the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Stable Stream Temperature Prediction for Different Basins Using Time Series Encoding and Temporal Convolutional Networks
Abstract. Flow temperature prediction is essential for assessing the health of river ecosystems. Water temperature data sets are often provided inconsistently in tasks that predict river water temperatures in different river basins, especially in different climatic regions. At the same time, spatial heterogeneity within different river basins significantly complicates water temperature prediction, which makes it challenging to establish a water temperature prediction model with strong generalization capabilities and stable prediction results. To solve this problem, the moving average encoding and DOY encoding of time series data into the time convolutional network model have been merged, thus constructing a time convolutional network model for time series data encoding (time-limited-TCN). The model effectively captured multimodal features of dynamic water temperature data from complex random time series, subsequently producing stable prediction results in different river basins. Thirteen hydrographic stations across four Bardeen rivers (Thames, Colorado, Mississippi and Sacramento) were used to test the proposed improved pre-temporal-TCN model and compare its performance with reference models (Air2Stream, Narx, Gru and Gboost). The results showed that the enhanced characteristics performed well in the river in the presence of human intervention, and that air temperature and DOY were important variables that influenced water temperature prediction. The proposed improved model shows that in cross-water water temperature prediction tasks, more stable and accurate prediction performance (average RMSE on the test set of at least 8.7 % better than the comparison model. Taking into account the characteristics and model performance, the proposed model should be a promising approach for the reconstruction of flow temperatures in several river basin data accumulation areas.
- Preprint
(2094 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2025-4550', Anonymous Referee #1, 23 Jan 2026
-
AC3: 'Reply on RC1', Lichen Su, 13 Sep 2026
Referee 1:
This manuscript addresses the important problem of stream water temperature prediction across multiple river basins, a task that is highly relevant for river ecosystem assessment and management. The authors propose an enhanced temporal convolutional network (TimENC-TCN) that incorporates moving average encoding and day-of-year (DOY) encoding to improve model generalization under data-scarce and heterogeneous conditions. The model is evaluated using observations from multiple hydrological stations across four major river basins and is compared against several established reference models. Overall, the study tackles a timely and challenging topic and presents a modeling framework with potential applicability to cross-basin stream temperature prediction. This is a good study with appropriate methods, and the manuscript is generally clear and easy to follow. I believe it should be suitable for publication once the following issues are addressed.
Response:
We sincerely thank the reviewer for the constructive and professional comments, as well as for the positive evaluation of our manuscript. We greatly appreciate the reviewer’s recognition of the significance of this study, the proposed TimENC-TCN model, and the overall clarity of the manuscript. Following the reviewer’s comments, we have carefully addressed each of the issues raised and made corresponding revisions throughout the manuscript. Detailed responses to each comment and the associated revisions are provided below.Major:
Several key conclusions in the manuscript appear insufficiently supported by the presented results. For example, the statement on line 273—“Therefore, it is reasonable to infer that at these stations where performance has declined, human factors have masked the effects of natural factors”—does not appear fully justified. This inference does not hold consistently across stations in the Sacramento River. Specifically, station J (RMSE=0.969 ºC) contradicts this pattern, and the RMSE values at stations H (1.29 °C) and B (1.189 °C) are quite similar, undermining a clear distinction. Additionally, the manuscript lacks clarity on the precise locations of stations I (Verona) and the other two Sacramento River stations, which is necessary to assess the validity of this conclusion.
Similarly, the claim on line 285—that the TimENC-TCN model demonstrates better ability to handle spatial heterogeneity in basins influenced by natural factors compared to those influenced by human factors—relies heavily on observations from a single station. This limits the strength of the conclusion. Moreover, the lower model performance at this station might be due to other factors such as measurement errors in water or air temperature data, or differences in input data volume, rather than human influence alone. Overall, clarifications on station locations, more consistent evidence across multiple stations, and a cautious interpretation of results are needed to strengthen these conclusions.
