Shifting Snowmelt Regime in a High-Latitude Asian Basin: Insights from the Songhua River Basin

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

General comments

This study analyses changes in snowmelt in the Songhua River Basin (in Northeast China). The analysis is based on the ERA5-Land dataset, which uses data from 1961 to 2020.  The results show a significant elevation-dependent decreasing trend in annual snowmelt, indicating a more significant decrease, particularly at lower elevations. The tendency of an increased snowmelt in early spring is followed by a decreasing snowmelt in April and May. Dominant factors controlling snowmelt and its seasonal shift are related to thermal factors and initial snow depth.

Overall, the study is clearly written and has a good structure. I have only two suggestions that can be considered before recommending the manuscript for publication:

1. While ERA5-Land reanalysis was used in many studies and proved to be a useful tool for many hydrological analyses, several limitations have also been reported, such as overestimating shallow snow and underestimating deep snow late in the season, or biases in estimates of air temperature or precipitation. The manuscript briefly discusses the point that ERA5 has some limitations, but does not provide any quantitative assessment of the magnitude of the errors, the accuracy of the estimates, or the uncertainty associated with the analyzed snow simulations. It will be very valuable for the readers to understand/see how reliable the analysis is. Leaving the reader with just a note that this can be evaluated in future studies is not sufficient and does not give credit to all the presented results.
2. The reasoning for using correlation analysis between reanalysis data and some observations is not clear. What kind of new understanding does it bring? Reanalysis is driven by some observations, so what information about the correlation is adding to the current knowledge? Would it not be better (e.g. also for addressing point 1) to use the observations to validate the ERA5 datasets used for the analysis of snowmelt changes? For example, do the total precipitation or air temperatures from ERA5 correspond with observed values in terms of magnitude, timing, and frequency?

 

Author Response

Response to Reviewer #1 on hydrology-4010576 “Shifting Snowmelt Regime in a High-latitude Asian Basin: Insights from the Songhua River Basin”

 

This study analyses changes in snowmelt in the Songhua River Basin (in Northeast China). The analysis is based on the ERA5-Land dataset, which uses data from 1961 to 2020.  The results show a significant elevation-dependent decreasing trend in annual snowmelt, indicating a more significant decrease, particularly at lower elevations. The tendency of an increased snowmelt in early spring is followed by a decreasing snowmelt in April and May. Dominant factors controlling snowmelt and its seasonal shift are related to thermal factors and initial snow depth.

Overall, the study is clearly written and has a good structure. I have only two suggestions that can be considered before recommending the manuscript for publication:

[Authors’ response]: Thank you for your positive feedback and for recognizing the clarity and structure of our manuscript. We greatly appreciate your overall assessment.

 

1. While ERA5-Land reanalysis was used in many studies and proved to be a useful tool for many hydrological analyses, several limitations have also been reported, such as overestimating shallow snow and underestimating deep snow late in the season, or biases in estimates of air temperature or precipitation. The manuscript briefly discusses the point that ERA5 has some limitations, but does not provide any quantitative assessment of the magnitude of the errors, the accuracy of the estimates, or the uncertainty associated with the analyzed snow simulations. It will be very valuable for the readers to understand/see how reliable the analysis is. Leaving the reader with just a note that this can be evaluated in future studies is not sufficient and does not give credit to all the presented results.

[Authors’ response]: Thank you for raising this valuable and constructive point. We fully agree that a more substantive discussion regarding the reliability of the dataset is essential for contextualizing our findings. As suggested, we have expanded the discussion in Section 4.4 “Uncertainty and Limitations” (lines 509–609 of the revised manuscript) to include a quantitative assessment. Specifically, we validated ERA5-Land snow depth against an independent long-term daily snow depth dataset for China.

2. The reasoning for using correlation analysis between reanalysis data and some observations is not clear. What kind of new understanding does it bring? Reanalysis is driven by some observations, so what information about the correlation is adding to the current knowledge? Would it not be better (e.g. also for addressing point 1) to use the observations to validate the ERA5 datasets used for the analysis of snowmelt changes? For example, do the total precipitation or air temperatures from ERA5 correspond with observed values in terms of magnitude, timing, and frequency?

[Authors’ response]: Thank you for your thoughtful comments. The purpose of conducting a correlation analysis between reanalysis data and observations is to isolate and quantify the statistical relationship between key climate drivers (such as temperature and precipitation) and the total snowmelt during melt events, while controlling for the influence of other meteorological factors. We acknowledge that reanalysis products are informed by observations, but this analysis still reveals the independent explanatory power of each variable within the reanalysis system, which is valuable for understanding the processes. Furthermore, as you correctly suggested, direct validation of reanalysis inputs is essential. Therefore, as noted in response to Comment 1, we have validated the data (see revised Section 4.4, lines 509–609).

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study has substantial practical significance, as it clarifies the spatiotemporal variations of snowmelt, snowmelt events, and their climatic drivers in the Songhua River Basin (SRB). This is particularly important under the context of rapid climate change. The manuscript also contains rich figures and is highly readable. However, several issues still require improvement. My suggestions are as follows:

Line 42
The transition from snowmelt-related hazards to the spatiotemporal distribution of snow is clearly inappropriate. A transitional sentence should be added to explain the relationship between snow distribution patterns and snowmelt hazards.

Line 59
The citation numbered 15 does not align with your argument and should be revised.

