Pith. sign in

REVIEW 3 major objections 6 minor 67 references

A Bayesian High-Resolution Transformer can map pan-Arctic sea ice concentration at 200 m resolution with calibrated uncertainty, using geographically-weighted weak supervision and decision-level fusion of Sentinel-1, RCM, and AMSR2 data.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 19:04 UTC pith:PHN47G2E

load-bearing objection Solid engineering with one genuinely external validation (R²=0.90 vs ASI), but the headline feature-detection and ECE results are partly circular because they are measured against the same NIC and NASA Team products used in training. the 3 major comments →

arxiv 2603.03503 v1 pith:PHN47G2E submitted 2026-03-03 cs.CV cs.LG

Geographically-Weighted Weakly Supervised Bayesian High-Resolution Transformer for 200m Resolution Pan-Arctic Sea Ice Concentration Mapping and Uncertainty Estimation using Sentinel-1, RCM, and AMSR2 Data

Mabel Heffring , Lincoln Linlin Xu This is my paper
classification cs.CV cs.LG
keywords sea ice concentrationpan-ArcticSentinel-1passive microwaveTransformerBayesian neural networkweak supervisionuncertainty quantification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that pan-Arctic sea ice concentration (SIC) can be mapped at 200 m resolution, far finer than existing operational products, by training a high-resolution Transformer with a geographically-weighted weakly supervised loss. The key idea is to supervise the model at the region level against a coarse SIC product (NASA Team) while weighting pure open water and ice pack more heavily than the ambiguous marginal ice zone. The model also treats its parameters as random variables (a Bayesian neural network) to produce calibrated per-pixel uncertainties, and fuses Sentinel-1, RCM, and AMSR2 at the decision level to combine SAR resolution with daily passive-microwave coverage. The authors report 0.70 overall ice/water feature-detection accuracy on Sentinel-1 and R²=0.90 against the ASI product while preserving pan-Arctic SIC patterns. If true, this would be a significant step toward operational high-resolution sea-ice charting with usable uncertainty.

Core claim

On the paper's own terms, the central claim is that a high-resolution Transformer with separate global and local attention modules, trained with a geographically-weighted weakly supervised L1 loss and endowed with Bayesian parameter distributions, produces pan-Arctic SIC maps at 200 m spatial resolution that preserve large-scale SIC patterns (R²=0.90 relative to ASI) and detect small features such as leads and floes (0.70 overall ice/water accuracy on Sentinel-1). The same uncertainty estimates are better calibrated than MC dropout or epoch ensembles, with the lowest ECE and standard deviations (0.01–0.33%) for Sentinel-1.

What carries the argument

The load-bearing pieces are: (1) a high-resolution Transformer built from a GloFormer (global among-token attention) and a LoFormer (local within-token attention) stacked four times; (2) a geographically-weighted weakly supervised L1 loss that averages predicted SIC within NASA Team concentration clusters and weights samples by the NIC ice chart, down-weighting the marginal ice zone; (3) a Bayes-by-Backpropagation treatment that places a Gaussian distribution over attention parameters and adds a KL loss, giving predictive means and variances from 30 variational inferences; and (4) decision-level fusion that layers Sentinel-1 over RCM over AMSR2 89.0 GHz estimates.

Load-bearing premise

The feature-detection evaluation assumes that adaptive thresholding of the NIC ice chart produces reliable ice/water binary masks at 200 m resolution, and that the 300 manually selected samples faithfully represent the difficulty of the mapping task; since the same charts also set the training weights, the 0.70 accuracy may not be independent of the training signal.

What would settle it

Take the three validation dates and compare the model's predicted 200 m ice/water boundaries against Sentinel-2 (or other) 10 m optical imagery with manual ice/water labelling. If the 0.70 overall accuracy does not survive when the reference masks are derived from independent high-resolution imagery rather than from adaptive-thresholded NIC charts, the claim that the model resolves 200 m features is weakened. A second check: compute the R² against an ASI product derived from a different sensor (e.g., SSMIS) to rule out AMSR2 circularity.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • If correct, the approach yields 200 m pan-Arctic SIC maps with daily coverage when SAR is available, enabling detection of leads and floes that are invisible in ~3–12 km passive-microwave products.
  • The geographically-weighted weak supervision suggests a general recipe for training high-resolution models from coarse, location-dependent labels: weight by the reliability of the label region rather than assuming uniform pixel accuracy.
  • The Bayesian calibration result implies that operational ice products could carry per-pixel uncertainty estimates that are statistically consistent with residuals, which is required for data assimilation and ice forecasting.
  • The decision-level fusion framework provides a simple way to combine heterogeneous sensors without forcing them into a common feature space.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A testable extension would be to validate the 200 m feature claims against independent very-high-resolution imagery (e.g., 10 m optical) on the same dates; the current 0.70 accuracy is measured against adaptive-thresholded NIC charts, which are themselves region-level manual products and also set the training weights, so the metric may partly reflect the training signal.
  • If the geographically-weighted weighting is doing the work, then the method should be especially sensitive to the NIC chart quality in the MIZ; a perturbation study that shuffles or removes the MIZ weighting would separate the contribution of weak supervision from that of the architecture.
  • The model is trained only on September minimum-extent dates; a natural next step is to test whether the same architecture transfers across freeze-up and melt seasons, or whether separate seasonal experts are needed.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The manuscript presents a Bayesian High-Resolution Transformer for 200 m resolution pan-Arctic sea ice concentration (SIC) mapping from Sentinel-1, RCM, and AMSR2. The architecture combines a global among-token transformer (GloFormer) and a local within-token transformer (LoFormer). Training uses a geographically-weighted weakly supervised L1 loss: the model is supervised at SIC-cluster level against NASA Team SIC, with sample weights derived from NIC ice charts. Bayesian inference is implemented via Bayes-by-Backpropagation, and compared with MC dropout and epoch ensembles. Decision-level fusion layers Sentinel-1 over RCM and AMSR2 SIC/uncertainty maps. The paper reports 0.70 overall feature detection accuracy for Sentinel-1, R²=0.90 against the ASI product after downsampling, and lower ECE for the Bayesian model.

