In this study, Xiong et al. applied a joint inversion downscaling method to downscale GRACE/GFO products to 0.5 degrees. The method extracts spatial information from hydrological simulations (WGHM) and applies spatial corrections to glaciers and lakes to account for components that WGHM insufficiently models. Then, the temporal consistency with GRACE/GFO is ensured to keep the mass conservation. The authors provide comprehensive evaluations in different aspects and compare them with some other downscaled/data assimilation products, and demonstrate the quality of SYSU TWSAv1.0.

This paper is generally well written and provides valuable contributions to the community. The special care of inland glaciers and large lakes is especially appreciated. In the end, it shows remarkable improvements in the glaciated regions compared to Gou and Soja (2024), who clearly stated that such regions are their product's main limitation. No major issues exist for this approach and dataset, but some descriptions are a bit too brief and need to be elaborated more for the audience to understand the product/approach better. Therefore, I would recommend publication of this paper after some rather minor revisions. The specific comments are as follows:

About Section 2 Data:

1. Line 87: There are multiple institutions that provide GRACE L2 data. Please specify which one you used.
2. Lines 97-98: Please specify if you used WaterGap v2.2d or 2.2e. If they are forced by GSWP3-W5E5, maybe you used v2.2e, right? In that case, please provide the reference to 2.2e (Müller Schmied et al., 2024).
3. Lines 121: Are all the 5-yr records continuous? Or some gaps are allowed in these records. If some gaps exist, how did you correct some jumps that may exist?

About Section 3 Joint Inversion Downscaling Framework

This section is important for the audience to understand the whole approach. Although I understand the method is based on their earlier publication (Xiong et al. 2025), I highly recommend that the authors expand this section to describe the methods in more detail. For example:

1. Line 145: Did you apply ICA to different WGHM components, such as groundwater and surface water, separately?
2. Lines 147-148: What do "larger singular values" mean? How many leading values did you use?
3. How did you apply the least-squares adjustment to all the spatial patterns to ensure mass conservation? Please describe them in more detail.

About Section 4 Results

1. Section 4.3: Did you remove other components, like soil moisture, from your downscaled TWSA products before comparing with in-situ groundwater data? If yes, please mention which datasets you used to remove them.
   
   
   
   
2. Line 335: It is good to see that SYSU has more variability than Gou and Soja (2024), which shows the possibility of involving more small-scale signals (also in Fig. 9). However, how would you argue if they include more signals or more noise? Moreover, GLWAS seems to have lower magnitudes in Fig. 8 than SYSU but rather similar degree variances in Fig. 9. What would be the reason behind this? As the data description paper is open for the data users from the hydrological community, more explanations about the spectral domain would be appreciated, so that the audience does not need much geodetic background knowledge to understand it.
3. Line 390: It’s fair that SYSU is closer to SHC solutions while Gou and Soja (2024) is closer to JPL mascons, as SYSU took SHC as inputs while Gou and Soja (2024) took JPL mascons as inputs. Please clarify this point, as the difference is unlikely to be caused by methodologies but rather by data differences.
4. 15: Did you convert all the datasets back to SHC d/o 90 and apply DDK3 filters to them as you did for Fig. 5? Both approaches are sound, but it's better to keep consistency. Moreover, Fig. 5 and Figs. 12-16 are based on different basin boundaries. Any specific reasons for doing so? Otherwise, I might recommend keeping consistent.
5. Starting from line 443 (GLDAS analysis). It is rather similar to the previous analysis and did not provide much more new information. Maybe move to the supplementary information.

General comments: Some of the figures have relatively low resolution. Please provide high-resolution figures and ideally vector figures for the final publication.

References

Müller Schmied et al. (2024). The global water resources and use model WaterGAP v2.2e: description and evaluation of modifications and new features. https://doi.org/10.5194/gmd-17-8817-2024

Xiong et al. (2025). High-Resolution Terrestrial Water Storage Anomalies and Components in China From GRACE/GFO via Joint Inversion Downscaling. https://doi.org/10.1029/2024WR038996

I have reviewed the manuscript titled " SYSU TWSA v1.0: Global High-Resolution Terrestrial Water Storage Anomalies via Satellite Gravimetry " authored by Yuhao Xiong et al. The manuscript is generally well written, but several major and minor issues should be addressed to improve readability and reproducibility. Below are some suggestions to enhance the manuscript:

Major Comments:

