pith. sign in

arxiv: 2605.30467 · v2 · pith:Z6WGFRHJnew · submitted 2026-05-28 · 💻 cs.CV

Clustering Guided Domain-Specific Pretrained Foundation Model for Very High-Resolution Arctic Remote Sensing

Amal S. Perera , Chandi Witharana , Elias Manos , Michael Pimenta , Anna K. Liljedahl This is my paper

Pith reviewed 2026-06-29 08:00 UTC · model grok-4.3

classification 💻 cs.CV
keywords Arctic remote sensingdomain-specific pretrainingmasked autoencoderVision Transformervery high resolution imageryself-supervised learningfoundation modelsimage curation
0
0 comments X

The pith

Domain-specific self-supervised pretraining on curated Arctic VHSR imagery produces more transferable representations for fine-scale mapping than general Earth observation foundation models.

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

The paper establishes that selecting a diverse set of roughly 3 million image chips from 267 TB of Vantor very-high-resolution Arctic imagery via affinity-propagation clustering on spectral and metadata features enables effective masked autoencoder pretraining of a Vision Transformer encoder. This produces weights that, when plugged into a location-aware detection and segmentation pipeline, raise foreground mean F1 scores by 5-8 points over an ImageNet-initialized baseline and by at least 15 points over Prithvi-EO-2.0 across four hand-labeled Arctic tasks. A sympathetic reader would care because the result isolates the effect of matching the pretraining data distribution to the target domain while holding architecture and objective fixed.

Core claim

By applying affinity-propagation clustering to spectral and acquisition-metadata descriptors, the authors curate a balanced corpus of approximately 3 million chips that reduces oversampling of repetitive scenes while retaining domain-wide diversity. They then pretrain a ViT-Large encoder on this corpus with a domain-adapted masked autoencoder objective. The resulting Arctic-specific weights, when integrated into an existing detection framework, deliver foreground mean F1 scores of 0.87, 0.72, 0.93, and 0.87 on infrastructure, ice-wedge polygons, retrogressive thaw slumps, and thermokarst lakes, respectively, outperforming both ImageNet and a general-purpose Earth-observation foundation model

What carries the argument

Affinity-propagation clustering workflow on spectral and acquisition-metadata descriptors that selects diverse pretraining chips, followed by masked autoencoder pretraining of the ViT-Large encoder.

If this is right

  • The domain-adapted encoder transfers to multiple fine-scale Arctic mapping tasks without task-specific architectural changes.
  • Gains appear consistently for infrastructure detection, ice-wedge polygon mapping, retrogressive thaw slump identification, and thermokarst lake delineation.
  • Regional-scale optimization of pretraining data distribution can improve reusability of the encoder while keeping model size and training objective unchanged.

Where Pith is reading between the lines

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

  • The same curation approach might improve domain-specific pretraining for other geographically constrained remote-sensing problems such as tropical or arid regions.
  • If curation quality matters more than sheer data volume, future work could test whether smaller but carefully balanced regional corpora rival much larger global collections.
  • The result invites direct comparison of clustering-based selection against other diversity metrics such as embedding-space coverage or uncertainty sampling.

Load-bearing premise

The clustering step successfully reduces repetitive sampling and that this curation step is what drives the measured performance gains on downstream tasks.

What would settle it

Train an identical ViT-Large MAE model on a random sample of the same 267 TB corpus without the clustering curation and measure whether the 5-8 point F1 advantage over the ImageNet baseline disappears on the four Arctic test sets.

read the original abstract

This study introduces a novel Arctic-focused remote sensing foundation model (RSFM) by combining diversity-aware regional-scale image curation with masked autoencoder (MAE) self-supervised pretraining of a Vision Transformer (ViT) encoder for very-high-spatial-resolution (VHSR) satellite image analysis. Spectral and acquisition-metadata descriptors were used in a scalable affinity-propagation clustering workflow to select approximately 3 million chips from 267 TB of Vantor VHSR imagery This curation strategy was designed to reduce oversampling of visually repetitive or low-information areas while preserving broad scene diversity across the study domain. We pretrained a ViT-Large encoder on the curated corpus using a domain-adapted MAE reconstruction objective, producing Arctic-specific transformer weights for downstream feature mapping. The pretrained encoder was integrated into an existing location-aware detection and segmentation framework and evaluated across four hand-labeled Arctic datasets. Compared to ImageNet-initialized ViT-Large baseline, Arctic MAE pretraining produced consistent improvements in foreground mean F1 scores of 0.87, 0.72, 0.93, and 0.87, for infrastructure, IWP, RTS, and TCNs, with approximately 5-8 percentage increase. The proposed model also outperformed Prithvi-EO-2.0 in all downstream comparisons, with the smallest gain corresponding to at least a 15 percentage improvement mean F1, suggesting that domain-specific self-supervised pretraining on curated Arctic VHSR imagery provides more transferable representations for fine-scale Arctic mapping than a general-purpose Earth observation foundation model. These results demonstrate that optimizing the pretraining data distribution at regional scale, while keeping the architecture and MAE objective fixed, can produce a reusable Arctic-domain encoder for multiple VHSR remote sensing applications.