Significance. If the results are accepted, the paper would offer one of the first demonstrations of 200 m pan-Arctic SIC mapping with formal Bayesian uncertainty quantification, and a practical decision-level fusion scheme. The manuscript has solid internal methodology: the loss and Bayesian objectives are standard, the comparison against MC dropout and epoch ensembles is systematic, and ablations on LoFormer and the geographic weighting are helpful. The ASI-based pan-Arctic R² is an independent check at coarse scale. However, the two signature quantitative claims — 0.70 feature detection accuracy and superior calibration — rely on evaluation references that overlap with the training signals (NIC chart and NASA Team SIC), so the central evidence needs re-evaluation before the claims can be accepted.

major comments (3)
  1. [§4.7.2, Table 5; §4.2, Eq. (5)] The headline feature-detection accuracy (0.70 overall, Table 5) is evaluated against ice/water masks produced by adaptive thresholding of the NIC ice chart (§4.7.2), while the same NIC chart is used to define the geographical weights GW in the training loss (§4.2, Eq. (5)). Because the model is explicitly trained to assign higher weight to open-water and ice-pack regions identified from the NIC chart, agreement with a NIC-derived binary mask partially reflects the training signal rather than independent 200 m feature-resolution skill. The manual selection of 300 'high quality' samples where thresholding is accurate further risks selection bias. The authors should validate feature detection against an independent high-resolution reference (e.g., expert-delineated SAR ice/water masks) and specify that the MIZ/ice-pack thresholds are not derived from the same NIC chart used for training.
  2. [§5.2, Eq. (12)] The calibration claim (Bayesian Transformer ECE 0.0018–0.0028, Table 3) is computed with y_i in Eq. (12) taken from the NASA Team SIC product, which is also the weak-supervision target in the loss (Eq. (5)) and the basis for the SIC clusters in Section 4.2. Low ECE against the training label is therefore partly a measure of fit to that label, not of out-of-sample calibration. The comparison among MC dropout, ensemble, and BBB is informative only if all are evaluated against an independent reference. Please report ECE against the ASI product (with resolution-matched aggregation) or another independent reference, and state whether the ranking persists.
  3. [§5.3.1 and §5.5] The claimed 200 m capability is supported qualitatively by visual examples (Figs. 10–11) and by the circular feature-detection metric, while the quantitative pan-Arctic accuracy R²=0.90 is computed after averaging to the 3125 m ASI grid (§5.5). Downsampled agreement with ASI cannot substantiate detection of 1–2-pixel floes. The paper should either provide a non-circular, resolution-aware evaluation at 200 m or explicitly separate the validated coarse-scale SIC accuracy from the exploratory fine-scale feature-detection results.
minor comments (6)
  1. [§4.2 / Fig. 1] The numerical GW values (10, 5, 8 in Fig. 1 caption) are not defined in the text; please specify the mapping from NIC chart categories to GW and the source of these values.
  2. [§4.3, Eq. (8)] The sentence 'Assuming a diagonal Gaussian variational posterior, p(ω)=N(0,I)' appears to be a typo: the variational posterior should be denoted q_θ(ω), with p(ω)=N(0,I) the prior.
  3. [Algorithm 1] The Update operation in Algorithm 1 is undefined; please specify how the SIC and uncertainty layers are combined (overwrite, weighted average, etc.).
  4. [§5.3.2, Table 6] The architecture comparison is trained on a single day (2021/09/10) whereas the main results use seven training days; please clarify whether Table 6 is intended as a controlled comparison or note the different training set.
  5. [§4.1.1 / §4.1.2] Architectural hyperparameters (patch size, token count, number of heads, hidden dimension, MLP widths) are not stated; please provide them for reproducibility.
  6. [§5.5] For the AMSR2 branch, validation against the ASI product is not fully independent because both use the 89 GHz channel; state this caveat when reporting per-sensor R².

Circularity Check

2 steps flagged

Feature-detection and calibration claims are partly circular: NIC-chart weights used in training/model selection also define the evaluation masks, and ECE is computed against the NASA Team product used as training labels; only the ASI R² is independent.

specific steps
  1. other [Section 4.2 (geographically-weighted loss and validation selection) and Section 4.7.2/Section 5.3.1 (feature detection evaluation)]
    "The geographically-dependent weight of each sample, GW, is determined according to the daily U.S NIC ice charts, where more confident regions like open water and ice pack are weighted higher than the less confident MIZ. ... This loss is also used to determine the best epoch during validation. ... Feature detection accuracy is assessed by manually selecting 300 high quality validation samples where adaptive thresholding of the NIC ice chart has achieved accurate ice and water binary masks."

    The same NIC-chart product supplies the per-sample loss weights (including model selection on validation) and the binary ground-truth masks used to compute the headline 0.70 feature-detection accuracy. A model trained and selected to prioritize NIC-confident regions (open water/ice pack) is scored against a mask produced by thresholding that same NIC chart, with MIZ/pack thresholds also keyed to the chart's regions. The metric therefore largely measures agreement with the source that already steered training, rather than independent skill at resolving 200m-scale leads/floes.

  2. fitted input called prediction [Section 4.7.1 (ECE) vs Section 4.2 (loss on NASA Team SIC)]
    "z_i=||y_i−ŷ_i||/σ_i where y_i is the NASA Team SIC, ŷ_i is the mean predicted SIC, and σ_i is the mean predicted standard deviation. ... the difference between the predicted SIC and NASA Team ground truth is measured for each SIC cluster/region instead of each high-resolution pixel."

    The model's loss (Eq. 5) is minimized against cluster-level NASA Team SIC, and the same loss selects the best epoch on validation. The reported ECE (Table 3) is computed from residuals to NASA Team SIC. Hence the 'superior calibration' result is at least partly a self-consistency check against the training-label product rather than a calibration test against an independent reference. The paper's only external benchmark, ASI, is used for R²/MAE but not for ECE.

full rationale

The derivation chain is not fully circular. The architecture (GloFormer/LoFormer), the Bayesian BBB extension, the ablation studies, and the pan-Arctic accuracy comparison against the independent ASI product (R²=0.90) stand on their own and provide genuine evidence. However, two of the paper's signature evaluation quantities are measured against the same sources that steer training: the 0.70 feature-detection accuracy uses NIC-chart-derived binary masks while NIC charts also set the geographical weights and validation-based model selection; the low ECE values are computed against NASA Team SIC, the very product used as the weak supervision label. These are real but partial reductions—the model is not pixel-fitted to the evaluation masks, and the 2025 test dates are temporally held out—so the appropriate score is a moderate 5, not a higher score that would imply the entire derivation is forced.

