Q1. Can the authors numerically label the findings in the abstract, e.g., (1), (2), etc.?

Q2. L35-100: Some points within this paragraph need improvement, as follows:

First, the authors discuss the significance of refining ASTER GDEM but do not clearly explain what specific improvements or innovations this study introduces compared to previous studies.

Second, the paragraph starts by emphasizing the importance of high-quality DEMs, but the latter part seems to drift towards specific technical details about the ASTER GDEM correction method without clearly linking these details back to the broader impact or significance.

I found that the authors assumed that integrating ASTER GDEM with other DEM products will inherently lead to better outcomes but did not provide evidence or references to justify this assumption. Please revisit and address this point carefully.

I found phrases such as “high-accuracy global control point dataset” and “automatic processing scheme” are used without clear definitions or explanations of what makes them superior or innovative. This is very important for the product’s validation in this work.

The significance of the study is stated multiple times (e.g., “of great significance,” “beneficial supplement”), but without concrete examples or data to support these claims, the statements lack impact.

In addition, the literature review conducted on the use of remote-sensing DEMs Earth science research and scientific applications, including hydrological modeling, climate change research, natural hazard assessment, and ecosystem management was not well reviewed, suggesting accuracy of watershed delineation (10.1016/j.ejrh.2022.101282), flood risk assessment (10.3389/fenvs.2023.1304845), water resources management (10.1016/j.scitotenv.2024.174289 and 10.1007/s00382-024-07319-7), disaster preparedness (10.1109/jstars.2024.3380514), and promote human resilience for coastal communities (10.1016/j.jenvman.2024.121375).

Q3. The study specifically excludes polar regions from the correction process due to challenges like high variability in ice sheets and flow rates. This exclusion limits the global applicability of the IC2-GDEM product and leaves a coverage gap, particularly for researchers focused on polar studies.

Q4. The potential for temporal inconsistencies between the ASTER GDEM data and the more recent ICESat-2 data is not fully discussed. In dynamic landscapes, such as areas experiencing rapid coastal erosion or land use changes, these temporal discrepancies could lead to inaccuracies in the corrected DEM, which the authors did not quantify or address adequately. Please revisit and provide reasonable discussions to address this point.

Q5. The authors acknowledged that the density of ICESat-2 observations varies significantly with latitude, but it does not thoroughly investigate how this variation impacts the accuracy of the DEM corrections. In low-latitude regions, where ICESat-2 data are sparser, the correction results might be less reliable, a factor that needs more detailed examination.

Q6. The authors briefly mention the challenges posed by dynamic landscapes, where changes between the times of data collection could lead to inconsistencies. However, it does not provide a detailed analysis or propose methods to mitigate these issues, which is crucial for applications in rapidly changing environments.

In general, please separate the Discussion from the Conclusion section and provide a more in-depth discussion based on qualitative results.

Q7. Please include a section on limitations and future work.

Q8. In the conclusion, please highlight the main findings with a brief description (suggest highlighting qualitative results), but please keep them short, direct, and concise. The current form is lengthy and difficult to follow.

The paper is well written, and I have only a few minor suggestions:

In section 3.4 (GDEM Elevation Correction), the part on random forest regression could be expanded with more details, such as parameter selection, and supported by additional references. Additionally, I couldn’t find the random forest in Figure 3, which raises some curiosity about its role in the overall correction process.

In section 4.1 or the abstract, it would be helpful to include more specific details about the new data, such as the spatial and temporal resolution. You might also consider creating a table summarizing the characteristics of the input and output data to allow users to quickly reference these features.

The discussion in section 4 is somewhat limited, especially in sections 4.1 and 4.2, where it primarily presents results. It might be beneficial to discuss why the Corrected ASTER DEM Product showed greater improvement in Europe compared to other regions. Also, it seems there is no LDEM data for Asia and Africa, which could introduce uncertainties in validation—this could be worth discussing as well.

Review of “Global DEM Product Generation by Correcting ASTER GDEM

Elevation with ICESat-2 Altimeter Data”

The manuscript describes the correction of ASTER GDEM data with ICESat-2 data to create a product with a lower RMSE (according to a validation dataset). The ASTER GDEM is a widely used dataset and improving the accuracy is a useful exercise. The journal is the right venue to publish an open-source dataset.

The manuscript is mostly written in correct English and grammar – but please refrain from using etc. The abstract alone contains three occasions using etc. and several others throughout the manuscript. Either it is important enough to spell out – then list the additional points. If it is not important, there is no etc. needed. The term adds unnecessary ambiguity.

I can mostly follow the manuscript and reasoning, but have some comments. I understand that this article has seen previous reviews. I suggest that some of these are added as caveats or critical thoughts.

1) Dataset description. A more detailed description of the ASTER GDEM is necessary. What is the time frame of acquisition? Is it reasonable to use an ICESat2 dataset to correct the data (ICESat2 likely postdates some of the scenes used in the generation of ASTER GDEM). Same with the validation dataset: The ICESat data likely predates the ASTER GDEM scenes. While it is not likely that the large number of validation points have changed and I don’t think there is an impact on the statistics – but it will be useful to give these relevant information and a word of caution. The years of the lidar DEMs are listed.

2) While the training and validation dataset are somewhat independent (training: ICESat2, validation: ICESat about 10-15 years earlier), they both exhibit the same data characteristic. I have no concerns about the data quality, but it is not an unbiased validation. The geographic location points are not the same, but both data are point measurements (albeit taken with different instruments).

3) I may have missed it, but how are the border effects of the individually-adjusted tiles treated? Each 1-degree tile is calibrated (or trained) individually and the adjustment parameters may be different than the neighboring parameters. This may cause (or not) a small offset at the boundaries of the tile. Initially, I thought there is a feathering approach used with a buffer (equations 1 to 5), but I am not certain that this point is clearly illustrated or explained. This is section 3.4.

4) Where is the list of attributes that are trained with RF? Is there an attribute importance list that describe the usefulness of these parameters. I see the description of the “GDEM Elevation Evaluation Attribute Set” in 3.2 and that is useful. It is not clear what is contained in the elevation correction model (e.g., is this using Nuth and Kaeaeb to make horizontal adjustment or is this just a z component?)

5) Along the same lines: Individually training each tile is useful and will allow to correct for local problems. The current description of the random forest training is a black-box approach – I did not see the parameters (or attributes) that are used for height adjustment (or I have missed them). Is this just the dH? In any case, it will be useful to include the adjustment parameter or vertical offset in a separate dataset. This will allow the user to see how much each tile has been adjusted. Most other global DEM datasets have additional quality data (e.g., Copernicus has the number of measurement or TanDEM-X pairs). The averaged adjusted dH value for each tile is a useful assessment metric.

6) I am wondering about the improved product. I can imagine that a post-processed ASTER GDEM has an reduced RMSE. But the core problem with optical data is the inherent noise (see also Purinton and Bookhagen, 2021). Even in vegetation-free areas, the correlation problems with optical data lead to lower quality DEMs. Is the noise level reduced by the random-forest filtering? A zoom in area of a characteristic area with sufficient detail would be useful (before and after correction). I point out that several studies now have shown that the Copernicus DEM is currently the best available DEM. It is difficult to compare optical and radar-based DEMs, because they measure different things. Different story again with a Lidar DEM.

I am not certain if Figure 4 is useful – at this scale you are not seeing any difference. WHAT would be useful is to show the elevation adjustments that have been done (in a divergent color scale).