Antagonism or synergy: Divergent surface water dynamics at the southern margin of the Eurasian permafrost

Zhang, Bo; Liao, Tiantian; Lu, Haitian; Li, Jianuo; Huang, Ziyan; Li, Jiuhui; Guo, Meng

doi:10.5194/egusphere-2026-1772

Preprints

https://doi.org/10.5194/egusphere-2026-1772

Preprints

29 Apr 2026

| 29 Apr 2026

Status: this preprint is open for discussion and under review for Hydrology and Earth System Sciences (HESS).

Antagonism or synergy: Divergent surface water dynamics at the southern margin of the Eurasian permafrost

Bo Zhang, Tiantian Liao, Haitian Lu, Jianuo Li, Ziyan Huang, Jiuhui Li, and Meng Guo

Abstract. Intensifying climate change and human activities are substantially altering frozen ground conditions, disrupting both surface water regimes and groundwater connectivity. The specific driving mechanisms behind these surface water shifts at the southern margin of the Eurasian permafrost, however, remain poorly quantified due to overlooked spatial heterogeneity. This study analyzed surface water dynamics in the Songhua River Zone (SHRZ) from 1988 to 2024 by integrating an improved water detection method with an interpretable geographical extreme gradient boosting framework coupled with shapley additive explanations. The results show a marked hydrological reversal from shrinkage to expansion around 2012. Expansion in the seasonal frozen ground region (24.77 %) significantly outpaced that in the permafrost region (9.38 %). Spatially explicit attribution identified a structural divergence in regulation mechanisms: the permafrost region is dominated by human activities (76.4 %), forming an "antagonistic" pattern where reservoir-driven expansion is constrained by environmental barriers. In contrast, the seasonal frozen ground region is governed by natural factors (72.4 %), exhibiting a "synergistic" pattern where climate and terrain jointly promote water expansion. Across distinct water types, natural factors control 93.6 % of lake dynamics, whereas human activities dominate river systems (71.0 %) and reservoirs (56.0 %). Furthermore, this surface water expansion occurred alongside accelerated groundwater depletion, suggesting that the surface recovery was achieved at the expense of subsurface storage. These findings demonstrate that surface water expansion does not equate to water security, highlighting the need for targeted surface-groundwater management strategies and prospective research integrating dynamic permafrost degradation processes to further elucidate these ecohydrological trade-offs.

Received: 30 Mar 2026 – Discussion started: 29 Apr 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 8723 KB)

Supplement (3569 KB)

Download & links

Bo Zhang, Tiantian Liao, Haitian Lu, Jianuo Li, Ziyan Huang, Jiuhui Li, and Meng Guo

Status: open (until 28 Aug 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2026-1772', Anonymous Referee #1, 18 May 2026 reply
Review of “Antagonism or synergy: Divergent surface water dynamics at the southern margin of the Eurasian permafrost” submitted to Hydrology and Earth System Sciences (HESS)
This manuscript investigates long-term dynamics of permanent and seasonal surface water across the Songhua River Zone (SHRZ) and evaluates how climatic and anthropogenic drivers shape these patterns. The study assembles multiple remote sensing and meteorological datasets, develops a water-body classification product (IOWDM-ENC), and applies trend analysis, Geographical-XGBoost (G-XGBoost), and SHAP interpretability to assess spatiotemporal changes and their underlying mechanisms.
The topic is timely and relevant for understanding hydrological responses to climate change and human pressures in cold-region environments. The dataset compilation is a valuable contribution. However, the manuscript currently faces major challenges in methodology reproducibility, clarity of data preprocessing, justification of analytical choices, robustness of interpretations, and alignment between methods and conclusions. Several figures require clearer explanation, and some statements overinterpret non-significant or insufficiently supported results. The paper would benefit from substantial revision to strengthen methodological transparency, contextualization in existing literature, and consistency in narrative.
For these reasons, I recommend extensive major revisions, with particular attention to the following points:
Major comments:
The authors should provide additional details on the Landsat imagery used (line 96), including spatial/temporal resolution and how these images were selected.

At line 107, the authors state that SR, SSR, and snowmelt were aggregated annually. How the remaining CEF variables were treated? Did they preserve their original temporal resolution?

