pith. machine review for the scientific record. sign in

arxiv: 2604.09922 · v1 · submitted 2026-04-10 · 💻 cs.LG · cs.CV

Recognition: unknown

K-STEMIT: Knowledge-Informed Spatio-Temporal Efficient Multi-Branch Graph Neural Network for Subsurface Stratigraphy Thickness Estimation from Radar Data

Maryam Rahnemoonfar, Zesheng Liu

Authors on Pith no claims yet

Pith reviewed 2026-05-10 16:57 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords graph neural networksradargramsice sheet stratigraphyphysical priorsmulti-branch networkssnow accumulationspatio-temporal modeling
0
0 comments X

The pith

Knowledge-informed graph network fuses weather model data into radar processing to cut ice layer thickness error by 21 percent.

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

The paper introduces K-STEMIT, a multi-branch graph neural network that learns spatial geometry from radargrams while adding temporal convolutions and physical constraints from a regional weather model. An adaptive fusion step dynamically weighs the branches so that the physical priors help correct for speckle noise and improve estimates when the network must extrapolate across space or time. Experiments show this combination lowers root mean squared error by 21.01 percent relative to conventional multi-branch versions at almost no extra computational cost. The resulting thickness maps support continuous tracking of snow accumulation across large polar regions. Without the physical priors, purely data-driven approaches produce unrealistic layer thicknesses outside the training distribution.

Core claim

K-STEMIT is a knowledge-informed spatio-temporal efficient multi-branch graph neural network that combines a geometric spatial learning framework, temporal convolution, and synchronized physical data from the Model Atmospheric Regional weather model, then uses adaptive feature fusion to integrate the branches. This produces more accurate subsurface stratigraphy thickness estimates from radargrams than existing knowledge-informed or non-knowledge-informed methods while preserving near-optimal efficiency.

What carries the argument

K-STEMIT, the knowledge-informed efficient multi-branch spatio-temporal graph neural network that incorporates physical weather model priors and adaptive feature fusion to combine spatial, temporal, and physical branches.

Load-bearing premise

The physical data synchronized from the Model Atmospheric Regional weather model must be accurate and relevant enough to provide useful constraints that improve generalization under spatial or temporal extrapolation.

What would settle it

Run the same radar test sets through K-STEMIT with the physical priors and adaptive fusion removed; if root mean squared error does not rise by roughly 21 percent or more, or if performance on new regions with mismatched weather data stays equally good, the central benefit claim is falsified.

Figures

Figures reproduced from arXiv: 2604.09922 by Maryam Rahnemoonfar, Zesheng Liu.

Figure 1
Figure 1. Figure 1: Diagram of capturing radargrams and generating corresponding labels. (a) How airborne radar sensor is used to capture the status of internal ice layers (Image adapted from [35]) (b) Example of a radargram captured by the airborne radar sensor. (c) Corresponding labeled images, where each ice layer is manually annotated. Each layer is formed in a certain year. ice and snow layers to collect high-resolution … view at source ↗
Figure 2
Figure 2. Figure 2: Architecture of our proposed K-STEMIT. Aggregate A temporal sequence of 𝑚 spatial graphs Compressed spatial graph with concatenated node features [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Dimension reduction in the spatial branch, where the graph sequence is compressed into a single graph with concatenated node features. reduction strategies help disentangle spatial and temporal dependencies, allowing each branch to operate on the most informative feature subsets while minimizing redundancy. For the spatial branch, we leverage the structural characteristics of radargrams: nodes located in t… view at source ↗
Figure 4
Figure 4. Figure 4: Diagram of Temporal Convolution Block. As shown in [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Qualitative results of our proposed K-STEMIT. The blue line is used to generate the graphs. The green line is the groundtruth (manually-labeled ice layers) and the red line is the model prediction [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Comparison between qualitative results of different graph models. The blue line is used to generate the graphs. The green line is the groundtruth (manually-labeled ice layers) and the red line is the model prediction. all error terms equally by normalizing each residual by its ground truth value. Evaluating 𝛿 (𝑘) 𝑗 across layers provides insights into the error consistency of the model, revealing whether d… view at source ↗
Figure 7
Figure 7. Figure 7: Relative MAE on individual ice layers for each model. As shown in [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
read the original abstract