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

2 major / 2 minor

Summary. The manuscript introduces an Arctic-specific remote sensing foundation model by applying affinity-propagation clustering on spectral and acquisition-metadata descriptors to curate ~3 million chips from 267 TB of Vantor VHSR imagery, then pretraining a ViT-Large encoder with a domain-adapted MAE objective. The resulting encoder is integrated into a location-aware detection/segmentation pipeline and evaluated on four hand-labeled Arctic datasets (infrastructure, IWP, RTS, TCNs), reporting foreground mean F1 improvements of 5-8 pp over an ImageNet-initialized ViT-Large baseline and at least 15 pp over Prithvi-EO-2.0, with the claim that optimizing the pretraining data distribution via curation yields more transferable representations than general-purpose EO models.

Significance. If the attribution to curation holds, the work would provide concrete evidence that regional-scale data curation can improve domain-specific foundation models for fine-scale remote sensing tasks in data-sparse environments such as the Arctic; the scale of the corpus and consistent multi-task gains constitute a useful empirical contribution to the literature on domain-adapted MAE pretraining.

major comments (2)
  1. [§4] §4 (downstream evaluation): the 5-8 pp F1 gains are reported only versus ImageNet and Prithvi-EO-2.0; no control MAE is trained on an equal-sized random sample drawn from the same 267 TB Vantor corpus without the affinity-propagation clustering step. This directly undermines attribution of the observed improvements to the curation workflow rather than domain shift alone, which is load-bearing for the central claim in the abstract and §5.
  2. [§3.2] §3.2 (curation workflow): the paper asserts that affinity propagation on spectral/metadata descriptors reduces oversampling while preserving diversity, yet provides no quantitative validation (e.g., diversity metrics or scene-type histograms) comparing the curated set to a random subsample; without this, the mechanism linking curation to downstream gains remains unverified.
minor comments (2)
  1. The manuscript should report the precise train/validation/test splits for each of the four downstream datasets, the number of random seeds or statistical tests used to establish significance of the F1 differences, and whether baseline models were re-trained or used off-the-shelf.
  2. [§3.3] Notation for the MAE reconstruction objective and the integration of the pretrained encoder into the detection framework could be clarified with an explicit equation or diagram reference in §3.3.

Simulated Author's Rebuttal

2 responses · 1 unresolved

Thank you for the constructive feedback. We provide point-by-point responses to the major comments and indicate planned revisions to the manuscript.

read point-by-point responses

  1. Referee: [§4] §4 (downstream evaluation): the 5-8 pp F1 gains are reported only versus ImageNet and Prithvi-EO-2.0; no control MAE is trained on an equal-sized random sample drawn from the same 267 TB Vantor corpus without the affinity-propagation clustering step. This directly undermines attribution of the observed improvements to the curation workflow rather than domain shift alone, which is load-bearing for the central claim in the abstract and §5.

    Authors: We agree that the absence of a random-sample control experiment limits the strength of attribution specifically to the affinity-propagation curation rather than to the use of Arctic VHSR data in general. The comparisons to ImageNet and Prithvi-EO-2.0 demonstrate benefits of domain-specific pretraining, but isolating the curation effect would require the proposed control. Given the significant computational resources needed to train an additional ViT-Large MAE, we are unable to perform this experiment. In the revised version, we will add a discussion of this limitation and adjust the claims in the abstract and conclusion to reflect that the results show the value of Arctic-specific MAE pretraining on curated data, without claiming isolated proof of the curation mechanism. revision: partial

  2. Referee: [§3.2] §3.2 (curation workflow): the paper asserts that affinity propagation on spectral/metadata descriptors reduces oversampling while preserving diversity, yet provides no quantitative validation (e.g., diversity metrics or scene-type histograms) comparing the curated set to a random subsample; without this, the mechanism linking curation to downstream gains remains unverified.