In addition, too many references are older than five years; more recent studies should be incorporated.

Line 84
Figure 1 is overly simple. It is recommended to add provincial boundaries to the China base map. Also, the color of “Rivers” in the legend does not match the color shown in the figure.

Line 85
For all datasets, you should clearly describe the download sources and processing procedures. Moreover, a validation section should be added to demonstrate whether these remote sensing datasets are reliable within the study area.

Line 91
“Based on data from the SRB from 1961 to 2020, this study defines the period from August 1 each year to July 31 of the following year as a hydrological year.”
What is the basis for this definition? Please add a reference.

Line 113
“For the definition of snowmelt events, a threshold of daily snowmelt ≥ 0.1 mm was applied, ……”
Why is the threshold set at 0.1 mm? Provide justification and add a citation.

Line 118
Why was the nearest-neighbor method used for resampling? Compared with other resampling methods, it is generally not optimal. Please explain.

Line 131
The definitions of j and i are missing and should be added.

Line 134
When introducing the Mann–Kendall (MK) test, please include relevant citations.

Line 180
Figure 2 appears overly complicated, which hinders readability. It is recommended to present Fig. 2(a) separately. The explanation for Fig. 2(d) is oversimplified—does the figure represent the maximum, minimum, or mean values during 1991–2020? This should be clarified.

Line 325
Many factors influence snowmelt. Why were MSD, ARF, and other variables selected? Provide justification and add supporting references. Also, using only correlation analysis seems too simplistic; consider employing more robust analytical methods if possible.

Line 353
The Discussion section is very weak, which I believe is the part that requires the most improvement. It is recommended to compare your results with previous studies to substantiate your findings. If differences arise, explain the reasons.
You should also compare with research conducted in other regions worldwide, and discuss how the unique geographic and climatic conditions of the SRB contribute to its distinct snowmelt characteristics.

Line 412
It is suggested to revise the heading to “Conclusions”.

Author Response

Response to Reviewer #2 on hydrology-4010576 “Shifting Snowmelt Regime in a High-latitude Asian Basin: Insights from the Songhua River Basin”

 

This study has substantial practical significance, as it clarifies the spatiotemporal variations of snowmelt, snowmelt events, and their climatic drivers in the Songhua River Basin (SRB). This is particularly important under the context of rapid climate change. The manuscript also contains rich figures and is highly readable. However, several issues still require improvement. My suggestions are as follows:

[Authors’ response]: Thank you for your thoughtful and constructive review of our manuscript. We sincerely appreciate your positive assessment of the study's practical significance, readability, and visual presentation, as well as the time and effort you have dedicated to providing these valuable suggestions. We have carefully considered all the points you raised and agree that addressing them will significantly strengthen the manuscript. In our revised version, we will incorporate detailed responses and make the necessary modifications

 

Line 42: The transition from snowmelt-related hazards to the spatiotemporal distribution of snow is clearly inappropriate. A transitional sentence should be added to explain the relationship between snow distribution patterns and snowmelt hazards.

[Authors’ response]: Thank you for pointing out the need for a clearer transition between the discussion on snowmelt-related hazards and the spatiotemporal distribution of snow. In response to your suggestion, we have added a bridging sentence at Lines 42-45 in the revised manuscript:

“The distribution characteristics of snow, such as range, depth, and duration, directly determine the potential scale and timing of meltwater release, and the changes in its distribution pattern caused by climate change will further reshape the risk pattern of related disasters.”

This sentence explicitly links snow distribution patterns to snowmelt hazards, thereby clarifying the logical progression and justifying the subsequent focus on spatiotemporal snow patterns. We believe this addition improves the coherence and flow of the argument. Please let us know if further adjustments are needed.

 

Line 59

The citation numbered 15 does not align with your argument and should be revised.

In addition, too many references are older than five years; more recent studies should be incorporated.

[Authors’ response]: Thank you for your valuable feedback. We have revised your comments on citation 15 accordingly. The original citation has been replaced by the following latest and relevant research:

15. Qi W, Feng L, Liu J, Yang H (2020) Snow as an important natural reservoir for runoff and soil moisture in Northeast China. J Geophys Res Atmos, 125, e2020JD033086. https://doi.org/10.1029/2020JD033086.

We believe that this statement is more consistent with the argument presented in this section. In addition, we have noticed your concern about the number of old references. Throughout the manuscript, we have included several recent studies (published within the past five years) to ensure that the literature review and supporting evidence are up-to-date. These supplementary contents have been shown in the reference section of the revised manuscript (such as lines 755-760 and 767-770).

 

Line 84

Figure 1 is overly simple. It is recommended to add provincial boundaries to the China base map. Also, the color of “Rivers” in the legend does not match the color shown in the figure.

[Authors’ response]: Thank you for your suggestion. We have revised Figure 1 accordingly by adding provincial boundaries to the base map of China, as recommended. Additionally, the color of “Rivers” in the legend has been corrected to match the display in the figure. The updated version can be found in line 85 of the revised manuscript.

 

Line 85

For all datasets, you should clearly describe the download sources and processing procedures. Moreover, a validation section should be added to demonstrate whether these remote sensing datasets are reliable within the study area.

[Authors’ response]: Thank you for the suggestion. As requested, we have added detailed download sources and processing descriptions for all datasets in Section 2.2 “Datasets” (lines 88–134). Furthermore, a validation subsection has been included in Section 4.4 “Uncertainty and Limitations” (lines 509–609) to quantitatively assess the reliability of the key remote sensing data within the study area. Specifically, we validated the ERA5-Land snow depth product against an independent long-term daily snow depth dataset for China.