Axiom & Free-Parameter Ledger

6 free parameters · 5 axioms · 0 invented entities

The method introduces no new physical entities. The key free parameters are the hand-set geographic weights and feature-detection thresholds; the main domain assumptions concern the validity of the coarse label products (NASA Team, NIC) and the chosen Bayesian restriction.

free parameters (6)
  • Geographical weights GW (open water, ice pack, MIZ) = Shown as GW=10, 5, 8 in Figure 1(c); not stated in text
    Hand-selected weights for the geographically-weighted loss (Eq 5); directly controls the training objective and is not derived from data or prior literature.
  • SIC cluster bin width = 10% SIC bins, K clusters from 0-100%
    Choice of bin width affects the aggregation in Eq (4) and the granularity of weak supervision; not optimized.
  • Feature-detection thresholds = 15% SIC in MIZ, 80% in ice pack
    Hand-set thresholds convert predicted SIC to binary ice/water masks (Sec 4.7.2); overall accuracy is sensitive to these values.
  • MC dropout rate = 10%
    Dropout probability used for the MC dropout baseline (Sec 4.4); affects the baseline comparison.
  • Variational inference sample count = 30
    Number of posterior samples used for predictive mean and variance (Sec 4.6); affects the uncertainty estimates.
  • Model precision tau in likelihood = Not specified
    Eq (6) introduces tau as precision; its value must balance the squared loss with the KL term in the total loss, but is never stated.
axioms (5)
  • domain assumption NASA Team SIC product is a valid weak-supervision signal at cluster level.
    Used as target in Eq (5); known to underestimate thin ice and have coarse resolution (Sec 3.2), which the loss is designed to tolerate.
  • domain assumption NIC ice charts provide reliable region-level confidence and a valid basis for 200m ice/water ground truth via adaptive thresholding.
    GW weights in Sec 4.2 and feature-detection ground truth in Sec 4.7.2 both derive from NIC; charts are region-level/manual, so this is a strong unverified assumption.
  • domain assumption ASI AMSR2 product is an appropriate independent benchmark for pan-Arctic SIC.
    Used for validation (Sec 3.2, 5.5); independent algorithm but same satellite family, with known systematic difference relative to NASA Team.
  • domain assumption Restricting Bayesian inference to attention parameters adequately captures epistemic uncertainty.
    Sec 4.3/Fig 1b; only attention weights are random; validity untested.
  • standard math Bayes theorem, KL divergence, softmax attention are standard math.
    Used in Eqs (2),(3),(7),(8); no dispute.

pith-pipeline@v1.3.0-alltime-deepseek · 27397 in / 14804 out tokens · 133030 ms · 2026-08-02T19:04:16.909579+00:00 · methodology

0 comments
read the original abstract

Although high-resolution mapping of pan-Arctic sea ice with reliable corresponding uncertainty is essential for operational sea ice concentration (SIC) charting, it is a difficult task due to key challenges, such as the subtle nature of ice signature features, inexact SIC labels, model uncertainty, and data heterogeneity. This study presents a novel Bayesian High-Resolution Transformer approach for 200 meter resolution pan-Arctic SIC mapping and uncertainty quantification using Sentinel-1, RADARSAT Constellation Mission (RCM), and Advanced Microwave Scanning Radiometer 2 (AMSR2) data. First, to improve small and subtle sea ice feature (e.g., cracks/leads, ponds, and ice floes) extraction, we design a novel high-resolution Transformer model with both global and local modules that can better discern the subtle differences in sea ice patterns. Second, to address low-resolution and inexact SIC labels, we design a geographically-weighted weakly supervised loss function to supervise the model at region level instead of pixel level, and to prioritize pure open water and ice pack signatures while mitigating the impact of ambiguity in the marginal ice zone (MIZ). Third, to improve uncertainty quantification, we design a Bayesian extension of the proposed Transformer model, treating its parameters as random variables to more effectively capture uncertainties. Fourth, to address data heterogeneity, we fuse three different data types (Sentinel-1, RCM, and AMSR2) at decision-level to improve both SIC mapping and uncertainty quantification. The proposed approach is evaluated under pan-Arctic minimum-extent conditions in 2021 and 2025. Results demonstrate that the proposed model achieves 0.70 overall feature detection accuracy using Sentinel-1 data, while also preserving pan-Arctic SIC patterns (Sentinel-1 R\textsuperscript{2} = 0.90 relative to the ARTIST Sea Ice product).

Figures

Figures reproduced from arXiv: 2603.03503 by Lincoln Linlin Xu, Mabel Heffring.