    Authors: We accept that quantitative validation of the curation workflow was not provided. We will revise the manuscript to include comparisons such as scene-type histograms derived from the spectral and metadata descriptors, as well as diversity metrics (e.g., cluster coverage or descriptor entropy) between the curated set and a random subsample of equivalent size. This will be added to §3.2 to better support the mechanism. revision: yes

standing simulated objections not resolved
  • The request for an additional full-scale MAE pretraining run on a random sample from the Vantor corpus, due to prohibitive computational costs.

Circularity Check

0 steps flagged

No circularity; empirical comparisons to external baselines

full rationale

The paper reports empirical pretraining of a ViT-Large MAE on a curated Arctic VHSR corpus followed by downstream F1 evaluations against ImageNet-initialized ViT-Large and Prithvi-EO-2.0. No equations, fitted parameters renamed as predictions, self-definitional steps, or load-bearing self-citations appear in the provided text. The curation workflow is an input to the experiment rather than a quantity derived from the reported results. The central claim therefore rests on external benchmarks and does not reduce to its own inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The abstract provides no explicit free parameters, axioms, or invented entities; the approach relies on standard ViT and MAE components plus a custom clustering pipeline whose internal parameters are not specified here.

pith-pipeline@v0.9.1-grok · 5868 in / 1256 out tokens · 41862 ms · 2026-06-29T08:00:44.691219+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

60 extracted references · 49 canonical work pages · 1 internal anchor

  1. [1]

    Evolutionary geomorphology: thresholds and nonlinearity in landform response to environmental change,

    J. D. Phillips, “Evolutionary geomorphology: thresholds and nonlinearity in landform response to environmental change,” Hydrol. Earth Syst. Sci., vol. 10, no. 5, pp. 731–742, Oct. 2006, doi: 10.5194/hess-10-731-2006

    work page doi:10.5194/hess-10-731-2006 2006

  2. [2]

    Interacting tipping elements increase risk of climate domino effects under global warming,

    N. Wunderling, J. F. Donges, J. Kurths, and R. Winkelmann, “Interacting tipping elements increase risk of climate domino effects under global warming,” Earth Syst. Dyn., vol. 12, no. 2, pp. 601–619, Jun. 2021, doi: 10.5194/esd-12-601-2021

    work page doi:10.5194/esd-12-601-2021 2021

  3. [3]

    Perspectives on tipping points in integrated models of the natural and human Earth system: cascading effects and telecoupling,

    C. L. E. Franzke et al., “Perspectives on tipping points in integrated models of the natural and human Earth system: cascading effects and telecoupling,” Environ. Res. Lett., vol. 17, no. 1, p. 015004, Jan. 2022, doi: 10.1088/1748-9326/ac42fd

    work page doi:10.1088/1748-9326/ac42fd 2022

  4. [4]

    A Scientific Perspective on Big Data in Earth Observation,

    C. Vaduva, M. Iapaolo, and M. Datcu, “A Scientific Perspective on Big Data in Earth Observation,” in Principles of Data Science, H. R. Arabnia, K. Daimi, R. Stahlbock, C. Soviany, L. Heilig, and K. Brüssau, Eds., Cham: Springer International Publishing, 2020, pp. 155–

    2020

  5. [5]

    doi: 10.1007/978-3-030-43981-1_8

    work page doi:10.1007/978-3-030-43981-1_8

  6. [6]

    Disturbances in North American boreal forest and Arctic tundra: impacts, interactions, and responses,

    A. C. Foster et al., “Disturbances in North American boreal forest and Arctic tundra: impacts, interactions, and responses,” Environ. Res. Lett., vol. 17, no. 11, p. 113001, Oct. 2022, doi: 10.1088/1748- 9326/ac98d7

    work page doi:10.1088/1748- 2022

  7. [7]

    Challenges and Solutions for Utilizing Earth Observations in the ‘Big Data’ era,

    L. Filchev, L. Pashova, V. Kolev, and S. Frye, “Challenges and Solutions for Utilizing Earth Observations in the ‘Big Data’ era,” Dec. 2018, doi: 10.5281/zenodo.2391937

    work page doi:10.5281/zenodo.2391937 2018

  8. [8]

