Vertical Accuracy Assessment of the ASTER, SRTM, GLO-30, and ATLAS in a Forested Environment

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Comments and Suggestions for Authors

I recommend accepting the manuscript in its current form. I appreciate the authors' efforts in addressing the feedback, and I believe the paper is now ready for publication.

Thank you for your attention to this matter.

Author Response

I am writing to express my deepest gratitude to you for your valuable time and effort in reviewing my paper. Your comments and suggestions have been invaluable to me and have greatly contributed to the improvement of my work.

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

The study is about the vertical accuracy of freely available DEM data. The study is well-structured. The authors can find suggestions to improve the quality of the manuscript in the attached file.

Comments

Part of the comments are based on the location of the page (P) and line (L):

1.      Please be careful about the order of abbreviations and their explanations. For example, P.2 L.47 no need to define again the DEM. Please check the manuscript.

2.      The authors should clarify the difference of the paper from the 28^th reference in the paper except having more study areas.

3.      Please evaluate the data based on each study area considering their unique characteristics.

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researchers. We have studied the comments carefully and have made corresponding revisions which we hope to meet with approval. An item-by-item response to the reviewers’ comments is provided as below:

Response to Reviewer 2:

Comment 1:

1. Please be careful about the order of abbreviations and their explanations. For example, P.2 L.47 no need to define again the DEM. Please check the manuscript.

Response 1:

Dear reviewer, thanks for your valuable comment.

I have noticed the problem of repeated definitions in the order of the abbreviations and its interpretation, which has been modified.

Comment 2:

2. The authors should clarify the difference of the paper from the 28^threference in the paper except having more study areas.

Response 2:

Dear reviewer, thanks for your valuable comment.

In [29], a comparative evaluation was conducted by contrasting the Digital Terrain Models (DTMs) derived from the Advanced Topographic Laser Altimeter System (ATLAS) ATL03 (Global Geolocated Photon Data) and ATL08 (Land and Vegetation Height) products with the G-LIHT (Goddard's LiDAR, Hyperspectral and Thermal) airborne LiDAR DTM. The study separately analyzed the errors in ATLAS data under different forest cover conditions concerning laser intensity and laser pointing angles. Qualitative and quantitative results indicated an average R² of 1.00 and an average RMSE of 0.75 m for all laser intensities and laser pointing types, highlighting the capability of ATL03 and ATL08 data in retrieving ground elevation. [29] focused solely on the impact of laser pointing angles and laser intensity on inverting ground terrain, without considering other factors influencing the results. In this study, we extend upon the findings of [28] to explore the accuracy of ATLAS, ASTER, SRTM and GLO-30 data as understory topography under varying elevations, slopes, aspects, and forest cover types in the forest research area.

Comment 3:

3. Please evaluate the data based on each study area considering their unique characteristics.

Response 3:

Dear reviewer, thanks for your valuable comment.

I have supplemented Table 3, providing error values for each evaluation metric within the three study areas. According to the results in Table 3, it can be observed that ICEsat-2 ATL03 exhibits the highest correlation with DTM data, and the differences in errors for it as an under-canopy DEM are relatively small across different study areas. In the Aiken study area, ASTER data performs the worst, with ME, RMSE, and STD errors of 7.95 m, 8.13 m, and 2.51 m, respectively. In the California study area, GLO-30 data shows the largest ME and RMSE errors, at 17.33 meters and 18.57 m, while ASTER data has the highest STD error of 7.74 m. In the Puerto Rico study area, ASTER data has the largest ME and RMSE errors, measuring 18.62 m and 20.28 m, and SRTM data has the highest STD error of 10.53 m. Across different study areas, the accuracy of the four products varies to some extent, with ICEsat-2 ATL03 data exhibiting the most stability, while the other three products are more influenced by different terrain and landform conditions. Consequently, future research should comprehensively analyze the performance of the four DEM data under different conditions to explore the reasons for the fluctuations in DEM products.

Table 3. Accuracy evaluation indicators for different study area

Reviewer 3 Report (New Reviewer)

Comments and Suggestions for Authors

In the article, the authors focused on the assessment of the vertical accuracy of ASTER, SRTM, GLO-30, and ATLAS, compared to the DTM dataset provided by G-LiHT in the forest study areas of the United States.

It can be concluded that the study is beneficial for potential readers of Forests magazine as well as for users of the mentioned height models.

I have the following comments on the methodology used in the study and on the content of the article:

-          in the introduction, there are studies that achieved different results in the assessment of the accuracy of elevation models. It must be noted that the mentioned accuracies are highly dependent on the choice of locations, their roughness and the density of the vegetation.