Subsurface stratigraphy contains important spatio-temporal information about accumulation, deformation, and layer formation in polar ice sheets. In particular, variations in internal ice layer thickness provide valuable constraints for snow mass balance estimation and projections of ice sheet change. Although radar sensors can capture these layered structures as depth-resolved radargrams, convolutional neural networks applied directly to radar images are often sensitive to speckle noise and acquisition artifacts. In addition, purely data-driven methods may underuse physical knowledge, leading to unrealistic thickness estimates under spatial or temporal extrapolation. To address these challenges, we develop K-STEMIT, a novel knowledge-informed, efficient, multi-branch spatio-temporal graph neural network that combines a geometric framework for spatial learning with temporal convolution to capture temporal dynamics, and incorporates physical data synchronized from the Model Atmospheric Regional physical weather model. An adaptive feature fusion strategy is employed to dynamically combine features learned from different branches. Extensive experiments have been conducted to compare K-STEMIT against current state-of-the-art methods in both knowledge-informed and non-knowledge-informed settings, as well as other existing methods. Results show that K-STEMIT consistently achieves the highest accuracy while maintaining near-optimal efficiency. Most notably, incorporating adaptive feature fusion and physical priors reduces the root mean-squared error by 21.01% with negligible additional cost compared to its conventional multi-branch variants. Additionally, our proposed K-STEMIT achieves consistently lower per-year relative MAE, enabling reliable, continuous spatiotemporal assessment of snow accumulation variability across large spatial regions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper proposes K-STEMIT, a knowledge-informed spatio-temporal efficient multi-branch graph neural network for subsurface stratigraphy thickness estimation from radar data. It integrates geometric spatial learning, temporal convolutions, physical priors synchronized from the Model Atmospheric Regional (MAR) weather model, and an adaptive feature fusion strategy. The central claim is that this yields the highest accuracy among state-of-the-art methods in both knowledge-informed and non-knowledge-informed settings, with a 21.01% RMSE reduction and negligible added cost relative to conventional multi-branch variants, while enabling reliable spatiotemporal snow accumulation assessment.

Significance. If the empirical claims hold under rigorous validation, the work could advance the use of graph neural networks with embedded physical knowledge for geophysical radar analysis, potentially improving constraints on ice-sheet mass balance and climate projections by better handling extrapolation and noise.

major comments (3)
  1. [Abstract and Experiments] Abstract and Experiments section: The headline claim of a 21.01% RMSE reduction via adaptive fusion and MAR physical priors is presented without dataset descriptions (e.g., radargram sources, spatial/temporal coverage, label acquisition), baseline specifications, error-bar reporting, statistical tests, or ablation breakdowns, rendering the central empirical result unverifiable.
  2. [Experiments and Results] Experiments and Results sections: No independent validation is provided that the synchronized MAR fields supply accurate, relevant, and non-redundant constraints on accumulation (e.g., cross-checks against in-situ stakes or alternative reanalysis products). This leaves open whether observed gains arise from the priors themselves or from the fusion mechanism alone, directly undermining the generalization and extrapolation claims.
  3. [Results] Results section: The reported accuracy improvements lack details on the number of runs, cross-validation strategy, or significance testing, so the 21.01% RMSE figure cannot be assessed for robustness or reproducibility.
minor comments (2)
  1. [Methods] Methods section: Expand on the precise formulation of the geometric spatial learning component and the adaptive feature fusion mechanism, including any equations or pseudocode.
  2. [Abstract] Abstract: The phrase 'near-optimal efficiency' is vague; quantify computational cost (e.g., FLOPs or inference time) relative to baselines.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for their constructive and detailed feedback on our manuscript. We have carefully reviewed each major comment and provide point-by-point responses below, indicating where revisions will be made to strengthen the presentation of our results.

read point-by-point responses

  1. Referee: [Abstract and Experiments] Abstract and Experiments section: The headline claim of a 21.01% RMSE reduction via adaptive fusion and MAR physical priors is presented without dataset descriptions (e.g., radargram sources, spatial/temporal coverage, label acquisition), baseline specifications, error-bar reporting, statistical tests, or ablation breakdowns, rendering the central empirical result unverifiable.

    Authors: We agree that additional details are necessary to make the empirical claims fully verifiable. The current manuscript provides a concise summary in the abstract and high-level experimental comparisons but omits explicit descriptions of radargram sources, spatial/temporal coverage, label acquisition procedures, baseline model specifications, error bars, statistical tests, and comprehensive ablation breakdowns. In the revised manuscript, we will expand the Experiments and Results sections to include these elements, ensuring the 21.01% RMSE reduction and related claims can be independently assessed. revision: yes

  2. Referee: [Experiments and Results] Experiments and Results sections: No independent validation is provided that the synchronized MAR fields supply accurate, relevant, and non-redundant constraints on accumulation (e.g., cross-checks against in-situ stakes or alternative reanalysis products). This leaves open whether observed gains arise from the priors themselves or from the fusion mechanism alone, directly undermining the generalization and extrapolation claims.

    Authors: This comment correctly identifies a gap in the current validation strategy. While the manuscript shows performance gains from integrating MAR priors via the adaptive fusion mechanism, it does not include new independent cross-checks (such as direct comparisons to in-situ stake measurements or other reanalysis products) to confirm the accuracy and non-redundancy of the MAR fields specifically for this task. We will revise the paper to add a dedicated discussion subsection referencing established literature on MAR model performance for polar accumulation, and we will clarify that the reported improvements stem from the combined knowledge-informed framework rather than isolating the priors alone. However, performing new cross-validations would require datasets and analyses outside the scope of the present study. revision: partial

  3. Referee: [Results] Results section: The reported accuracy improvements lack details on the number of runs, cross-validation strategy, or significance testing, so the 21.01% RMSE figure cannot be assessed for robustness or reproducibility.

    Authors: We acknowledge that the current Results section does not explicitly report the number of independent runs, the cross-validation strategy, or statistical significance tests. In the revised manuscript, we will include these details, specifying the experimental protocol (e.g., number of runs with different random seeds), the cross-validation approach used, and results from appropriate significance tests to demonstrate the robustness and reproducibility of the reported accuracy improvements, including the 21.01% RMSE reduction. revision: yes

standing simulated objections not resolved
  • Independent cross-validation of the MAR physical priors against in-situ stakes or alternative reanalysis products, as this would require access to additional external datasets and new experiments beyond the current study scope.