Line 91

“Based on data from the SRB from 1961 to 2020, this study defines the period from August 1 each year to July 31 of the following year as a hydrological year.”

What is the basis for this definition? Please add a reference.

[Authors’ response]: Thank you for your question regarding the definition of the hydrological year. The period from August 1 to July 31 of the following year was selected to better align with the hydrological characteristics of the SRB, particularly in relation to snowmelt processes.The reference supporting this definition is:

Tan XJ, Wu ZN, Mu XM, et al. (2019) Spatiotemporal changes in snow cover over China during 1960–2013. Atmos Res, 218, 183–194. https://doi.org/10.1016/j.atmosres.2018.11.018.

We have added this reference in the revised manuscript (Lines 112-114) .

 

Line 113

“For the definition of snowmelt events, a threshold of daily snowmelt ≥ 0.1 mm was applied, ……”

Why is the threshold set at 0.1 mm? Provide justification and add a citation.

[Authors’ response]: Thank you for your question regarding the selection of the 0.1 mm daily snowmelt threshold. In the revised manuscript (lines 137–144), we have added the following justification:

"For the definition of snowmelt events, a threshold of daily snowmelt ≥ 0.1 mm was applied, and consecutive periods meeting this condition with a minimum duration of one day were identified as independent events. The daily melting threshold of ≥ 0.1 millimeters was selected to sensitively capture the beginning of spring snowmelt and weak snowmelt signals, and to maintain the continuity of the snowmelt period for accurate hydrological analysis. This method is consistent with common practice in mid- to high-latitude snow research, where 0.1 mm of water column per day is widely used as a baseline sensitivity threshold for detecting melt occurrence while minimizing noise [24,25]." The references cited are:

[24] Gorodetskaya et al. (2023), who used a similar threshold for detecting surface melt in Antarctic Peninsula studies.

[25] Puggaard et al. (2025), who applied a 0.1 mm/day melt threshold in Greenland Ice Sheet melt analysis.

 

Line 118

Why was the nearest-neighbor method used for resampling? Compared with other resampling methods, it is generally not optimal. Please explain.

[Authors’ response]: Thank you for your comment regarding the use of the Nearest Neighbor method for resampling. We agree that Nearest Neighbor interpolation is often not the optimal choice in many interpolation scenarios, particularly where smooth gradients are expected. In our study, we thoroughly evaluated three common interpolation methods—Nearest Neighbor, Bilinear, and Inverse Distance Weighting—specifically for snow depth and snowmelt data from ERA5-Land over the Songhua River Basin. The results indicated that all three methods performed very well, with high accuracy (R² > 0.968 for snow depth and R² = 1.0 for snowmelt) and low errors (RMSE ~3.52-3.58 mm for snow depth and ~0.10-0.11 mm for snowmelt). The differences among the methods were marginal in our application. Given the comparable performance, we selected the Nearest Neighbor approach based on the following considerations: Physical conservatism, Numerical stability, Computational efficiency、Thus, while Nearest Neighbor may not be universally optimal, it proved to be a suitable and justified choice for our specific data and research context.. We have added this explanation in the revised manuscript (lines 161-178) and appreciate your feedback, which helped us clarify our methodological reasoning.

 

Line 131

The definitions of j and i are missing and should be added.

[Authors’ response]: Thank you for pointing out the missing definitions of indices j and i in the equation. We have added their definitions in the revised manuscript at line 200, clarifying that they represent the grid row and column indices, respectively.

 

Line 134

When introducing the Mann–Kendall (MK) test, please include relevant citations.

[Authors’ response]: Thank you for your suggestion regarding the inclusion of citations for the Mann-Kendall (MK) test. As recommended, we have added the relevant foundational references in the revised manuscript (lines 202-203):

Kendall, M.G. (1975) Rank Correlation Methods, 4th ed. Charles Griffin, London.

Mann, H.B. (1945) Nonparametric tests against trend. Econometrica, 13(3), 245–259. https://doi.org/10.2307/1907187.

 

Line 180

Figure 2 appears overly complicated, which hinders readability. It is recommended to present Fig. 2(a) separately. The explanation for Fig. 2(d) is oversimplified—does the figure represent the maximum, minimum, or mean values during 1991–2020? This should be clarified.

[Authors’ response]: Thank you for your valuable feedback. We have carefully revised the figures and their descriptions in response to your comments. Regarding your first point, we agree that the original Figure 2 was overly complex. To improve clarity, we have separated it into two distinct figures: Figure 4 (showing the temporal trend of annual snowmelt from 1961 to 2020) and Figure 5 (presenting the spatial and elevational distributions). Concerning your second point on the need for clarification, we have updated the description of the proportional change maps (now in Figure 5c and 5d). The caption now explicitly states that these panels show the proportional change in the average annual snowmelt for the period 1991–2020 relative to the baseline period 1961–1990. This revision ensures the meaning is precise and unambiguous. The specific modifications can be found in Lines 280–285 of the revised manuscript.

 

Line 325

Many factors influence snowmelt. Why were MSD, ARF, and other variables selected? Provide justification and add supporting references. Also, using only correlation analysis seems too simplistic; consider employing more robust analytical methods if possible.

