the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 20 Aug 2025
Evolution and drivers of the large-scale surface freeze-back onset in Siberian permafrost regions with the ERA5-Land reanalysis
Abstract. Permafrost is a defining feature of the Arctic and sub-Arctic environments. The extent of the permafrost region accounts for about a quarter of the Northern Hemisphere’s terrestrial surface. While most research on permafrost has focused on the summer season—when the active layer is thawed and carbon emissions peak—the late shoulder season, marking the transition season between summer and winter and ending by surface freeze-back at the large scale, has received less attention. Yet, about 14 % of the annual mean methane emissions from the permafrost occur during the refreezing period of the active layer. Understanding the seasonality, interannual variability and long-term trends of the surface freeze-back is therefore crucial to better constrain the high-latitude atmospheric carbon budget and improve Earth System Model projections. In this study, we analyze the evolution of surface freeze-back onset from 1950 to 2020 using the ERA5-Land reanalysis (0.1° spatial resolution) over a large region of Siberia encompassing the four main permafrost types. We find that surface freeze-back onset has been delayed by five days on average over that 70-yr period. Through spatial regression modeling, we show that, while several climatic and geographic factors influence freeze-back timing, the driving factor at the large scale (~kilometers) is the date when the 2-meter air temperature first falls below 0 °C, followed by the snow cover depth. These findings complement previous research that focused on the small scale (~meters), which emphasized the importance of the vegetation type and the snow cover characteristics at these spatial scales. Our results provide new insights into changes during the late shoulder season in one of the world’s fastest-warming regions and identify key variables to monitor for improving sub-seasonal forecasts that could become relevant to infrastructure upgrade and logistics planning in permafrost-affected areas.
- Preprint
(2111 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 30 Oct 2025)
- RC1: 'Comment on egusphere-2025-3680', Anonymous Referee #1, 04 Oct 2025 reply
-
RC2: 'Comment on egusphere-2025-3680', Anonymous Referee #2, 23 Oct 2025
reply
Review of „Evolution and drivers of the large-scale surface freeze-back onset in Siberian permafrost regions with the ERA5-Land reanalysis“ by Osy et al.
The article by Osy et al. investigates temporal changes and drivers of surface freeze-back timing across central Siberian permafrost regions over the period 1950–2020. The authors analyze ERA5-Land reanalysis data (spatial resolution ~9 km) and apply statistical regression models to identify long-term trends and potential controlling variables of surface freeze-back. The study also discusses implications for ecosystems, carbon cycling, and potential applications for predicting ground freeze-back.
Regarding the temporal trends, the authors find that the onset of surface freeze-back has been delayed by approximately five days across the study region over the 70-year period, with some differences among permafrost types. While a later freeze-back is an expected response to a warming climate, the reported regional differences could indeed be interesting. However, the analysis remains rather limited in depth—being based on only four linear trends derived from regional averages—and I am not fully convinced by the explanations offered for the differences among permafrost types (see detailed comments below).
Concerning the spatial drivers, the study concludes that surface freezing in the ERA5-Land dataset is primarily controlled by the timing of subzero air temperatures and the onset of an insulating snow cover. The authors interpret this as being in contrast with field-scale studies, which emphasize the dominant role of vegetation and snow characteristics in ground freezing. However, I would argue that the reported results are not particularly surprising and are, in fact, broadly consistent with expectations given the scale and nature of reanalysis data (see comment below).
The manuscript is generally well understandable, and the applied methods appear technically sound. The presentation quality of the findings, in particular the figures, is relatively poor and visually not appealing. In addition, some sections could be better structured; for example, methodological details currently placed in the Results section would be more appropriate in the Methods section.
In summary, while the manuscript addresses an interesting topic and applies established datasets and methods, it provides limited new understanding of permafrost processes and several key methodological and interpretative aspects are insufficiently substantiated. Below, I am providing specifc points of critique to substantiate this evaluation. Overall, I do not consider the manuscript suitable for publication in The Cryosphere in its current form.