Figure 1
Figure 1. Figure 1: Overview of our key innovations for pan-Arctic SIC mapping with corresponding uncertainty quantification. The proposed model utilizes a (a) High-Resolution Transformer architecture, 𝑓 𝜔 (.), with an among-token Transformer block for modeling global context (GloFormer) and a within-token Transformer block for modeling local detail (LoFormer). The high-resolution Transformer becomes a (b) Bayesian Neural Net… view at source ↗
Figure 2
Figure 2. Figure 2: Theoretical Comparison of Epistemic Uncertainty Quantification Approaches. Figure (a) illustrates the Bayesian Neural Network (BNN) where the model parameters are assumed to be probabilistic. Figure (b) illustrates MC dropout where red denotes masked nodes and green denotes masked connections. With each variational inference the dropout configuration changes, introducing randomness into the neural network.… view at source ↗
Figure 3
Figure 3. Figure 3: Overview of data downloaded and processed for September 4th, 2021. Includes Sentinel-1 (a) HH, (b) HV, and (c) HH and HV cross-polarization, RCM (f) HH, (g) HV, and (h) HH and HV cross-polarization, AMSR2 (d) 89.0 GHz H and (e) 89.0 GHz V data, as well as (i) NIC Ice Chart and (j) NASA Team SIC products. Algorithm 1 Decision-Level Data Fusion for Pan-Arctic SIC and Uncertainty Mapping 1: Input: 𝑆𝑆𝑒𝑛𝑡𝑖𝑛𝑒𝑙−1… view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of predicted SIC (left), the ice/water binary ground truth mask derived using adaptive thresholding of the NIC ice chart (middle), and corresponding SAR HV (right) for one 256 x 256 validation sample. 4.7.2. Feature Detection Accuracy Feature detection accuracy is assessed by manually se￾lecting 300 high quality validation samples where adaptive thresholding of the NIC ice chart has achieved acc… view at source ↗
Figure 5
Figure 5. Figure 5: Pan-Arctic SIC from the Bayesian High-Resolution Transformer for validation data on September 4th, 18th, and 27th, 2021. SIC derived from (a) Sentinel-1, (b) RCM, and (c) AMSR2, and (d) is the final fused map of all data sources with Sentinel-1 layered on top, followed by RCM and AMSR2. SIC estimates are visually compared to (e) NASA Team SIC and (f) NIC Ice Chart [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: (a) Fused Pan-Arctic SIC map and corresponding uncertainty from the Bayesian High-Resolution Transformer for validation data on September 4th, 18th, and 27th, 2021. SIC and uncertainty are derived from (b) Sentinel-1, (c) RCM, and (d) AMSR2, and (e) is the final fused uncertainty map of all data sources with Sentinel-1 layered on top, followed by RCM and AMSR2. Bayesian Transformer across data types suppor… view at source ↗
Figure 7
Figure 7. Figure 7: Pan-Arctic SIC from the Bayesian High-Resolution Transformer for test data on September 3rd, 10th, and 30th, 2025. SIC derived from (a) Sentinel-1, (b) RCM, and (c) AMSR2, and (d) is the final fused map of all data sources with Sentinel-1 layered on top, followed by RCM and AMSR2. SIC estimates are visually compared to (e) NASA Team SIC and (f) NIC Ice Chart [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: (a) Fused Pan-Arctic SIC map and corresponding uncertainty from the Bayesian High-Resolution Transformer for test data on September 3rd, 10th, and 30th, 2025. SIC and uncertainty are derived from (b) Sentinel-1, (c) RCM, and (d) AMSR2, and (e) is the final fused uncertainty map of all data sources with Sentinel-1 layered on top, followed by RCM and AMSR2. given by the Bayesian Transformer model. The least … view at source ↗
Figure 9
Figure 9. Figure 9: Comparison of uncertainty quantification approaches for Sentinel-1, RCM, and AMSR2. Uncertainty estimates are shown using (a) MC Dropout, (b) Epoch Ensemble, and (c) BBB implemented in the Bayesian Transformer for September 4th, 2021. The distribution of uncertainty for all validation samples are shown using (d) Gaussian Kernel Density curves for each dataset with corresponding mean and standard deviation … view at source ↗
Figure 10
Figure 10. Figure 10: Local visual comparison of SIC derived from Sentinel-1 on September 4th, 2021, where (a) NASA Team SIC, (b) Sentinel-1 HV Imagery, (c) Sentinel-1 HH Imagery, (d) Deterministic Transformer SIC, (e) Mean Monte Carlo Dropout SIC, (f) Mean Epoch Ensemble SIC, and (g) Mean Bayesian Transformer SIC (our approach) [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Local visual comparison of SIC derived from RCM on September 4th, 2021, where (a) NASA Team SIC, (b) RCM HV Imagery, (c) RCM HH Imagery, (d) Deterministic Transformer SIC, (e) Mean Monte Carlo Dropout SIC, (f) Mean Epoch Ensemble SIC, and (g) Mean Bayesian Transformer SIC (our approach). : Preprint submitted to Elsevier Page 16 of 23 [PITH_FULL_IMAGE:figures/full_fig_p016_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Local visual comparison of SIC in MIZ region derived from Sentinel-1 on September 4th, 2021 using (a) U-Net, (b) HRNet, (c) ConvNeXt, (d) Swin Transformer, (e) SegFormer, and (f) High-Resolution (HR) Transformer (our approach). SIC predictions are compared to (g) SAR HV and (h) NASA Team. Our approach detects small sea ice floes and ice boundaries in the MIZ while maintaining large-scale SIC patterns [PI… view at source ↗
Figure 13
Figure 13. Figure 13: Local visual comparison of SIC in ice pack region derived from Sentinel-1 on September 4th, 2021 using (a) U-Net, (b) HRNet, (c) ConvNeXt, (d) Swin Transformer, (e) SegFormer, and (f) High-Resolution (HR) Transformer (our approach). SIC predictions are compared to (g) SAR HV and (h) NASA Team. Our approach detects small leads/cracks in ice pack while maintaining large-scale SIC patterns [PITH_FULL_IMAGE:… view at source ↗
Figure 14
Figure 14. Figure 14: Local visual comparison of SIC in near open water region derived from Sentinel-1 on September 4th, 2021 using (a) U-Net, (b) HRNet, (c) ConvNeXt, (d) Swin Transformer, (e) SegFormer, and (f) High-Resolution (HR) Transformer (our approach). SIC predictions are compared to (g) SAR HV and (h) NASA Team. Our approach detects thin ice without over estimation. : Preprint submitted to Elsevier Page 18 of 23 [PI… view at source ↗
Figure 15
Figure 15. Figure 15: Local visual comparison of SIC in MIZ region derived from RCM on September 18th, 2021 using (a) U-Net, (b) HRNet, (c) ConvNeXt, (d) Swin Transformer, (e) SegFormer, and (f) High-Resolution (HR) Transformer (our approach). SIC predictions are compared to (g) SAR HV and (h) NASA Team. Our approach detects small sea ice floes and ice boundaries in the MIZ while maintaining large-scale SIC patterns [PITH_FUL… view at source ↗
Figure 16
Figure 16. Figure 16: Local visual comparison of SIC in ice pack region derived from RCM on September 18th, 2021 using (a) U-Net, (b) HRNet, (c) ConvNeXt, (d) Swin Transformer, (e) SegFormer, and (f) High-Resolution (HR) Transformer (our approach). SIC predictions are compared to (g) SAR HV and (h) NASA Team. Our approach detects small leads/cracks in ice pack while maintaining large-scale SIC patterns [PITH_FULL_IMAGE:figure… view at source ↗
Figure 17
Figure 17. Figure 17: Local visual comparison of SIC in near open water region derived from RCM on September 18th, 2021 using (a) U-Net, (b) HRNet, (c) ConvNeXt, (d) Swin Transformer, (e) SegFormer, and (f) High-Resolution (HR) Transformer (our approach). SIC predictions are compared to (g) SAR HV and (h) NASA Team. Our approach detects thin ice while mitigating the impact of incidence angle effects and thermal banding noise. … view at source ↗
Figure 18
Figure 18. Figure 18: Predicted Bayesian Transformer SIC accuracy vs. ASI SIC scatter plot for all 2021 validation samples. SIC estimates are aggregated into 11 bins ranging from 0% to 100% for visualization, while the R2, RMSE, and MAE are calculated using all individual SIC estimates. There was stronger agreement across Sentinel-1, RCM, and AMSR2 within the validation datasets compared to the test datasets. However, the over… view at source ↗
Figure 20
Figure 20. Figure 20: Training data distribution of NASA Team SIC in MIZ and Ice Pack according to the NIC Ice Chart. 6. Conclusion Accurate daily pan-Arctic SIC mapping with corre￾sponding uncertainty quantification is crucial for monitoring climate change, supporting climate adaptation for Northern communities, and ensuring safe navigation as sea ice extents continue to decrease. However, this is a challenging task due to th… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

67 extracted references · 17 canonical work pages

  1. [1]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in ":" * " " * FUNCTION f...

  2. [2]

    , author Pourpanah, F

    author Abdar, M. , author Pourpanah, F. , author Hussain, S. , author Rezazadegan, D. , author Liu, L. , author Ghavamzadeh, M. , author Fieguth, P. , author Cao, X. , author Khosravi, A. , author Acharya, U.R. , author Makarenkov, V. , author Nahavandi, S. , year 2021 . title A review of uncertainty quantification in deep learning: Techniques, applicatio...