Circularity Check

0 steps flagged

No significant circularity; empirical claims rest on external comparisons

full rationale

The paper introduces K-STEMIT as a GNN architecture incorporating physical priors from the MAR weather model and an adaptive fusion strategy. Performance claims (e.g., 21.01% RMSE reduction) are presented as outcomes of comparative experiments against baselines and variants, not as derivations or predictions that reduce by construction to fitted parameters or self-citations. No equations, uniqueness theorems, or ansatzes are shown that equate the claimed gains to the model's own inputs. The central results depend on held-out test data and external benchmarks, making the derivation chain self-contained against those comparisons.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim depends on the utility of synchronized physical weather data and the effectiveness of the proposed multi-branch architecture; no free parameters or invented physical entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Physical data from the Model Atmospheric Regional weather model can be accurately synchronized with radar observations and supplies useful priors for thickness estimation.
    Invoked to justify improved performance under extrapolation.

pith-pipeline@v0.9.0 · 5583 in / 1192 out tokens · 71696 ms · 2026-05-10T16:57:50.752574+00:00 · methodology

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

66 extracted references · 52 canonical work pages · 2 internal anchors

  1. [1]

    Chapter 12 - glaciers and ice sheets, in: Jol, H.M

    Arcone, S.A., 2009. Chapter 12 - glaciers and ice sheets, in: Jol, H.M. (Ed.), Ground Penetrating Radar Theory and Applications. Elsevier, Amsterdam, pp. 361–392. URL:https://www.sciencedirect.com/science/article/pii/B9780444533487000120, doi:https: //doi.org/10.1016/B978-0-444-53348-7.00012-0

    work page doi:10.1016/b978-0-444-53348-7.00012-0 2009

  2. [2]

    Cresis airborne radars and platforms for ice and snow sounding

    Arnold, E., Leuschen, C., Rodriguez-Morales, F., Li, J., Paden, J., Hale, R., Keshmiri, S., 2020. Cresis airborne radars and platforms for ice and snow sounding. Annals of Glaciology 61, 58–67. doi:10.1017/aog.2019.37

    work page doi:10.1017/aog.2019.37 2020

  3. [3]

    Automatic enhancement and detection of layering in radar sounder data based on a local scale hidden markov model and the viterbi algorithm

    Carrer, L., Bruzzone, L., 2017. Automatic enhancement and detection of layering in radar sounder data based on a local scale hidden markov model and the viterbi algorithm. IEEE Trans. Geosci. Remote Sens. 55, 962–977. doi:10.1109/TGRS.2016.2616949

    work page doi:10.1109/tgrs.2016.2616949 2017

  4. [4]

    Scalablespatiotemporalgraphneuralnetworks,in:ProceedingsoftheAAAIconference on artificial intelligence, pp

    Cini,A.,Marisca,I.,Bianchi,F.M.,Alippi,C.,2023. Scalablespatiotemporalgraphneuralnetworks,in:ProceedingsoftheAAAIconference on artificial intelligence, pp. 7218–7226

    2023

  5. [5]

    Fausto, R., Fettweis, X., 2020

    Delhasse, A., Kittel, C., Amory, C., Hofer, S., van As, D., S. Fausto, R., Fettweis, X., 2020. Brief communication: Evaluation of the near- surface climate in era5 over the greenland ice sheet. The Cryosphere 14, 957–965. URL:https://tc.copernicus.org/articles/14/ 957/2020/, doi:10.5194/tc-14-957-2020

    work page doi:10.5194/tc-14-957-2020 2020

  6. [6]

    Parsimonious neural networks learn interpretable physical laws

    Desai, S., Strachan, A., 2021. Parsimonious neural networks learn interpretable physical laws. Scientific Reports 11, 12761. URL: https://doi.org/10.1038/s41598-021-92278-w, doi:10.1038/s41598-021-92278-w

    work page doi:10.1038/s41598-021-92278-w 2021

  7. [7]

    When will arctic sea ice disappear? projections of area, extent, thickness, and volume

    Diebold, F.X., Rudebusch, G.D., Göbel, M., Goulet Coulombe, P., Zhang, B., 2023. When will arctic sea ice disappear? projections of area, extent, thickness, and volume. Journal of Econometrics 236, 105479. URL:https://www.sciencedirect.com/science/article/ pii/S0304407623001951, doi:https://doi.org/10.1016/j.jeconom.2023.105479

    work page doi:10.1016/j.jeconom.2023.105479 2023

  8. [8]

    Estimating the greenland ice sheetsurfacemassbalancecontributiontofuturesealevelriseusingtheregionalatmosphericclimatemodelmar

    Fettweis, X., Franco, B., Tedesco, M., Van Angelen, J., Lenaerts, J.T., van den Broeke, M.R., Gallée, H., 2013. Estimating the greenland ice sheetsurfacemassbalancecontributiontofuturesealevelriseusingtheregionalatmosphericclimatemodelmar. TheCryosphere7,469–489