- Study area: While the study area spans a gradient from continuous permafrost to permafrost-free regions, I do not consider it representative of the circum-Arctic permafrost domain. Consequently, the conclusions drawn from this region may not be transferable to other permafrost environments. The selected area is highly continental, characterized by strong temperature seasonality and comparatively high aridity, which differ markedly from coastal permafrost regions. These climatic conditions likely influence the timing of surface freeze-back—through faster autumn cooling and distinct snow characteristics—relative to other permafrost settings. Given that the analysis relies entirely on reanalysis data, it is unclear why the study was restricted to central Siberia. A circum-Arctic analysis should have been feasible in terms of both data availability and computational cost, and would have provided more robust and generalizable insights.
- Study design and variable selection: Both the dependent variable (timing of surface freeze-back) and most of the predictor variables (air temperature, snow depth, soil moisture) are derived from the ERA5-Land reanalysis. As such, the statistical model primarily reflects the internal parameterizations and physics of the IFS/CHTESSEL land-surface scheme, rather than providing independent insights into real-world processes. It is therefore unsurprising that the most influential predictors of freeze-back timing in the analysis are also derived from key variables (air temperature and snow depth) within the reanalysis system itself.
- Statistical regression: The authors chose to fit regression models using large-scale spatial averages (Section 3.1) and long-term temporal averages (Section 3.2). This approach results in a substantial loss of variability—and therefore information—contained in the original dataset (as the authors acknowledge in Line 227). Given that the statistical models are computationally inexpensive to fit, it is unclear why the analysis was not conducted at the grid-cell level. Such an approach would still have allowed for spatial aggregation for interpretation. Similarly, retaining interannual variability in the predictor variables for the spatial regressions would likely have led to more robust and meaningful results.
- Interpretation of findings: At several points, I find the authors’ interpretations unconvincing or insufficiently supported:
- Lines 237ff: The authors attribute the weaker trend in continuous permafrost zones to lower ground temperatures, suggesting that these make the ground “less responsive” to atmospheric forcing, whereas non-continuous permafrost is “more responsive.” The term “responsiveness” is not clearly defined, and no physical mechanism is provided to substantiate this claim. Moreover, if the authors hypothesize that permafrost temperature influences freeze-back timing, it would have been appropriate to include permafrost temperature as a predictor variable in the regression analysis to test this hypothesis directly.
- Lines 350ff: The explanation linking weaker freeze-back trends in continuous permafrost regions to snow characteristics is not plausible. First, snow it not only spatially variable in the continous permafrost zone, but also in non-continous zones, which can actually be the cause for the non-continuity of permafrost. Second, a higher spatial variability of surface freeze-back would not explain, why the temporal trend is lower, compared to a region with less spatial variabilty. Third, the subgrid variablity of snow and surface conditions is not represented in the ERA5-Land product used to fit the statistical model: How should the statistical model „learn“ this, if its training data (the ERA5 product) does not include these processes?
- Practical implications: In Section 4.4, the authors propose that their regression models could support practical applications by predicting freeze-back timing from seasonal forecasts of air temperature and snow depth (e.g., from the ECMWF IFS system). However, this argument is not convincing. Seasonal forecast systems already provide soil temperature outputs, from which the timing of surface freeze-back could be derived directly. This would be both simpler and likely more accurate than using a regionally tuned statistical model based on reanalysis-derived variables.
Citation: https://doi.org/10.5194/egusphere-2025-3680-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 2,215 | 42 | 16 | 2,273 | 47 | 65 |
- HTML: 2,215
- PDF: 42
- XML: 16
- Total: 2,273
- BibTeX: 47
- EndNote: 65
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Authors presented evolution and drivers of the surface freeze-back onset in Siberian permafrost regions based on the ERA5-Land reanalysis. The results show a delay in the surface freeze-back onset of approximately 5 days since 1950. The most significant controlling factors were identified as near-surface air temperature and snow depth. Furthermore, the positive trend in the freeze-back onset was found to be weakest in regions with continuous permafrost.
However, in its current form, the manuscript does not appear to contribute substantial new insights to the scientific understanding of how environmental changes interact with soil freeze-back processes. This limitation stems partly from the somewhat subjective and potentially problematic approach taken in the selection and testing of candidate predictors. The paper could be significantly strengthened through more in-depth spatial analysis. General comments and specific suggestions are provided below for the authors' consideration.