    2021

  3. [3]

    , author Prigent, C

    author Aires, F. , author Prigent, C. , author Rossow, W.B. , year 2004 . title Neural network uncertainty assessment using bayesian statistics: A remote sensing application . journal Neural computation volume 16 , pages 2415--2458

    2004

  4. [4]

    , author Hosking, J.S

    author Andersson, T.R. , author Hosking, J.S. , author Pérez-Ortiz, M. , author Paige, B. , author Elliott, A. , author Russell, C. , author Law, S. , author Jones, D.C. , author Wilkinson, J. , author Phillips, T. , author Byrne, J. , author Tietsche, S. , author Sarojini, B.B. , author Blanchard-Wrigglesworth, E. , author Aksenov, Y. , author Downie, R....

    2021

  5. [5]

    , author Scott, K.A

    author Asadi, N. , author Scott, K.A. , author Komarov, A.S. , author Buehner, M. , author Clausi, D.A. , year 2020 . title Evaluation of a neural network with uncertainty for detection of ice and water in sar imagery . journal IEEE Transactions on Geoscience and Remote Sensing volume 59 , pages 247--259

    2020

  6. [6]

    , author Pahlavani, P

    author Bigdeli, B. , author Pahlavani, P. , author Amirkolaee, H.A. , year 2021 . title An ensemble deep learning method as data fusion system for remote sensing multisensor classification . journal Applied Soft Computing volume 110 , pages 107563 . https://www.sciencedirect.com/science/article/pii/S1568494621004841, :https://doi.org/10.1016/j.asoc.2021.107563

    arXiv 2021

  7. [7]

    , author Gloersen, P

    author Cavalieri, D.J. , author Gloersen, P. , author Campbell, W.J. , year 1984 . title Determination of sea ice parameters with the nimbus 7 smmr . journal Journal of Geophysical Research: Atmospheres volume 89 , pages 5355--5369 . https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1029/JD089iD04p05355, :https://doi.org/10.1029/JD089iD04p05355

    work page doi:10.1029/jd089id04p05355 1984

  8. [8]

    , author Brunsdon, C

    author Charlton, M. , author Brunsdon, C. , author Fotheringham, A. , year 1998 . title Geographically weighted regression-modelling spatial non-stationarity . journal Journal of The Royal Statistical Society, Series D: The Statistician

    1998

  9. [9]

    , author Chen, K

    author Chen, B. , author Chen, K. , author Yang, M. , author Zou, Z. , author Shi, Z. , year 2025 a. title Heterogeneous mixture of experts for remote sensing image super-resolution . journal IEEE Geoscience and Remote Sensing Letters volume 22 , pages 1--5 . :10.1109/LGRS.2025.3557928

    arXiv 2025

  10. [10]

    , author Yao, S

    author Chen, S. , author Yao, S. , author Wang, Z. , author Chen, S. , year 2025 b. title Sea ice concentration prediction for arctic shipping: A transformer-based approach with latent uncertainty modeling , in: booktitle 2025 6th International Conference on Big Data & Artificial Intelligence & Software Engineering (ICBASE) , pp. pages 476--479 . :10.1109...

    arXiv 2025

  11. [11]

    , author Patel, M

    author Chen, X. , author Patel, M. , author Xu, L. , author Chen, Y. , author Scott, K.A. , author Clausi, D.A. , year 2024 . title Weakly supervised learning for pixel-level sea ice concentration extraction using ai4arctic sea ice challenge dataset . journal IEEE Geoscience and Remote Sensing Letters volume 21 , pages 1--5 . :10.1109/LGRS.2023.3338061

    arXiv 2024

  12. [12]

    , author Scott, K.A

    author Chen, X. , author Scott, K.A. , author Xu, L. , author Jiang, M. , author Fang, Y. , author Clausi, D.A. , year 2023 a. title Uncertainty-incorporated ice and open water detection on dual-polarized sar sea ice imagery . journal IEEE Transactions on Geoscience and Remote Sensing volume 61 , pages 1--13 . :10.1109/TGRS.2022.3233871

    arXiv 2023

  13. [13]

    , author Valencia, R

    author Chen, X. , author Valencia, R. , author Soleymani, A. , author Scott, K.A. , year 2023 b. title Predicting sea ice concentration with uncertainty quantification using passive microwave and reanalysis data: A case study in baffin bay . journal IEEE Transactions on Geoscience and Remote Sensing volume 61 , pages 1--13 . :10.1109/TGRS.2023.3250164

    arXiv 2023

  14. [14]

    , author Yue, B

    author Clausi, D. , author Yue, B. , year 2004 . title Comparing cooccurrence probabilities and markov random fields for texture analysis of sar sea ice imagery . journal IEEE Transactions on Geoscience and Remote Sensing volume 42 , pages 215--228 . :10.1109/TGRS.2003.817218

    arXiv 2004

  15. [15]

    , year 2001

    author Clausi, D.A. , year 2001 . title Comparison and fusion of co-occurrence, G abor and MRF texture features for classification of SAR sea-ice imagery . journal Atmosphere-Ocean volume 39 , pages 183--194 . :10.1080/07055900.2001.9649675

    arXiv 2001

  16. [16]

    , author Scott, K.A

    author Cooke, C.L.V. , author Scott, K.A. , year 2019 . title Estimating sea ice concentration from sar: Training convolutional neural networks with passive microwave data . journal IEEE Transactions on Geoscience and Remote Sensing volume 57 , pages 4735--4747 . :10.1109/TGRS.2019.2892723

    arXiv 2019

  17. [17]

    , author Clausi, D

    author Deng, H. , author Clausi, D. , year 2005 . title Unsupervised segmentation of synthetic aperture radar sea ice imagery using a novel markov random field model . journal IEEE Transactions on Geoscience and Remote Sensing volume 43 , pages 528--538 . :10.1109/TGRS.2004.839589

    arXiv 2005

  18. [18]

    , author Lu, H

    author Ding, R. , author Lu, H. , author Liu, M. , year 2025 . title Denseformer-moe: A dense transformer foundation model with mixture of experts for multi-task brain image analysis . journal IEEE Transactions on Medical Imaging volume 44 , pages 4037--4048 . :10.1109/TMI.2025.3551514

    arXiv 2025

  19. [19]

    , author Workneh, A.C

    author Dugassa, D.D. , author Workneh, A.C. , author Yersaw, B.T. , author Sedeta, G.E. , author Chane, M.B. , author Tola, S.Y. , author Ousman, S.A. , author Birhan, Z.A. , year 2026 . title Deep learning for epistemic uncertainty in smap-derived soil moisture estimates over the kulfo watershed, ethiopia . journal Science of Remote Sensing volume 13 , p...

