Fusing Ice Surface Temperature With the AI4Arctic Dataset for Improved Deep Learning-Based Sea Ice Mapping
Bibliographic record
Abstract
Arctic sea ice mapping is vital for supporting marine navigation, climate monitoring, and efforts by northern communities to adapt to variable ice conditions. Automated mapping approaches can leverage freely accessible satellite data to supplement navigational ice charts, improve operational forecasting, and produce high-resolution sea ice parameter estimates. The AI4Arctic dataset enables deep learning-based mapping using synthetic aperture radar (SAR), passive microwave (PM), and reanalysis data. However, SAR and PM can struggle to resolve ice features due to ambiguous textures, atmospheric effects, and sensor limitations. To provide complementary data, an 84-scene VIIRS dataset is co-registered with AI4Arctic to evaluate whether ice surface temperature (IST) measurements can improve estimation of sea ice concentration (SIC), stage of development, and floe size. Input- and feature-level fusion methods, based on the U-Net architecture, are explored. Models are evaluated using the SIC R2 coefficient and SOD/FLOE F1-score, as well as predicted sea ice maps. Additionally, an alternative SIC accuracy score is introduced to assist with evaluating marginal ice predictions. Incorporating IST improves performance across all models compared to the AI4Arctic baseline; this includes single-encoder, dual-encoder, and multi-decoder U-Nets. Results highlight significant improvements in the prediction of open water under conditions with low-incidence angle, high atmospheric moisture, and wind roughening. Overall, the best performing dual-encoder model, DUE-Net-V, improves predictions by 2.18- 5.01% across all metrics, relative to the baseline. These results support integrating IST in deep learning workflows and highlight the potential for next-generation thermal-infrared sensors to improve automated sea ice mapping.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.002 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".