[Authors’ response]: Thank you for your insightful comment regarding variable selection and analytical methods. The key variables (MSD, ARF, MWS, AMPT, ASSD, MRHU) were selected based on their established physical roles in snowpack energy balance and melt dynamics, as detailed in Lines 184-192. They comprehensively represent snow storage (MSD), radiative (ASSD), turbulent (MWS, MRHU), and advective (ARF, AMPT) energy inputs. References:

10. Li X, Cui P, Zhang X Q, et al., 2024. Intensified warming suppressed the snowmelt in the Tibetan Plateau[J/OL]. Advances in Climate Change Research, 15(3): 452-463. https://doi.org/10.1016/j.accre.2024.06.005.
11. Li S, Luo C, Lu H (2025) The spatial and temporal distribution of rain-on-snow events and their driving factors in China. Hydrol Process, 39, e70098. https://doi.org/10.1002/hyp.70098.
12. Tian L, Li H, Li F, et al. (2018) Identification of key influence factors and an empirical formula for spring snow-melt-runoff: a case study in mid-temperate zone of northeast China. Sci Rep, 8, 16950. https://doi.org/10.1038/s41598-018-35282-x.

We agree that moving beyond simple correlation is important. Therefore, our analysis of the identified snowmelt events employed a robust, multi-method framework. As described in Lines 448-519, we used Pearson correlation, partial correlation, and multiple regression analyses to evaluate the associations.

 

Line 353

The Discussion section is very weak, which I believe is the part that requires the most improvement. It is recommended to compare your results with previous studies to substantiate your findings. If differences arise, explain the reasons.

You should also compare with research conducted in other regions worldwide, and discuss how the unique geographic and climatic conditions of the SRB contribute to its distinct snowmelt characteristics.

[Authors’ response]: Thank you for your valuable suggestion. We all agree that strengthening the discussion section is crucial. In response, we have made significant revisions and extensions to this section in order to provide a more in-depth analysis of our results. Specifically, as described in the revised manuscript (lines 541-562 and 571-584 of the revised manuscript).

Line 412

It is suggested to revise the heading to “Conclusions”.

[Authors’ response]: Thank you for your suggestion regarding Line 412. The heading has been revised to "Conclusions" as recommended.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The MS presents a comprehensive, basin-scale analysis of snowmelt changes in the Songhua River Basin (SRB) from 1961–2020 using ERA5-Land reanalysis data. It documents a clear shift toward earlier, shorter, and less intense snowmelt, with strong elevation dependence and changing drivers (temperature → initial snow depth → rain-on-snow). The topic is highly relevant for Hydrology, especially for high-latitude Asian basins that are under-represented in the literature. The writing is generally clear, figures are informative, and the event-scale perspective is a genuine contribution. However, the current version has several critical scientific and methodological shortcomings that prevent acceptance in its present form. The most serious issues are (1) complete lack of validation or uncertainty discussion of the core variable (snowmelt from ERA5-Land), (2) insufficient justification and robustness checks for the event definition, and (3) over-confident attribution claims that are not supported by the correlation-only analysis.

Major Comments

1. Absence of validation/uncertainty assessment of ERA5-Land snowmelt.This is the single most critical flaw. The entire study hinges on daily snowmelt derived from ERA5-Land, yet the authors provide zero validation against in-situ observations, remote-sensing products (e.g., GlobSnow, CMC, MODIS SWE), or even published basin-scale snow studies in Northeast China. ERA5-Land is known to overestimate snow depth and have timing biases in complex terrain and forested regions (exactly the case for the SRB). Without any uncertainty estimate or sensitivity test, all conclusions about magnitude, trends, and event characteristics remain questionable. → Mandatory: Include a quantitative validation of ERA5-Land snow depth and, if possible, derived snowmelt against independent data (station records from Heilongjiang/Jilin provinces, or at least published regional studies). At minimum, provide a thorough discussion of known ERA5-Land biases in high-latitude Asia and a sensitivity analysis (e.g., using different thresholds or another reanalysis).
2. Snowmelt event definition is arbitrary and not robust. The choice of ≥0.1 mm/day with no minimum duration and no separation of minor mid-winter melts from major spring events is problematic. This likely inflates event frequency in early decades and makes inter-decadal comparison questionable. → Should be:

• Test sensitivity of all event metrics (frequency, TSM, PDSM, etc.) to thresholds (e.g., 0.5 mm, 1 mm, 2 mm and 2–5 consecutive days).
• Consider defining “major spring melt events” separately from minor mid-winter events (e.g., events occurring after 1 February and contributing >70% of annual melt).
• Report how many events are single-day, 2-day, etc., and how this distribution changed over time.

3.   Over-interpretation of Pearson correlations as “dominant controls” and causal  attribution Statements such as “early melt events were primarily driven by thermal factors (r=0.71–0.72)” and “final complete ablation events were governed by initial snow depth (r≥0.89)” are not supported by simple bivariate correlations alone, especially when variables are strongly collinear (temperature, radiation, rain-on-snow, snow depth). → Should be:

• Replace causal language with “strongest correlation” or “most closely associated”.
• Perform partial correlation or multiple regression analysis to disentangle collinear drivers.
• Consider dominance analysis or relative importance metrics.