-          some studies dealing with the accuracy of vertical models are missing, e.g.:

-          https://aimt.cz/index.php/aimt/article/view/1034/12

-          In Chapter 2, the selection of locations for assessing the accuracy of the altimetry is not sufficiently justified. Location Aiken elevations range from 111 m to 154 m, Mendocino area elevations range from 84 m to 653 m.   Are the listed locations representative in terms of terrain fragmentation (slope gradients) and vegetation surface density for the US or for the world? Complete the explanation!

-          in equations (1) - (4) the heights are indicated by symbols x, y. But this is contrary to international standards for indicating the position of points in mapping and geodesy. In general, x and y are used to denote the position of a point and heights are denoted by z and/or h. I recommend changing the notation and using indexing.

-          Figure 3 illustrates the frequency histograms of height differences between the four DEM products and the DTM when using DTM as the reference truth. Explain in more detail the procedure for dividing errors into statistical frequency classes in Figure 3.

After corrections based on the comments above, I recommend the article accept for publication.

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researchers. We have studied the comments carefully and have made corresponding revisions which we hope to meet with approval. An item-by-item response to the reviewers’ comments is provided as below:

Response to Reviewer 3:

Comment 1:

1. In the introduction, there are studies that achieved different results in the assessment of the accuracy of elevation models. It must be noted that the mentioned accuracies are highly dependent on the choice of locations, their roughness and the density of the vegetation.

Response 1:

Dear reviewer, thanks for your valuable comment.

At the corresponding citation positions in the text, I have included factors related to the accuracy results of the studies:

In [19], Mahesh et al. utilized ICESat's GLAS LiDAR data to evaluate the vertical accuracy of AW3D30 v3, SRTM DEM V3, Carto DEM V3R1, and ASTER GDEM V2 under different coastal geomorphic conditions.

In [20], Bhardwaj Ashutosh et al. analyzed the data accuracy of ALOS and TanDEM-X (90m) DEMs under different rugged terrain conditions.

In [21], Liu et al. used ICESat-2 elevation points to assess the vertical accuracy of SRTM-1 DEM, SRTM-3 DEM, ASTER GDEM2, AW3D30 DEM, and TanDEM under various terrain morphological features (elevation, slope, and aspect) and land cover types.

In [22], Vassilaki et al. employed airborne LiDAR data to validate the vertical accuracy of TanDEM-X DEM, AW3D30, ASTER, SRTM (v001), and SRTM (v003) under different latitudes, terrain types, and land cover conditions.

In [22], Le et al. used airborne LiDAR-generated DEM and DSM as a reference, evaluating the vertical accuracy of ASTER, AW3D30, MERIT, SRTM, and TanDEM-X 90m as DEM and DSM in forested areas compared to residential areas.

In reference 27, Zhu et al. selected study areas such as HARV, SCBI, SJER, and Duke Forest SoAP, using airborne LiDAR data to assess the vertical accuracy of ICESat-2 data under daytime, nighttime conditions, and different slope situations.

In [29], Xing and Huang et al. utilized airborne data to validate the elevation accuracy of ICESat-2 elevation data in the forested area of Aiken, South Carolina, under various laser pointing angles and intensity types.

Comment 2:

2.Some studies dealing with the accuracy of vertical models are missing.

Response 2:

Dear reviewer, thanks for your valuable comment.

As your suggestion, we added the reference [18], which evaluates DEM products using traditional geodetic methods. The introduction of the research situation is improved.

Comment 3:

3.In Chapter 2, the selection of locations for assessing the accuracy of the altimetry is not sufficiently justified. Location Aiken elevations range from 111 m to 154 m, Mendocino area elevations range from 84 m to 653 m.   Are the listed locations representative in terms of terrain fragmentation (slope gradients) and vegetation surface density for the US or for the world? Complete the explanation!

Response 3:

Dear reviewer, thanks for your valuable comment.

The understory terrain in the United States comprises primarily gentle slopes, steep hillsides, deep valleys, and high-altitude areas. Additionally, different regions may host varying types of vegetation, ranging from coniferous forests to deciduous forests, directly influencing the characteristics of the understory terrain. The three study areas exhibit significant differences in geographical location, climate, forest cover types, topography, and elevation ranges, essentially encompassing the majority of understory terrain conditions. These study areas possess unique features in terms of topography, weather conditions, and forest cover types, leading to diverse impacts on the acquisition and analysis of terrain elevation data. Hence, the selected study areas in this paper can be considered representative. The study selected open-source and high-precision airborne G-LiHT data as the research data, which was provided by the G-LiHT team. Nevertheless, this research has not yet delved into polar regions, and future investigations will further explore these areas to obtain more comprehensive experimental results.

