the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 17 Jun 2026
Regional Variations in Drivers of Active Layer Thickness: A Site-Scale Analysis Across Northern Hemisphere Permafrost
Abstract. In recent decades, permafrost degradation across the Northern Hemisphere has accelerated markedly under climate warming. However, the characteristics and driving mechanisms of this degradation at the sample-plot scale remain poorly understood. This study classifies Northern Hemisphere permafrost into three types: the Circum-Arctic permafrost region (CAP), the Sub-Circum-Arctic permafrost region (SCAP), and the Qinghai–Tibet Plateau (QTP) permafrost region. Based on data from 785 active layer thickness (ALT) monitoring sites we found that the mean value of ALT in CAP, SCAP and QTP region was 84.9 ± 1.56 cm, 200 ± 8.99 cm and 224 ± 7.26 cm, respectively. Based on 291 sites with more than five years of ALT records, we found that 60 % of the sites exhibit an increasing trend in ALT (35 % with statistically significant increases), while 40 % show a decreasing trend (11 % with statistically significant decreases). The rates of ALT changing vary considerably among different permafrost regions and with higher increasing rate in QTP (3.23 cm yr⁻¹, n=36) and followed by CAP (1.48 cm yr⁻¹, n=58) and SCAP (1.19 cm yr⁻¹, n=38). Results from the Partial least squares path modelling (PLS-PM) indicate that, at the site scale, soil characteristics exert a stronger influence on ALT than air temperature, particularly in the CAP and QTP regions. In SCAP, precipitation is the most important factor driving ALT changes, as higher precipitation can transfer heat into the soil and affect soil temperature. Vegetation also plays a significant role in SCAP, where denser vegetation can generate a warming effect. Snow cover shows limited influence on ALT across all monitoring sites. This study offers a systematic review of permafrost degradation and its driving mechanisms at the monitoring-site scale.
- Preprint
(4485 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 29 Jul 2026)
-
RC1: 'Comment on egusphere-2026-841', Anonymous Referee #1, 18 Jun 2026
reply
-
AC1: 'Reply on RC1', Yutong Lin, 27 Jun 2026
reply
We sincerely thank the reviewer for the time and effort dedicated to reviewing our manuscript. We are encouraged by the reviewer's positive remarks regarding the timeliness of our study and the value of our compiled dataset.
The reviewer rightly points out that our manuscript faced key challenges regarding methodological clarity, data interpretation, and statistical rigor. We agree with this diagnosis and have systematically addressed these issues. In the revised manuscript, we have implemented comprehensive improvements, including clarifying the trend analysis and sample size, revising the abstract to better reflect the role of vegetation, removing repetitive statements in the Introduction, adding recommended references, clarifying regional delineation procedures, specifying the temporal scale of ALT, reconciling inconsistent site numbers, discussing spatial resolution limitations in Section 4.4, supplementing the trend analysis with Mann-Kendall tests and sensitivity analysis, clarifying collinearity diagnostics using VIF, improving figure readability, providing robust explanations for decreasing ALT trends with multiple factors and literature support, reorganizing Section 4.3.1 to explain why GST must be considered, and expanding Section 4.4 to discuss the coarse resolution of NDVI and snow cover data as a limitation.
We believe these systematic revisions directly resolve the issues raised and will significantly enhance the scientific rigor and transparency of our study. Detailed point-by-point responses to each comment are provided in the attached supplementary document.
-
AC1: 'Reply on RC1', Yutong Lin, 27 Jun 2026
reply
-
RC2: 'Comment on egusphere-2026-841', Anonymous Referee #2, 18 Jun 2026
reply
This manuscript compiles active layer thickness (ALT) data from 785 monitoring sites across the Northern Hemisphere to analyze spatial distribution patterns, multi-year trends, and the relative importance of different environmental drivers at the site scale. The authors classify the study regions into three broad categories: the Circum-Arctic permafrost region (CAP), the Sub-Circum-Arctic permafrost region (SCAP), and the Qinghai–Tibet Plateau (QTP) region. Using Extreme Gradient Boosting (XGBoost) for importance ranking and Partial Least Squares Path Modeling (PLS-PM) for causal pathway analysis, the study concludes that soil characteristics exert a stronger control on ALT than air temperature at the local scale. The manuscript presents a substantial data compilation effort, which is highly valuable for the permafrost community. The application of PLS-PM to resolve direct and indirect environmental pathways is a welcome step toward shifting from simple univariate correlation analyses to multivariate causal networks. However, the manuscript has several fundamental scientific and physical-mechanistic flaws that must be resolved before publication.