All driver datasets begin after 2000, while Landsat images span 1988-2024. The temporal mismatch between trend estimation for surface water (1988–2024) and for drivers (2000–2020) should be addressed. The authors should either justify this inconsistency or harmonize the time periods. In addition, datasets in Table 1 have heterogeneous spatial resolutions. How were they spatially harmonized and resampled? Plase, provide this information in the Materials and Method section.

Line 124 introduces the WI+VI paradigm, which needs a brief conceptual explanation. What does WI+VI stand for, what is the extraction index, and why is this method chosen?

Several classification thresholds, such as the 10% cloud coverage or the 75% water-frequency threshold used to differentiate seasonal and permanent water, require justification. Which criteria did the authors use to determine these choices?

Lines 159–160 state that “a significant excursion from zero” indicates a shift in hydrological regime. The authors should quantify what constitutes a “significant excursion”.

The authors should report the numerical slope ranges used to classify grid-cell trends as significant or insignificant increase/decrease, or as stable. It should also be clarified whether “insignificant” refers to statistical insignificance or to a change of very small magnitude.

The identification of the breakpoint around 2012 should be supported with statistical evidence, such as confidence intervals, or comparisons with alternative breakpoint detection approaches that justify selecting 2012 as a meaningful regime shift.

The authors should expand their explanation of the G-XGBoost framework, including parameter choices, training-validation strategy, and how spatial heterogeneity is incorporated.

Lines 194-196 describe improvements in User’s Accuracy (UA) and Producer’s Accuracy (PA), but the underlying values are not reported. The authors should present the actual UA and PA metrics, not only the percentage changes.

Lines 198-199 state that the proposed method has a “distinct advantage in temporal continuity” relative to JRC-GSW. This appears overstated, as the current JRC-GSW coverage (ending in 2021) reflects its latest release rather than an inherent limitation, and the product is periodically updated. The authors should clarify whether the advantage pertains to near-real-time operational updating of IOWDM-ENC. It would also be valuable to compare detection accuracy across different water-body types against the JRC product.

The interpretation of Fig. 3 as visual confirmation of improved noise suppression should be revised. The comparison appears to rely on single-date imagery (Sentinel-2 and Jilin-1), which reflect specific acquisition times and may not capture seasonal or ephemeral surface water dynamics. As such, discrepancies between the imagery and the mapped products may reflect temporal sampling differences rather than classification performance. A more robust visual validation would require temporally consistent reference data (e.g., multi-temporal composites or independent water occurrence products) rather than single-scene imagery.

In Fig. 4, the authors should define CV and explain how it quantifies data fluctuation. Also, what is the cause of the peaks in surface water in 1998?

The manuscript refers to groundwater depletion in both the abstract and results, but it is unclear how this inference is derived from the presented analysis. How is groundwater depletion inferred from the analysis? Were climate vs anthropogenic contributions separated?

The manuscript occasionally interprets non-significant trends as meaningful increases or decreases. While the direction of estimated slopes can be reported, non-significant results (p ≥ 0.05) should not be described as evidence of change. For example, in Fig. 5 panles a and b, all linear regressions are non-significant (p ≥ 0.05), indicating that no statistically supported trend is detected. The authors should therefore revise the wording throughout the manuscript to clearly differentiate between statistically significant trends and non-significant directional tendencies, and avoid drawing interpretive conclusions from the latter.

Minor comments:
In Fig. 1, the black boundaries delineating the permafrost and seasonal frozen ground regions appear open. The authors should clarify whether this is intentional or correct the boundaries so they are fully closed.

Lines 108–109 (“quantifying the depth… grid box”) are unclear. The authors should revise the sentence to improve clarity.

Acronyms such as MNDWI, NDWI, EVI should be defined upon their first appearance in the manuscript.

Line 127 ends with “and” which suggests that the sentence may be incomplete. The authors should revise or complete the sentence.

The caption for Fig. 2 could be more descriptive. It currently reiterates only the panel titles already shown within the figure. The authors should expand the caption to better explain the figure content.

The caption for Fig. 5 should explain the meaning of the single and double stars used in the cross-correlation shown in panel d. It should also clarify the labels along the x-axis (e.g., what 1a, 2a, 3a represent).

