the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 27 May 2025
Insights into spatiotemporal dynamics of riverine water age and its controlling factors in five contrasting subtropical headwater catchments, South-Central China using stable water isotopes
Abstract. Understanding river water age and its controlling factors are fundamental for comprehending catchment hydrological and biogeochemical processes. However, how do landscape characteristics and climate properties control the spatiotemporal heterogeneity of riverine water age remain to be further clarified in subtropical headwater catchments. This study used stable isotopic ratios (δD and δ18O) from five contrasting headwater catchments. The study explored time-variable young (Fyw) and new (Fnew) water fraction among five contrasting headwater catchments from the upper reaches of Xiu River, located within the Poyang Lake catchment of South-Central China using stable isotopes (δD and δ18O) from 2021 to 2023. The isotopic compositions of precipitation exhibited greater fluctuations than those of river water from five sub-catchments. The lower slopes (3.78 to 6.63) and intercepts (-13.12 to 2.65) of linear regression correlations between δD and δ18O were observed in river water compared to the global and local meteoric water lines, indicating significant evaporation effects on river water. The young water fraction (Fyw) showed considerable spatial variability ranging from 0.07 to 0.21 among five sub-catchments, suggesting the dominant recharge of groundwater to the river. The pronounced temporal variations of Fyw highlighted its susceptibility to short-term hydroclimatic change. Random forest models revealed that precipitation (25.48±5.41 %) and potential evapotranspiration (27.84±6.62 %) were the primary drivers for young and new water generation. Furthermore, Fyw was significantly influenced by upstream inflows (24.21±0.71 %), whereas Fnew was more susceptible to the influence of percentage of forest (22.63 %) and cropland (29.85 %). Shapley Additive Explanations reveal a significant negative correlation between river area and Fyw, and a significant positive correlation between agricultural area proportion and Fnew. Combined with the dynamic variations in the Fyw and Fnew, these results indicated that the regulatory function of riparian zones played a crucial role in young water generation, while land use changes significantly altered the process of new water generation. Our findings suggest that intensified evapotranspiration and increased precipitation will significantly impact the generation of riverine young and new water in the context of global warming and land use type changes.
- Preprint
(12831 KB) - Metadata XML
-
Supplement
(162 KB) - BibTeX
- EndNote
Status: closed (peer review stopped)
-
RC1: 'Comment on Kaiwen Li et al.', Francesc Gallart, 17 Jun 2025
The authors propose an analysis of both the young and new water fractions in a set of five middle-sized basins in the headwaters of the Xiu River in South-Central China, along with the relationships of these fractions with the physiographic and land use characteristics of the catchments.
The main strength of this work is the large number of river water samples analysed. The methods used for investigating the young and new water fractions are adequate although the likely dependence of young water fraction on river discharge could have been analysed using known methods.
The main and serious weakness of the study is the limitation of precipitation characteristics to a single measuring and sampling station for a set of middle-mountain basins covering and area of more than 1000 km2. This station is located near the centre of the studied catchments, but significant parts of these catchments are located farther than 40 km from it. Furthermore, the isotopic signature of precipitation events observed for this station is highly scattered, as the fitted sinusoid explains only 22% of the variance.
Using the data from this precipitation station for analysing the young water fraction of these catchments means assuming that the seasonal variation of the isotopic signature of precipitation water is similar around all the area in these catchments, a risky assumption taking into account the long distances and the relevant variability of altitudes.
Furthermore, using data from this precipitation station to analyse the new water fractions of these catchments means assuming a one-week synchrony of sampled precipitation and sampled river waters across the network. This assumption cannot be maintained, not only because the above objections but also because the hydrographs do not show the necessary synchronism at the event scale.
Citation: https://doi.org/10.5194/egusphere-2025-1739-RC1 -
CC1: 'Reply on RC1', Kaiwen Li, 09 Jul 2025
Dear Editor and reviewers,
We would like to thank editor and three reviewers for valuable and constructive comments, which are very helpful to the improvement of the manuscript. We have learned much from the reviewers’ comments, which are fair, encouraging and constructive. After carefully studying these comments and advice, we have made corresponding changes in revised model for the revised manuscript. The following paragraphs respond to the specific comments, the original review comments are listed first in their originals, followed by our responses in Blod.
If you have any question about this manuscript, please don’t hesitate to contact us.
Sincerely yours,
Kaiwen Li and Huawu Wu
Comments:
The authors propose an analysis of both the young and new water fractions in a set of five middle-sized basins in the headwaters of the Xiu River in South-Central China, along with the relationships of these fractions with the physiographic and land use characteristics of the catchments.
The main strength of this work is the large number of river water samples analysed. The methods used for investigating the young and new water fractions are adequate although the likely dependence of young water fraction on river discharge could have been analysed using known methods.
The main and serious weakness of the study is the limitation of precipitation characteristics to a single measuring and sampling station for a set of middle-mountain basins covering and area of more than 1000 km2. This station is located near the center of the studied catchments, but significant parts of these catchments are located farther than 40 km from it. Furthermore, the isotopic signature of precipitation events observed for this station is highly scattered, as the fitted sinusoid explains only 22% of the variance.
Response: Thanks for your constructive and valuable comments on the focus of this study. Firstly, this study represents the first attempt to conduct the high-solution measurements of stable isotope from precipitation and river water in this basin. We agree with your comments on the representativeness of precipitation samples in the basin with the area of more than 1000 km2. This sampling station is located the center of the studied catchments and two far basin of Zhajin and Panxi Basin were found significant long distance about 40 km from it. To avoid the potential effect of precipitation sampling representativeness and reduce the uncertainty of the estimation of the new water fraction (Fnew), we focus on three basins (Hangkou, Fengxiang, and Huangsha basins) scattering around the precipitation isotope observation. Hence, we exclude the far basins of Zhajin and Panxi Basin.