General comments
1) Candidate predictors
The authors dedicate a substantial portion of the manuscript to discussing the importance of the selected candidate predictors. However, the results indicate that the influence of most predictors (3 out of 5) is not statistically significant. While the coarse scale and inherent biases of the ERA5-Land dataset may explain why such influences are not captured, I strongly recommend that the authors condense this section, particularly the discussions concerning vegetation type, groundwater content, and latitude. Additionally, the authors mention the potential importance of snowfall timing but do not provide further analysis on this point.
2) Manuscript Structure
Abstract: Please consider making the abstract more concise by removing information that is either well-known (e.g., the spatial resolution of ERA5-Land and general statements on permafrost extent) or not directly relevant to the study's core findings (e.g., methane emissions). The abstract should function as a summary and contain only the most critical points of the article.
Introduction: A significant part of the introduction (2 out of 5 paragraphs) focuses on the carbon cycle feedback. While the permafrost-carbon cycle is undoubtedly important, the introduction would benefit from a better balance. Please consider reformulating this section to also adequately highlight other crucial processes significantly affected by the freeze-back onset, such as hydrological cycles and phenology.
Furthermore, the extensive body of research on surface freeze/thaw status derived from remote sensing (e.g., Kim et al., 2012, 2017) is overlooked and should be incorporated into the literature review.
Results: The readability of this section is compromised because it frequently incorporates methodological explanations (e.g., L233-235, 251-255, 273-274). These details should be relocated to the Methods section to allow the Results to focus solely on the findings and their implications.
3) Nomenclature
The nomenclature for some variables is unconventional. For instance, in Equation 4, "Qno snow" could be changed to the more standard notation "Qsf" (for snow-free conditions) or "Qsc" (for snow-covered conditions). Additionally, the variable "dayair temp" in Equation 6 is not commonly used and should be revised for clarity.
4) Methology
Daily mean ground temperature: is defined as the arithmetic average of the temperature at 00:00 and 12:00 UTC. Given that ERA5-Land provides hourly soil temperature data, could the authors clarify why only two specific timesteps were selected for this calculation? It would be more representative to use the full diurnal cycle
Snow Cover: While snow insulation plays an important role in the soil thermal regime at seasonal and annual scales, its relevance to the freeze-back onset is less clear, as this transition typically occurs before a snow depth exceeding 14 cm is established. In my opinion, the freeze-back onset may be more closely linked to early snowfall events, which can cool the ground via the latent heat during snowmelt.
The use of a 14 cm snow depth as an insulating threshold (based on Eq. 5) depends on several factors, including snow thermal conductivity, the chosen insulation threshold (e.g., 5%), and the values of parameters k and u. Although the use of typical values is a reasonable starting point, additional sensitivity experiments are needed to assess the influence of these parameters and improve the robustness of the algorithm. Furthermore, please provide a reference for the snow thermal conductivity value of 0.25 W m-1 K-1—this value appears relatively low for a snow density of 350 kg m-3.
5) Spatial analysis
This study could be significantly enhanced by additional spatial analysis. For instance, mapping the onset trend with significant patterns across the entire study area would be particularly informative.
Specific Comments
L28: Cohen et al., 2020? This is not a permafrost mapping research
L84: Consider using the updated referece of Obu et al., 2019.
L199: Typo here?
L224: 1950-200?
References:
Kim, Y., Kimball, J. S., Zhang, K., and McDonald, K. C.: Satellite detection of increasing Northern Hemisphere non-frozen seasons from 1979 to 2008: Implications for regional vegetation growth, Remote Sensing of Environment, 121, 472–487, 2012.
Kim, Y., Kimball, J. S., Glassy, J., and Du, J.: An extended global Earth system data record on daily landscape freeze–thaw status determined from satellite passive microwave remote sensing, 2017.
Obu, J., Westermann, S., Bartsch, A., Berdnikov, N., Christiansen, H. H., Dashtseren, A., Delaloye, R., Elberling, B., Etzelmüller, B., Kholodov, A., Khomutov, A., Kääb, A., Leibman, M. O., Lewkowicz, A. G., Panda, S. K., Romanovsky, V., Way, R. G., Westergaard-Nielsen, A., Wu, T., Yamkhin, J., and Zou, D.: Northern Hemisphere permafrost map based on TTOP modelling for 2000–2016 at 1 km2 scale, Earth-Science Reviews, 193, 299–316, 2019.