Response:
We sincerely thank the reviewer for this insightful and constructive comment. We agree that attributing the relatively lower model performance at certain stations solely to human factors was insufficiently supported by the presented results and could lead to an overly strong interpretation.In response to the reviewer’s concern, we re-examined the natural, hydrological, and anthropogenic characteristics of the relevant stations in the Sacramento River basin. We have added a schematic representation of the station locations and surrounding hydrological environments in the Methods section to provide clearer spatial and hydrological context (Line 80 - 85). In particular, station I (Verona) is located downstream of the confluence of the Sacramento and Feather Rivers. Its stream temperature may be affected by multiple interacting factors, including tributary interactions, water-management activities associated with the Central Valley Project (U.S. Bureau of Reclamation, 2024a), and temperature-management measures implemented to protect winter-run Chinook salmon (U.S. Bureau of Reclamation, 2024b). The corresponding reference and figure has been added to the revised manuscript.
In addition, the observation period includes the 2012–2016 California drought (California Department of Food and Agriculture, 2021), which was one of the most severe droughts in California's history. During this period, prolonged dry conditions substantially affected surface-water deliveries in the Sacramento and San Joaquin valleys, while water-management actions were implemented to respond to the drought and protect fish and wildlife. These conditions may have altered hydrological processes and the relationship between air temperature, seasonality, and stream temperature during the study period. The corresponding reference has been added to the revised manuscript.
Therefore, rather than attributing the relatively lower model performance directly to human influence, we have revised the discussion to emphasize the potential role of temporal non-stationarity and the combined effects of hydrological, water-management, and climatic conditions. We also acknowledge that other factors, including measurement uncertainty and differences in data availability, may contribute to the observed differences in model performance. Accordingly, we have removed the original generalized statement that human factors masked the effects of natural factors and have revised the manuscript.
The original text:
“This station is located on the Sacramento River, which is managed for water allocation by the Central Valley Project. The temperature at this station is influenced by human factors. Therefore, it is reasonable to infer that at these stations where performance has declined, human factors have masked the effects of natural factors, thereby weakening the direct explanatory power of temperature, DOY, or DOY encoding.”
has been revised to:
“A possible explanation is the temporal non-stationarity of the river thermal regime at this site. Verona is located downstream of the Sacramento–Feather River confluence, where the mixing of water from the two rivers may itself contribute to variations in stream water temperature. In addition, the site is situated within the broader Sacramento River system, where water management under the Central Valley Project (U.S. Bureau of Reclamation, 2024a) and temperature-management measures implemented to protect winter-run Chinook salmon (U.S. Bureau of Reclamation, 2024b) may further contribute to temporal variability in hydrological and thermal conditions. In addition, the observation period covered the 2012–2016 California drought (California Department of Food and Agriculture, 2021), which may have altered hydrological conditions and thermal responses. These factors may have introduced discrepancies between training and testing periods, reducing model generalization capability. Further investigation incorporating hydrological variables such as discharge and reservoir operation data is required.”
We have also revised the statement in the conclusion to avoid drawing a generalized distinction between naturally and anthropogenically influenced stations.
The original statement:
“As described in Sect. 4.1, this indicates that the TimENC-TCN model using DOY and air temperature as inputs has a better ability to handle spatial heterogeneity in basins influenced by natural environmental factors, while stations influenced by human factors require additional feature learning.”
has been revised to:
“As described in Sect. 4.1, these results suggest that the TimENC-TCN model using DOY and air temperature as inputs can effectively capture water temperature variability at most stations, whereas the relatively lower performance at some stations may be associated with differences in data availability, measurement uncertainty, and complex or temporally variable hydrological conditions. Therefore, additional hydrological and water-management-related features may be beneficial for improving model generalization at stations with more complex thermal regimes.”
These revisions provide a more cautious interpretation of the results and avoid attributing differences in model performance to human influence alone. We appreciate the reviewer’s comment, which helped us clarify the spatial and hydrological context of the study sites and improve the interpretation of model generalization across heterogeneous river environments.
Minor:
Line 43- 45. I suggest rephrasing the sentence in lines 43–45 by removing the initial “And” in the second sentence for better flow.
Response:
Thank you for this helpful suggestion. Following the reviewer’s recommendation, we have removed the initial “And” from the second sentence to improve the flow and readability of the paragraph.