    2013

  9. [9]

    Brief communication: Reduction in the future greenland ice sheet surface melt with the help of solar geoengineering

    Fettweis, X., Hofer, S., Séférian, R., Amory, C., Delhasse, A., Doutreloup, S., Kittel, C., Lang, C., Van Bever, J., Veillon, F., Irvine, P., 2021. Brief communication: Reduction in the future greenland ice sheet surface melt with the help of solar geoengineering. The Cryosphere 15, 3013–3019. URL:https://tc.copernicus.org/articles/15/3013/2021/, doi:10.5...

    work page doi:10.5194/tc-15-3013-2021 2021

  10. [10]

    Greenland and Antarctica Ice Sheet Mass Changes and Effects on Global Sea Level

    Forsberg, R., Sørensen, L., Simonsen, S., 2017. Greenland and Antarctica Ice Sheet Mass Changes and Effects on Global Sea Level. Springer International Publishing, Cham. pp. 91–106. URL:https://doi.org/10.1007/978-3-319-56490-6_5, doi:10.1007/ 978-3-319-56490-6_5

    work page doi:10.1007/978-3-319-56490-6_5 2017

  11. [11]

    Edouard Grave, Armand Joulin, and Nicolas Usunier

    Gehring, J., Auli, M., Grangier, D., Yarats, D., Dauphin, Y.N., 2017. Convolutional sequence to sequence learning.arXiv:1705.03122

    work page arXiv 2017

  12. [13]

    Ultra-widebandradarsforremotesensingofsnow and ice, in: IEEE MTT-S International Microwave and RF Conference, pp

    Gogineni,S.,Yan,J.B.,Gomez,D.,Rodriguez-Morales,F.,Paden,J.,Leuschen,C.,2013b. Ultra-widebandradarsforremotesensingofsnow and ice, in: IEEE MTT-S International Microwave and RF Conference, pp. 1–4. doi:10.1109/IMaRC.2013.6777743

    work page doi:10.1109/imarc.2013.6777743 2013

  13. [14]

    Inductive Representation Learning on Large Graphs

    Hamilton, W.L., Ying, R., Leskovec, J., 2018. Inductive representation learning on large graphs.arXiv:1706.02216

    work page Pith review arXiv 2018

  14. [15]

    The era5 global reanalysis

    Hersbach,H.,Bell,B.,Berrisford,P.,Hirahara,S.,Horányi,A.,Muñoz-Sabater,J.,Nicolas,J.,Peubey,C.,Radu,R.,Schepers,D.,etal.,2020. The era5 global reanalysis. Quarterly Journal of the Royal Meteorological Society 146, 1999–2049

    2020

  15. [16]

    Hierarchy-aware graph neural network and inverse-variance reinforcement learning for drug recom- mendation

    Hou, W., Liu, X., Li, L., Fu, C., 2026. Hierarchy-aware graph neural network and inverse-variance reinforcement learning for drug recom- mendation. Neurocomputing 676, 132989. URL:https://www.sciencedirect.com/science/article/pii/S0925231226003863, doi:https://doi.org/10.1016/j.neucom.2026.132989

    work page doi:10.1016/j.neucom.2026.132989 2026

  16. [17]

    Novel deep learning framework for rainfall forecasting integrating generative adversarial and spatiotemporal graph neural networks

    Humphries, U.W., Waqas, M., Ahmad, S., 2026. Novel deep learning framework for rainfall forecasting integrating generative adversarial and spatiotemporal graph neural networks. Results in Engineering 29, 108616. URL:https://www.sciencedirect.com/science/ article/pii/S2590123025046602, doi:https://doi.org/10.1016/j.rineng.2025.108616

    work page doi:10.1016/j.rineng.2025.108616 2026

  17. [18]

    Snow radar layer tracking using iterative neural network approach, in: IGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, pp

    Ibikunle, O., Paden, J., Rahnemoonfar, M., Crandall, D., Yari, M., 2020. Snow radar layer tracking using iterative neural network approach, in: IGARSS 2020 - 2020 IEEE International Geoscience and Remote Sensing Symposium, pp. 2960–2963. doi:10.1109/IGARSS39084. 2020.9323957

    work page doi:10.1109/igarss39084 2020

  18. [19]

    Ai-ready snow radar echogram dataset (sred) for climate change monitoring,

    Ibikunle, O., Talasila, H., Varshney, D., Li, J., Paden, J., Rahnemoonfar, M., 2025. Ai-ready snow radar echogram dataset (sred) for climate change monitoring. URL:https://arxiv.org/abs/2505.00786,arXiv:2505.00786

    work page arXiv 2025

  19. [20]

    Snow radar echogram layer tracker: Deep neural networks for radar data from nasa operation icebridge, in: 2023 IEEE Radar Conference (RadarConf23), pp

    Ibikunle, O., Talasila, H.M., Varshney, D., Paden, J.D., Li, J., Rahnemoonfar, M., 2023. Snow radar echogram layer tracker: Deep neural networks for radar data from nasa operation icebridge, in: 2023 IEEE Radar Conference (RadarConf23), pp. 1–6. doi:10.1109/ RadarConf2351548.2023.10149734

    work page arXiv 2023

  20. [21]