4. Rain-on-snow (RoS) analysis is underdeveloped The abstract and discussion emphasize RoS as a “key driver”, but the only evidence is accumulated rainfall during events when Tavg > 0°C. This is a very crude proxy. → Pls

• Define proper RoS events (liquid precipitation on days with snow cover present).
• Quantify frequency, intensity, and contribution of genuine RoS events to total melt in early vs. late periods.

5. The introduction and conclusion stress implications for spring floods and agricultural irrigation, but the results actually show reduced late spring (April–May) melt → less meltwater recharge exactly when irrigation demand peaks in Northeast China. This important (and counter-intuitive) consequence is never discussed. Discuss the implications of reduced late-spring meltwater availability for the well-known “spring drought” problem in the Songhua/Sanjiang Plain.

Minor Comments

1. Define SDAP/SDABP clearly in the abstract or earlier.
2. English polishing needed in several places (e.g., “which is yet limited in previous work”).
3. Units: snow depth is given in mm in Figure 7 but elsewhere seems to be cm → standardize.
4. Provide the actual number of grid cells (or effective sample size) used in correlation analysis.
5. The claim that snowmelt dates advanced “9, 6, and 2 days earlier” for 25%, 50%, 75% cumulative levels needs the baseline period clearly stated in the abstract.

Author Response

Response to Reviewer #3 on hydrology-4010576 “Shifting Snowmelt Regime in a High-latitude Asian Basin: Insights from the Songhua River Basin”

 

The MS presents a comprehensive, basin-scale analysis of snowmelt changes in the Songhua River Basin (SRB) from 1961–2020 using ERA5-Land reanalysis data. It documents a clear shift toward earlier, shorter, and less intense snowmelt, with strong elevation dependence and changing drivers (temperature → initial snow depth → rain-on-snow). The topic is highly relevant for Hydrology, especially for high-latitude Asian basins that are under-represented in the literature. The writing is generally clear, figures are informative, and the event-scale perspective is a genuine contribution. However, the current version has several critical scientific and methodological shortcomings that prevent acceptance in its present form. The most serious issues are (1) complete lack of validation or uncertainty discussion of the core variable (snowmelt from ERA5-Land), (2) insufficient justification and robustness checks for the event definition, and (3) over-confident attribution claims that are not supported by the correlation-only analysis.

[Authors’ response]: Thank you for your time and constructive comments on our manuscript. We appreciate the positive assessment of the relevance, clarity, and contribution of our work, as well as the thorough identification of key issues that need to be addressed.

 

Major Comments

1. Absence of validation/uncertainty assessment of ERA5-Land snowmelt.This is the single most critical flaw. The entire study hinges on daily snowmelt derived from ERA5-Land, yet the authors provide zero validation against in-situ observations, remote-sensing products (e.g., GlobSnow, CMC, MODIS SWE), or even published basin-scale snow studies in Northeast China. ERA5-Land is known to overestimate snow depth and have timing biases in complex terrain and forested regions (exactly the case for the SRB). Without any uncertainty estimate or sensitivity test, all conclusions about magnitude, trends, and event characteristics remain questionable. → Mandatory: Include a quantitative validation of ERA5-Land snow depth and, if possible, derived snowmelt against independent data (station records from Heilongjiang/Jilin provinces, or at least published regional studies). At minimum, provide a thorough discussion of known ERA5-Land biases in high-latitude Asia and a sensitivity analysis (e.g., using different thresholds or another reanalysis).

[Authors’ response]: Thank you for raising this critical point regarding the validation of ERA5-Land snow data. In response to your comment, we have added a dedicated subsection, “4.4 Uncertainty and Limitations,” in the revised manuscript (lines 590–609). A quantitative validation of ERA5-Land snow depth against the independent Long-term Series of Daily Snow Depth Dataset in China. Results indicate that while ERA5-Land shows a systematic high bias—which may lead to overestimated melt magnitudes—it exhibits strong consistency in spatiotemporal patterns (mean spatial correlation = 0.77) and effectively captures daily and annual trends (R² = 0.81 and 0.75, respectively).

 

2. Snowmelt event definition is arbitrary and not robust. The choice of ≥0.1 mm/day with no minimum duration and no separation of minor mid-winter melts from major spring events is problematic. This likely inflates event frequency in early decades and makes inter-decadal comparison questionable. → Should be:

• Test sensitivity of all event metrics (frequency, TSM, PDSM, etc.) to thresholds (e.g., 0.5 mm, 1 mm, 2 mm and 2–5 consecutive days).
• Consider defining “major spring melt events” separately from minor mid-winter events (e.g., events occurring after 1 February and contributing >70% of annual melt).
• Report how many events are single-day, 2-day, etc., and how this distribution changed over time.

[Authors’ response]: Thank you for raising this important point regarding the definition and robustness of snowmelt events. We agree that the choice of threshold and the structure of event identification can significantly influence the resulting metrics and their inter-decadal comparability. We have addressed your concerns through a comprehensive sensitivity analysis and additional investigations as detailed below.

1. Sensitivity Analysis of Detection Thresholds

In direct response to your suggestion, we have added a dedicated sensitivity analysis in the revised manuscript (Section 4.4 "Uncertainty and Limitations," lines 609–625). This analysis systematically tests the impact of using different daily melt-rate thresholds (0.1, 0.5, 1.0, and 2.0 mm/day) on key event metrics, including frequency, duration, and cumulative volume. The results confirm that the chosen threshold directly influences these characteristics. However, the comparison demonstrates that our selected threshold of 0.1 mm/day is optimal for our study's objectives, which focus on capturing the initial onset and temporal continuity of melt processes crucial for phenological and hydrological analyses. This threshold minimizes the omission of early melt signals while maintaining robustness against potential data noise, thereby ensuring a reliable basis for our long-term analysis.