1. The spatial categorization of Northern Hemisphere permafrost into CAP, SCAP, and QTP is structurally problematic because it overlooks distinct permafrost typologies that possess fundamentally different hydrothermal and morphological behaviors. For example, grouping the mountain permafrost in the European Alps in mid-latitude, high-relief mountain sites into either CAP or SCAP is physically unjustifiable. Mountain permafrost is characterized by steep topography, alpine weather regimes, and blocky/debris-covered rock glaciers with thin sediment cover. At these sites, ALT is extremely thick (often >3 to 5 m), highly heterogeneous, and extraordinarily sensitive to altitude, aspect, and the timing of snowpack onset/melt. As for maritime permafrost (e.g., Scandinavia): Scandinavian permafrost (such as in southern and northern Norway) is highly influenced by the maritime climate of the North Atlantic Current. This permafrost is generally warm (ground temperatures close to 0 °C), wet, and insulated by deep winter snow. Consequently, the active layer is significantly thicker than that in continental high-latitude continuous permafrost (e.g., central Siberia) at comparable latitudes. Its response to climate warming is governed by a complex coupling of winter insulation and rainfall-driven sensible heat transfer. By merging these unique alpine and maritime environments into generic “CAP” or “SCAP” pools, their distinctive thermodynamic responses are masked. The authors must refine their spatial grouping or, at a minimum, explicitly discuss and analyze the mountain (Alps) and maritime (Scandinavia) sites as separate sub-populations to reflect their specific environmental controls (refer to Isaksen et al., 2007; Zenklusen Mutter and Phillips, 2012; Luo et al., 2016).
2. On Page 20, Lines 6–7, the authors state: “The 40% monitoring sites show decreasing trend of ALT may because that the warming climate not yet be sufficient to induce widespread permafrost degradation in these areas”. This statement reveals a fundamental misunderstanding of permafrost degradation morphology in ice-rich terrains. In continuous high-latitude permafrost (CAP), the upper permafrost (transition layer) often contains massive ground ice. Under climate warming, the thaw of this excess ice leads to thaw settlement (ground surface subsidence). As the ground surface settles downward alongside the descending thaw table, the measured ALT—which is strictly the distance from the subsided ground surface to the thaw table—remains stable or even decreases. Thus, a stable or decreasing ALT trend at ice-rich sites does not indicate “insufficient warming” or a lack of degradation; rather, it is a morphological indicator of severe thermokarst development and underground ice loss. The authors must correct this interpretation throughout the manuscript and cite appropriate literature documenting this process (e.g., Shiklomanov et al., 2013; Luo et al., 2016; Luo et al., 2023).
3. The paper is framed as a “Site-Scale Analysis”, yet the datasets used to drive the statistical models (Table 1) are regional or global-scale gridded products: Meteorological data (MAT, TP, LST) are extracted from ERA5-Land at 0.1° resolution (∼9 km grid cells). Vegetation data (NDVI) are from GIMMS NDVI3g at an 8 km spatial resolution. Soil properties are from SoilGrids at 250 m resolution. At the actual site scale (typically a 100 m×100 m CALM grid or a single borehole footprint), the surface energy balance and thermal conductivity are controlled by micro-scale factors: localized microtopography (slope/aspect), local soil organic layer thickness, micro-vegetation canopy, and localized snow redistribution. Extracting an 8 km or 9 km grid value to represent a localized monitoring plot introduces severe spatial representativeness errors. This mismatch is highly likely to artificially suppress the apparent influence of temperature/precipitation while overestimating the relative importance of static soil properties (like soil depth or sand content). The authors must discuss this major scale limitation and its impact on the PLS-PM model path coefficients.
