the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Jan 2025
SDUST2024MSS_AO: a MSS model of the Arctic Ocean derived from CryoSat-2 SAR altimeter data
Abstract. Due to the seasonal and perennial sea ice coverage in the Arctic Ocean, determining the sea surface height (SSH) is more difficult compared to mid- and low-latitude regions. This has resulted in a lack of high-precision, high-resolution mean sea surface (MSS) models for the Arctic Ocean. This paper focuses on the SSH in the ice-covered regions of the Arctic Ocean, using CryoSat-2 SAR mode altimeter data and MODIS images to develop a feature and threshold optimization method based on waveform characteristic method, combining mutual information and the F1 Score. This method detects lead observations in Baseline-E CryoSat-2 ice products with a precision of 90.79 % and a recall of 85.25 %. Using CryoSat-2 SAR mode altimeter data from July 2010 to December 2023, the lead observations are divided into 5-km grids for each month, and gross error observations in each grid are removed according to the two-sigma principle. The mean SSH for each month is calculated to establish a monthly mean SSH time series within each grid. Then, using least square estimation (LSE), a new MSS model with a grid size of 5 km is constructed, named SDUST 2024 MSS of Arctic Ocean (SDUST2024MSS_AO). The SDUST2024MSS_AO model is compared with four internationally renowned MSS models (CLS2022, DTU21, UCL13, and SDUST2020) and validated using ICESat-2 altimeter data, demonstrating the reliability of the SDUST2024MSS_AO model. The SDUST2024MSS_AO model data are available at https://doi.org/10.5281/zenodo.13624487 (Liu et al., 2024).
- Preprint
(1524 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 29 Mar 2025)
-
RC1: 'Comment on essd-2025-2', Anonymous Referee #1, 01 Feb 2025
reply
The comment was uploaded in the form of a supplement: https://essd.copernicus.org/preprints/essd-2025-2/essd-2025-2-RC1-supplement.pdf
-
RC2: 'Comment on the supplementary materials of essd-2025-2.', Anonymous Referee #1, 03 Feb 2025
reply
The previously submitted review file was incomplete.
We have now provided additional comments to supplement the original review.
The authors are kindly requested to review the updated comments.
-
AC1: 'Reply on RC2', X. Liu, 24 Feb 2025
reply
Response to Reviewer Comments
Dear Reviewer,
We sincerely thank the reviewers for their time, effort, and thoughtful comments on our manuscript. Your valuable feedback has played a crucial role in improving the quality of our work. Below, we provide a point-by-point response to address each of your comments in detail. We have carefully considered all suggestions, and where appropriate, revised the manuscript accordingly to clarify our methods and enhance the overall presentation of the study.Please note that, due to the limitations of the current online review system, we are unable to attach the updated manuscript directly within this response. However, all suggested revisions have been incorporated, and additional adjustments have been made to further strengthen the manuscript.
_____________________________________________
Point-by-Point Responses
1. In Line 47, "MODIS" is used as an abbreviation. It would be helpful to spell out the full name of the acronym the first time it appears in the text, followed by the abbreviation in parentheses.
Response: Thank you for your valuable suggestion. We have revised the manuscript accordingly by spelling out the full name of MODIS the first time it appears in the text, followed by the abbreviation in parentheses.
2. In Line 84, the reference period for the CLS2022 model is incorrectly stated as spanning from 1993 to 2021. The correct reference period is from 2002 to 2020. Additionally, the data used in the CLS2022 model does not include T/P, ERS-2, GFO, Envisat, or Sentinel-3A. Please verify and correct this information.Response: Thank you for your careful review. We have corrected the reference period for the CLS2022 model to 2002–2020 as suggested. Additionally, we have removed the statement indicating that the CLS2022 model includes T/P, ERS-2, GFO, Envisat, or Sentinel-3A.
3.In Lines 88-89, the reference period for the DTU21 model is incorrectly stated as spanning from 1993 to 2020. The correct reference period is from 1993 to 2012. Additionally, the data used in the DTU21 model does not include ERS-1, ICESat, Geosat, or Sentinel-3A, but does include SARAL/AltiKa. Please verify and correct this information.