    Current status of Landsat program, science, and applications,

    M. A. Wulder et al., “Current status of Landsat program, science, and applications,” Remote Sens. Environ., vol. 225, pp. 127–147, May 2019, doi: 10.1016/j.rse.2019.02.015

    work page doi:10.1016/j.rse.2019.02.015 2019

  9. [9]

    Scalable big earth observation data mining algorithms: a review,

    N. Sisodiya, N. Dube, O. Prakash, and P. Thakkar, “Scalable big earth observation data mining algorithms: a review,” Earth Sci. Inform., vol. 16, no. 3, pp. 1993–2016, Sep. 2023, doi: 10.1007/s12145-023-01032- 5

    work page doi:10.1007/s12145-023-01032- 1993

  10. [10]

    Onboard Information Fusion for Multisatellite Collaborative Observation: Summary, challenges, and perspectives,

    G. Gao, L. Yao, W. Li, L. Zhang, and M. Zhang, “Onboard Information Fusion for Multisatellite Collaborative Observation: Summary, challenges, and perspectives,” IEEE Geosci. Remote Sens. Mag., vol. 11, no. 2, pp. 40–59, Jun. 2023, doi: 10.1109/MGRS.2023.3274301. IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. XX, 202X

    work page doi:10.1109/mgrs.2023.3274301 2023

  11. [11]

    Remote Sensing and Geospatial Analysis in the Big Data Era: A Survey,

    E. Dritsas and M. Trigka, “Remote Sensing and Geospatial Analysis in the Big Data Era: A Survey,” Remote Sens., vol. 17, no. 3, p. 550, Jan. 2025, doi: 10.3390/rs17030550

    work page doi:10.3390/rs17030550 2025

  12. [12]

    Artificial Intelligence to Advance Earth Observation: A review of models, recent trends, and pathways forward,

    D. Tuia et al., “Artificial Intelligence to Advance Earth Observation: A review of models, recent trends, and pathways forward,” IEEE Geosci. Remote Sens. Mag., vol. 13, no. 4, pp. 119–141, Dec. 2025, doi: 10.1109/MGRS.2024.3425961

    work page doi:10.1109/mgrs.2024.3425961 2025

  13. [13]

    Price, I., Sanchez-Gonzalez, A., Alet, F., Ewalds, T., El- Kadi, A., Stott, J., Mohamed, S., Battaglia, P

    K. He, X. Chen, S. Xie, Y. Li, P. Dollar, and R. Girshick, “Masked Autoencoders Are Scalable Vision Learners,” in 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, LA, USA: IEEE, Jun. 2022, pp. 15979–15988. doi: 10.1109/CVPR52688.2022.01553

    work page doi:10.1109/cvpr52688.2022.01553 2022

  14. [14]

    Masked Autoencoders in Computer Vision: A Comprehensive Survey,

    Z. Zhou and X. Liu, “Masked Autoencoders in Computer Vision: A Comprehensive Survey,” IEEE Access, vol. 11, pp. 113560–113579, 2023, doi: 10.1109/ACCESS.2023.3323383

    work page doi:10.1109/access.2023.3323383 2023

  15. [15]

    An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,

    A. Dosovitskiy et al., “An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale,” in 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3- 7, 2021, OpenReview.net, 2021. Accessed: Nov. 15, 2024. [Online]. Available: https://openreview.net/forum?id=YicbFdNTTy

    2021

  16. [16]

    A multi-scale segmentation/object relationship modelling methodology for landscape analysis,

    C. Burnett and T. Blaschke, “A multi-scale segmentation/object relationship modelling methodology for landscape analysis,” Ecol. Model., vol. 168, no. 3, pp. 233–249, Oct. 2003, doi: 10.1016/S0304- 3800(03)00139-X

    work page doi:10.1016/s0304- 2003

  17. [17]

    Object representations at multiple scales from digital elevation models,

    L. Drăguţ and C. Eisank, “Object representations at multiple scales from digital elevation models,” Geomorphology, vol. 129, no. 3, pp. 183–189, Jun. 2011, doi: 10.1016/j.geomorph.2011.03.003

    work page doi:10.1016/j.geomorph.2011.03.003 2011

  18. [18]

    Visualizing Scale-Domain Manifolds: A Multiscale Geo- Object-Based Approach,

    G. J. Hay, “Visualizing Scale-Domain Manifolds: A Multiscale Geo- Object-Based Approach,” in Scale Issues in Remote Sensing, John Wiley & Sons, Ltd, 2014, pp. 139–169. doi: 10.1002/9781118801628.ch08

    work page doi:10.1002/9781118801628.ch08 2014

  19. [19]