Line 75: I suggest including a figure showing the locations of the stations to help readers visualize the geographical differences among them. This would clarify how the stations’ distinct environments may influence the results.
Response:
Thank you for this valuable suggestion. Following the reviewer’s recommendation, we have added a figure showing the locations of all study stations and their surrounding hydrological environments. This figure provides a clearer spatial context for the geographical differences among the stations and helps readers better understand how their distinct environmental settings may influence the results. The figure was generated using QGIS based on geographic and hydrological data obtained from the corresponding official sources, which are now also documented in the revised manuscript. Besides,the following content has been added to the manuscript.
“The locations of the 13 hydrological stations and their surrounding hydrological environments were mapped using QGIS. Station coordinates were obtained from the corresponding official hydrological data sources, while river and geographic background data were obtained from Natural Earth.”
Line 86, Table 1: I suggest including the watershed area for each station to provide additional context on the catchment characteristics.
Response:
Thank you for this helpful suggestion. Following the reviewer’s recommendation, we have added the watershed area for each study station to the revised Table 1. The watershed-area data were obtained from official sources and are provided to offer additional context regarding the catchment size and hydrological characteristics of each station.
Line 100: In the Methods section, it is important to include the approach used to evaluate model performance with gradual sample removal, detailing how this analysis was conducted.
Response:
Thank you for this helpful suggestion. Following the reviewer’s recommendation, we have revised and reorganized Sect. 3.3 Model Performance Comparison, in the Methods section. The section has been divided into two subsections, namely Model performance metrics and Assessment of model robustness, to improve its structure and readability. In the Assessment of model robustness subsection, we have provided a detailed description of the gradual sample removal experiment, including the sample removal ratio, removal procedure, and model performance evaluation method. These revisions provide a clearer and more reproducible description of the experimental procedure (Lines 258–272).
Line 190: References are missing for the developers of Air2Stream and the other models mentioned. Please include appropriate citations to acknowledge the original sources.
Response:
Thank you for pointing this out. Following the reviewer’s recommendation, we have added the appropriate original references for Air2Stream and the other models mentioned in the manuscript to acknowledge their original development and provide the necessary citations.
Line 259: The table showing “Option Feature input Feature output” appears to be missing a caption. I believe this is Table 3 and recommend adding an appropriate caption for clarity.
Response:
Thank you for carefully checking the manuscript. The table referred to by the reviewer is indeed Table 3. We have added an appropriate caption to Table 3 in the revised manuscript to clarify its content and improve its readability.
Line 315: The authors tested the model’s generalization performance by removing samples at Purfleet station (total data volume = 5718). While this approach is appropriate, the study’s conclusions would be strengthened by including at least one additional station in the generalization analysis, such as Cisco station (total data volume = 2100).
Response:
Thank you for this valuable suggestion. Following the reviewer’s recommendation, we have added a gradual sample removal experiment for the Catchment bridge station (total data volume = 2,100) to further evaluate the generalization performance of the model across different stations.
During the revision, we aidentified a labeling error in the original manuscript, where the total data volumes of the Cisco station and Catchment bridge station were incorrectly matched. Specifically, the data volume of 2,100 observations belonging to the Catchment bridge station was incorrectly attributed to the Cisco station. This was a labeling error rather than an error in data processing. We have carefully checked and corrected the relevant station names and data-volume information throughout the revised manuscript.
Accordingly, the revised analysis includes gradual sample removal experiments at both the Purfleet and Catchment bridge stations, providing additional evidence for evaluating the generalization capability of the TimENC-TCN model under different data-volume and station conditions.
Line 337: I suggest rephrasing this sentence to: “Meanwhile, the case of the Verona station and experiments on the drivers of stream temperature changes (Alger et al., 2021; Wade et al., 2023) indicate that introducing other features might be necessary at stations with significant human interference.”
Response:
Thank you for this helpful suggestion. We agree with the reviewer’s recommendation and have revised the sentence accordingly. The revised wording provides a more cautious interpretation of the results and better reflects the potential need for additional features when predicting stream temperature at stations with complex environmental influences.
For Figures 2, 5, 7, and 8, I suggest increasing the font size of the axis values to improve readability.
