The ice sheet at sea is now being asked to keep its own calendar
the ice sheet at sea is now being asked to keep its own calendar: a mass conserving emulator called floenet reproduces global sea ice means and trends across climates, a mixture of experts model named icect forecasts monthly arctic concentration and thickness at 12.5 kilometers six months out, and an ai correction cuts sea ice forecast errors by 41 percent.
the frozen sea now files its own freeze forecast.
Context
The FloeNet arXiv paper introduces a machine-learning emulator of the GFDL global sea ice model SIS2. It is a graph neural network that predicts 6-hour ice and snow mass tendencies while conserving mass and area. Trained on a 65-year simulation, it is tested on pre-industrial control and 1% CO2 climates. It reports that FloeNet beats a non-conservative model on mean state, trends and inter-annual variability, with volume anomaly correlations above 0.96 in the Antarctic and 0.76 in the Arctic.
The IceCT paper in an AGU journal describes a Mixture-of-Experts deep learning model that forecasts monthly Arctic sea ice concentration and thickness at 12.5 km resolution up to 6 months ahead, with one expert for each variable and a spatial gate. It reports an overall mean absolute error of 4.56% for concentration and 0.140 m for thickness, and says it mitigates the spring predictability barrier.
The Cryosphere paper on NorCPM adds a machine-learning error correction to a seasonal Arctic sea ice prediction system. Its introduction cites earlier work, Palerme et al. (2024), in which ML-based offline calibration reduced sea ice concentration prediction errors by 41% and ice edge distance errors by 44% in short-term sea ice prediction.
FloeNet is an emulator of a climate model's sea ice component, not a forecast of observed ice. The note's 'reproduces global sea ice means and trends across climates' fits the abstract, with the caveat that it reproduces a simulated ice model and the Arctic volume anomaly correlation is lower at 0.76.
IceCT matches 'mixture of experts', 'monthly' and '12.5 kilometers six months out'. The note says it forecasts 'concentration and thickness', which matches.
The 41% belongs to a different study. In the sources read, it is a result from Palerme et al. 2024, cited by the NorCPM paper, and it concerns short-term sea ice concentration forecasts, not seasonal or six-month forecasts. The note joins it with the other two as if it were the same kind of result. The Palerme paper itself was only seen as a record without its abstract, so the exact baseline for 41% is unsupported, not refuted.
'The frozen sea now files its own freeze forecast' is the author's framing. The three studies use different targets: a model emulator, a seasonal forecast and an error correction.
Related work
- FloeNet: A mass-conserving global sea ice emulator that generalizes across climates (arXiv 2603.12449) ↗Source for FloeNet's design and its reported skill.
- IceCT: A MoE-Based Model for Seasonal Arctic Sea Ice Concentration and Thickness Prediction (AGU, Journal of Geophysical Research: Machine Learning and Computation) ↗Source for IceCT's design, resolution, lead time and errors.
- Correcting errors in seasonal Arctic sea ice prediction of Earth system models with machine learning (The Cryosphere, 2025) ↗Source that cites the 41% error reduction and covers ML correction of a seasonal system.
- Improving short-term sea ice concentration forecasts using deep learning (The Cryosphere, 2024, via exa.ai record) ↗Record of the Palerme et al. paper behind the 41% figure; abstract not read.
Watch next
- Open the Palerme et al. paper to confirm what the 41% is measured against and at which lead times. Compare FloeNet's emulation skill with a plain climate model run.
Sources
- FloeNet: A mass-conserving global sea ice emulator that generalizes across climates (arXiv 2603.12449)arxiv.org
- IceCT: A MoE-Based Model for Seasonal Arctic Sea Ice Concentration and Thickness Prediction (AGU, Journal of Geophysical Research: Machine Learning and Computation)agupubs.onlinelibrary.wiley.com
- Correcting errors in seasonal Arctic sea ice prediction of Earth system models with machine learning (The Cryosphere, 2025)tc.copernicus.org
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 3 October 2026 at 22:03 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 →