    Object based image analysis for remote sensing,

    T. Blaschke, “Object based image analysis for remote sensing,” ISPRS J. Photogramm. Remote Sens., vol. 65, no. 1, pp. 2–16, Jan. 2010, doi: 10.1016/j.isprsjprs.2009.06.004

    work page doi:10.1016/j.isprsjprs.2009.06.004 2010

  20. [20]

    2025.Approaches to Automated NACE Coding of German Business Activity Descriptions

    G. Castilla and G. J. Hay, “Image objects and geographic objects,” in Object-Based Image Analysis: Spatial Concepts for Knowledge-Driven Remote Sensing Applications, T. Blaschke, S. Lang, and G. J. Hay, Eds., Berlin, Heidelberg: Springer, 2008, pp. 91–110. doi: 10.1007/978-3- 540-77058-9_5

    work page doi:10.1007/978-3- 2008

  21. [21]

    Geographic Object-Based Image Analysis - Towards a new paradigm,

    T. Blaschke et al., “Geographic Object-Based Image Analysis - Towards a new paradigm,” ISPRS J. Photogramm. Remote Sens. Off. Publ. Int. Soc. Photogramm. Remote Sens., vol. 87, no. 100, pp. 180– 191, Jan. 2014, doi: 10.1016/j.isprsjprs.2013.09.014

    work page doi:10.1016/j.isprsjprs.2013.09.014 2014

  22. [22]

    Domain Adaptation for the Classification of Remote Sensing Data: An Overview of Recent Advances,

    D. Tuia, C. Persello, and L. Bruzzone, “Domain Adaptation for the Classification of Remote Sensing Data: An Overview of Recent Advances,” IEEE Geosci. Remote Sens. Mag., vol. 4, no. 2, pp. 41–57, Jun. 2016, doi: 10.1109/MGRS.2016.2548504

    work page doi:10.1109/mgrs.2016.2548504 2016

  23. [23]

    Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources,

    X. X. Zhu et al., “Deep Learning in Remote Sensing: A Comprehensive Review and List of Resources,” IEEE Geosci. Remote Sens. Mag., vol. 5, no. 4, pp. 8–36, Dec. 2017, doi: 10.1109/MGRS.2017.2762307

    work page doi:10.1109/mgrs.2017.2762307 2017

  24. [24]

    Multisource and Multitemporal Data Fusion in Remote Sensing: A Comprehensive Review of the State of the Art,

    P. Ghamisi et al., “Multisource and Multitemporal Data Fusion in Remote Sensing: A Comprehensive Review of the State of the Art,” IEEE Geosci. Remote Sens. Mag., vol. 7, no. 1, pp. 6–39, Mar. 2019, doi: 10.1109/MGRS.2018.2890023

    work page doi:10.1109/mgrs.2018.2890023 2019

  25. [25]

    Prithvi-EO-2.0: A Versatile Multitemporal Foundation Model for Earth Observation Applications,

    D. Szwarcman et al., “Prithvi-EO-2.0: A Versatile Multitemporal Foundation Model for Earth Observation Applications,” IEEE Trans. Geosci. Remote Sens., vol. 64, pp. 1–20, 2026, doi: 10.1109/TGRS.2025.3642610

    work page doi:10.1109/tgrs.2025.3642610 2026

  26. [26]

    Foundation Models for Generalist Geospatial Artificial Intelligence,

    J. Jakubik et al., “Foundation Models for Generalist Geospatial Artificial Intelligence,” Nov. 08, 2023, arXiv: arXiv:2310.18660. doi: 10.48550/arXiv.2310.18660

    work page doi:10.48550/arxiv.2310.18660 2023

  27. [27]

    SatMAE: pre-training transformers for temporal and multi-spectral satellite imagery,

    Y. Cong et al., “SatMAE: pre-training transformers for temporal and multi-spectral satellite imagery,” in Proceedings of the 36th International Conference on Neural Information Processing Systems, in NIPS ’22. Red Hook, NY, USA: Curran Associates Inc., Nov. 2022, pp. 197–211

    2022

  28. [28]

    Bokhovkin, S

    C. J. Reed et al., “Scale-MAE: A Scale-Aware Masked Autoencoder for Multiscale Geospatial Representation Learning,” in 2023 IEEE/CVF International Conference on Computer Vision (ICCV), Paris, France: IEEE, Oct. 2023, pp. 4065–4076. doi: 10.1109/ICCV51070.2023.00378

    work page doi:10.1109/iccv51070.2023.00378 2023

  29. [29]