Secondly, we re-examined the influence of various driving factors on Fyw (young water fraction) and Fnew in three basins using a Random Forest (RF) model combined with SHAP (Shapley Additive Explanations). Compared to previous findings in the original manuscript, our results showed an increased and significantly negative influence of forest cover on Fyw, as well as a stronger and negatively correlated impact of drainage density. The RF model further highlights the negative correlation between Fyw and both drainage density and riparian area, reinforcing our earlier conclusion that riparian zones play a critical role in decreasing Fyw due to their strong water storage capacity. Additionally, we found that cropland area exerted a significant negative influence on Fnew, which differs from our earlier conclusion of a positive correlation. Further analysis revealed that the interaction contribution between cropland proportion and potential evapotranspiration (PET) is positively associated with Fnew. This may suggest that in regions with both high PET and extensive cropland, irrigation practices may enhance water input and rapidly replenish the system with new water. Consistent with our previous conclusions in the original manuscript, the agricultural conversion of riparian zones reduces their buffering capacity, leading to increased Fnew in the basin. This finding further underscores the role of evapotranspiration in regulating hydrological responses in the study area.
Furthermore, we applied a 1-year moving time window, shifted in 1-month increments, to fit the precipitation stable isotope time series using a sinusoidal function over a 2-year period. Our results showed that, although the magnitude of precipitation variability was relatively limited, the fitting performance improved markedly under both approaches. This enhancement suggests that temporal smoothing and weighting can better capture the underlying seasonal signals in the isotope data by reducing short-term fluctuations. Therefore, we consider our analytical framework and the resulting interpretations to be robust and scientifically justified.
Finally, a sinusoidal function was applied to the δD dataset to capture its annual cyclic variation. Although the resulting coefficient of determination (R² = 0.22) appears relatively low, such values are common in tropical-subtropical monsoon regions and likely reflect high-frequency disturbances and non-periodic variability in precipitation processes. This potential mechanism controlling on the precipitation isotopes would be explored deeply in future researches on precipitation isotope. One contributing factor is sub-cloud evaporation, whereby raindrops undergo partial evaporation during descent in the humid monsoon environment (Gou et al., 2022; Wu et al., 2023), diminishing the original isotopic signal and increasing data discretization, thus weakening the seasonal fit.
Using the data from this precipitation station for analysing the young water fraction of these catchments means assuming that the seasonal variation of the isotopic signature of precipitation water is similar around all the area in these catchments, a risky assumption taking into account the long distances and the relevant variability of altitudes.
Response: Thanks for your valuable comments on the representativeness of precipitation data to analyze the young water fraction. We selected three basins around the precipitation isotope observation, located in the central location (Figure 1), which well represent the isotopic variations of precipitation in these small basins (Hangkou, Fengxiang, and Huangsha basin).
Furthermore, using data from this precipitation station to analyse the new water fractions of these catchments means assuming a one-week synchrony of sampled precipitation and sampled river waters across the network. This assumption cannot be maintained, not only because the above objections but also because the hydrographs do not show the necessary synchronism at the event scale.
Response: We sincerely thank the reviewer for the insightful comments regarding the representativeness of precipitation samples in the entire catchment. As noted above, to minimize the influence of spatial variability in precipitation, we restricted our analysis to three catchments located near the precipitation sampling site. This spatial alignment helps ensure that the precipitation input and streamflow response are sufficiently synchronized, thereby satisfying the necessary synchronism at the event scale.
References:
Wu H., et al. 2023. Atmospheric processes control the stable isotopic variability of precipitation in the middle–lower reaches of the Yangtze River Basin, East Asian monsoon region. Journal of Hydrology, 623: 129835.
Gou J., et al., 2022. Relationship between precipitation isotopic compositions and synoptic atmospheric circulation patterns in the lower reach of the Yangtze River. Journal of Hydrology, 605: 127289.
-
CC1: 'Reply on RC1', Kaiwen Li, 09 Jul 2025
-
RC2: 'Comment on egusphere-2025-1739', Anonymous Referee #2, 28 Jul 2025
Li et al. report an assessment of an interesting 2-year water isotope collection across five headwater catchments in South-Central China. While the data collection is impressive and the data could be used to reveal interesting new findings I found several flaws in the analysis that in my opinion should disqualify the manuscript from publication in HESS:
- Statistical analyses
The authors use a random forest approach on a dataset where n=5 which provides an output but I’d like to highlight that this is not something that should be done. Random Forest Models with such small amounts of input data are likely to be overfitting, without the respective quality metrics the analysis it is hard to assess the overall model robustness. Similarly, the use of SHAP values is not meaningful on such a limited dataset. Both of these tools were developed and should only be used in large datasets.
The use of time-variant young water fractions is not necessarily reliable across a two-year data collection and violates some of the key assumptions outlined in Kirchner 2016. When splitting the data in subsets new water fraction calculations should be used instead (and exclusively) as it has some important advantages over Fyw.
- Literature Review
The manuscript misses the more recent significant amount of literature on young and new water fractions. Some citations used within the manuscript are missing in the reference list (e.g., Kirchner 2016). Before resubmission, a decent literature review has to be done to capture previous findings on young and new water fractions across other regions.