    arXiv 2026

  20. [20]

    , author Chanussot, J

    author Fauvel, M. , author Chanussot, J. , author Benediktsson, J. , year 2006 . title Decision fusion for the classification of urban remote sensing images . journal IEEE Transactions on Geoscience and Remote Sensing volume 44 , pages 2828--2838 . :10.1109/TGRS.2006.876708

    arXiv 2006

  21. [21]

    , author Sheffield, J

    author Ferguson, C.R. , author Sheffield, J. , author Wood, E.F. , author Gao, H. , year 2010 . title Quantifying uncertainty in a remote sensing-based estimate of evapotranspiration over continental usa . journal International Journal of Remote Sensing volume 31 , pages 3821--3865 . https://doi.org/10.1080/01431161.2010.483490, :10.1080/01431161.2010.483...

    arXiv 2010

  22. [22]

    , author Lynch, H.J

    author Gonçalves, B.C. , author Lynch, H.J. , year 2021 . title Fine-scale sea ice segmentation for high-resolution satellite imagery with weakly-supervised cnns . journal Remote Sensing volume 13 . https://www.mdpi.com/2072-4292/13/18/3562, :10.3390/rs13183562

    work page doi:10.3390/rs13183562 2021

  23. [23]

    , author Colin, A

    author de Gélis, I. , author Colin, A. , author Longépé, N. , year 2021 . title Prediction of categorized sea ice concentration from sentinel-1 sar images based on a fully convolutional network . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing volume 14 , pages 5831--5841 . :10.1109/JSTARS.2021.3074068

    arXiv 2021

  24. [24]

    , author Zhao, Y

    author He, J. , author Zhao, Y. , author Yang, D. , author Wang, H. , author Deng, X. , year 2025 . title Physically constrained spatiotemporal deep learning model for fine-scale, long-term arctic sea ice concentration prediction . journal IEEE Transactions on Geoscience and Remote Sensing volume 63 , pages 1--21 . :10.1109/TGRS.2025.3549823

    arXiv 2025

  25. [25]

    , author Zhao, Y

    author He, J. , author Zhao, Y. , author Yang, S. , author Wu, W. , author Wang, J. , author Deng, X. , year 2026 . title Explainable spatiotemporal deep learning for subseasonal super-resolution forecasting of arctic sea ice concentration during the melting season . journal ISPRS Journal of Photogrammetry and Remote Sensing volume 232 , pages 1--17 . htt...

    work page doi:10.1016/j.isprsjprs.2025.11.027 2026

  26. [26]

    , author Gottschling, N.M

    author Iagaru, D. , author Gottschling, N.M. , year 2023 . title Uncertainty quantification with deep ensemble methods for super-resolution of sentinel 2 satellite images . journal Physical Sciences Forum volume 9 . https://www.mdpi.com/2673-9984/9/1/4, :10.3390/psf2023009004

    work page doi:10.3390/psf2023009004 2023

  27. [27]

    , author Yang, F

    author Jakubik, J. , author Yang, F. , author Blumenstiel, B. , author Scheurer, E. , author Sedona, R. , author Maurogiovanni, S. , author Bosmans, J. , author Dionelis, N. , author Marsocci, V. , author Kopp, N. , author Ramachandran, R. , author Fraccaro, P. , author Brunschwiler, T. , author Cavallaro, G. , author Bernabe-Moreno, J. , author Longépé, ...

    Pith/arXiv arXiv 2025

  28. [28]

    , author Taleghan, S.A

    author Jalayer, S. , author Taleghan, S.A. , author Lima, R.P.d. , author Vahedi, B. , author Hughes, N. , author Banaei-Kashani, F. , author Karimzadeh, M. , year 2025 . title Enhancing and Interpreting Deep Learning for Sea Ice Charting using the AutoICE Benchmark . journal Remote Sensing Applications: Society and Environment volume 38 , pages 101538 . ...

    arXiv 2025

  29. [29]

    , author Yang, D

    author Jia, H. , author Yang, D. , author Deng, W. , author Wei, Q. , author Jiang, W. , year 2021 . title Predicting land surface temperature with geographically weighed regression and deep learning . journal Wiley Interdisciplinary Reviews: Data Mining and Knowledge Discovery volume 11 , pages e1396

    2021

  30. [30]

    , author Clausi, D.A

    author Jiang, M. , author Clausi, D.A. , author Xu, L. , year 2022 . title Sea-ice mapping of radarsat-2 imagery by integrating spatial contexture with textural features . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing volume 15 , pages 7964--7977 . :10.1109/JSTARS.2022.3205849

    arXiv 2022

  31. [31]

    , author Guo, B

    author Jiang, Z. , author Guo, B. , author Zhao, H. , author Jiang, Y. , author Sun, Y. , year 2024 . title Sicformer: A 3d-swin transformer for sea ice concentration prediction . journal Journal of Marine Science and Engineering volume 12 . https://www.mdpi.com/2077-1312/12/8/1424, :10.3390/jmse12081424

    work page doi:10.3390/jmse12081424 2024

  32. [32]

    , author Dierking, W

    author Khachatrian, E. , author Dierking, W. , author Chlaily, S. , author Eltoft, T. , author Dinessen, F. , author Hughes, N. , author Marinoni, A. , year 2023 . title Sar and passive microwave fusion scheme: A test case on sentinel-1/amsr-2 for sea ice classification . journal Geophysical Research Letters volume 50 , pages e2022GL102083 . https://agupu...

    work page doi:10.1029/2022gl102083 2023

  33. [33]

    , author Komarov, S.A

    author Komarov, A.S. , author Komarov, S.A. , author Buehner, M. , year 2022 . title Ocean surface wind speed retrieval from the radarsat constellation mission co- and cross-polarization images without wind direction input . journal IEEE Transactions on Geoscience and Remote Sensing volume 60 , pages 1--15 . :10.1109/TGRS.2021.3137127

    arXiv 2022

  34. [34]

    , author Shen, H

    author Li, T. , author Shen, H. , author Yuan, Q. , author Zhang, L. , year 2020 . title Geographically and temporally weighted neural networks for satellite-based mapping of ground-level pm2.5 . journal ISPRS Journal of Photogrammetry and Remote Sensing volume 167 , pages 178--188 . https://www.sciencedirect.com/science/article/pii/S0924271620301775, :ht...

    work page doi:10.1016/j.isprsjprs.2020.06.019 2020

  35. [35]