1. Analysis of Event Duration Distribution and Change

As you recommended, we have explicitly analyzed how the duration of snowmelt events has changed over time. This new analysis is presented in the manuscript (lines 414–434) and summarized in Figure 15. It reveals distinct seasonal shifts in event structure between 1961–1990 and 1991–2020:

In the early season (Sep-Jan): The total number of events decreased, with the most pronounced declines occurring in **longer-duration events (e.g., 7+ day events).

In the late season (Feb-May): The total number of events increased, driven almost entirely by a significant rise in short-duration events (1-6 days), while longer-duration events slightly decreased or remained stable.

This detailed breakdown directly addresses your concern by showing not just if the number of events changed, but how their internal structure (duration distribution) has evolved, providing a more nuanced understanding of changing melt patterns.

1. Regarding the Separation of Minor Winter and Major Spring Events

We acknowledge the value of distinguishing between minor mid-winter melts and major spring melt events. While our current event definition (using the 0.1 mm/day threshold) is designed to capture the full continuum of melt processes—which is essential for assessing early phenological triggers and hydrological connectivity—we recognize that defining "major spring events" separately could offer additional insights for certain applications.

In this revision, we have addressed the spirit of this suggestion through the seasonal analysis mentioned above (point 2). By analyzing the Snow Depth Accumulation Period (SDAP, Sep-Jan) and Snow Depth Ablation Period (SDABP, Feb-May) separately, we effectively distinguish between earlier/milder melt events and the core spring melt season. Furthermore, our finding that events in the SDABP contribute the dominant share (e.g., >70%) of the annual melt volume inherently identifies this season as comprising the "major" melt events. We have clarified this interpretation in the text.

 

3. Over-interpretation of Pearson correlations as “dominant controls” and causal attribution Statements such as “early melt events were primarily driven by thermal factors (r=0.71–0.72)” and “final complete ablation events were governed by initial snow depth (r≥0.89)” are not supported by simple bivariate correlations alone, especially when variables are strongly collinear (temperature, radiation, rain-on-snow, snow depth). → Should be:

• Replace causal language with “strongest correlation” or “most closely associated”.
• Perform partial correlation or multiple regression analysis to disentangle collinear drivers.
• Consider dominance analysis or relative importance metrics.

[Authors’ response]: Thank you for your insightful comment regarding the over-interpretation of bivariate correlations. We fully agree that collinearity among variables such as temperature, radiation, rain-on-snow, and snow depth can make it misleading to attribute causality or dominance based solely on Pearson correlation coefficients. We have revised the manuscript (lines 448–491). The revised section now bases its interpretation on a multi-method statistical approach. We evaluated the associations using Pearson correlation, partial correlation, and multiple regression analysis specifically to disentangle the effects of collinear drivers. This allows us to distinguish between strong bivariate associations and independent statistical relationships.

 

4. Rain-on-snow (RoS) analysis is underdeveloped The abstract and discussion emphasize RoS as a “key driver”, but the only evidence is accumulated rainfall during events when Tavg > 0°C. This is a very crude proxy. → Pls

• Define proper RoS events (liquid precipitation on days with snow cover present).
• Quantify frequency, intensity, and contribution of genuine RoS events to total melt in early vs. late periods.

[Authors’ response]: Thank you for raising this important point regarding the analysis of rain-on-snow (ROS) events. We agree that the proxy used in our original analysis was crude. In the revised manuscript, we have addressed this by first clarifying our definitions (Lines 154-157): We specify that while a standard ROS event requires liquid precipitation onto existing snow cover, our study does not directly identify such events. Instead, we analyze a defined subclass termed "rain-on-snow melt (ROSm) events," which are broader snowmelt events during which at least 0.1 mm of daily rainfall occurs. To quantify the contribution of genuine ROS conditions to melt, we performed a stratified analysis of these ROSm events by season (Snow Depth Accumulation Period, SDAP: Sep-Jan; Snow Depth Ablation Period, SDABP: Feb-May) and by rainfall intensity (Lines 499-519; Figure 18).

 

5. The introduction and conclusion stress implications for spring floods and agricultural irrigation, but the results actually show reduced late spring (April–May) melt → less meltwater recharge exactly when irrigation demand peaks in Northeast China. This important (and counter-intuitive) consequence is never discussed. Discuss the implications of reduced late-spring meltwater availability for the well-known “spring drought” problem in the Songhua/Sanjiang Plain.

Minor Comments

1. Define SDAP/SDABP clearly in the abstract or earlier.

[Authors’ response]: Thank you for this suggestion. We have clearly defined the snow season phases earlier in the revised manuscript (Lines 114-117) as follows: "Based on the snow cover phenology in Northeast China [22,23], the snow season is divided into two distinct phases: the Snow Depth Accumulation Period (SDAP, spanning September 1st to January 31st) and the Snow Depth Ablation Period (SDABP, spanning February 1st to May 31st)."

 

2. English polishing needed in several places (e.g., “which is yet limited in previous work”).

[Authors’ response]: Thank you for highlighting the need for language polishing. We have revised the specific phrase you noted, and have conducted a thorough language edit throughout the manuscript to improve clarity and expression. For instance, the phrasing in question on Line 20 has been revised to "that have been underexplored in previous work." We appreciate your careful reading.