- Serious doubts on the meaningful use of LLMs
On a first glimpse the manuscript is well written, however a lot of the text is scientifically meaningless. I pick here two examples, but this is a widespread issue of the manuscript: «The isotopic data of river water in the study area were scattered along both sides of the LMWL (Figure 3), indicating that river water may be of meteoric origin (Bugna et al., 2020).» or «The variability in isotopic composition of precipitation was larger than that of river water, illustrating that the damping of the precipitation signatures due to mixing and dispersion within the subsurface.» I have some serious doubts on the origin of such statements, and I suspect that these are the artefacts of a misuse of LLMs. A very obvious example can be found in the author’s reply to reviewer 1 that starts with “We would like to thank editor and three reviewers for valuable and constructive comments, which are very helpful to the improvement of the manuscript.»
Citation: https://doi.org/10.5194/egusphere-2025-1739-RC2 -
AC1: 'Reply on RC2', Huawu Wu, 22 Aug 2025
Comments:
Li et al. report an assessment of an interesting 2-year water isotope collection across five headwater catchments in South-Central China. While the data collection is impressive and the data could be used to reveal interesting new findings I found several flaws in the analysis that in my opinion should disqualify the manuscript from publication in HESS:
Statistical analyses
The authors use a random forest approach on a dataset where n=5 which provides an output but I’d like to highlight that this is not something that should be done. Random Forest Models with such small amounts of input data are likely to be overfitting, without the respective quality metrics the analysis it is hard to assess the overall model robustness. Similarly, the use of SHAP values is not meaningful on such a limited dataset. Both of these tools were developed and should only be used in large datasets.
Response: We sincerely appreciate the Reviewer’s concerns regarding the potential limitations of applying random forest and SHAP analysis on very small datasets (e.g., n = 5). We fully acknowledge that, under such conditions, there is indeed a high risk of model overfitting and a lack of robustness, and we agree that such approaches would not be appropriate in practice. However, we would like to clarify that in our study the dataset used for the random forest model and SHAP analysis was substantially larger, which referred to large time-variable amounts of weekly events for two years from five field observations. Specifically, we employed a one-year moving window with a monthly step to calculate time-variable young water fractions (Fyw_60), which generated n = 60 independent data points. The motivation for using this time-variable approach is that the isotopic composition of streamflow can vary substantially from year to year due to the increasing frequency of extreme hydro-climatic events, changes in land use, and reservoir construction. In particular, during our study period both El Niño and La Niña events occurred, further amplifying interannual variability. Under such circumstances, simply fitting a sinusoidal curve to represent a multi-year average isotope cycle would not be reasonable. Instead, the time-variable calculation allowed us to explicitly capture these interannual dynamics. This methodological choice was explicitly described in the original manuscript (line 181: “Additionally, F*yw_60 will represent the 60 individual time-variable young water fraction results using a 1-year calculation window that was shifted in 1-month intervals.”). These 60 data points formed the basis for both the random forest analysis and SHAP interpretation, as also presented in Figures 6 and 7.
Furthermore, following the suggestion of Reviewer #1 to reduce the number of catchments under consideration, the total number of data points used in the revised analysis became n = 36 (12 values per catchment for three catchments). While the revised dataset (n = 36) is admittedly smaller than the original n = 60, it still remains within a range where random forest models can provide meaningful insights. Previous methodological work has shown that random forest can be applied effectively even in relatively small datasets, provided that appropriate validation procedures are implemented1. Indeed, in hydrology and related environmental sciences, random forest has been successfully applied with sample sizes of similar order2. To ensure robustness, we carefully re-examined our analysis using repeated training and cross-validation, and confirmed that the model outputs are interpretable under this sample size. Regarding SHAP analysis, while we recognize that uncertainty increases with fewer samples, SHAP remains a valuable tool for identifying relative feature importance and directionality of effects, rather than precise quantitative contributions, even with moderate datasets3.
The use of time-variant young water fractions is not necessarily reliable across a two-year data collection and violates some of the key assumptions outlined in Kirchner 2016. When splitting the data in subsets new water fraction calculations should be used instead (and exclusively) as it has some important advantages over Fyw.
Response: We sincerely appreciate the Reviewer’s concern regarding the reliability of time-variant young water fraction (Fyw) estimates and the potential violation of assumptions outlined in Kirchner (2016). We fully agree that the conventional calculation of Fyw relies on fitting sinusoidal functions to the isotopic cycles of precipitation and streamflow, and that the method assumes a relatively stationary system over the analyzed period. Under very short records or strongly non-stationary conditions, applying this method indiscriminately could indeed lead to misleading results.
However, in our study the moving-window approach was carefully designed to remain consistent with the principles of the original method. Specifically, each Fyw estimate was derived from one-year segments of isotope data in both precipitation and streamflow, allowing the fitted sine waves to capture the seasonal cycle within each window. This design ensured that the analysis was performed over periods long enough to resolve the annual isotopic signal, while also accommodating potential interannual variability. In this sense, the method does not violate the underlying assumptions of sinusoidal fitting, but rather applies them in a temporally localized way. Importantly, such an approach has precedent in the literature: for instance, Stockinger et al. (2019) applied a 1-year moving window to a 4.5-year isotope time series and demonstrated that Fyw values can vary over time in response to both extreme events (e.g., the 2015 European heatwave) and seasonal dynamics4.
Furthermore, we acknowledge that new water fraction (Fnew) provides complementary insights and certain advantages over Fyw, particularly when data are subdivided into shorter periods, as the Reviewer correctly points out. In recognition of this, our study also explicitly calculated Fnew using the same moving-window approach, and compared its behavior with Fyw. We emphasize in the revised manuscript that Fnew is a valuable diagnostic, and that interpreting both Fnew and Fyw together provides a more comprehensive understanding of streamflow isotope dynamics. Thus, while our use of time-variant Fyw is methodologically consistent and supported by prior studies, we agree that Fnew is indispensable in such analyses, and we have revised the text to better highlight this point.
Literature Review
The manuscript misses the more recent significant amount of literature on young and new water fractions. Some citations used within the manuscript are missing in the reference list (e.g., Kirchner 2016). Before resubmission, a decent literature review has to be done to capture previous findings on young and new water fractions across other regions.
Response: We thank the reviewer for pointing this out. We have carefully reviewed the recent literature on young and new water fractions and have updated these literatures in this revised manuscript. Relevant studies from various regions have been included to better contextualize our work and highlight similarities and differences with previous findings, which help to explain the mechanisms on the young water fractions. Additionally, some missing references, such as Kirchner (2016), have been carefully rechecked and added to the reference list and correctly cited in the text. We believe that these revisions strengthen the background and discussion of our study.
Serious doubts on the meaningful use of LLMs
On a first glimpse the manuscript is well written, however a lot of the text is scientifically meaningless. I pick here two examples, but this is a widespread issue of the manuscript: «The isotopic data of river water in the study area were scattered along both sides of the LMWL (Figure 3), indicating that river water may be of meteoric origin (Bugna et al., 2020).» or «The variability in isotopic composition of precipitation was larger than that of river water, illustrating that the damping of the precipitation signatures due to mixing and dispersion within the subsurface.» I have some serious doubts on the origin of such statements, and I suspect that these are the artefacts of a misuse of LLMs. A very obvious example can be found in the author’s reply to reviewer 1 that starts with “We would like to thank editor and three reviewers for valuable and constructive comments, which are very helpful to the improvement of the manuscript.
Response: We sincerely thank the Reviewer for the careful reading of our manuscript and for raising these serious concerns. We take this comment very seriously, and we would like to respond in detail as follows.
Firstly, regarding the suspicion of artefacts from a misuse of large language models (LLMs), we would like to clarify that none of the scientific content in our manuscript was generated by LLMs. All statements and interpretations are based on our own analysis of the original isotope data, combined with insights from peer-reviewed literatures, which is consistent to the previous studies in Poyang Lake regions8.
Secondly, regarding the examples cited by the Reviewer, we acknowledge that some of our original wordings may have lacked sufficient precision, which could have given the impression of being scientifically vague. We have now thoroughly revised these sections to ensure that the interpretations are explicitly linked to hydrological processes and supported by appropriate references throughout the whole manuscript. For instance, the discussion of the Local Meteoric Water Line (LMWL) is a standard and essential component of isotope hydrology, as it helps identify meteoric origin and potential evaporative effects. Our original phrasing followed the style of earlier isotope studies5–8, but we recognize that it was not clearly contextualized in our manuscript. In the revision, we explicitly explain why river water plots close to or deviates from the LMWL, and how the lower variability in stream water isotopes compared to precipitation reflects mixing and storage processes in the catchment that dampen precipitation signals. These revisions ensure that the discussion is scientifically accurate, clearly justified, and directly relevant to hydrological interpretation.
Lastly, we sincerely acknowledge that the opening sentence in our previous response to Reviewer #1 (“We would like to thank editor and three reviewers …”) followed a general template and may have created the impression of artificial or generic text. This was an oversight on our part, and we apologize for any confusion it may have caused. We have carefully rewritten the responses in the current revision to ensure that all replies are specific, customized, and directly address each Reviewer’s point in detail.
In summary, we have re-evaluated the entire manuscript to remove any ambiguous statements, revised the discussion to be more scientifically rigorous and hydrologically grounded, and ensured that both the manuscript and our responses are entirely original and transparent.
Reference
1 Genuer, R.; Poggi, J.-M.; Tuleau-Malot, C. Variable Selection Using Random Forests. Pattern Recognit. Lett. 2010, 31 (14), 2225–2236. https://doi.org/10.1016/j.patrec.2010.03.014.
2 Lutz, S. R.; Krieg, R.; Müller, C.; Zink, M.; Knöller, K.; Samaniego, L.; Merz, R. Spatial Patterns of Water Age: Using Young Water Fractions to Improve the Characterization of Transit Times in Contrasting Catchments. Water Resour. Res. 2018, 54 (7), 4767–4784. https://doi.org/10.1029/2017WR022216.
3 Lundberg, S. M.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions.
4 Stockinger, M. P.; Bogena, H. R.; Lücke, A.; Stumpp, C.; Vereecken, H. Time Variability and Uncertainty in the Fraction of Young Water in a Small Headwater Catchment. Hydrol. Earth Syst. Sci. 2019, 23 (10), 4333–4347. https://doi.org/10.5194/hess-23-4333-2019.
5 Vystavna, Y.; Harjung, A.; Monteiro, L. R.; Matiatos, I.; Wassenaar, L. I. Stable Isotopes in Global Lakes Integrate Catchment and Climatic Controls on Evaporation. Nat. Commun. 2021, 12 (1), 7224. https://doi.org/10.1038/s41467-021-27569-x.
6 Gou, J.; Qu, S.; Guan, H.; Shi, P.; Su, Z.; Lin, Z.; Liu, J.; Zhu, J. Relationship between Precipitation Isotopic Compositions and Synoptic Atmospheric Circulation Patterns in the Lower Reach of the Yangtze River. J. Hydrol. 2022, 605, 127289. https://doi.org/10.1016/j.jhydrol.2021.127289.
7 Wang, L.; Liu, W.; Xu, Z.; Zhang, J. Water Sources and Recharge Mechanisms of the Yarlung Zangbo River in the Tibetan Plateau: Constraints from Hydrogen and Oxygen Stable Isotopes. J. Hydrol. 2022, 614, 128585. https://doi.org/10.1016/j.jhydrol.2022.128585.
8 Zhan, L.; Chen, J.; Zhang, S.; Li, L.; Huang, D.; Wang, T. Isotopic Signatures of Precipitation, Surface Water, and Groundwater Interactions, Poyang Lake Basin, China. Environ. Earth Sci. 2016, 75 (19), 1307. https://doi.org/10.1007/s12665-016-6081-8.
Citation: https://doi.org/10.5194/egusphere-2025-1739-AC1
Status: closed (peer review stopped)
-
RC1: 'Comment on Kaiwen Li et al.', Francesc Gallart, 17 Jun 2025
The authors propose an analysis of both the young and new water fractions in a set of five middle-sized basins in the headwaters of the Xiu River in South-Central China, along with the relationships of these fractions with the physiographic and land use characteristics of the catchments.
The main strength of this work is the large number of river water samples analysed. The methods used for investigating the young and new water fractions are adequate although the likely dependence of young water fraction on river discharge could have been analysed using known methods.
The main and serious weakness of the study is the limitation of precipitation characteristics to a single measuring and sampling station for a set of middle-mountain basins covering and area of more than 1000 km2. This station is located near the centre of the studied catchments, but significant parts of these catchments are located farther than 40 km from it. Furthermore, the isotopic signature of precipitation events observed for this station is highly scattered, as the fitted sinusoid explains only 22% of the variance.
Using the data from this precipitation station for analysing the young water fraction of these catchments means assuming that the seasonal variation of the isotopic signature of precipitation water is similar around all the area in these catchments, a risky assumption taking into account the long distances and the relevant variability of altitudes.
Furthermore, using data from this precipitation station to analyse the new water fractions of these catchments means assuming a one-week synchrony of sampled precipitation and sampled river waters across the network. This assumption cannot be maintained, not only because the above objections but also because the hydrographs do not show the necessary synchronism at the event scale.
Citation: https://doi.org/10.5194/egusphere-2025-1739-RC1 -
CC1: 'Reply on RC1', Kaiwen Li, 09 Jul 2025
Dear Editor and reviewers,
We would like to thank editor and three reviewers for valuable and constructive comments, which are very helpful to the improvement of the manuscript. We have learned much from the reviewers’ comments, which are fair, encouraging and constructive. After carefully studying these comments and advice, we have made corresponding changes in revised model for the revised manuscript. The following paragraphs respond to the specific comments, the original review comments are listed first in their originals, followed by our responses in Blod.
If you have any question about this manuscript, please don’t hesitate to contact us.
Sincerely yours,
Kaiwen Li and Huawu Wu
Comments:
The authors propose an analysis of both the young and new water fractions in a set of five middle-sized basins in the headwaters of the Xiu River in South-Central China, along with the relationships of these fractions with the physiographic and land use characteristics of the catchments.
The main strength of this work is the large number of river water samples analysed. The methods used for investigating the young and new water fractions are adequate although the likely dependence of young water fraction on river discharge could have been analysed using known methods.
The main and serious weakness of the study is the limitation of precipitation characteristics to a single measuring and sampling station for a set of middle-mountain basins covering and area of more than 1000 km2. This station is located near the center of the studied catchments, but significant parts of these catchments are located farther than 40 km from it. Furthermore, the isotopic signature of precipitation events observed for this station is highly scattered, as the fitted sinusoid explains only 22% of the variance.
Response: Thanks for your constructive and valuable comments on the focus of this study. Firstly, this study represents the first attempt to conduct the high-solution measurements of stable isotope from precipitation and river water in this basin. We agree with your comments on the representativeness of precipitation samples in the basin with the area of more than 1000 km2. This sampling station is located the center of the studied catchments and two far basin of Zhajin and Panxi Basin were found significant long distance about 40 km from it. To avoid the potential effect of precipitation sampling representativeness and reduce the uncertainty of the estimation of the new water fraction (Fnew), we focus on three basins (Hangkou, Fengxiang, and Huangsha basins) scattering around the precipitation isotope observation. Hence, we exclude the far basins of Zhajin and Panxi Basin.
Secondly, we re-examined the influence of various driving factors on Fyw (young water fraction) and Fnew in three basins using a Random Forest (RF) model combined with SHAP (Shapley Additive Explanations). Compared to previous findings in the original manuscript, our results showed an increased and significantly negative influence of forest cover on Fyw, as well as a stronger and negatively correlated impact of drainage density. The RF model further highlights the negative correlation between Fyw and both drainage density and riparian area, reinforcing our earlier conclusion that riparian zones play a critical role in decreasing Fyw due to their strong water storage capacity. Additionally, we found that cropland area exerted a significant negative influence on Fnew, which differs from our earlier conclusion of a positive correlation. Further analysis revealed that the interaction contribution between cropland proportion and potential evapotranspiration (PET) is positively associated with Fnew. This may suggest that in regions with both high PET and extensive cropland, irrigation practices may enhance water input and rapidly replenish the system with new water. Consistent with our previous conclusions in the original manuscript, the agricultural conversion of riparian zones reduces their buffering capacity, leading to increased Fnew in the basin. This finding further underscores the role of evapotranspiration in regulating hydrological responses in the study area.
Furthermore, we applied a 1-year moving time window, shifted in 1-month increments, to fit the precipitation stable isotope time series using a sinusoidal function over a 2-year period. Our results showed that, although the magnitude of precipitation variability was relatively limited, the fitting performance improved markedly under both approaches. This enhancement suggests that temporal smoothing and weighting can better capture the underlying seasonal signals in the isotope data by reducing short-term fluctuations. Therefore, we consider our analytical framework and the resulting interpretations to be robust and scientifically justified.
Finally, a sinusoidal function was applied to the δD dataset to capture its annual cyclic variation. Although the resulting coefficient of determination (R² = 0.22) appears relatively low, such values are common in tropical-subtropical monsoon regions and likely reflect high-frequency disturbances and non-periodic variability in precipitation processes. This potential mechanism controlling on the precipitation isotopes would be explored deeply in future researches on precipitation isotope. One contributing factor is sub-cloud evaporation, whereby raindrops undergo partial evaporation during descent in the humid monsoon environment (Gou et al., 2022; Wu et al., 2023), diminishing the original isotopic signal and increasing data discretization, thus weakening the seasonal fit.
Using the data from this precipitation station for analysing the young water fraction of these catchments means assuming that the seasonal variation of the isotopic signature of precipitation water is similar around all the area in these catchments, a risky assumption taking into account the long distances and the relevant variability of altitudes.
Response: Thanks for your valuable comments on the representativeness of precipitation data to analyze the young water fraction. We selected three basins around the precipitation isotope observation, located in the central location (Figure 1), which well represent the isotopic variations of precipitation in these small basins (Hangkou, Fengxiang, and Huangsha basin).
Furthermore, using data from this precipitation station to analyse the new water fractions of these catchments means assuming a one-week synchrony of sampled precipitation and sampled river waters across the network. This assumption cannot be maintained, not only because the above objections but also because the hydrographs do not show the necessary synchronism at the event scale.
Response: We sincerely thank the reviewer for the insightful comments regarding the representativeness of precipitation samples in the entire catchment. As noted above, to minimize the influence of spatial variability in precipitation, we restricted our analysis to three catchments located near the precipitation sampling site. This spatial alignment helps ensure that the precipitation input and streamflow response are sufficiently synchronized, thereby satisfying the necessary synchronism at the event scale.
References:
Wu H., et al. 2023. Atmospheric processes control the stable isotopic variability of precipitation in the middle–lower reaches of the Yangtze River Basin, East Asian monsoon region. Journal of Hydrology, 623: 129835.
Gou J., et al., 2022. Relationship between precipitation isotopic compositions and synoptic atmospheric circulation patterns in the lower reach of the Yangtze River. Journal of Hydrology, 605: 127289.
-
CC1: 'Reply on RC1', Kaiwen Li, 09 Jul 2025
-
RC2: 'Comment on egusphere-2025-1739', Anonymous Referee #2, 28 Jul 2025
Li et al. report an assessment of an interesting 2-year water isotope collection across five headwater catchments in South-Central China. While the data collection is impressive and the data could be used to reveal interesting new findings I found several flaws in the analysis that in my opinion should disqualify the manuscript from publication in HESS:
- Statistical analyses
The authors use a random forest approach on a dataset where n=5 which provides an output but I’d like to highlight that this is not something that should be done. Random Forest Models with such small amounts of input data are likely to be overfitting, without the respective quality metrics the analysis it is hard to assess the overall model robustness. Similarly, the use of SHAP values is not meaningful on such a limited dataset. Both of these tools were developed and should only be used in large datasets.
The use of time-variant young water fractions is not necessarily reliable across a two-year data collection and violates some of the key assumptions outlined in Kirchner 2016. When splitting the data in subsets new water fraction calculations should be used instead (and exclusively) as it has some important advantages over Fyw.
- Literature Review
The manuscript misses the more recent significant amount of literature on young and new water fractions. Some citations used within the manuscript are missing in the reference list (e.g., Kirchner 2016). Before resubmission, a decent literature review has to be done to capture previous findings on young and new water fractions across other regions.
- Serious doubts on the meaningful use of LLMs
On a first glimpse the manuscript is well written, however a lot of the text is scientifically meaningless. I pick here two examples, but this is a widespread issue of the manuscript: «The isotopic data of river water in the study area were scattered along both sides of the LMWL (Figure 3), indicating that river water may be of meteoric origin (Bugna et al., 2020).» or «The variability in isotopic composition of precipitation was larger than that of river water, illustrating that the damping of the precipitation signatures due to mixing and dispersion within the subsurface.» I have some serious doubts on the origin of such statements, and I suspect that these are the artefacts of a misuse of LLMs. A very obvious example can be found in the author’s reply to reviewer 1 that starts with “We would like to thank editor and three reviewers for valuable and constructive comments, which are very helpful to the improvement of the manuscript.»
Citation: https://doi.org/10.5194/egusphere-2025-1739-RC2 -
AC1: 'Reply on RC2', Huawu Wu, 22 Aug 2025
Comments:
Li et al. report an assessment of an interesting 2-year water isotope collection across five headwater catchments in South-Central China. While the data collection is impressive and the data could be used to reveal interesting new findings I found several flaws in the analysis that in my opinion should disqualify the manuscript from publication in HESS:
Statistical analyses
The authors use a random forest approach on a dataset where n=5 which provides an output but I’d like to highlight that this is not something that should be done. Random Forest Models with such small amounts of input data are likely to be overfitting, without the respective quality metrics the analysis it is hard to assess the overall model robustness. Similarly, the use of SHAP values is not meaningful on such a limited dataset. Both of these tools were developed and should only be used in large datasets.