    Full article: GeoAI: spatially explicit artificial intelligence techniques for geographic knowledge discovery and beyond

    “Full article: GeoAI: spatially explicit artificial intelligence techniques for geographic knowledge discovery and beyond.” Accessed: May 15,

  30. [30]

    Available: https://www.tandfonline.com/doi/full/10.1080/13658816.2019.168450 0

    [Online]. Available: https://www.tandfonline.com/doi/full/10.1080/13658816.2019.168450 0

    work page doi:10.1080/13658816.2019.168450 2019

  31. [31]

    Position: mission critical - satellite data is a distinct modality in machine learning,

    E. Rolf, K. Klemmer, C. Robinson, and H. Kerner, “Position: mission critical - satellite data is a distinct modality in machine learning,” in Proceedings of the 41st International Conference on Machine Learning, in ICML’24, vol. 235. Vienna, Austria: JMLR.org, Jul. 2024, pp. 42691–42706

    2024

  32. [32]

    When Remote Sensing Meets Foundation Model: A Survey and Beyond,

    C. Huo et al., “When Remote Sensing Meets Foundation Model: A Survey and Beyond,” Remote Sens., vol. 17, no. 2, p. 179, Jan. 2025, doi: 10.3390/rs17020179

    work page doi:10.3390/rs17020179 2025

  33. [33]

    A Survey on Remote Sensing Foundation Models: From Vision to Multimodality,

    Z. Huang et al., “A Survey on Remote Sensing Foundation Models: From Vision to Multimodality,” Mar. 28, 2025, arXiv: arXiv:2503.22081. doi: 10.48550/arXiv.2503.22081

    work page doi:10.48550/arxiv.2503.22081 2025

  34. [34]

    Foundation Models for Remote Sensing and Earth Observation: A survey,

    A. Xiao et al., “Foundation Models for Remote Sensing and Earth Observation: A survey,” IEEE Geosci. Remote Sens. Mag., vol. 13, no. 4, pp. 297–324, Dec. 2025, doi: 10.1109/MGRS.2025.3576766

    work page doi:10.1109/mgrs.2025.3576766 2025

  35. [35]

    A raster version of the Circumpolar Arctic Vegetation Map (CAVM),

    M. K. Raynolds et al., “A raster version of the Circumpolar Arctic Vegetation Map (CAVM),” Remote Sens. Environ., vol. 232, p. 111297, 2019

    2019

  36. [36]

    Detection of a planetary system orbiting the eclipsing polar HU Aqr

    D. A. Walker et al., “The Circumpolar Arctic vegetation map,” J. Veg. Sci., vol. 16, no. 3, pp. 267–282, 2005, doi: 10.1111/j.1654- 1103.2005.tb02365.x

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.1111/j.1654- 2005

  37. [37]

    Pan-Arctic ice-wedge degradation in warming permafrost and its influence on tundra hydrology,

    A. K. Liljedahl et al., “Pan-Arctic ice-wedge degradation in warming permafrost and its influence on tundra hydrology,” Nat. Geosci., vol. 9, p. 312, 2016

    2016

  38. [38]

    Automated Detection of Retrogressive Thaw Slumps in the High Arctic Using High-Resolution Satellite Imagery,

    C. Witharana et al., “Automated Detection of Retrogressive Thaw Slumps in the High Arctic Using High-Resolution Satellite Imagery,” Remote Sens., vol. 14, no. 17, Art. no. 17, Jan. 2022, doi: 10.3390/rs14174132

    work page doi:10.3390/rs14174132 2022

  39. [39]

    Recent trends and remaining challenges for optical remote sensing of Arctic tundra vegetation: A review and outlook,

    A. Beamish et al., “Recent trends and remaining challenges for optical remote sensing of Arctic tundra vegetation: A review and outlook,” Remote Sens. Environ., vol. 246, p. 111872, Sep. 2020, doi: 10.1016/j.rse.2020.111872

    work page doi:10.1016/j.rse.2020.111872 2020

  40. [40]

    Impacts of climate-induced permafrost degradation on vegetation: A review,

    X.-Y. Jin et al., “Impacts of climate-induced permafrost degradation on vegetation: A review,” Adv. Clim. Change Res., vol. 12, no. 1, pp. 29– 47, Feb. 2021, doi: 10.1016/j.accre.2020.07.002

    work page doi:10.1016/j.accre.2020.07.002 2021

  41. [41]