    , author Hsu, C.Y

    author Li, W. , author Hsu, C.Y. , author Tedesco, M. , year 2024 . title Advancing arctic sea ice remote sensing with ai and deep learning: Opportunities and challenges . journal Remote Sensing volume 16 . https://www.mdpi.com/2072-4292/16/20/3764, :10.3390/rs16203764

    work page doi:10.3390/rs16203764 2024

  36. [36]

    , author Karimzadeh, M

    author Pires de Lima, R. , author Karimzadeh, M. , year 2023 . title Model ensemble with dropout for uncertainty estimation in sea ice segmentation using sentinel-1 sar . journal IEEE Transactions on Geoscience and Remote Sensing volume 61 , pages 1--15 . :10.1109/TGRS.2023.3331276

    arXiv 2023

  37. [37]

    , author Guo, H

    author Liu, H. , author Guo, H. , author Zhang, L. , year 2014 . title SVM -based sea ice classification using textural features and concentration from RADARSAT-2 dual-pol ScanSAR data . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing volume 8 , pages 1601--1613 . :10.1109/JSTARS.2014.2365215

    arXiv 2014

  38. [38]

    , author Lin, Y

    author Liu, Z. , author Lin, Y. , author Cao, Y. , author Hu, H. , author Wei, Y. , author Zhang, Z. , author Lin, S. , author Guo, B. , year 2021 . title Swin transformer: Hierarchical vision transformer using shifted windows , in: booktitle Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV) , pp. pages 10012--10022

    2021

  39. [39]

    , author Mao, H

    author Liu, Z. , author Mao, H. , author Wu, C.Y. , author Feichtenhofer, C. , author Darrell, T. , author Xie, S. , year 2022 . title A convnet for the 2020s . https://arxiv.org/abs/2201.03545, arXiv:2201.03545 http://arxiv.org/abs/2201.03545

    Pith/arXiv arXiv 2022

  40. [40]

    , author Pedersen, L.T

    author Malmgren-Hansen, D. , author Pedersen, L.T. , author Nielsen, A.A. , author Kreiner, M.B. , author Saldo, R. , author Skriver, H. , author Lavelle, J. , author Buus-Hinkler, J. , author Krane, K.H. , year 2021 . title A convolutional neural network architecture for sentinel-1 and amsr2 data fusion . journal IEEE Transactions on Geoscience and Remot...

    arXiv 2021

  41. [41]

    , author Álvaro Moreno-Martínez , author Campos-Taberner, M

    author Martínez-Ferrer, L. , author Álvaro Moreno-Martínez , author Campos-Taberner, M. , author García-Haro, F.J. , author Muñoz-Marí, J. , author Running, S.W. , author Kimball, J. , author Clinton, N. , author Camps-Valls, G. , year 2022 . title Quantifying uncertainty in high resolution biophysical variable retrieval with machine learning . journal Re...

    arXiv 2022

  42. [42]

    , author Cooper, G

    author Naeini, M.P. , author Cooper, G. , author Hauskrecht, M. , year 2015 . title Obtaining well calibrated probabilities using bayesian binning , in: booktitle Proceedings of the AAAI conference on artificial intelligence

    2015

  43. [43]

    , author Clausi, D.A

    author Ochilov, S. , author Clausi, D.A. , year 2012 . title Operational sar sea-ice image classification . journal IEEE Transactions on Geoscience and Remote Sensing volume 50 , pages 4397--4408 . :10.1109/TGRS.2012.2192278

    arXiv 2012

  44. [44]

    , author Minasny, B

    author Padarian, J. , author Minasny, B. , author McBratney, A. , year 2022 . title Assessing the uncertainty of deep learning soil spectral models using monte carlo dropout . journal Geoderma volume 425 , pages 116063 . https://www.sciencedirect.com/science/article/pii/S0016706122003706, :https://doi.org/10.1016/j.geoderma.2022.116063

    arXiv 2022

  45. [45]

    , author Scott, K.A

    author Radhakrishnan, K. , author Scott, K.A. , author Clausi, D.A. , year 2021 . title Sea ice concentration estimation: Using passive microwave and sar data with a u-net and curriculum learning . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing volume 14 , pages 5339--5351 . :10.1109/JSTARS.2021.3076109

    arXiv 2021

  46. [46]

    , author Li, X

    author Ren, Y. , author Li, X. , author Wang, Y. , year 2025 . title Sicnet _season v1.0: a transformer-based deep learning model for seasonal arctic sea ice prediction by incorporating sea ice thickness data . journal Geoscientific Model Development volume 18 , pages 2665--2678 . https://gmd.copernicus.org/articles/18/2665/2025/, :10.5194/gmd-18-2665-2025

    work page doi:10.5194/gmd-18-2665-2025 2025

  47. [47]

    , author Li, X

    author Ren, Y. , author Li, X. , author Zhang, W. , year 2022 . title A data-driven deep learning model for weekly sea ice concentration prediction of the pan-arctic during the melting season . journal IEEE Transactions on Geoscience and Remote Sensing volume 60 , pages 1--19 . :10.1109/TGRS.2022.3177600

    arXiv 2022

  48. [48]

    , author Fischer, P

    author Ronneberger, O. , author Fischer, P. , author Brox, T. , year 2015 . title U-net: Convolutional networks for biomedical image segmentation , in: editor Navab, N. , editor Hornegger, J. , editor Wells, W.M. , editor Frangi, A.F. (Eds.), booktitle Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015 , publisher Springer Internati...

    2015

  49. [49]

    , author Ries, D

    author Sahay, R. , author Ries, D. , author Zollweg, J.D. , author Brinton, C.G. , year 2021 . title Hyperspectral image target detection using deep ensembles for robust uncertainty quantification , in: booktitle 2021 55th Asilomar Conference on Signals, Systems, and Computers , pp. pages 1715--1719 . :10.1109/IEEECONF53345.2021.9723384

    arXiv 2021

  50. [50]

    , author Zhu, X.X

    author Schmitt, M. , author Zhu, X.X. , year 2016 . title Data fusion and remote sensing: An ever-growing relationship . journal IEEE Geoscience and Remote Sensing Magazine volume 4 , pages 6--23 . :10.1109/MGRS.2016.2561021

    arXiv 2016

  51. [51]

    , author Kaleschke, L

    author Spreen, G. , author Kaleschke, L. , author Heygster, G. , year 2008 . title Sea ice remote sensing using amsr-e 89-ghz channels . journal Journal of Geophysical Research: Oceans volume 113 . :https://doi.org/10.1029/2005JC003384

    work page doi:10.1029/2005jc003384 2008

  52. [52]