Response: We sincerely appreciate the Reviewer’s concerns regarding the potential limitations of applying random forest and SHAP analysis on very small datasets (e.g., n = 5). We fully acknowledge that, under such conditions, there is indeed a high risk of model overfitting and a lack of robustness, and we agree that such approaches would not be appropriate in practice. However, we would like to clarify that in our study the dataset used for the random forest model and SHAP analysis was substantially larger, which referred to large time-variable amounts of weekly events for two years from five field observations. Specifically, we employed a one-year moving window with a monthly step to calculate time-variable young water fractions (Fyw_60), which generated n = 60 independent data points. The motivation for using this time-variable approach is that the isotopic composition of streamflow can vary substantially from year to year due to the increasing frequency of extreme hydro-climatic events, changes in land use, and reservoir construction. In particular, during our study period both El Niño and La Niña events occurred, further amplifying interannual variability. Under such circumstances, simply fitting a sinusoidal curve to represent a multi-year average isotope cycle would not be reasonable. Instead, the time-variable calculation allowed us to explicitly capture these interannual dynamics. This methodological choice was explicitly described in the original manuscript (line 181: “Additionally, F*yw_60 will represent the 60 individual time-variable young water fraction results using a 1-year calculation window that was shifted in 1-month intervals.”). These 60 data points formed the basis for both the random forest analysis and SHAP interpretation, as also presented in Figures 6 and 7.
Furthermore, following the suggestion of Reviewer #1 to reduce the number of catchments under consideration, the total number of data points used in the revised analysis became n = 36 (12 values per catchment for three catchments). While the revised dataset (n = 36) is admittedly smaller than the original n = 60, it still remains within a range where random forest models can provide meaningful insights. Previous methodological work has shown that random forest can be applied effectively even in relatively small datasets, provided that appropriate validation procedures are implemented1. Indeed, in hydrology and related environmental sciences, random forest has been successfully applied with sample sizes of similar order2. To ensure robustness, we carefully re-examined our analysis using repeated training and cross-validation, and confirmed that the model outputs are interpretable under this sample size. Regarding SHAP analysis, while we recognize that uncertainty increases with fewer samples, SHAP remains a valuable tool for identifying relative feature importance and directionality of effects, rather than precise quantitative contributions, even with moderate datasets3.
The use of time-variant young water fractions is not necessarily reliable across a two-year data collection and violates some of the key assumptions outlined in Kirchner 2016. When splitting the data in subsets new water fraction calculations should be used instead (and exclusively) as it has some important advantages over Fyw.
Response: We sincerely appreciate the Reviewer’s concern regarding the reliability of time-variant young water fraction (Fyw) estimates and the potential violation of assumptions outlined in Kirchner (2016). We fully agree that the conventional calculation of Fyw relies on fitting sinusoidal functions to the isotopic cycles of precipitation and streamflow, and that the method assumes a relatively stationary system over the analyzed period. Under very short records or strongly non-stationary conditions, applying this method indiscriminately could indeed lead to misleading results.
However, in our study the moving-window approach was carefully designed to remain consistent with the principles of the original method. Specifically, each Fyw estimate was derived from one-year segments of isotope data in both precipitation and streamflow, allowing the fitted sine waves to capture the seasonal cycle within each window. This design ensured that the analysis was performed over periods long enough to resolve the annual isotopic signal, while also accommodating potential interannual variability. In this sense, the method does not violate the underlying assumptions of sinusoidal fitting, but rather applies them in a temporally localized way. Importantly, such an approach has precedent in the literature: for instance, Stockinger et al. (2019) applied a 1-year moving window to a 4.5-year isotope time series and demonstrated that Fyw values can vary over time in response to both extreme events (e.g., the 2015 European heatwave) and seasonal dynamics4.
Furthermore, we acknowledge that new water fraction (Fnew) provides complementary insights and certain advantages over Fyw, particularly when data are subdivided into shorter periods, as the Reviewer correctly points out. In recognition of this, our study also explicitly calculated Fnew using the same moving-window approach, and compared its behavior with Fyw. We emphasize in the revised manuscript that Fnew is a valuable diagnostic, and that interpreting both Fnew and Fyw together provides a more comprehensive understanding of streamflow isotope dynamics. Thus, while our use of time-variant Fyw is methodologically consistent and supported by prior studies, we agree that Fnew is indispensable in such analyses, and we have revised the text to better highlight this point.
Literature Review
The manuscript misses the more recent significant amount of literature on young and new water fractions. Some citations used within the manuscript are missing in the reference list (e.g., Kirchner 2016). Before resubmission, a decent literature review has to be done to capture previous findings on young and new water fractions across other regions.
Response: We thank the reviewer for pointing this out. We have carefully reviewed the recent literature on young and new water fractions and have updated these literatures in this revised manuscript. Relevant studies from various regions have been included to better contextualize our work and highlight similarities and differences with previous findings, which help to explain the mechanisms on the young water fractions. Additionally, some missing references, such as Kirchner (2016), have been carefully rechecked and added to the reference list and correctly cited in the text. We believe that these revisions strengthen the background and discussion of our study.