    Tundra vegetation change and impacts on permafrost,

    M. M. P. D. Heijmans et al., “Tundra vegetation change and impacts on permafrost,” Nat. Rev. Earth Environ., vol. 3, no. 1, pp. 68–84, Jan. 2022, doi: 10.1038/s43017-021-00233-0

    work page doi:10.1038/s43017-021-00233-0 2022

  42. [42]

    Reviews and syntheses: Effects of permafrost thaw on Arctic aquatic ecosystems,

    J. E. Vonk et al., “Reviews and syntheses: Effects of permafrost thaw on Arctic aquatic ecosystems,” Biogeosciences, vol. 12, no. 23, pp. 7129–7167, Dec. 2015, doi: 10.5194/bg-12-7129-2015

    work page doi:10.5194/bg-12-7129-2015 2015

  43. [43]

    Emergent biogeochemical risks from Arctic permafrost degradation,

    K. R. Miner et al., “Emergent biogeochemical risks from Arctic permafrost degradation,” Nat. Clim. Change, vol. 11, no. 10, pp. 809– 819, Oct. 2021, doi: 10.1038/s41558-021-01162-y

    work page doi:10.1038/s41558-021-01162-y 2021

  44. [44]

    The costs of Arctic infrastructure damages due to permafrost degradation,

    D. A. Streletskiy, S. Clemens, J.-P. Lanckman, and N. I. Shiklomanov, “The costs of Arctic infrastructure damages due to permafrost degradation,” Environ. Res. Lett., vol. 18, no. 1, p. 015006, Jan. 2023, doi: 10.1088/1748-9326/acab18

    work page doi:10.1088/1748-9326/acab18 2023

  45. [45]

    Permafrost thaw-related infrastructure damage costs in Alaska are projected to double under medium and high emission scenarios,

    E. Manos, C. Witharana, and A. K. Liljedahl, “Permafrost thaw-related infrastructure damage costs in Alaska are projected to double under medium and high emission scenarios,” Commun. Earth Environ., vol. 6, no. 1, pp. 1–11, Mar. 2025, doi: 10.1038/s43247-025-02191-7

    work page doi:10.1038/s43247-025-02191-7 2025

  46. [46]

    Iterative self- organizing SCEne-LEvel sampling (ISOSCELES) for large-scale IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. XX, 202X building extraction,

    B. Swan, M. Laverdiere, H. L. Yang, and A. Rose, “Iterative self- organizing SCEne-LEvel sampling (ISOSCELES) for large-scale IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. XX, 202X building extraction,” GIScience Remote Sens., vol. 59, no. 1, pp. 1–16, Dec. 2022, doi: 10.1080/15481603.2021.2006433

    work page doi:10.1080/15481603.2021.2006433 2022

  47. [47]

    Pan-Arctic Permafrost Landform and Human-Built Infrastructure Feature Detection With Vision Transformers and Location Embeddings,

    A. S. Perera et al., “Pan-Arctic Permafrost Landform and Human-Built Infrastructure Feature Detection With Vision Transformers and Location Embeddings,” IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens., vol. 19, pp. 7858–7879, 2026, doi: 10.1109/JSTARS.2025.3648673

    work page doi:10.1109/jstars.2025.3648673 2026

  48. [48]

    Circum-Arctic map of permafrost and ground-ice conditions,

    J. L. Brown, O. J. F. Jr, J. A. Heginbottom, and E. S. Melnikov, “Circum-Arctic map of permafrost and ground-ice conditions,” U.S. Geological Survey, 45, 1997. doi: 10.3133/cp45

    work page doi:10.3133/cp45 1997

  49. [49]

    Convolutional Neural Networks for Automated Built Infrastructure Detection in the Arctic Using Sub-Meter Spatial Resolution Satellite Imagery,

    E. Manos, C. Witharana, M. R. Udawalpola, A. Hasan, and A. K. Liljedahl, “Convolutional Neural Networks for Automated Built Infrastructure Detection in the Arctic Using Sub-Meter Spatial Resolution Satellite Imagery,” Remote Sens., vol. 14, no. 11, Art. no. 11, Jan. 2022, doi: 10.3390/rs14112719

    work page doi:10.3390/rs14112719 2022

  50. [50]

    Hyperparameter Optimization for Large-Scale Remote Sensing Image Analysis Tasks: A Case Study Based on Permafrost Landform Detection Using Deep Learning,

