model predictions for a satellite image. (a) Satellite image from Pollença bay in the island of Mallorca, a part of the training set. Image © 2022 Planet Labs PBC (b) Ground truth data for the benthic habitats in Pollença bay. (c) Habitat classification from the CAMELE model.

Giménez-Romero, À., Ferchichi, D., Moreno-Spiegelberg, P., Sintes, T., and Matías, M. A. (2025). A generalizable deep learning framework for large-scale mapping of seagrass habitats. Ecological Indicators, 180, 114349. https://doi.org/10.1016/j.ecolind.2025.114349

• El Diario