4. The PLS-PM model results indicate that air temperature has a weaker direct effect on ALT than soil characteristics, which the authors interpret as temperature being a secondary control at the site scale. From a heat-transfer perspective, this statistical decoupling is physically mediated by phase-change latent heat (the zero-curtain effect) at the active layer base. Air temperature acts as the upper boundary thermal forcing (driving force). The soil column, regulated by soil moisture (SWC) and soil sand content (SSC), acts as a thermodynamic filter (thermal conductivity and heat capacity). When down-propagating thermal waves reach the ice-rich active layer base, the thermal energy is consumed by the latent heat of ice melting. This physical thermal barrier decouples the direct statistical correlation between air temperature and thaw depth. Therefore, temperature is not “less important”; its signal is physically masked by the moisture-phase-change coupling. This physical mechanism must be integrated into the discussion (Section 4.3) to ground the statistical PLS-PM results in cryospheric thermodynamics.
For these reasons, I recommend a Major Revision. Below are detailed evaluations and specific comments to guide the authors in revising their work.
Citation: https://doi.org/10.5194/egusphere-2026-841-RC2 -
AC2: 'Reply on RC2', Yutong Lin, 27 Jun 2026
reply
We sincerely thank the reviewer for the time and effort dedicated to reviewing our manuscript. We are encouraged by the reviewer's positive remarks regarding the substantial data compilation effort and the application of PLS-PM to resolve direct and indirect environmental pathways.
The reviewer rightly points out that our manuscript faces several fundamental scientific and physical-mechanistic challenges regarding spatial categorization, interpretation of decreasing ALT trends, scale mismatch between site-scale analysis and coarse-resolution datasets, and the physical mechanism underlying the statistical PLS-PM results. We agree with this diagnosis and have systematically addressed these issues. In the revised manuscript, we have: (1) clarified that our regional classification follows Brown et al. (1997) and explained that only one site from Scandinavia is present with no sites from the Alps, adding a discussion of this limitation in Sections 2.1 and 4.4; (2) revised Section 4.1 to incorporate the thaw settlement mechanism in ice-rich permafrost (Shiklomanov et al., 2013; Luo et al., 2023); (3) added a thorough discussion in Section 4.4 acknowledging the spatial scale mismatch and its impact on PLS-PM results; and (4) integrated the physical mechanism of phase-change latent heat at the active layer base into Section 4.3.1, citing Dobinski (2011) and Groenke et al. (2024).
We believe these systematic revisions directly resolve the issues raised and will significantly enhance the scientific rigor and physical-mechanistic basis of our study.
Detailed point-by-point responses to each comment are provided in the attached supplementary document.
-
AC2: 'Reply on RC2', Yutong Lin, 27 Jun 2026
reply
Data sets
ALT sites CALM https://www2.gwu.edu/~calm/
ALT sites GTN-P http://gtnpdatabase.org/activelayers
This study addresses the regional variability, trends, and drivers of ALT across Northern Hemisphere permafrost zones. With a dataset of 785 sites and a multi-method approach (XGBoost and PLS-PM), the authors demonstrate clear spatial heterogeneity among CAP, SCAP, and QTP regions. The findings offer robust evidence to inform future permafrost projections. The manuscript is recommended for acceptance after minor revisions.
Page 1:
Lines 19-21: Was the statistical analysis restricted exclusively to the sampling sites exhibiting an increasing trend in ALT, or were all sampling sites across the three regions incorporated? Judging from the reported sample sizes, it appears that only the sites with upward trends were considered.
Lines 24-25: Vegetation also intercepts incoming solar radiation, exerting both shading and insulating buffering effects, which helps maintain relatively cooler soil temperatures.
Page 2-3:
Lines 13-14, 5-6: Repetitive wording is observed between the closing lines of the first and second paragraphs on page 2. It is recommended that one instance be deleted to improve conciseness.
Page 3:
Lines 6-7, 18-19: Similarly, the two expressions are essentially consistent in meaning. Rather than repeating this summary at the end of each paragraph, we suggest that the authors integrate it into a more appropriate position based on the specific content of the two paragraphs.