Response:
Thank you for this helpful suggestion. Following the reviewer’s recommendation, we have increased the font size of the axis tick labels in Figures 2, 5, 7, and 8 to improve their readability and visual clarity.
In addition, we have carefully reviewed the entire manuscript and refined its overall logic and presentation to further improve its clarity, coherence, and readability.
Reference
California Department of Food and Agriculture: Report on the 2012–2016 California Drought, Office of Public Affairs, 2021, https://plantingseedsblog.cdfa.ca.gov/index.php/2021/03/16/report-on-the-2012-2016-california-drought/.
U.S. Bureau of Reclamation: Central Valley Project, California-Great Basin Region, updated 8 April 2024, 2024a, https://www.usbr.gov/mp/cvp/.
U.S. Bureau of Reclamation: 2024 Sacramento River Temperature Management Plan, California-Great Basin Region, 2024b, https://www.usbr.gov/mp/bdo/docs/2024-sac-riv-temp-mgt-plan.pdf.
Citation: https://doi.org/10.5194/egusphere-2025-4550-AC3
-
AC3: 'Reply on RC1', Lichen Su, 13 Sep 2026
-
CEC1: 'Comment on egusphere-2025-4550 - No compliance with the policy of the journal', Juan Antonio Añel, 11 Feb 2026
Dear authors,
Unfortunately, after checking your manuscript, it has come to our attention that it does not comply with our "Code and Data Policy".
https://www.geoscientific-model-development.net/policies/code_and_data_policy.html
You have not published openly and without restrictions all the models, code, input data, and output data relative to your study. This does not comply with the policy of the journal, which explicitly requests the publication of all the elements necessary to replicate a manuscript in one of the repositories accepted according to the policy, before its submission to the journal. Because of it, your manuscript should have never been accepted for Discussions or peer-review in our journal. Therefore, the current situation is irregular, as the GMD review process depends on reviewers and community commentators being able to access, during the discussion phase, the code and data on which a manuscript depends.
Please, therefore, publish your code and data in one of the appropriate repositories and reply to this comment with the relevant information (link and a permanent identifier for it (e.g. DOI)) as soon as possible. We cannot have manuscripts under discussion that do not comply with our policy.
The 'Code and Data Availability’ section must also be modified to cite the new repository locations, and corresponding references added to the bibliography.
I must note that if you do not fix this problem, we cannot continue with the peer-review process or accept your manuscript for publication in GMD.
Juan A. Añel
Geosci. Model Dev. Executive EditorCitation: https://doi.org/10.5194/egusphere-2025-4550-CEC1 -
AC1: 'Reply on CEC1', Lichen Su, 15 Feb 2026
Dear Prof. Áñel,
Thank you very much for your careful review of our submission and for bringing this issue to our attention.
We sincerely apologize for not fully complying with the GMD code and data policy at the time of submission. We understand the importance of ensuring unrestricted public access to all model code, input data, and output data required to reproduce our results.
We have now taken the following actions to address this issue:
-
Model code has been deposited in Zenodo, and is publicly accessible at:
DOI: https://doi.org/10.5281/zenodo.18650137 -
Input data and Output data required to reproduce the simulations have been archived at:
DOI: https://doi.org/10.5281/zenodo.18650137 - Raw data: The raw data can be obtained from the following link:
The observed data for river stations in the UK are available at: https://environment.data.gov.uk/hydrology/explore#/landing (last access: 23 September 2025)
The observed data for river stations in the USA are available at: https://maps.waterdata.usgs.gov/mapper/index.html (last access: 23 September 2025)
The observed climate data for meteorological stations in the UK are available at: https://www.metoffice.gov.uk/research/climate/maps-and-data/historic-station-data (last access: 23 September 2025)
The observed climate data for meteorological stations in the USA are available at: https://www.ncei.noaa.gov/ (last access: 23 September 2025)
Some air temperature data were obtained from the National Aeronautics and Space Administration (NASA) Langley Research Center's Prediction Of Worldwide Energy Resources (POWER) project funded through the NASA Earth Science Division. The data were obtained from the POWER Project's Hourly v2.7.2 version on 2025/6/4.