 

3. Units: snow depth is given in mm in Figure 7 but elsewhere seems to be cm → standardize.

[Authors’ response]: Thank you for your careful attention to the units. We have thoroughly checked the manuscript and figures and confirm that snow depth is consistently reported in millimeters (mm) throughout the document, including in Figure 7. The use of mm is standard for the hydrological analysis presented. We appreciate you prompting this verification.

 

4. Provide the actual number of grid cells (or effective sample size) used in correlation analysis.

[Authors’ response]: Thank you for the comment. In our correlation analysis, the actual number of independent data points (effective sample size) for each grid cell is the length of the time series (60 years, 1961-2020). Thus, the correlations at each of the 1,050 grid points were calculated based on N=60 annual values (e.g., yearly count or magnitude of events). The total of 680,070 events refers to the sum of all events across all grids and years, which was used for compositing and categorization, but not as N in the per-grid correlation calculations. We have clarified this in the revised manuscript (Lines 448-450).

 

5. The claim that snowmelt dates advanced “9, 6, and 2 days earlier” for 25%, 50%, 75% cumulative levels needs the baseline period clearly stated in the abstract.

[Authors’ response]: Thank you for your comment. The baseline period (1961-1990) has been clearly stated in the revised abstract, as seen in Line 23: “Relative to the baseline period (1961-1990), the snowmelt dates...” This provides the necessary context for the reported advancements of 9, 6, and 2 days.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Thank you for your valuable contribution. The manuscript addresses an important hydrological topic and clearly demonstrates significant shifts in snowmelt timing and intensity in the Songhua River Basin. The long observation period (1961–2020) and event-scale snowmelt analysis are major strengths. However, I recommend Major Revision before the manuscript is suitable for publication. Specific points are listed below:


Data and Methodological Transparency

• The study relies heavily on ERA5-Land reanalysis snowmelt data, but the processing workflow is not sufficiently described.

  • How is snowmelt computed in ERA5-Land? (energy-balance model? diagnostic subtraction?)

  • What specific variables were extracted and how were they converted to melt depth?

  • Any bias corrections or validation using ground measurements?

Currently, reproducibility is limited because the reader cannot replicate snowmelt time series or event identification based on the information given. A detailed description (or flowchart) of the snowmelt extraction method and software tools used is required.

Event Definition and Detection

While daily melt ≥0.1 mm is a common threshold, please:

• Provide justification with references

• Discuss sensitivity to detection threshold

This will improve robustness and reduce methodological uncertainty.

Validation and Uncertainty Analysis

The manuscript acknowledges ERA5-Land bias indirectly, but no validation against ground observations is provided. Snow depth gauges exist in the region — even limited comparison would greatly enhance confidence.

Also consider:

• Quantifying spatial/temporal uncertainty

• Discussing implications for event-scale metrics (especially early-season melt that is small in magnitude)

Strengthen Hydrological Implications

The Discussion suggests consequences for water management and flood risk, but these are not supported by quantitative analysis. Suggestions:

• Relate snowmelt timing changes to observed streamflow records

• Discuss agricultural water shortages with regional references

• Add citations supporting compound flood statements

The contribution is significant and original, but transparency in methods and uncertainty is essential for scientific reproducibility. Addressing the comments above will greatly strengthen the manuscript.

Author Response

Response to Reviewer #4 on hydrology-4010576 “Shifting Snowmelt Regime in a High-latitude Asian Basin: Insights from the Songhua River Basin”

 

Thank you for your valuable contribution. The manuscript addresses an important hydrological topic and clearly demonstrates significant shifts in snowmelt timing and intensity in the Songhua River Basin. The long observation period (1961–2020) and event-scale snowmelt analysis are major strengths. However, I recommend Major Revision before the manuscript is suitable for publication. Specific points are listed below:

[Authors’ response]: Thank you very much for taking the time to review our manuscript and for your constructive comments. We sincerely appreciate your positive assessment of the importance of our topic, the long-term data, and the event-scale analysis. We also agree that your suggestions will significantly strengthen the manuscript. We have carefully considered all your points and will revise the manuscript accordingly. A point-by-point response to your specific comments is provided below, detailing the changes we have made or intend to make.

 

Data and Methodological Transparency

The study relies heavily on ERA5-Land reanalysis snowmelt data, but the processing workflow is not sufficiently described.

How is snowmelt computed in ERA5-Land? (energy-balance model? diagnostic subtraction?)

What specific variables were extracted and how were they converted to melt depth?

Any bias corrections or validation using ground measurements?

Currently, reproducibility is limited because the reader cannot replicate snowmelt time series or event identification based on the information given. A detailed description (or flowchart) of the snowmelt extraction method and software tools used is required.

[Authors’ response]: Thank you for this important question regarding methodological transparency. You are correct that further details are required for reproducibility. In the ERA5-Land reanalysis system, snowmelt is computed diagnostically from the snow energy and mass balance within the ECMWF's HTESSEL land surface model. For this study, the specific variables extracted were the daily fields of `snowdepth` (m of water equivalent) and `snowmelt` (m of liquid water equivalent). The original values in meters were multiplied by 1000 to convert to millimeters (mm) for analysis, with no additional scaling or bias correction applied.