    Deep hybrid wavelet network for ice boundary detection in radra imagery, in: IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium, pp

    Kamangir, H., Rahnemoonfar, M., Dobbs, D., Paden, J., Fox, G., 2018. Deep hybrid wavelet network for ice boundary detection in radra imagery, in: IGARSS 2018 - 2018 IEEE International Geoscience and Remote Sensing Symposium, pp. 3449–3452. doi:10.1109/IGARSS. 2018.8518617

    work page doi:10.1109/igarss 2018

  21. [22]

    Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang

    Karniadakis, G.E., Kevrekidis, I.G., Lu, L., Perdikaris, P., Wang, S., Yang, L., 2021. Physics-informed machine learning. Nature Reviews Physics 3, 422–440. URL:https://doi.org/10.1038/s42254-021-00314-5, doi:10.1038/s42254-021-00314-5

    work page doi:10.1038/s42254-021-00314-5 2021

  22. [23]

    Adam: A Method for Stochastic Optimization

    Kingma, D.P., Ba, J., 2017. Adam: A method for stochastic optimization.arXiv:1412.6980. Zesheng Liu, Maryam Rahnemoonfar:Preprint submitted to ElsevierPage 18 of 20 Knowledge-Informed Spatio-Temporal Efficient Multi-Branch Graph Neural Network

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  23. [24]

    Semi-Supervised Classification with Graph Convolutional Networks

    Kipf, T.N., Welling, M., 2016. Semi-supervised classification with graph convolutional networks. CoRR abs/1609.02907. URL:http: //arxiv.org/abs/1609.02907,arXiv:1609.02907

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  24. [25]

    Annualgreenlandaccumulationrates(2009–2012)fromairbornesnowradar

    Koenig, L.S., Ivanoff, A., Alexander, P.M., MacGregor, J.A., Fettweis, X., Panzer, B., Paden, J.D., Forster, R.R., Das, I., McConnell, J.R., Tedesco,M.,Leuschen,C.,Gogineni,P.,2016. Annualgreenlandaccumulationrates(2009–2012)fromairbornesnowradar. TheCryosphere 10, 1739–1752. URL:https://tc.copernicus.org/articles/10/1739/2016/, doi:10.5194/tc-10-1739-2016

    work page doi:10.5194/tc-10-1739-2016 2016

  25. [26]

    Icebridge snow radar l1b geolocated radar echo strength profiles

    Leuschen, C., Panzer, B., Gogineni, P., Rodriguez, F., Paden, J., Li, J., 2011/2024. Icebridge snow radar l1b geolocated radar echo strength profiles. Boulder, Colorado USA: National Snow and Ice Data Center. Digital media. Accessed on 2024

    2011

  26. [27]

    Mast-gnn: A multimodal adaptive spatio-temporal graph neural network for airspace complexity prediction

    Li, B., Li, Z., Chen, J., Yan, Y., Lv, Y., Du, W., 2024. Mast-gnn: A multimodal adaptive spatio-temporal graph neural network for airspace complexity prediction. Transportation Research Part C: Emerging Technologies 160, 104521. URL:https://www.sciencedirect.com/ science/article/pii/S0968090X24000421, doi:https://doi.org/10.1016/j.trc.2024.104521

    work page doi:10.1016/j.trc.2024.104521 2024

  27. [28]

    Learning spatio-temporal patterns of polar ice layers with physics-informed graph neural network

    Liu, Z., Rahnemoonfar, M., 2024. Learning spatio-temporal patterns of polar ice layers with physics-informed graph neural network. URL: https://arxiv.org/abs/2406.15299,arXiv:2406.15299

    work page arXiv 2024

  28. [29]

    Grit: Graph transformer for internal ice layer thickness prediction, in: IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium, pp

    Liu, Z., Rahnemoonfar, M., 2025a. Grit: Graph transformer for internal ice layer thickness prediction, in: IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium, pp. 1–5. doi:10.1109/IGARSS55030.2025.11243115

    work page doi:10.1109/igarss55030.2025.11243115 2025

  29. [30]

    Locateandextend:ageometricdeeplearningstrategyforpredictingpolaricelayerstructuresusinggraph neural networks, in: Pattern Recognition and Prediction XXXVI, SPIE

    Liu,Z.,Rahnemoonfar,M.,2025b. Locateandextend:ageometricdeeplearningstrategyforpredictingpolaricelayerstructuresusinggraph neural networks, in: Pattern Recognition and Prediction XXXVI, SPIE. p. 1346402

  30. [31]

    Pdd-agent: Multimodal large language model-driven ai agent for enhanced plant disease diagnosis

    Liu, Z., Rahnemoonfar, M., 2025c. St-grit: Spatio-temporal graph transformer for internal ice layer thickness prediction, in: 2025 IEEE International Conference on Image Processing (ICIP), pp. 1109–1114. doi:10.1109/ICIP55913.2025.11084445

    work page doi:10.1109/icip55913.2025.11084445 2025

  31. [32]