During our testing process, we discovered that some computers were unable to access Zenodo. Consequently, we have provided supplementary GitHub links (URL: https://github.com/LcSu2025/Stable-Stream-Temperature-Prediction-TimENC-TCN-model) as an alternative. All repositories provide unrestricted public access.
The revised Code and Data Availability section now reads as follows:
The model generating the results presented herein is archived on Zenodo: https://doi.org/10.5281/zenodo.18650137, 2026 (Su., 2026a) and GitHub https://github.com/LcSu2025/Stable-Stream-Temperature-Prediction-TimENC-TCN-model (Su., 2026b). Further guidance on running the model and conducting ablation experiments is provided in the code repository's README file. The complete pre-processed dataset used for model training is archived on https://doi.org/10.5281/zenodo.18650137, 2026 (Su., 2026a), alongside experimental results.
And we add the corresponding references to the bibliography as required:
Su, L.: Stable Stream Temperature Prediction for Different Basins Using Time Series Encoding and Temporal Convolutional Networks: TimENC-TCN model. Zenodo. https://doi.org/10.5281/zenodo.18650137, 2026a.
Su, L.: Stable-Stream-Temperature-Prediction-TimENC-TCN-model. Github. https://github.com/LcSu2025/Stable-Stream-Temperature-Prediction-TimENC-TCN-model, 2026b.
We greatly appreciate the opportunity to correct this oversight. Thank you again for your guidance and for maintaining the high standards of GMD.
With best regards,
Lichen Su
On behalf of all co-authorsCitation: https://doi.org/10.5194/egusphere-2025-4550-AC1 -
CEC2: 'Reply on AC1', Juan Antonio Añel, 15 Feb 2026
Dear authors,
Many thanks for your reply. Unfortunately, it does not fully address the issues previously pointed out. We can not accept as links for the data webpages of NOAA, MetOffice, and similar. You must store the specific data you have used in one of the repositories that we accept, not link generic webpages.
Also, you state that some computers have problems accessing Zenodo. I do not quite understand by this. Problems to access Zenodo are obviously not on the side of Zenodo. In any case, we can not accept GitHub pages, as they are not trusted for long-term scientific archival. Actually, GitHub itself recommends to use Zenodo for such purposes. Therefore, please, remove any link to GitHub, as it does not serve the purpose of the policy of the journal.
Please, reply to this comment addressing the above mentioned issues, and include in your reply the text for a new "Code and Data Availability" section for your manuscript with the information for the permanent repositories for all the code and data necessary to replicate your study.
I must insist that if you do not solve the mentioned issues, we can not continue with the review and publication process in GMD.
Juan A. Añel
Geosci. Model Dev. Executive Editor
Citation: https://doi.org/10.5194/egusphere-2025-4550-CEC2 -
AC2: 'Reply on CEC2', Lichen Su, 16 Feb 2026
Dear Prof. Áñel,
We are most grateful for your valuable feedback. We have implemented the following changes:
1. All original data and their sources are now archived on Zenodo: https://doi.org/10.5281/zenodo.18654575 (Su., 2026b), replacing the web links.
2. The relevant GitHub links have been removed.The revised Code and Data Availability section now reads as follows:
The model generating the results presented herein is archived on Zenodo: https://doi.org/10.5281/zenodo.18650137, 2026 (Su., 2026a). Further guidance on running the model and conducting ablation experiments is provided in the code repository's README file. The original dataset is archived on Zenodo: https://doi.org/10.5281/zenodo.18654575 (Su., 2026b). The complete pre-processed dataset used for model training is archived on https://doi.org/10.5281/zenodo.18650137, 2026 (Su., 2026a), alongside experimental results.
And we add the corresponding references to the bibliography as required:
Su, L.: Stable Stream Temperature Prediction for Different Basins Using Time Series Encoding and Temporal Convolutional Networks: TimENC-TCN model. Zenodo. https://doi.org/10.5281/zenodo.18650137, 2026a.
Su, L.: Stable Stream Temperature Prediction for Different Basins Using Time Series Encoding and Temporal Convolutional Networks: TimENC-TCN model Original Dataset. Zenodo. https://doi.org/10.5281/zenodo.18654575 , 2026b.
