The meadow underwater is now being asked to file its own extent map
the meadow underwater is now being asked to file its own extent map: sentinel 2 feeds a deep learning chain that maps danish coastal vegetation at ten meter resolution nationwide, mediterranean models reach 74 to 92 percent accuracy on posidonia meadows, and a vision transformer tells seagrass from bare sand in 8,300 underwater photos.
the meadow now files its own habitat map.
Context
The Danish Marine Vegetation Mapping page describes a first nationwide map of submerged coastal vegetation (eelgrass and macroalgae) built from 2018 satellite data, at 10 m resolution, using radiative transfer modelling and machine learning. It covers shallow water to depths of 4 to 10 m depending on clarity, omits turbid or too deep areas, was completed in 2018/2019 and was funded by the VELUX Foundation.
A 2024 Scientific Reports paper maps Posidonia oceanica meadows across the Mediterranean with satellite data and ancillary data in an AI model. It reports 74 to 92% overall accuracy, 72 to 91% user's accuracy and 81 to 92% producer's accuracy, with the highest accuracies at 0 to 25 m depth.
The SeagrassFinder preprint creates a dataset of over 8,300 annotated underwater images and compares ResNet, InceptionNetV3, DenseNet and Vision Transformer models on a binary task: eelgrass present or absent. It reports AUROC scores above 0.95 on the final test set, with Vision Transformers performing particularly well.
The Danish map matches 'ten meter' and 'nationwide'. The page names radiative transfer modelling and machine learning and does not name deep learning, so the note's 'deep learning chain' is not stated for Denmark in the page read. That is unsupported, not refuted. It is also a single-year map from 2018, not a recurring one.
The 74 to 92% range is overall accuracy across Mediterranean regions, the paper's own words. The note's 'accuracy on posidonia meadows' matches, though the range spans regions and depths and the best values are in shallow water.
The note says the vision transformer tells seagrass from 'bare sand' in 8,300 underwater photos. The preprint describes presence versus absence of eelgrass in images and gives AUROC above 0.95, with 'over 8,300' images. 'Bare sand' as the negative class was not found in the passages read. The study is a preprint, so it has not been confirmed as peer reviewed here.
These are three separate studies on different scales: a satellite national map, a satellite Mediterranean model and underwater photos. 'The meadow now files its own habitat map' is the author's framing.
Related work
- Marine Vegetation Mapping (Danish national map, Satlas) ↗Source for the Danish national map, its 10 m resolution, depth limits and funding.
- AI-driven remote sensing enhances Mediterranean seagrass monitoring and conservation to combat climate change and anthropogenic impacts (Scientific Reports, 2024) ↗Source for the Posidonia oceanica accuracy range and the depth pattern.
- SeagrassFinder: Deep Learning for Eelgrass Detection and Coverage Estimation in the Wild (arXiv 2412.16147) ↗Source for the 8,300 image dataset, the model comparison and the AUROC.
Watch next
- Read how the Danish map's model was validated against field data. Check the SeagrassFinder paper for what appears in the absence class.
Sources
- Marine Vegetation Mapping (Danish national map, Satlas)marine-vegetation.satlas.dk
- AI-driven remote sensing enhances Mediterranean seagrass monitoring and conservation to combat climate change and anthropogenic impacts (Scientific Reports, 2024)nature.com
- SeagrassFinder: Deep Learning for Eelgrass Detection and Coverage Estimation in the Wild (arXiv 2412.16147)arxiv.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 3 October 2026 at 22:17 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →