the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 May 2026
An Ensemble Dataset of Permafrost Thaw Conditions for Northern High Latitudes from Open Satellite Data
Abstract. Permafrost in the northern high latitudes is experiencing accelerated warming, casting important implications for carbon–climate feedback and ecosystem stability. Integrating three open-access permafrost products, this study generates an ensemble permafrost dataset of permafrost percent (PP) and mean annual ground temperature (MAGT), and proposes a Permafrost Thaw Index (PTI) to measure thaw vulnerability in a two-tier ranking system. Sixteen open-access datasets derived primarily from satellite observations are ensembled to represent environmental conditions, including land surface temperature (LST), vegetation dynamics, snow cover, freeze–thaw state, soil properties, and topography. An ensemble machine learning approach, XGBoost, is employed to predict PTI from these datasets with an overall accuracy of 91.8%. Thermal variables, particularly LST, LST trends, and frozen days, contribute most strongly to PTI prediction. The PTI map reveals clear latitudinal gradients and regional variations in alignment with eco-climatic transitions. The total permafrost coverage in the northern high latitudes (> 45°N) is approximately 18.7 million km2. About 41.4% remains highly stable, while nearly half is subject to high thaw pressure. The most stable conditions persist in the High Arctic tundra of North America, the Arctic East Siberia, and the high-elevation alpine tundra of North Asia. The highest vulnerability occurs along the southern margins dominated by boreal and montane forests. A comparison of the predicted PTI with borehole records from 26 established stations reveals strong agreement (Spearman's r = 0.69). This study demonstrates the value of multi-source open satellite data in permafrost research. The proposed PTI framework provides a scalable approach for monitoring permafrost dynamics and supporting climate impact assessments over the northern high latitudes. The ensemble products of this study, including the PTI map and the ensemble land cover and permafrost distributions (PP and MAGT), can be freely downloaded at https://doi.org/10.5281/zenodo.19148960.
- Preprint
(3449 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 16 Jul 2026)
-
CC1: 'Comment on essd-2026-299', Robert Way, 21 May 2026
reply
-
CC2: 'Reply on CC1', Shuying Zang, 10 Jun 2026
reply
Dear Dr. Way,
Thank you for your constructive comments, the relevant literature you suggested, and references to the Nordicana D Network, which provides additional borehole sites to enrich our product evaluation. Your extensive experience in cryosphere and northern environment research is highly valued and greatly appreciated. In the attached response letter, please find our point-by-point responses to your comments. We will also revise the manuscript accordingly after the review window is closed.
Best,
Susan Wang
University of South Carolina
(on behalf of all authors)
-
CC3: 'Reply on CC2', Robert Way, 10 Jun 2026
reply
I would like to acknowledge that the authors have provided a quite meaningful response to the comments that I provided on their manuscript. It is certainly laudable to see this effort put in for a non-reviewer comment. The intent of the comments from me was to highlight some of the challenges with some of the existing products that are currently being used in place of the IPA map. There are prominent issues present in these maps in areas like valley bottoms in the Yukon which are prone to inversions and coastal environments like those described in our work in Labrador. I do agree with the authors that the median of the existing ensemble will present a more accurate representation of permafrost conditions than each of those individual maps may illustrate. As a coauthor of Obu et al who was involved in analysis on its performance in Subarctic Canada, I believe it is fair to say that I was not expecting this map was going to be used instead of the IPA map in Canada by so many studies in the time since it was published.
I also appreciate the authors going through the effort to note the challenges in these products in the region I work in (Labrador). However, from my perspective, our early mapping efforts like those in Way and Lewkowicz (2016) were prone to some of the same issues that the global products were - insufficient representation of permafrost along the coast. This issue is not only pertinent to peatlands but also in many wind-exposed coastal tundra environments, especially with marine deposits that are prone to ground ice formation (see e.g. Chiasson and Allard, 2022). We have unpublished work that suggests that improvements to the representation of snow redistribution along the coast is critical to representing permafrost in these areas so although we agree that improved peatland distribution maps would help, we do think there are systemic issues with some of these global maps in not capturing coastal climate conditions appropriately.
We certainly appreciate the authors willingness to discuss these important issues. Thanks again for their contributions.
Chiasson, A., and Allard, M. 2022. Thermal contraction crack polygons in Nunavik (northern Quebec): Distribution and development of polygonal patterned ground. Permafrost and Periglacial Processes, 33(3): 195–213. doi:10.1002/ppp.2150.Citation: https://doi.org/10.5194/essd-2026-299-CC3 -
AC1: 'Reply on CC3', Shuying Zang, 10 Jun 2026
reply
Dr. Way,
Thank you again for your supportive comments. We have downloaded the article by Chiasson and Allard (2022) to gain a better understanding of the unique characteristics of coastal permafrost associated with peat, localized snow cover and ground ice, etc. In the revised manuscript, we will comprehensively discuss these environmental impacts along coastlines on global permafrost products.
Best,
Susan Wang (on behalf of all co-authors)
Citation: https://doi.org/10.5194/essd-2026-299-AC1
-
AC1: 'Reply on CC3', Shuying Zang, 10 Jun 2026
reply
-
CC3: 'Reply on CC2', Robert Way, 10 Jun 2026
reply
-
CC2: 'Reply on CC1', Shuying Zang, 10 Jun 2026
reply
Data sets
Ensemble Dataset of Permafrost Thaw Conditions, Northern High Latitudes (>45°N) Dianfan Guo et al. https://doi.org/10.5281/zenodo.19148960
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 216 | 49 | 21 | 286 | 14 | 9 |
- HTML: 216
- PDF: 49
- XML: 21
- Total: 286
- BibTeX: 14
- EndNote: 9
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
To the authors of Duo et al.
Certainly there is value in collating together globally datasets of permafrost conditions for use by the broader community so I applaud the authors in attempting to do so. However, there are also challenges that emerge when averaging or presenting an ensemble of datasets that suffer from the same flaws in similar geographic areas. In these cases, it is best to hatch out or not present these regions where there was no source data or where there are known problems rather than to have these issues extend throughout the analysis.
I'll call attention to one particular region of my research interest, Labrador, northeast Canada, where we've noted significant problems with representing coastal permafrost in previous works (Way and Lewkowicz, 2016; Way et al. 2018; Wang et al, 2023). Of particular relevance was the work by Wang et al (2023) which directly compared against Obu et al (2021) and others, finding a complete misrepresentation of the distribution of permafrost in southeastern Labrador. Part of this was because of poor land cover products but other parts of it relate to not representing coastal tundra snow redistribution well. This current work seems to reproduce this challenge across its maps. This issue is less prominent on the IPA map which does consider geomorphological observations in its derivation. Although lower resolution, the IPA map is still the gold standard for most areas in the discontinuous zone in Canada.
It is also notable when the datasets do not necessarily use all available public borehole sources. For example, Nordicana D includes a wide variety of boreholes from locations in northeastern Canada not well-represented by the data used by the authors.
I do understand the need for global efforts, but it's important to remember that these products get used by others and in some cases this can result in real harm during environmental assessment processes if whole geographic regions have had their permafrost or ground ice conditions mischaracterized. This is all to say that far more effort needs to be given to improving models/maps by examining where they have failed and not simply averaging out or reproducing the same issues identified with prior efforts. At the very least, a more expansive literature search is needed to identify the areas where these existing models/maps fail to accurately characterize permafrost at broad regional scales.
Way, R.G., and Lewkowicz, A.G. 2016. Modelling the spatial distribution of permafrost in Labrador–Ungava using the temperature at the top of permafrost. Canadian Journal of Earth Sciences, 53(10): 1010–1028. doi:10.1139/cjes-2016-0034.
Way, R.G., Lewkowicz, A.G., and Zhang, Y. 2018. Characteristics and fate of isolated permafrost patches in coastal Labrador, Canada. The Cryosphere, 12(8): 2667–2688. doi:10.5194/tc-12-2667-2018.