1. The manuscript presents a joint-inversion framework that combines GRACE/GFO constraints with WGHM spatial patterns and predefined mascon groups. However, the level of methodological novelty relative to existing approaches (e.g., data assimilation and machine learning-based downscaling) is not sufficiently clarified. The current presentation makes it difficult to distinguish whether the proposed method represents a fundamentally new approach or an alternative formulation of existing constrained inversion strategies. To improve clarity, the authors are encouraged to:
   1. Provide a clearer conceptual comparison with: 1) EnKF-based assimilation approaches. 2) Machine learning/statistical downscaling approaches.
   2. Explicitly state: 1) What problem existing methods fail to solve. 2) How the proposed method uniquely addresses it.
2. The proposed framework relies heavily on spatial basis functions derived from WGHM (Sect. 3.1), while only the temporal evolution is adjusted using GRACE/GFO observations. This implies that any structural bias in WGHM spatial patterns is directly propagated into the final product. Although briefly acknowledged in the conclusion, this limitation is not quantitatively assessed. This is a critical issue because it challenges the independence and physical reliability of the dataset. The authors should:
   1. Quantify the sensitivity of the results to WGHM spatial patterns
   2. Identify regions where WGHM is known to perform poorly and evaluate SYSU performance there.
   3. Discuss how spatial biases may affect glacier- and groundwater-dominated regions.
3. The manuscript states that WGHM outputs forced by the GSWP3-W5E5 dataset were used for the period 2002–2022. However, the standard GSWP3-W5E5 forcing dataset is typically available only up to 2019. It is therefore unclear how WGHM outputs were generated for the period 2020–2022. The authors should clarify whether an extended forcing dataset was used, and if so, provide details on its source and consistency.
4. A substantial portion of the validation relies on comparisons with GRACE/GFO data, which are already used as constraints in the inversion process. This raises concerns about circularity, particularly in basin-scale comparisons (Sect. 4.1), spectral-domain analysis (Sect. 4.4.1), and parts of the spatial-domain evaluation. While the manuscript acknowledges some of these limitations, the interpretation of results still tends to suggest validation rather than consistency. The authors should:
   1. Clearly distinguish between consistency checks (with GRACE/GFO), and Independent validation
   2. Rephrase conclusions where necessary to avoid overinterpretation.
   3. Strengthen truly independent validation (e.g., groundwater, water balance) with more rigorous analysis
5. The comparison with groundwater wells (Sect. 4.3) is based solely on correlation coefficients between TWSA and groundwater levels. While this approach is understandable given data limitations, it has important shortcomings as: 1) Groundwater levels are not directly equivalent to total water storage. 2) No conversion to storage is performed. 3) Correlation alone does not ensure physical consistency. As a result, the reported improvement (67.7% of wells) may not necessarily reflect improved physical realism. To improve this section, the authors should:
   1. Clarify the limitations of using correlation-based validation more explicitly.
   2. Perform regional or hydrogeological stratification (e.g., by aquifer type or climate zone)
   3. Assess statistical significance of improvements
   4. Moderate claims regarding physical consistency
6. The manuscript does not provide a comprehensive uncertainty analysis of the SYSU TWSA product, which is a major limitation for a data paper. Currently, the evaluation focuses on performance metrics (RMSE, NSE, CC), but does not address: 1) Propagation of GRACE/GFO measurement errors; 2) Uncertainty introduced by: ICA decomposition, Selection of spatial basis functions, and Mascon group definitions; and 3) Spatial variability of uncertainty. To address this, The authors are encouraged to:
   1. Provide uncertainty estimates where feasible (e.g., spatial error patterns or variance fields).
   2. Discuss how uncertainties propagate through the inversion framework
   3. Include sensitivity analyses for key methodological choices
   4. Clarify the expected level of confidence for different regions or conditions
7. The dataset is provided at 0.5° spatial resolution, but the effective resolution of GRACE/GFO observations remains on the order of ~300 km. While the downscaling enhances spatial detail, the manuscript does not clearly distinguish between nominal grid resolution, and effective information content. This may lead to misinterpretation of fine-scale signals by users. The authors should clearly state the effective resolution of the SYSU product; emphasize limitations of interpreting grid-scale variability; and discuss how much independent information exists at sub-basin scales
8. There is insufficient analysis of high-frequency variability (signal vs noise). In the spectral analysis, the SYSU product is shown to retain more high-degree variability compared to some other products (Sect. 4.4.1). However, it is not clear whether this additional variability represents meaningful hydrological signal or amplified noise. This distinction is critical for evaluating the usefulness of the dataset. The authors should:
   1. Discuss whether the high-frequency variability corresponds to physically meaningful features or potential noise
   2. Discuss potential noise amplification due to the inversion process
   3. Provide guidance on how such high-resolution signals should be interpreted
9. Although the manuscript compares SYSU with several products (e.g., GLWS, Gou & Soja 2024, GLDAS CLSM DA), the analysis remains largely metric-based and does not provide sufficient diagnostic insight into: 1) Why certain methods perform better in specific regions. 2) Under which conditions each method fails. To improve this section, the authors should provide regional analysis (e.g., humid vs arid, glacierized vs non-glacierized regions). Analyze differences in trend, seasonal amplitude, phase, and discuss strengths and weaknesses of each approach in specific hydrological regimes
10. There is inconsistency in GRACE data resolution (degree/order). The metadata of the NetCDF file indicates that the original GRACE data are “CSR SH d/o 60”, whereas the manuscript states that Level-2 data up to degree and order 90 were used. This discrepancy should be clarified, as it directly affects the effective resolution and information content of the dataset.