    A. S. Perera, C. Witharana, E. Manos, and A. K. Liljedahl, “Hyperparameter Optimization for Large-Scale Remote Sensing Image Analysis Tasks: A Case Study Based on Permafrost Landform Detection Using Deep Learning,” IEEE Access, vol. 12, pp. 43062– 43077, 2024, doi: 10.1109/ACCESS.2024.3379142

    work page doi:10.1109/access.2024.3379142 2024

  51. [51]

    Pan- Alaskan permafrost tundra capillary network detection and graph theoretic analysis from <1 meter resolution satellite imagery (2011 - 2025)

    M. Pimenta, C. Witharana, A. Perera, A. Liljedahl, and E. Manos, “Pan- Alaskan permafrost tundra capillary network detection and graph theoretic analysis from <1 meter resolution satellite imagery (2011 - 2025).” 2026. doi: 10.18739/A24Q7QS3R

    work page doi:10.18739/a24q7qs3r 2011

  52. [52]

    An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm,

    E. Dinerstein et al., “An Ecoregion-Based Approach to Protecting Half the Terrestrial Realm,” BioScience, vol. 67, no. 6, pp. 534–545, 2017

    2017

  53. [53]

    The Spectral Image Processing System (SIPS)— Interactive visualization and analysis of imaging spectrometer data,

    F. A. Kruse et al., “The Spectral Image Processing System (SIPS)— Interactive visualization and analysis of imaging spectrometer data,” Remote Sens. Environ., vol. 44, no. 2–3, pp. 145-163, 1993

    1993

  54. [54]

    Brightness-normalized partial least squares regression for hyperspectral data,

    H. Feilhauer, G. P. Asner, R. E. Martin, and S. Schmidtlein, “Brightness-normalized partial least squares regression for hyperspectral data,” J. Quant. Spectrosc. Radiat. Transf., vol. 111, no. 12–13, pp. 1947-1957, 2010, doi: 10.1016/j.jqsrt.2010.03.007

    work page doi:10.1016/j.jqsrt.2010.03.007 1947

  55. [55]

    Mahalanobis kernel for the classification of hyperspectral images,

    M. Fauvel, A. Villa, J. Chanussot, and J. A. Benediktsson, “Mahalanobis kernel for the classification of hyperspectral images,” in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS, Honolulu, HI, USA, 2010, pp. 3724-3727,. doi: 10.1109/IGARSS.2010.5651956

    work page doi:10.1109/igarss.2010.5651956 2010

  56. [56]

    Quantifying and mapping biodiversity and ecosystem services: Utility of a multi-season NDVI based Mahalanobis distance surrogate,

    J. Krishnaswamy, K. S. Bawa, K. N. Ganeshaiah, and M. C. Kiran, “Quantifying and mapping biodiversity and ecosystem services: Utility of a multi-season NDVI based Mahalanobis distance surrogate,” Remote Sens. Environ., vol. 113, no. 4, pp. 857-867, 2009

    2009

  57. [57]

    Attention-aware spectral difference representation for hyperspectral anomaly detection,

    W. Zhang, H. Guo, S. Liu, and S. Wu, “Attention-aware spectral difference representation for hyperspectral anomaly detection,” Remote Sens., vol. 15, no. 10, Art. no. 2652, 2023, doi: 10.3390/rs15102652

    work page doi:10.3390/rs15102652 2023

  58. [58]

    Frontera: The Evolution of Leadership Computing at the National Science Foundation,

    D. Stanzione, J. West, R. T. Evans, T. Minyard, O. Ghattas, and D. K. Panda, “Frontera: The Evolution of Leadership Computing at the National Science Foundation,” in Practice and Experience in Advanced Research Computing, in PEARC ’20. New York, NY, USA: Association for Computing Machinery, Jul. 2020, pp. 106–111. doi: 10.1145/3311790.3396656

    work page doi:10.1145/3311790.3396656 2020

  59. [59]

    On Wasserstein two-sample testing and related families of nonparametric tests,

    A. Ramdas, N. Garcia, and M. Cuturi, “On Wasserstein two-sample testing and related families of nonparametric tests,” Entropy, vol. 19, no. 2, p. 47, 2017, doi: 10.3390/e19020047

    work page doi:10.3390/e19020047 2017

  60. [60]

    I. T. Jolliffe, Principal Component Analysis, 2nd ed. New York, NY, USA: Springer, 2002

    2002