    Radiostratigraphy and age structure of the greenland ice sheet

    MacGregor, J.A., Fahnestock, M.A., Catania, G.A., Paden, J.D., Prasad Gogineni, S., Young, S.K., Rybarski, S.C., Mabrey, A.N., Wagman, B.M., Morlighem, M., 2015. Radiostratigraphy and age structure of the greenland ice sheet. Journal of Geophysical Research: Earth Surface 120, 212–241. URL:https://agupubs.onlinelibrary.wiley.com/doi/abs/10.1002/2014JF0032...

    work page doi:10.1002/2014jf003215 2015

  32. [33]

    The climate modelling primer

    McGuffie, K., Henderson-Sellers, A., 2014. The climate modelling primer. John Wiley & Sons

    2014

  33. [34]

    An introduction to three-dimensional climate modeling

    Morel, P., 1988. An introduction to three-dimensional climate modeling

    1988

  34. [35]

    Greenland ice sheet stratigraphy

    NASA Scientific Visualization Studio, 2015. Greenland ice sheet stratigraphy. NASA Scientific Visualization Studio. Available online

    2015

  35. [36]

    Automated mapping of local layer slope and tracing of internal layers in radio echograms

    Panton, C., 2013. Automated mapping of local layer slope and tracing of internal layers in radio echograms. Annals of Glaciology 55. doi:10.3189/2014AoG67A048

    work page doi:10.3189/2014aog67a048 2013

  36. [37]

    Evolvegcn: Evolving graph convolutional networks for dynamic graphs

    Pareja, A., Domeniconi, G., Chen, J., Ma, T., Suzumura, T., Kanezashi, H., Kaler, T., Leiserson, C.E., 2019. Evolvegcn: Evolving graph convolutional networks for dynamic graphs. CoRR abs/1902.10191. URL:http://arxiv.org/abs/1902.10191,arXiv:1902.10191

    work page arXiv 2019

  37. [38]

    15 - ice core studies, in: PATERSON, W

    PATERSON, W., 1994. 15 - ice core studies, in: PATERSON, W. (Ed.), The Physics of Glaciers (Third Edition). third edition ed.. Pergamon, Amsterdam, pp. 378–409. URL:https://www.sciencedirect.com/science/article/pii/B9780080379449500212, doi:https://doi.org/10.1016/B978-0-08-037944-9.50021-2

    work page doi:10.1016/b978-0-08-037944-9.50021-2 1994

  38. [39]

    Feature-enhanced graph neural network with multiple attention for molecular property prediction

    Qin, B., Zhu, X., Fan, C.Y., Xue, X., Wang, M.M., Tang, H.Y., 2026. Feature-enhanced graph neural network with multiple attention for molecular property prediction. Neurocomputing 669, 132426. URL:https://www.sciencedirect.com/science/article/pii/ S092523122503098X, doi:https://doi.org/10.1016/j.neucom.2025.132426

    work page doi:10.1016/j.neucom.2025.132426 2026

  39. [40]

    Deep multi-scale learning for automatic tracking of internal layers of ice in radar data

    Rahnemoonfar, M., Yari, M., Paden, J., Koenig, L., Ibikunle, O., 2021. Deep multi-scale learning for automatic tracking of internal layers of ice in radar data. Journal of Glaciology 67, 39–48. doi:10.1017/jog.2020.80

    work page doi:10.1017/jog.2020.80 2021

  40. [41]

    In: IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, pp

    Rahnemoonfar, M., Zalatan, B., 2024. Physics-informed machine learning for deep ice layer tracing in sar images, in: IGARSS 2024 - 2024 IEEE International Geoscience and Remote Sensing Symposium, pp. 6938–6942. doi:10.1109/IGARSS53475.2024.10641831

    work page doi:10.1109/igarss53475.2024.10641831 2024

  41. [42]

    Parameterization of melt rate and surface temperature in the greenland ice sheet

    Reeh, N., 1991. Parameterization of melt rate and surface temperature in the greenland ice sheet. Polarforschung 59, 113–128

    1991

  42. [43]

    Structured sequence modeling with graph convolutional recurrent networks

    Seo, Y., Defferrard, M., Vandergheynst, P., Bresson, X., 2016. Structured sequence modeling with graph convolutional recurrent networks. arXiv:1612.07659

    work page arXiv 2016

  43. [44]

    Bidirectional spatial–temporal traffic data imputation via graph attentionrecurrentneuralnetwork

    Shen, G., Zhou, W., Zhang, W., Liu, N., Liu, Z., Kong, X., 2023. Bidirectional spatial–temporal traffic data imputation via graph attentionrecurrentneuralnetwork. Neurocomputing531,151–162. URL:https://www.sciencedirect.com/science/article/pii/ S0925231223001558, doi:https://doi.org/10.1016/j.neucom.2023.02.017

    work page doi:10.1016/j.neucom.2023.02.017 2023

  44. [45]

    Massbalanceofthegreenlandicesheetfrom1992to2018

    Shepherd,A.,Ivins,E.,Rignot,E.,Smith,B.,vandenBroeke,M.,Velicogna,I.,Whitehouse,P.,Briggs,K.,Joughin,I.,Krinner,G.,Nowicki, S., Payne, T., Scambos, T., Schlegel, N., A, G., Agosta, C., Ahlstrøm, A., Babonis, G., Barletta, V.R., Bjørk, A.A., Blazquez, A., Bonin, J., Colgan, W., Csatho, B., Cullather, R., Engdahl, M.E., Felikson, D., Fettweis, X., Forsberg,...