Lines 260–269 refer to Figs. 9, 10, and 11, although based on their order in the text they should correspond to Figs. 6, 7, and 8. The authors should revise the figure numbering to follow the figures’ order of appearance in the text.

The manuscript makes extensive use of acronyms, which become difficult to track. The authors should restate variable definitions when they first appear in the Results section and in figure captions to help reader comprehension.

More detailed captions should be provided for all figures to help readers interpret the visual information.

The authors should explain the meaning of the single and double stars used in Fig. 8.

In line 306, the phrase "around 2009" is repeated twice; the authors should revise for conciseness.

Reply
Citation: https://doi.org/10.5194/egusphere-2026-1772-RC1
- AC1: 'Reply on RC1', Bo Zhang, 27 May 2026 reply
  
  We sincerely thank the referee for the time and effort dedicated to reviewing our manuscript. We are encouraged by the referee’s positive remarks regarding the timeliness of our study and the value of our compiled dataset.
  The referee rightly points out that our manuscript faced key challenges regarding data preprocessing clarity, the justification of analytical choices, the robustness of interpretations, and the alignment between methods and conclusions. We agree with this diagnosis and will systematically address these issues. In the revised manuscript, we will implement comprehensive improvements structured directly around the four core dimensions highlighted by the referee:
  1. Enhancing Methodology Reproducibility and Clarity of Data Preprocessing
  (1) Detailed Image Selection: We will expand Section 2.2.1 to clarify the exact spatial (30 m) and temporal (16-day) resolutions of the Landsat imagery, specifying our quality filtering criteria (e.g., <10% cloud cover limit).
  (2) Data Harmonization: We will explicitly detail the preprocessing steps (Section 2.3), including the bilinear and nearest-neighbor resampling methods used to unify heterogeneous resolutions to a 1 km grid.
  (3) Algorithmic Disclosure: We will disclose the G-XGBoost algorithmic structure in Section 2.3.5, reporting the exact, independently-calibrated optimal hyperparameters (including bandwidth, learning rate, tree depth, and estimators) for both sub-regions in a clear comparative table.
  2. Justifying Analytical Choices
  (1) Temporal Scaling Rationale: We will explain the rationale behind our dual-timeframe strategy. Reconstructing the long-term historical trajectory (1988–2024) using the continuous Landsat archive is necessary to ensure the robustness and reliability of our surface water area trend and breakpoint analysis. This extended timeframe is required to capture the full trajectory of the hydrological reversal from shrinkage to expansion around 2012. Conversely, our quantitative attribution modeling (G-XGBoost) must be restricted to 2000-2020, as the key spatial drivers are restricted to the post-2000 era. This dual-timeframe approach follows previous remote sensing hydrological studies (Wang et al., 2020; Liang et al., 2024)
  (2) Threshold Justification: We will provide robust physical and literature-based justifications for our threshold choices, explaining why the 10% cloud cover limit and the 75% water-frequency threshold represent widely adopted standards in the remote sensing community (Sections 2.2.1 and 2.3.3).
  3. Strengthening Robustness of Interpretations and Differentiating Non-Significant Results
  (1) Addressing Overinterpretation of Non-Significant Trends: We completely agree with the referee's critique. To ensure strict statistical rigor, we will conduct a text-wide review to remove directional terms (such as "decline", "increase", or "trend") for any fitted slopes where p≥0.05. These will be strictly described as stable fluctuations (e.g., "fluctuated without a statistically significant trend"), and no interpretive or mechanistic conclusions will be drawn from them.
  (2) Statistical Breakpoint Verification: We will address the lack of significance testing in our original algorithm-driven CUSUM approach. While the binary segmentation algorithm mathematically located the point of maximum cumulative deviation, we will now implement a rigorous Statistical CUSUM Test evaluated by 1,000 bootstrap permutations (p<0.05) and cross-verified by Pettitt’s test (Section 2.3.4), providing robust statistical evidence for the 2012 regime shift.
  (3) Addressing Overclaiming of Groundwater Status: To ensure scientific precision and prevent overstating the regional condition, we will replace the term "groundwater depletion" with "groundwater storage decline" (or "groundwater storage loss") throughout the text, as the regional aquifers are experiencing a steady decline rather than absolute exhaustion. We will also clarify in Section 3.3 that this decline is a direct observation from the GRACE-derived GWSA dataset, rather than a modeled inference.
  4. Improving Alignment between Methods and Conclusions and Enhancing Figure Clarity
  (1) Type-Specific Accuracy Evaluation: To align our surface water classification with our validation, we will perform a stratified accuracy assessment (User's and Producer's accuracies) specifically for rivers, lakes, and reservoirs against the JRC product, to be included in the Supplementary Information.
  (2) Figure and Caption Upgrades: We will clarify the visual information by defining the Coefficient of Variation (CV) in Figure 4 and explaining the significance stars (* p < 0.05, ** p < 0.01) in Figures 5 and 8. All figure captions will be expanded to be fully self-explanatory.
  We believe these planned systematic revisions directly resolve the issues raised and will significantly enhance the scientific rigor and transparency of our study. Please find the more detailed point-by-point responses in the attached Supplementary document.
  References
  Wang, X., Xiao, X., Zou, Z., Dong, J., Qin, Y., Doughty, R. B., Menarguez, M. A., Chen, B., Wang, J., Ye, H., Ma, J., Zhong, Q., Zhao, B., Li, B.: Gainers and losers of surface and terrestrial water resources in China during 1989–2016, Nat Commun, 11, 3471, https://doi.org/10.1038/s41467-020-17103-w, 2020.
  Liang, H., Zhou, Y., Cui, Y., Dong, J., Gao, Z., Liu, B., and Xiao, X.: Is satellite-observed surface water expansion a good signal to China’s largest granary?, Agric. Water Manage., 303, 109039, https://doi.org/10.1016/j.agwat.2024.109039, 2024.
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2026-1772-AC1
RC2: 'Comment on egusphere-2026-1772', Anonymous Referee #2, 04 Aug 2026 reply

