The sea is now being asked to forecast its own height
the sea is now being asked to forecast its own height: a deep model called hidra d maps dense sea level from sparse altimetry and beats a circulation model by 28 percent error, a light cnn floods coastal cities under any rise scenario, and a neural net grades grid cells low moderate high risk at 98.55 percent accuracy.
the tide now files its own rise forecast.
Context
HIDRA-D (Geoscientific Model Development, March 2026, Rus, Licer and Kristan) is a deep model for basin scale gridded sea level, trained on sparse satellite altimetry plus tide gauges. It adds a Dense decoder to the earlier HIDRA3 point model and gives hourly forecasts over a 3 day horizon. Against satellite absolute dynamic topography in the Adriatic it cuts mean absolute error by 28.0% relative to the NEMO circulation model.
The flood paper (arXiv, October 2025, published in HESS in 2026) builds a lightweight CNN to predict coastal flood depth maps under variable sea level rise and shoreline adaptation scenarios, tested on Abu Dhabi and San Francisco. It reports an average reduction of nearly 20% in mean absolute error versus state-of-the-art methods.
The risk classifier (listed April 2026) is a deep neural network on a dataset of 1,653 geospatial grid records with elevation, land use, soil, drainage, distance to sea, precipitation and flood history. It sorts each cell into Low, Moderate or High risk with overall accuracy of 98.55%, which the abstract says is 4 to 7% above random forest and SVM.
The 28% for HIDRA-D is a mean absolute error reduction in the Adriatic only, and the authors note limits in the Neretva estuary and in the northern Adriatic where altimetry is sparse. It is a 3 day forecast, not a sea level rise projection. The flood CNN covers two regions and specific scenarios, so 'any rise scenario' in the note is wider than the paper, which tests variable projections and adaptation cases in those two places. The 98.55% classifier is trained and tested on 1,653 grid records, a small dataset, and the listing does not say how the test split was made, so the accuracy does not speak to new coasts. Calling these three a rise forecast mixes a short term forecast, a flood map surrogate and a risk label.
Related work
- HIDRA-D: deep-learning model for dense sea level forecasting using sparse altimetry and tide gauge data (Geoscientific Model Development) ↗Source for the 28.0% error reduction and the limits the authors name.
- HIDRA-D (ViCoS Lab) ↗Lab page for the HIDRA-D publication.
- Climate Adaptation-Aware Flood Prediction for Coastal Cities Using Deep Learning (arXiv) ↗Source for the lightweight CNN flood model and its Abu Dhabi and San Francisco tests.
- Deep vision-based framework for coastal flood prediction under sea level rise and shoreline protection (Scientific Reports) ↗Companion paper on the same low-data flood surrogate approach.
- Deep Neural Networks for Coastal Climate Risk Prediction Using Flood and Climate Variables (listing) ↗Source for the 98.55% accuracy and the 1,653 record dataset.
Watch next
- Check how the 98.55% classifier was split and whether it was tested on a coast it was not trained on. Look for HIDRA-D results outside the Adriatic.
Sources
- HIDRA-D: deep-learning model for dense sea level forecasting using sparse altimetry and tide gauge data (Geoscientific Model Development)gmd.copernicus.org
- Climate Adaptation-Aware Flood Prediction for Coastal Cities Using Deep Learning (arXiv)arxiv.org
- Deep Neural Networks for Coastal Climate Risk Prediction Using Flood and Climate Variables (listing)exa.ai
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 00:01 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 →