    , author Wulf, T

    author Stokholm, A. , author Wulf, T. , author Kucik, A. , author Saldo, R. , author Buus-Hinkler, J. , author Hvidegaard, S.M. , year 2022 . title Ai4seaice: Toward solving ambiguous sar textures in convolutional neural networks for automatic sea ice concentration charting . journal IEEE Transactions on Geoscience and Remote Sensing volume 60 , pages 1--...

    arXiv 2022

  53. [53]

    , author Scott, K.A

    author Tamber, M.S. , author Scott, K.A. , author Pedersen, L.T. , year 2022 . title Accounting for label errors when training a convolutional neural network to estimate sea ice concentration using operational ice charts . journal IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing volume 15 , pages 1502--1513 . :10.1109/JSTAR...

    arXiv 2022

  54. [54]

    , author Sun, K

    author Wang, J. , author Sun, K. , author Cheng, T. , author Jiang, B. , author Deng, C. , author Zhao, Y. , author Liu, D. , author Mu, Y. , author Tan, M. , author Wang, X. , author Liu, W. , author Xiao, B. , year 2020 . title Deep high-resolution representation learning for visual recognition . https://arxiv.org/abs/1908.07919, arXiv:1908.07919 http:/...

    Pith/arXiv arXiv 2020

  55. [55]

    , author Scott, K.A

    author Wang, L. , author Scott, K.A. , author Clausi, D.A. , year 2016 a. title Improved sea ice concentration estimation through fusing classified sar imagery and amsr-e data . journal Canadian Journal of Remote Sensing volume 42 , pages 41--52 . https://doi.org/10.1080/07038992.2016.1152547, :10.1080/07038992.2016.1152547, arXiv:https://doi.org/10.1080/...

    arXiv 2016

  56. [56]

    , author Scott, K.A

    author Wang, L. , author Scott, K.A. , author Clausi, D.A. , year 2017 . title Sea ice concentration estimation during freeze-up from sar imagery using a convolutional neural network . journal Remote Sensing volume 9 . https://www.mdpi.com/2072-4292/9/5/408, :10.3390/rs9050408

    work page doi:10.3390/rs9050408 2017

  57. [57]

    , author Scott, K.A

    author Wang, L. , author Scott, K.A. , author Xu, L. , author Clausi, D.A. , year 2016 b. title Sea ice concentration estimation during melt from dual-pol sar scenes using deep convolutional neural networks: A case study . journal IEEE Transactions on Geoscience and Remote Sensing volume 54 , pages 4524--4533 . :10.1109/TGRS.2016.2543660

    arXiv 2016

  58. [58]

    , author Yang, W

    author Wang, W. , author Yang, W. , author Wang, L. , author Wang, G. , author Lei, R. , year 2025 a. title Seasonal forecasting of Pan - Arctic sea ice with state space model . journal npj Climate and Atmospheric Science volume 8 , pages 172 . https://www.nature.com/articles/s41612-025-01058-0, :10.1038/s41612-025-01058-0

    work page doi:10.1038/s41612-025-01058-0 2025

  59. [59]

    , author Zhao, C

    author Wang, W. , author Zhao, C. , author Wu, Y. , year 2024 . title Spatial weighting — an effective incorporation of geological expertise into deep learning models . journal Geochemistry volume 84 , pages 126212 . https://www.sciencedirect.com/science/article/pii/S0009281924001375, :https://doi.org/10.1016/j.chemer.2024.126212

    arXiv 2024

  60. [60]

    , author Li, X.M

    author Wang, Y.R. , author Li, X.M. , year 2021 . title Arctic sea ice cover data from spaceborne synthetic aperture radar by deep learning . journal Earth System Science Data volume 13 , pages 2723--2742 . https://essd.copernicus.org/articles/13/2723/2021/, :10.5194/essd-13-2723-2021

    work page doi:10.5194/essd-13-2723-2021 2021

  61. [61]

    , author Zuo, R

    author Wang, Z. , author Zuo, R. , author Kreuzer, O.P. , year 2025 b. title Uncertainty Quantification of Deep Learning Algorithms for Lithological Mapping . journal Mathematical Geosciences https://doi.org/10.1007/s11004-025-10235-z, :10.1007/s11004-025-10235-z

    work page doi:10.1007/s11004-025-10235-z 2025

  62. [62]

    , author Yusong, W

    author Wenbo, W. , author Yusong, W. , author Xue, D. , author Xiaotong, J. , author Yida, K. , author Xiangli, W. , year 2015 . title Sea ice classification of SAR image based on wavelet transform and gray level co-occurrence matrix , in: booktitle 2015 Fifth International Conference on Instrumentation and Measurement, Computer, Communication and Control...

    work page doi:10.1109/imccc.2015.29 2015

  63. [63]

    , author Notz, D

    author Wernecke, A. , author Notz, D. , author Kern, S. , author Lavergne, T. , year 2024 . title Estimating the uncertainty of sea-ice area and sea-ice extent from satellite retrievals . journal The Cryosphere volume 18 , pages 2473--2486 . https://tc.copernicus.org/articles/18/2473/2024/, :10.5194/tc-18-2473-2024

    work page doi:10.5194/tc-18-2473-2024 2024

  64. [64]

    , author Buus-Hinkler, J

    author Wulf, T. , author Buus-Hinkler, J. , author Singha, S. , author Shi, H. , author Kreiner, M.B. , year 2024 . title Pan-arctic sea ice concentration from sar and passive microwave . journal The Cryosphere volume 18 , pages 5277--5300 . https://tc.copernicus.org/articles/18/5277/2024/, :10.5194/tc-18-5277-2024

    work page doi:10.5194/tc-18-5277-2024 2024

  65. [65]

    , author Wang, W

    author Xie, E. , author Wang, W. , author Yu, Z. , author Anandkumar, A. , author Alvarez, J.M. , author Luo, P. , year 2021 . title Segformer: Simple and efficient design for semantic segmentation with transformers . https://arxiv.org/abs/2105.15203, arXiv:2105.15203 http://arxiv.org/abs/2105.15203

    Pith/arXiv arXiv 2021

  66. [66]

    , author Xu, W

    author Yuan, W. , author Xu, W. , year 2021 . title Neighborloss: A loss function considering spatial correlation for semantic segmentation of remote sensing image . journal IEEE Access volume 9 , pages 75641--75649 . :10.1109/ACCESS.2021.3082076

    arXiv 2021

  67. [67]

    , author Zhang, J

    author Zeng, Y. , author Zhang, J. , author \ van Genderen\ , J. , year 2006 . title Comparison and analysis of remote sensing data fusion techniques at feature and decision levels , in: editor Skidmore, A. , editor Kerle, N. (Eds.), booktitle Proceedings of the ISPRS Commission VII Symposium , publisher International Institute for Geo-Information Science...

    2006