Response: Thank you for your valuable suggestion. We have corrected the reference period for the DTU21 model to 1993–2012 as suggested. Additionally, we have removed the statement indicating that the DTU21 model includes ERS-1, ICESat, Geosat, and Sentinel-3A data, and we have added a statement specifying that it includes SARAL/AltiKa data.
4. The UCL2013 model is outdated. Please consider using the more recent MSS model, mss_sio_32.1, instead.
Response: Thank you for your valuable suggestion. mss_sio_32.1 is indeed a newer model. However, we chose to use the UCL2013 model because both the SDUST2024MSS_AO model and UCL2013 model use the same data source, which is the single CryoSat-2 dataset, making them highly comparable in terms of data origin. Additionally, the spatial coverage of the mss_sio_32.1 model only reaches up to 82°N, making it less suitable for comparison in this study. Furthermore, the other three models used in this study (DTU21, CLS2022, and SDUST2020) are relatively recent models, all of which have been thoroughly validated and are scientifically reliable. Therefore, considering the higher relevance and direct comparability with these models, we believe it is unnecessary to include mss_sio_32.1 in the comparisons at this point.
5. In Lines 120-127, a more detailed analysis of the characteristics and differences of these 14 waveform features is needed. This will allow for a clearer comparison and understanding of their individual properties.
Response: Thank you for your valuable suggestion. The manuscript has already provided references that detail the sources and specific characteristics of these 14 waveform features, which readers can refer to for more information. Furthermore, the detailed characteristics of these waveform features are not the primary focus of this study; instead, this study aims to compare these features. Therefore, we believe the current description sufficiently meets the research requirements.
6. In Line 130, could you clarify why selecting the appropriate features and thresholds for lead detection is crucial? A brief explanation of their impact on detection accuracy would be helpful.
Response: Thank you for your insightful suggestion. In this study, mutual information is employed to select features, aiming to quantify the nonlinear correlation between features and leads while eliminating redundant features to enhance lead detection performance. Additionally, the threshold is optimized using the F1 score, which balances precision and recall to determine the optimal trade-off point for maximizing overall performance. This approach avoids the subjective bias of traditional manually set thresholds and ensures the robustness of detection results in the complex Arctic environment.
This explanation has been added to the paragraph following Line 130 in the revised manuscript.
7. In Line 132, could you explain why the method combining mutual information and the F1 score was chosen? Providing the rationale behind this selection would be beneficial.
Response: Thank you for your valuable suggestion. We chose the method combining mutual information and the F1 score primarily because mutual information effectively measures the nonlinear correlation between features and leads, providing a comprehensive assessment of feature relevance. Meanwhile, the F1 score, by balancing precision and recall, offers a more reasonable threshold selection criterion compared to a single metric such as precision, preventing performance degradation caused by an overemphasis on one aspect. The combination of these two methods ensures that the detection model achieves both effectiveness (optimal feature selection) and high performance (optimal threshold determination).
8. In Line 136, each parameter in the formula should be explained in detail, as should the parameters in the other formulas throughout the paper. This will help readers better understand their significance and application.
Response: Thank you for your suggestion. The parameters in the formula on Line 136, have already been explained in detail in Lines 134–135.
9. In Line 165, could you provide a clearer explanation of what is meant by the "gridded approach"? A brief description of this methodology would help readers understand how it is applied in the context of this study.
Response: We apologize for the lack of clarity in our description of the "gridded approach." What we intended to express is that the lead data are processed using a Cartesian grid with a fixed spatial resolution of 5 km × 5 km as the fundamental processing unit. We have revised the corresponding description in the manuscript.
10. The results presented in this paper should be discussed and analyzed in greater detail, particularly those shown in Table 1, Table 3, and Figures 4 and 5. A more thorough interpretation of these results would enhance the paper's clarity and depth.
Response: Thank you for your valuable suggestion. We have provided a more detailed discussion and analysis of the results presented in the paper, particularly those shown in Table 1, Table 3, and Figures 4 and 5. In the revised version, we have included more in-depth explanations to enhance the clarity and depth of the paper.
11. In Figure 4, why is the F1 score for the 14 waveform features not provided? Additionally, could you clarify what the horizontal coordinate represents in the figure?
