Mapping the distribution of seagrass meadows from space with deep convolutional neural networks

Seagrass meadows play a vital role in supporting coastal communities by promoting biodiversity, mitigating coastal erosion and contributing to local economies. Nowadays these
ecosystems face significant threats, including habitat loss and degradation or climate change. This has lead the United Nations to recognize the urgency of conserving marine
ecosystems, highlighting the need for evidence-based conservation strategies and high-quality monitoring. However, traditional monitoring approaches are often
time-consuming, labor-intensive, and costly, limiting their scalability and effectiveness. Here we present a deep learning framework based on convolutional neural networks to identify
Posidonia oceanica meadows in the Mediterranean Sea using satellite imagery. We demonstrate the model's generalization capability and robustness introducing appropiate metrics, which
is far beyond present limited approaches. We show that our model is capable of providing reliable estimates for the distribution of the considered habitats and accurate measures for
their extension area. Our work contributes to a future development of a reliable map of the distribution of Posidonia oceanica meadows in the Mediterranean Sea and showcases the
transformative potential of remote sensing and machine learning technologies in marine habitat monitoring.