    work page doi:10.1038/s41586-019-1855-2 2020

  45. [46]

    The era5 global reanalysis from 1940 to 2022

    Soci, C., Hersbach, H., Simmons, A., Poli, P., Bell, B., Berrisford, P., Horányi, A., Muñoz-Sabater, J., Nicolas, J., Radu, R., et al., 2024. The era5 global reanalysis from 1940 to 2022. Quarterly Journal of the Royal Meteorological Society 150, 4014–4048

    2024

  46. [47]

    Explainable spatio-temporal graph neural networks, in: Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pp

    Tang, J., Xia, L., Huang, C., 2023. Explainable spatio-temporal graph neural networks, in: Proceedings of the 32nd ACM International Conference on Information and Knowledge Management, pp. 2432–2441

    2023

  47. [48]

    Regression networks for calculating englacial layer thickness, in: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pp

    Teisberg, T.O., Schroeder, D.M., MacKie, E.J., 2021. A machine learning approach to mass-conserving ice thickness interpolation, in: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pp. 8664–8667. doi:10.1109/IGARSS47720.2021.9555002

    work page doi:10.1109/igarss47720.2021.9555002 2021

  48. [49]

    A deep learning perspective to atmospheric correction of satellite images

    Varshney, D., Ibikunle, O., Paden, J., Rahnemoonfar, M., 2022. Learning snow layer thickness through physics defined labels, in: IGARSS 2022-2022IEEEInternationalGeoscienceandRemoteSensingSymposium,pp.1233–1236. doi:10.1109/IGARSS46834.2022.9884370. Zesheng Liu, Maryam Rahnemoonfar:Preprint submitted to ElsevierPage 19 of 20 Knowledge-Informed Spatio-Temp...

    work page doi:10.1109/igarss46834.2022.9884370 2022

  49. [50]

    3943–3952

    Varshney,D.,Rahnemoonfar,M.,Yari,M.,Paden,J.,2020.Deepicelayertrackingandthicknessestimationusingfullyconvolutionalnetworks, in: 2020 IEEE International Conference on Big Data (Big Data), pp. 3943–3952. doi:10.1109/BigData50022.2020.9378070

    work page doi:10.1109/bigdata50022.2020.9378070 2020

  50. [51]

    Regression networks for calculating englacial layer thickness, in: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pp

    Varshney, D., Rahnemoonfar, M., Yari, M., Paden, J., 2021a. Regression networks for calculating englacial layer thickness, in: 2021 IEEE International Geoscience and Remote Sensing Symposium IGARSS, pp. 2393–2396. doi:10.1109/IGARSS47720.2021.9553596

    work page doi:10.1109/igarss47720.2021.9553596 2021

  51. [52]

    Deep learning on airborne radar echograms for tracing snow accumulation layers of the greenland ice sheet

    Varshney, D., Rahnemoonfar, M., Yari, M., Paden, J., Ibikunle, O., Li, J., 2021b. Deep learning on airborne radar echograms for tracing snow accumulation layers of the greenland ice sheet. Remote Sensing 13. URL:https://www.mdpi.com/2072-4292/13/14/2707, doi:10.3390/rs13142707

    work page doi:10.3390/rs13142707 2072

  52. [53]

    Refining ice layer tracking through wavelet combined neural networks, in: ICML 2021 Workshop on Tackling Climate Change with Machine Learning

    Varshney, D., Yari, M., Chowdhury, T., Rahnemoonfar, M., 2021c. Refining ice layer tracking through wavelet combined neural networks, in: ICML 2021 Workshop on Tackling Climate Change with Machine Learning. URL:https://www.climatechange.ai/papers/ icml2021/49

    2021

  53. [54]

    Skip-wavenet:awaveletbasedmulti-scale architecture to trace snow layers in radar echograms

    Varshney,D.,Yari,M.,Ibikunle,O.,Li,J.,Paden,J.,Gangopadhyay,A.,Rahnemoonfar,M.,2024. Skip-wavenet:awaveletbasedmulti-scale architecture to trace snow layers in radar echograms. Environmental Data Science 3, e39. doi:10.1017/eds.2024.25

    work page doi:10.1017/eds.2024.25 2024

  54. [55]

    Explainable spatio-temporal graph neural networks for multi-site photovoltaic energy production

    Verdone, A., Scardapane, S., Panella, M., 2024. Explainable spatio-temporal graph neural networks for multi-site photovoltaic energy production. Applied Energy 353, 122151. URL:https://www.sciencedirect.com/science/article/pii/S0306261923015155, doi:https://doi.org/10.1016/j.apenergy.2023.122151

    work page doi:10.1016/j.apenergy.2023.122151 2024

  55. [56]

    Surface mass balance model intercomparison for the greenland ice sheet

    Vernon, C.L., Bamber, J., Box, J., Van den Broeke, M., Fettweis, X., Hanna, E., Huybrechts, P., 2013. Surface mass balance model intercomparison for the greenland ice sheet. The Cryosphere 7, 599–614

    2013

  56. [57]