Response: Thank you for your suggestion. In Figure 4, we have provided the F1 scores for the five waveform features that were selected through the filtering process. As mentioned in Lines 196–197 of the manuscript, features with mutual information values lower than the average of all features were filtered out in two rounds, leaving Sigma_0, MP, LTPP, SA, and PP as the final selected features for further analysis. Since the discarded features were no longer included in the study, their F1 scores were not presented in the figure. Regarding the horizontal coordinate in Figure 4, taking Figure 4(a) as an example, it represents the values of Sigma_0. When Sigma_0 reaches 15.2, the corresponding F1 score attains its maximum value.
12. The different MSS models discussed in the paper (e.g., DTU21, CLS2022, and SDUST2024MSS_AO) have different reference periods, which may affect the comparability of their results. Specifically, DTU21 covers 1993–2012, CLS2022 spans 2002–2020, and SDUST2024MSS_AO covers 2010–2023. The paper does not adequately address how these temporal discrepancies are handled to ensure that comparisons between the models are valid. A clearer explanation of how to manage these time variations is necessary.
Response: Thank you for your valuable comments. The MSS models represent the mean sea level over a specific period, and changes in sea level are typically not drastic between consecutive periods. Therefore, while there may be an overall shift between MSS models with different reference periods, such system errors have a limited impact on the model comparison in this study.In our comparative analysis, we used mean bias and standard deviation (SD) to assess the differences between SDUST2024MSS_AO and the other MSS models. The mean bias is used to measure the overall system error, while SD mainly reflects the dispersion of local variations, and thus is not directly affected by the overall system error. Additionally, the difference map (Figure 7) provides a clear visual representation of the spatial bias between the models. If the system error is uniform, the difference map will show a consistent overall shift, and the SD will not increase significantly. However, if there is a large spatial variation in the system error, the SD will increase accordingly.
By combining the analysis of mean bias, SD, and the difference map (Figure 7), we are able to effectively distinguish between system errors and local variations, allowing us to reasonably evaluate the reliability of SDUST2024MSS_AO and its similarity to other MSS models.
We have further clarified this point in the manuscript to enhance the explanation of the impact of different reference periods on the model comparisons.
13. There is mention of discrepancies in model results due to differences in data coverage, such as the lack of CryoSat-2 data coverage north of 81.5°N. While this is a valid concern, the paper should discuss the extent to which these gaps in data coverage may affect the overall model accuracy, particularly in the central Arctic.Response: Thank you for your valuable comments. SDUST2024MSS_AO uses CryoSat-2 data in the region north of 81.5°N, while the four models compared in this study use different data sources in this area. For example, SDUST2020 does not use CryoSat-2 data in this region; although CLS2022 uses CryoSat-2 data, the data for this area have been corrected to ensure consistency with the reference period of the mid-latitude regions; UCL2013 and DTU21 also use CryoSat-2 data in this region. Therefore, the differences observed in the difference map (Figure 7) north of 81.5°N primarily reflect the impacts of differences in data coverage and data processing methods among the models. Notably, the smaller difference between SDUST2024MSS_AO, UCL2013, and DTU21 indicates better agreement between models using the same data source in this region. Additionally, the difference map (Figure 7) clearly illustrates the impact of data coverage differences on model accuracy, further demonstrating how data gaps or variations in data processing affect the model results. We believe that the difference map (Figure 7) effectively conveys this point, and we have discussed this in the manuscript.
Once again, we sincerely thank the reviewers for their constructive comments and insightful suggestions. We believe these revisions have significantly improved the manuscript, and we look forward to receiving your further feedback.
Best regards,
Xin LiuCitation: https://doi.org/10.5194/essd-2025-2-RC2
Â
Citation: https://doi.org/10.5194/essd-2025-2-AC1
-
AC1: 'Reply on RC2', X. Liu, 24 Feb 2025
reply
-
RC2: 'Comment on the supplementary materials of essd-2025-2.', Anonymous Referee #1, 03 Feb 2025
reply
Data sets
SDUST2024MSS_AO: a mean sea surface model of the Arctic Ocean based on CryoSat-2 SAR altimeter data Xin Liu https://zenodo.org/records/13624488
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 148 | 24 | 8 | 180 | 3 | 3 |
- HTML: 148
- PDF: 24
- XML: 8
- Total: 180
- BibTeX: 3
- EndNote: 3
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1