1) Regarding the intensification of climate and its link to surface and groundwater resources that the Authors mention (e.g., "Intensifying climate change and human activities are substantially altering frozen ground conditions, disrupting both surface water regimes and groundwater connectivity."), please discuss how strong is the intensification and which water source is the strongest; for example, based on the study by Koutosyiannis (2020; doi:10.5194/hess-24-3899-2020), the impact of the over-exploitation of groundwater resources is identified as one of the most severe anthropogenic actions with direct effect in the sea-level rise.
2) The Authors mention that "Across distinct water types, natural factors control 93.6% of lake dynamics, whereas human activities dominate river systems (71.0%) and reservoirs (56.0%)."; please further explain how did the Authors estimate the 93.6% of natural factors impact and the 6.6% for the human activities by also considering the uncertainty error in the maps themselves (please calculate this error and include it in the estimations with surplus values).
3) Please show Figures 4 and 5 along with rainfall and runoff satellite maps, taken for example from products such as IMERG (NASA, 2020; https://catalogue.ceda.ac.uk/uuid/47c32530265d4d6e8fdb6c08b2330371) and ERA5 (Hersbach et al., 2020; doi.org/10.1002/qj.3803), to justify the changes in surface water area and availability. Also, additional data analysis may be required to justify the very large peaks in these Figures, what is this sudden break in Figures 4's timseries, please further explain what are the shaded areas in their curves and how are they calculated.
4) A length of at least 30 years is required to capture the long-term climatic variability of the key hydrological-cycle dynamics (Dimitriadis et al., 2021; doi:10.3390/hydrology8020059); the Authors may use this info to justify the adequate length of the 37 years of data-maps they apply (i.e., 1988 to 2024) at the SHRZ. Additionally, please enhance the information on the data analysis by including Tables regarding the spatiotemporal timeseries used in the analysis with primary information (e.g., resolution, exact time-range, missing maps, etc.), spatiotemporal primary statistics (e.g., total mean, standard-deviation, skewness, kurtosis, etc.), seasonal statistics (e.g., primary statistics for each season), autocorrelation structures, etc; this could help the Authors identify several similarities among the data statistics and perform a pooled-analysis that can reveal additional interesting conclusions.
5) Please consider simplifying each conclusion since they seem very difficult to understand, and specify what is their practical meaning, where they can be used and under what conditions, etc.:
a) "The IOWDM-ENC demonstrated more superior applicability than existing methods in the SHRZ, achieving an overall accuracy of 99.03%."; what is a practical application for this.
b) "The Geographical-XGBoost-SHAP framework effectively resolved the stationarity limitations of global-scale models. Rather than assuming a uniform influence, this spatially explicit GeoAI framework quantitatively identifying specific locations where factors function as promoters or inhibitors. This capability provides a reliable technical basis for revealing localized hydrological controls that vary significantly across this complex permafrost margin."; please note that data cannot be stationary or non-stationary but rather a model can be either stationary or non-stationary, and one can select both models and then decide which one best describes the observed series (see for example, discussion in Serinaldi and Kilsby, 2015; doi:10.1016/j.advwatres.2014.12.013).
c) "Surface water dynamics are governed by distinct regulation mechanisms. At the regional scale, a divergence exists between the antagonistic pattern in the permafrost region (where permanent water expansion is constrained by environmental barriers) and the synergistic pattern in the seasonal frozen ground region (where climate, terrain, and human factors amplify expansion). At the typological scale, permanent lake dynamics are primarily controlled by natural factors (55.7% CEF and 37.9% TSF). In contrast, anthropogenic factors dominate both RCF (71.0%) and RP (56.0%), while exerting opposing influences by constraining RCF expansion but promoting RP expansion."; this is quite confusing (please simplify), what do the Authors mean by "distinct regulation mechanisms (are they not natural ones, and why they are distinct).
d) "The hydrological trade-off emerged between surface water expansion and groundwater storage depletion. The observed increase in surface water area coincides with a decline in GWSA, suggesting that surface water expansion has occurred at the potential expense of subsurface storage. This trade-off is particularly significant in the seasonal frozen ground region, highlighting the urgent need for surface-groundwater management strategies tailored to the distinct regulation mechanisms identified in this study."; this can be quite confusing since I would expect when the groundwater resources increase, and after a threshold point, the surface ones would also start to increase (please further explain the practical meaning of this).
6) Please further explain how Equations 1-4 are related to the water surface area and availability estimations extracted from the maps; please show the equations for the whole methodology followed in this study so the Readers can better comprehend it.
7) Please perform a strong polishing of the text by following the SI standards units and regulations (for example, please do not use italics in the Equations 1-4 when more that 1 letter is used for a symbol).