Once again, thank you for your support and assistance.
With best regards,
Lichen Su
On behalf of all co-authorsCitation: https://doi.org/10.5194/egusphere-2025-4550-AC2 -
CEC3: 'Reply on AC2', Juan Antonio Añel, 16 Feb 2026
Dear authors,
Many thanks for addressing the outstanding issues. We can consider now the current version of your manuscript in compliance with the Code and Data policy of the journal.
Juan A. Añel
Geosci. Model Dev. Executive Editor
Citation: https://doi.org/10.5194/egusphere-2025-4550-CEC3
-
CEC3: 'Reply on AC2', Juan Antonio Añel, 16 Feb 2026
-
AC2: 'Reply on CEC2', Lichen Su, 16 Feb 2026
-
-
AC1: 'Reply on CEC1', Lichen Su, 15 Feb 2026
-
RC2: 'Comment on egusphere-2025-4550', Anonymous Referee #2, 27 Oct 2026
Review of paper egusphere-2025-4550 entitled "Stable Stream Temperature Prediction for Different Basins Using Time Series Encoding and Temporal Convolutional Networks" by Lichen Su and Wei Zhao
The authors describe a data-driven model for predicting stream temperatures from measured air temperatures using a temporal convolutional network (TCN). This network is additionally fed with moving averages of air temperature and an encoding for the day of the year (DOY). The method is compared with other already exisiting physically-based and data-driven approaches for predicting stream temperature and the proposed TCN-methods yields the best results in the method comparison.
My overall impression is that the paper would benefit from a clearer and more detailed description of the method as well as a more thorough discussion of the applicability of the proposed model. More detailed comments are given below:
1) When I went through the paper for the first time, it was unclear to me, whether the TCN was a method that the authors actively developed or whether it was an application of an already existing method. It turned out to be the latter case and that the novelty is the application of TCN to predict stream temperatures with encodings for DOY and moving averages of air temperature. I think it should be made a bit clearer throughout the paper that TCN itself was not actively developed but applied and what are the novelties compared to existing approaches.
2) From the methodological description it does not become clear to me, whether the station data of all four basins were used to train one model or whether the training was done seperately for all four basins, i.e. each basin has its own prediciton model for the corresponding stations. I think this is an important point that needs to be clarified in the methods section, especially because it affects also the practical application of the proposed method.
3) Although the alternative data-driven models (e.g. NARX, GBOOST, etc.) are shortly described, it is unclear what input data were used for training these models. Were the training data exactly the same as for the TCN-model, or was only air temperature used? Please explain this more clearly.
4) It is mentioned that the location of the stations, which is relevant for describing the spatial dependencies within a catchment, are encoded in the TCN-model (lines 94-97). However, I do not find any specific explanation how this is actually achieved and a better description within the paper would be helpful here. Equations 3 and 4 would suggest that the sine/cosine encoding is very similar for all locations, so how does it carry information on the spatial location? This point is also relevant in connection to point 2), i.e., if there is only one model with all station data, how would the spatial encoding of stations from another basin be helpful for the model?
5) An important point that is not part of the analysis and discussion is the transferability of the model to other catchments. Could the trained model be used for predicting stream temperatures at other locations within the basin(s) or even for locations in catchments that were not used in the training phase? Or is the prediction capability of the model only limited to the locations that were used for training the model? I think this is an important implication for the applicability of the model for practical purposes and should be discussed in more detail. If the model is restricted to the training data locations this is also a clear disadvante compared to physically-based models which by definition should be transferable to other settings.
6) Line 297: "In 100 repeated experiments..." This approach is not explained in the methodology section beforehand. How did these 100 experiments differ, how where they conducted? Please explain this in more detail. Otherwise results cannot be interpreted correctly.
7) Section 4.1 discusses the performance of the model using three different combinations of input features. This is mainly done in terms of (global) performance indicators (RMSE, MAE, NSE) which is, of course, a valid approach. In addition, it would be interesting so see whether and how the input features DOY and averaged air temperature change the course of the output time series. Do they e.g., systematically reduce the noise in the output time series or do they e.g., improve the capability of the model to capture changes within the time series?