 

Event Definition and Detection

While daily melt ≥0.1 mm is a common threshold, please:

Provide justification with references

Discuss sensitivity to detection threshold

This will improve robustness and reduce methodological uncertainty.

[Authors’ response]: Thank you for your question regarding the selection of the 0.1 mm daily snowmelt threshold. In the revised manuscript (lines 137–144), we have added the following justification:

"For the definition of snowmelt events, a threshold of daily snowmelt ≥ 0.1 mm was applied, and consecutive periods meeting this condition with a minimum duration of one day were identified as independent events. The daily melting threshold of ≥ 0.1 millimeters was selected to sensitively capture the beginning of spring snowmelt and weak snowmelt signals, and to maintain the continuity of the snowmelt period for accurate hydrological analysis. This method is consistent with common practice in mid- to high-latitude snow research, where 0.1 mm of water column per day is widely used as a baseline sensitivity threshold for detecting melt occurrence while minimizing noise [24,25]." The references cited are:

[24] Gorodetskaya et al. (2023), who used a similar threshold for detecting surface melt in Antarctic Peninsula studies.

[25] Puggaard et al. (2025), who applied a 0.1 mm/day melt threshold in Greenland Ice Sheet melt analysis.

In direct response to your suggestion, we have added a dedicated sensitivity analysis in the revised manuscript (Section 4.4 "Uncertainty and Limitations," lines 609–625). This analysis systematically tests the impact of using different daily melt-rate thresholds (0.1, 0.5, 1.0, and 2.0 mm/day) on key event metrics, including frequency, duration, and cumulative volume. The results confirm that the chosen threshold directly influences these characteristics. However, the comparison demonstrates that our selected threshold of 0.1 mm/day is optimal for our study's objectives, which focus on capturing the initial onset and temporal continuity of melt processes crucial for phenological and hydrological analyses. This threshold minimizes the omission of early melt signals while maintaining robustness against potential data noise, thereby ensuring a reliable basis for our long-term analysis.

 

Validation and Uncertainty Analysis

The manuscript acknowledges ERA5-Land bias indirectly, but no validation against ground observations is provided. Snow depth gauges exist in the region — even limited comparison would greatly enhance confidence.

[Authors’ response]: Thank you for raising this critical point regarding the validation of ERA5-Land snow data. In response to your comment, we have added a dedicated subsection, “4.4 Uncertainty and Limitations,” in the revised manuscript (lines 590–609). A quantitative validation of ERA5-Land snow depth against the independent Long-term Series of Daily Snow Depth Dataset in China. Results indicate that while ERA5-Land shows a systematic high bias—which may lead to overestimated melt magnitudes—it exhibits strong consistency in spatiotemporal patterns (mean spatial correlation = 0.77) and effectively captures daily and annual trends (R² = 0.81 and 0.75, respectively).

 

Also consider:

Quantifying spatial/temporal uncertainty

Discussing implications for event-scale metrics (especially early-season melt that is small in magnitude)

Strengthen Hydrological Implications

The Discussion suggests consequences for water management and flood risk, but these are not supported by quantitative analysis. Suggestions:

Relate snowmelt timing changes to observed streamflow records

Discuss agricultural water shortages with regional references

Add citations supporting compound flood statements

[Authors’ response]: Thank you for your valuable suggestion. We all agree that strengthening the discussion section is crucial. In response, we have made significant revisions and extensions to this section in order to provide a more in-depth analysis of our results. Specifically, as described in the revised manuscript (lines 541-562 and 571-584 of the revised manuscript).

44. Wei P, Hao L, Fu Q, et al. (2024) Analysis of spring drought in Northeast China from the perspective of atmosphere, snow cover, and soil. CATENA, 236, 107715. https://doi.org/10.1016/j.catena.2023.107715.
45. Chen X, Li X, Jiang B, et al. (2023) Prediction of spring agricultural drought using machine learning algorithms in the southern Songnen Plain.
46. Wei P, Su Y, Fu Q, et al. (2025) Study on the driving factors of spring agricultural drought in Northeast China from the perspective of atmosphere and snow cover. Agric Water Manag, 317, 109620. https://doi.org/10.1016/j.agwat.2025.109620.
47. Kumar N, Kar KK, Srivastava S, et al. (2025) Trends and causal structures of rain-on-snow flooding. J Hydrol, 662(Part B), 133938. https://doi.org/10.1016/j.jhydrol.2025.133938.
48. Li Y, Sun F, Chen Y, et al. (2025) Unraveling the complexities of rain-on-snow events in High Mountain Asia. npj Clim Atmos Sci, 8, 118. https://doi.org/10.1038/s41612-025-00943-y.

 

The contribution is significant and original, but transparency in methods and uncertainty is essential for scientific reproducibility. Addressing the comments above will greatly strengthen the manuscript.

[Authors’ response]: Thank you for your thoughtful feedback on our manuscript. We sincerely appreciate your recognition of the significance and originality of our contribution, as well as your emphasis on the importance of transparency in methods and uncertainty analysis for scientific reproducibility.We fully agree with your comments and believe that addressing these points will indeed strengthen the manuscript.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

All the comments were addressed.

Author Response

Comment1: All the comments were addressed.

[Authors’ response]: Thank you for your positive feedback. We are pleased to hear that all comments have been addressed to your satisfaction. We appreciate the time and expertise you have contributed to the review of our manuscript.