Reply

Citation: https://doi.org/10.5194/egusphere-2026-1772-RC2

Bo Zhang, Tiantian Liao, Haitian Lu, Jianuo Li, Ziyan Huang, Jiuhui Li, and Meng Guo

Supplement

https://doi.org/10.5194/egusphere-2026-1772-supplement

Bo Zhang, Tiantian Liao, Haitian Lu, Jianuo Li, Ziyan Huang, Jiuhui Li, and Meng Guo

Viewed

Total article views: 439 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
292	124	23	439	40	23	26

HTML: 292
PDF: 124
XML: 23
Total: 439
Supplement: 40
BibTeX: 23
EndNote: 26

Views and downloads (calculated since 29 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	41	5	2	48
May 2026	179	73	15	267
Jun 2026	21	13	3	37
Jul 2026	29	15	2	46
Aug 2026	22	18	1	41

Cumulative views and downloads (calculated since 29 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	41	5	2	48
May 2026	179	73	15	267
Jun 2026	21	13	3	37
Jul 2026	29	15	2	46
Aug 2026	22	18	1	41

Viewed (geographical distribution)

Total article views: 412 (including HTML, PDF, and XML) Thereof 412 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 18 Aug 2026

Short summary

To reveal how climate and humans alter water resources, we analyzed 36 years of satellite data at the Eurasian permafrost margin using explainable artificial intelligence. We found surface water expanded rapidly after 2012, driven by human activities in permafrost zones and natural factors in seasonally frozen areas. Crucially, this surface water boom masks severe groundwater depletion, warning that visible water expansion does not equal true water security without integrated management.


Total:	0
HTML:	0
PDF:	0
XML:	0