Page 4:
Lines 4-6: Give the references
Page 5:
Line 10: ALT is the variable that the authors intend to analyze; therefore, it should not be used as a criterion for regional delineation.
Lines 11-14: Figure 1 appears to have been directly adopted from Obu et al. (2019). In the Methods section, the authors should further elaborate on the specific procedures used to delineate the CAP, SCAP, and QTP regions. In particular, details regarding the overlay processing between the QTP and other permafrost classification layers should be clearly described.
Page 6:
Line 4: The authors should explicitly specify the temporal scale for the ALT
Line 7: The authors state on page 4 that the GTN-P database contains 305 sites, yet here is 406. This inconsistency should be reconciled.
Page 7:
Line 7: The abstract states "291 sites," whereas the Methods and Results sections report "219 sites." We recommend that the authors verify the correct number and ensure consistency throughout the manuscript.
Page 8:
Table 1: It should be noted that the marked differences in spatial resolution among the datasets employed here may introduce substantial deviations when comparing the results with those obtained from previous site-based observational studies. In particular, the coarse resolution of climatic forcing data is likely to severely attenuate the detected contribution of climatic factors.
Line 3: The trend analysis is based solely on linear regression with p < 0.05, but the authors have not adequately addressed issues related to variable series lengths, missing data, autocorrelation, or nonlinear temporal dynamics. Since the inclusion criterion permits sites with as few as five consecutive years of observations, the derived linear trends may lack stability. It is advisable to incorporate Mann–Kendall tests and Sen's slope estimates, and to evaluate the sensitivity of the results to short time series, with explicit discussion of how record length may affect trend significance.
Page 9:
Line 6: How are data with collinearity handled?
Page 12:
Line 11-12: The comparison results are not clearly evident because the analysis for SCAP and QTP appears to have been omitted. The authors should include the corresponding analyses for these two regions. Alternatively, the authors should refer to Figure 3 in this paragraph to support the statement.
Page 14:
Lines 15-16: Should this not be GST (Ground Surface Temperature) instead?
Page 15-16:
Fig. 4: The font size and the significance asterisks in Figure 4 should be enlarged to improve readability. The term "Hot figures" should be revised to "Heatmaps" for accuracy and standard terminology.
Page 17:
Line 8: The PLS-PM approach serves as a critical method for interpreting the driving mechanisms in this study. However, the model specifications differ across the three regions (CAP, SCAP, and QTP), with varying sets of observed variables (e.g., TEMP). The authors should provide a clearer and more consistent justification for: (1) the construction of latent variables (e.g., which specific indicators constitute "soil," "temperature," "topography," "snow," etc.); (2) the criteria used for retaining or excluding particular observed variables; (3) the rationale for path directions and whether they are grounded in prior theoretical frameworks; and (4) why the model structures are not uniform across the three regions, and whether this regional differentiation is methodologically justified or imposes limitations on cross-regional comparisons.
Page 18:
Lines 19-20: The supplementary materials referred to in the manuscript could not be found.
Page 19:
Fig.6: The readability of the figures needs improvement: the font size is too small, and the line colors should be darkened for better contrast.
Page 20:
Lines 6-8: The current reasoning appears tenuous and requires either stronger empirical evidence or a more cautious interpretation.
Line 10: The use of the word "supports" in this context is not accurate / is imprecise.
Page 20:
Lines 21-23: Give the references
Page 22:
Line 7: air temperature or soil temperature?
Lines 21-24: This sentence should be rephrased, as the current wording is imprecise and lacks sufficient evidentiary support.
Lines 17-24: The current text does not clearly explain why GST must be taken into account. We suggest that the authors rephrase/reorganize this section to improve clarity and logical flow.
Page 24:
Lines 1 and 16: In the CAP and SCAP regions, vegetation and snow cover are recognized as important indicators influencing permafrost dynamics. However, the current results show that neither factor has a statistically significant effect. This may be largely attributable to the coarse resolution of the datasets used, rather than reflecting actual field conditions—especially given that the analysis is not based on site-specific monitoring data. The authors should provide a thorough discussion of this limitation in the Discussion section.