Serious doubts on the meaningful use of LLMs
On a first glimpse the manuscript is well written, however a lot of the text is scientifically meaningless. I pick here two examples, but this is a widespread issue of the manuscript: «The isotopic data of river water in the study area were scattered along both sides of the LMWL (Figure 3), indicating that river water may be of meteoric origin (Bugna et al., 2020).» or «The variability in isotopic composition of precipitation was larger than that of river water, illustrating that the damping of the precipitation signatures due to mixing and dispersion within the subsurface.» I have some serious doubts on the origin of such statements, and I suspect that these are the artefacts of a misuse of LLMs. A very obvious example can be found in the author’s reply to reviewer 1 that starts with “We would like to thank editor and three reviewers for valuable and constructive comments, which are very helpful to the improvement of the manuscript.
Response: We sincerely thank the Reviewer for the careful reading of our manuscript and for raising these serious concerns. We take this comment very seriously, and we would like to respond in detail as follows.
Firstly, regarding the suspicion of artefacts from a misuse of large language models (LLMs), we would like to clarify that none of the scientific content in our manuscript was generated by LLMs. All statements and interpretations are based on our own analysis of the original isotope data, combined with insights from peer-reviewed literatures, which is consistent to the previous studies in Poyang Lake regions8.
Secondly, regarding the examples cited by the Reviewer, we acknowledge that some of our original wordings may have lacked sufficient precision, which could have given the impression of being scientifically vague. We have now thoroughly revised these sections to ensure that the interpretations are explicitly linked to hydrological processes and supported by appropriate references throughout the whole manuscript. For instance, the discussion of the Local Meteoric Water Line (LMWL) is a standard and essential component of isotope hydrology, as it helps identify meteoric origin and potential evaporative effects. Our original phrasing followed the style of earlier isotope studies5–8, but we recognize that it was not clearly contextualized in our manuscript. In the revision, we explicitly explain why river water plots close to or deviates from the LMWL, and how the lower variability in stream water isotopes compared to precipitation reflects mixing and storage processes in the catchment that dampen precipitation signals. These revisions ensure that the discussion is scientifically accurate, clearly justified, and directly relevant to hydrological interpretation.
Lastly, we sincerely acknowledge that the opening sentence in our previous response to Reviewer #1 (“We would like to thank editor and three reviewers …”) followed a general template and may have created the impression of artificial or generic text. This was an oversight on our part, and we apologize for any confusion it may have caused. We have carefully rewritten the responses in the current revision to ensure that all replies are specific, customized, and directly address each Reviewer’s point in detail.
In summary, we have re-evaluated the entire manuscript to remove any ambiguous statements, revised the discussion to be more scientifically rigorous and hydrologically grounded, and ensured that both the manuscript and our responses are entirely original and transparent.
Reference
1 Genuer, R.; Poggi, J.-M.; Tuleau-Malot, C. Variable Selection Using Random Forests. Pattern Recognit. Lett. 2010, 31 (14), 2225–2236. https://doi.org/10.1016/j.patrec.2010.03.014.
2 Lutz, S. R.; Krieg, R.; Müller, C.; Zink, M.; Knöller, K.; Samaniego, L.; Merz, R. Spatial Patterns of Water Age: Using Young Water Fractions to Improve the Characterization of Transit Times in Contrasting Catchments. Water Resour. Res. 2018, 54 (7), 4767–4784. https://doi.org/10.1029/2017WR022216.
3 Lundberg, S. M.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions.
4 Stockinger, M. P.; Bogena, H. R.; Lücke, A.; Stumpp, C.; Vereecken, H. Time Variability and Uncertainty in the Fraction of Young Water in a Small Headwater Catchment. Hydrol. Earth Syst. Sci. 2019, 23 (10), 4333–4347. https://doi.org/10.5194/hess-23-4333-2019.
5 Vystavna, Y.; Harjung, A.; Monteiro, L. R.; Matiatos, I.; Wassenaar, L. I. Stable Isotopes in Global Lakes Integrate Catchment and Climatic Controls on Evaporation. Nat. Commun. 2021, 12 (1), 7224. https://doi.org/10.1038/s41467-021-27569-x.
6 Gou, J.; Qu, S.; Guan, H.; Shi, P.; Su, Z.; Lin, Z.; Liu, J.; Zhu, J. Relationship between Precipitation Isotopic Compositions and Synoptic Atmospheric Circulation Patterns in the Lower Reach of the Yangtze River. J. Hydrol. 2022, 605, 127289. https://doi.org/10.1016/j.jhydrol.2021.127289.
7 Wang, L.; Liu, W.; Xu, Z.; Zhang, J. Water Sources and Recharge Mechanisms of the Yarlung Zangbo River in the Tibetan Plateau: Constraints from Hydrogen and Oxygen Stable Isotopes. J. Hydrol. 2022, 614, 128585. https://doi.org/10.1016/j.jhydrol.2022.128585.
8 Zhan, L.; Chen, J.; Zhang, S.; Li, L.; Huang, D.; Wang, T. Isotopic Signatures of Precipitation, Surface Water, and Groundwater Interactions, Poyang Lake Basin, China. Environ. Earth Sci. 2016, 75 (19), 1307. https://doi.org/10.1007/s12665-016-6081-8.
Citation: https://doi.org/10.5194/egusphere-2025-1739-AC1
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 287 | 51 | 19 | 357 | 30 | 8 | 21 |
- HTML: 287
- PDF: 51
- XML: 19
- Total: 357
- Supplement: 30
- BibTeX: 8
- EndNote: 21
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1