    Stgformer: Efficient spatiotemporal graph transformer for traffic forecasting

    Wang, H., Chen, J., Pan, T., Dong, Z., Zhang, L., Jiang, R., Song, X., 2024. Stgformer: Efficient spatiotemporal graph transformer for traffic forecasting. URL:https://arxiv.org/abs/2410.00385,arXiv:2410.00385

    work page arXiv 2024

  57. [58]

    Stgnet: A spatio-temporal graph neural network for motion prediction in autonomous driving

    Wang, X., Liu, L., 2026. Stgnet: A spatio-temporal graph neural network for motion prediction in autonomous driving. Neurocomputing 676, 133063. URL:https://www.sciencedirect.com/science/article/pii/S0925231226004601, doi:https://doi.org/10. 1016/j.neucom.2026.133063

    work page arXiv 2026

  58. [59]

    Introduction to three-dimensional climate modeling

    Washington, W.M., Parkinson, C., 2005. Introduction to three-dimensional climate modeling. University science books

    2005

  59. [60]

    Acomprehensivesurveyongraphneuralnetworks

    Wu,Z.,Pan,S.,Chen,F.,Long,G.,Zhang,C.,Yu,P.S.,2021. Acomprehensivesurveyongraphneuralnetworks. IEEETrans.NeuralNetw. Learn. Syst. 32, 4–24. doi:10.1109/TNNLS.2020.2978386

    work page doi:10.1109/tnnls.2020.2978386 2021

  60. [61]

    A review of graph neural networks for brain diseases analy- sis

    Yang, H., Huang, R., Ye, S., Zhang, P., Guo, Y., Pan, S., Zhang, Y., 2026. A review of graph neural networks for brain diseases analy- sis. Neurocomputing , 133174URL:https://www.sciencedirect.com/science/article/pii/S0925231226005710, doi:https: //doi.org/10.1016/j.neucom.2026.133174

    work page doi:10.1016/j.neucom.2026.133174 2026

  61. [62]

    Airborne snow radar data simulation with deep learning and physics-driven methods

    Yari, M., Ibikunle, O., Varshney, D., Chowdhury, T., Sarkar, A., Paden, J., Li, J., Rahnemoonfar, M., 2021. Airborne snow radar data simulation with deep learning and physics-driven methods. IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. 14, 12035–12047. doi:10.1109/JSTARS.2021.3126547

    work page doi:10.1109/jstars.2021.3126547 2021

  62. [63]

    Multi-scale and temporal transfer learning for automatic tracking of internal ice layers, in: IGARSS2020-2020IEEEInternationalGeoscienceandRemoteSensingSymposium,pp.6934–6937

    Yari, M., Rahnemoonfar, M., Paden, J., 2020. Multi-scale and temporal transfer learning for automatic tracking of internal ice layers, in: IGARSS2020-2020IEEEInternationalGeoscienceandRemoteSensingSymposium,pp.6934–6937. doi:10.1109/IGARSS39084.2020. 9323758

    work page doi:10.1109/igarss39084.2020 2020

  63. [64]

    Prediction of annual snow accumulation using a recurrent graph convolutional approach, in: IGARSS2023-2023IEEEInternationalGeoscienceandRemoteSensingSymposium,pp.5344–5347

    Zalatan, B., Rahnemoonfar, M., 2023a. Prediction of annual snow accumulation using a recurrent graph convolutional approach, in: IGARSS2023-2023IEEEInternationalGeoscienceandRemoteSensingSymposium,pp.5344–5347. doi:10.1109/IGARSS52108.2023. 10283236

    work page doi:10.1109/igarss52108.2023 2023

  64. [65]

    Prediction of deep ice layer thickness using adaptive recurrent graph neural networks, in: 2023 IEEE International Conference on Image Processing (ICIP), pp

    Zalatan, B., Rahnemoonfar, M., 2023b. Prediction of deep ice layer thickness using adaptive recurrent graph neural networks, in: 2023 IEEE International Conference on Image Processing (ICIP), pp. 2835–2839. doi:10.1109/ICIP49359.2023.10222391

    work page doi:10.1109/icip49359.2023.10222391 2023

  65. [66]

    Snow radar echogram layer tracker: Deep neural networks for radar data from nasa operation icebridge, in: 2023 IEEE Radar Conference (RadarConf23), pp

    Zalatan, B., Rahnemoonfar, M., 2023c. Recurrent graph convolutional networks for spatiotemporal prediction of snow accumulation using airborne radar, in: 2023 IEEE Radar Conference (RadarConf23), pp. 1–6. doi:10.1109/RadarConf2351548.2023.10149562

    work page doi:10.1109/radarconf2351548.2023.10149562 2023

  66. [67]

    Graph neural networks: A review of methods and applications,

    Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., Wang, L., Li, C., Sun, M., 2020. Graph neural networks: A review of methods and applications. AIOpen1,57–81. URL:https://www.sciencedirect.com/science/article/pii/S2666651021000012,doi:https: //doi.org/10.1016/j.aiopen.2021.01.001. Zesheng Liu, Maryam Rahnemoonfar:Preprint submitted to ElsevierPage 20 of 20

    work page doi:10.1016/j.aiopen.2021.01.001 2020