Minor Comments:

• Overall, several sentences are overly long and would benefit from being split for clarity.

• The manuscript states that GRACE/GFO Level-2 data up to degree and order 90 were used; however, the data source (e.g., CSR, GFZ, or JPL) is not clearly specified in Section 2.1. Although CSR RL06.2 is mentioned later in the data availability section, this information should be explicitly stated in the methodology. In addition, given that higher-degree coefficients (beyond ~60) are typically noise-dominated, the rationale for using degree/order 90 should be clarified.

• There is incorrect description of spatial resolution. The dataset is described in the manuscript as having a spatial resolution of 0.5°, which is consistent with the grid dimensions (360 × 720). However, the NetCDF metadata specifies “Spatial_resolution = 50 km”, which is not strictly correct for a global latitude–longitude grid and may be misleading. This should be revised to “0.5 degree” or clarified appropriately.
• The NetCDF metadata does not clearly specify the physical units of the TWSA variable. This information is essential for data usability and should be explicitly included.
• The manuscript refers to the application of a GIA correction using the ICE-6G model, whereas the NetCDF metadata specifies “ICE 6G-D”. It is unclear whether these refer to the same model variant, and the naming should be made consistent and explicitly clarified.
• The manuscript refers to the use of GRACE/GFO Level-2 data, which are provided as spherical harmonic coefficients rather than gridded fields. However, the dataset is described at a 0.5° resolution. The authors should clearly distinguish between the native resolution of Level-2 data and the derived gridded product after processing and downscaling.
• The dataset metadata clearly lists missing GRACE/GFO months, which is appreciated. However, the manuscript does not explicitly describe how these gaps are handled in the dataset (e.g., removed, interpolated, or retained as discontinuities). Clarifying this aspect is important, particularly for time-series analyses and derivative-based quantities such as TWF.
• The methodology combines spherical harmonic representations with mascon-based spatial groupings. However, the consistency between these two representations is not fully explained. The authors should clarify how mascon groups are integrated within the spherical harmonic inversion framework.
• (Page 1, lines ~17–25) The use of strong claims such as “strong agreement” and “competitive overall accuracy” appears somewhat overstated given the validation limitations. Consider moderating the language (e.g., “generally good agreement”, “demonstrates competitive performance”).
• (Page 2, lines ~52) The statement that “models still exhibit large uncertainties” would benefit from more precise qualification or additional references.
• (Page 6, lines ~144–148) The term “dominant patterns” (ICA decomposition) is not sufficiently defined. Please specify:

                  - how many components were retained

                  - the selection criterion used

• (Page 7, lines ~163–170) The selection of specific lake groups (e.g., Great Lakes, Lake Victoria, Tibetan Plateau lakes) should be justified more explicitly.
• (Page 8, lines ~180–188) While limitations are acknowledged, it would help to explicitly clarify that basin-scale comparison represents a consistency check rather than independent validation.
• Figures 4, 11, 12 are overly dense and difficult to interpret. Consider simplifying or splitting panels, or increase the font sizes.
• In multi-panel figures (e.g., Figure 11), the large number of panels combined with small font sizes reduces readability. It is difficult to clearly distinguish spatial patterns and labels at standard viewing scale. 1) Suggested improvements: Increase font size for legends and annotations. 2) Consider splitting large composite figures into multiple figures. 3) Provide zoomed-in regional examples for key areas (e.g., glacierized or arid regions)
• In several spatial maps (e.g., Figures 6, and 11), the color contrast is insufficient to clearly distinguish variations in low-signal regions, particularly in arid areas (e.g., northern Africa and the Middle East). The current color scale compresses small-magnitude variations, making it difficult to visually interpret differences between products. The authors are encouraged to use a diverging color scale centered at zero (for anomaly fields); Adjust the color range to better resolve low-amplitude signals; and Consider using non-linear scaling or percentile-based limits to enhance contrast in low-variance regions. Ensure that color choices remain perceptually uniform and colorblind-friendly.
• The manuscript includes a large number of figures (16 Figures) but only one table. This creates an imbalance between visual and quantitative presentation. The authors are encouraged to include additional tables summarizing key metrics (e.g., RMSE, NSE, CC, R² across basin sizes and regions) to facilitate clearer comparison and reproducibility.

Sincerely,