Minor comments:
- Line 69: "...and it performs well in mitigating spatial heterogeneity." I am not convinced that the presented results really support a conclusion on the handling of spatial heterogeneity by the proposed method. Can you describe this in more detail?
- Line 96: "...or distance between codes." It is difficult to understand what is ment by this subsentence.
- Line 100: "..data denoising, ..." Please explain in more detail, how the raw data were denoised.
- Line 115: Please explain in more detail, what you mean with "distance correlation coefficient".
- Figure 2: Please provide explanations of the variables in the figure caption.
- Line 255: Please define what you mean with "ablation experiments". This is not a common term in my opinion.
- Figure 8: The colors in the legend to not fully match the line colors in the figure.Citation: https://doi.org/10.5194/egusphere-2025-4550-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,120 | 446 | 179 | 1,745 | 100 | 146 |
- HTML: 1,120
- PDF: 446
- XML: 179
- Total: 1,745
- BibTeX: 100
- EndNote: 146
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript addresses the important problem of stream water temperature prediction across multiple river basins, a task that is highly relevant for river ecosystem assessment and management. The authors propose an enhanced temporal convolutional network (TimENC-TCN) that incorporates moving average encoding and day-of-year (DOY) encoding to improve model generalization under data-scarce and heterogeneous conditions. The model is evaluated using observations from multiple hydrological stations across four major river basins and is compared against several established reference models. Overall, the study tackles a timely and challenging topic and presents a modeling framework with potential applicability to cross-basin stream temperature prediction. This is a good study with appropriate methods, and the manuscript is generally clear and easy to follow. I believe it should be suitable for publication once the following issues are addressed.
Major: Several key conclusions in the manuscript appear insufficiently supported by the presented results. For example, the statement on line 273—“Therefore, it is reasonable to infer that at these stations where performance has declined, human factors have masked the effects of natural factors”—does not appear fully justified. This inference does not hold consistently across stations in the Sacramento River. Specifically, station J (RMSE=0.969 ºC) contradicts this pattern, and the RMSE values at stations H (1.29 °C) and B (1.189 °C) are quite similar, undermining a clear distinction. Additionally, the manuscript lacks clarity on the precise locations of stations I (Verona) and the other two Sacramento River stations, which is necessary to assess the validity of this conclusion.
Similarly, the claim on line 285—that the TimENC-TCN model demonstrates better ability to handle spatial heterogeneity in basins influenced by natural factors compared to those influenced by human factors—relies heavily on observations from a single station. This limits the strength of the conclusion. Moreover, the lower model performance at this station might be due to other factors such as measurement errors in water or air temperature data, or differences in input data volume, rather than human influence alone. Overall, clarifications on station locations, more consistent evidence across multiple stations, and a cautious interpretation of results are needed to strengthen these conclusions.
Minor:
Line 43- 45. I suggest rephrasing the sentence in lines 43–45 by removing the initial “And” in the second sentence for better flow.
Line 75: I suggest including a figure showing the locations of the stations to help readers visualize the geographical differences among them. This would clarify how the stations’ distinct environments may influence the results.
Line 86, Table 1: I suggest including the watershed area for each station to provide additional context on the catchment characteristics.
Line 100: In the Methods section, it is important to include the approach used to evaluate model performance with gradual sample removal, detailing how this analysis was conducted.
Line 190: References are missing for the developers of Air2Stream and the other models mentioned. Please include appropriate citations to acknowledge the original sources.
Line 259: The table showing “Option Feature input Feature output” appears to be missing a caption. I believe this is Table 3 and recommend adding an appropriate caption for clarity.
Line 315: The authors tested the model’s generalization performance by removing samples at Purfleet station (total data volume = 5718). While this approach is appropriate, the study’s conclusions would be strengthened by including at least one additional station in the generalization analysis, such as Cisco station (total data volume = 2100).
Line 337: I suggest rephrasing this sentence to: “Meanwhile, the case of the Verona station and experiments on the drivers of stream temperature changes (Alger et al., 2021; Wade et al., 2023) indicate that introducing other features might be necessary at stations with significant human interference.”
For Figures 2, 5, 7, and 8, I suggest increasing the font size of the axis values to improve readability.