Comment 4:

4. In equations (1) - (4) the heights are indicated by symbols x, y. But this is contrary to international standards for indicating the position of points in mapping and geodesy. In general, x and y are used to denote the position of a point and heights are denoted by z and/or h. I recommend changing the notation and using indexing.

Response 4:

Dear reviewer, thanks for your valuable comment.

I have replaced x and y with z and h, in the formula, "z_i" represents the DTM surface elevation provided by G-LIHT, "h_i" represents the actual elevation value of different DEM data (ATLAS, ASTER, SRTM and GLO-30), "n" represents the number of study data points, ( is the average of different DEM data.

Comment 5:

5. Figure 3 illustrates the frequency histograms of height differences between the four DEM products and the DTM when using DTM as the reference truth. Explain in more detail the procedure for dividing errors into statistical frequency classes in Figure 3.

Response 5:

Dear reviewer, thanks for your valuable comment.

Figure 3 depicts the frequency histograms of height differences between the four DEM products and the DTM when using the DTM as the reference standard. In this figure, frequency represents the quantity or relative proportion of data points with specific height differences. The height of each histogram bar indicates the frequency of data points within the corresponding height difference range, i.e., the number of occurrences of data points in that range within the overall dataset. By observing the frequency histograms, insights into the distribution of height differences across different ranges can be gained, providing quantitative information about the dataset.

Thank you again for your valuable advices! We hope these modifications could make our paper better!

Kind regards,

All the authors.

Round 2

Reviewer 2 Report (New Reviewer)

Comments and Suggestions for Authors

The manuscript can be published in present form.

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors

I have carefully reviewed your manuscript titled "Vertical Accuracy Assessment of ASTER, SRTM, GLO-30, and ATLAS in Forested Environment" and find it to be well-organized and well-written. The problem description, literature review, and conclusion are particularly well-executed. However, there are a few points that need clarification and additional details, especially regarding the accuracy assessment methodology.

Methodology for Accuracy Assessment:

It would be beneficial to provide more details on the accuracy assessment methodology. Specifically, how many points were used for the accuracy tests? Did you utilize all the available rasters for calculation, or was there a subset chosen for assessment? A clearer description of the sampling strategy and the number of points used would enhance the transparency of your methodology.

Use of Metrics Like LE90:

The manuscript would benefit from an explanation of why metrics such as LE90 (vertical accuracy at 90% confidence level) were not employed in the accuracy assessment. Including a brief discussion on the choice of evaluation metrics and their relevance to the study could strengthen the methodological foundation of your research.

Elevation Difference Threshold (Line 478):

The manuscript mentions an "elevation difference threshold" in line 478. It would be helpful to elaborate on how this threshold was defined and its significance in the context of your study. Additionally, providing a reference or justification for the chosen threshold value would add credibility to this aspect of the methodology.

Addressing these points will enhance the clarity and completeness of your manuscript, ensuring that readers and fellow researchers can better understand and reproduce your accuracy assessment methodology.

Overall, your work makes a valuable contribution to the field, and these suggested revisions are intended to refine the presentation of your methodology for a more comprehensive understanding.

Thank you for considering these comments. I look forward to reviewing the revised version of your manuscript.

Best regards,

Author Response

Dear Editors and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript. Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to our researchers. We have studied the comments carefully and have made corresponding revisions which we hope to meet with approval. An item-by-item response to the reviewers’ comments is provided as below:

Response to Reviewer 1:

In this study, G-LiHT DTM data was selected as the reference truth to serve as a benchmark for comparative analysis. The analysis focused on the accuracy of DEM products generated within forested areas, specifically comparing footprint-based DEM products represented by ICESat-2 ATLAS with area-based DEM products represented by ASTER v003, SRTM v003, and GLO-30. The chosen methodology exhibits a logical structure, maintains high digital quality, and benefits from a well-founded historical context. The reference materials used in the study are comprehensive and ample.

Response:

Dear Reviewer, thanks for the prudent review and the kind recognition.

Comment 1:

1. It would be beneficial to provide more details on the accuracy assessment methodology. Specifically, how many points were used for the accuracy tests? Did you utilize all the available rasters for calculation, or was there a subset chosen for assessment? A clearer description of the sampling strategy and the number of points used would enhance the transparency of your methodology.

Response 1:

Dear reviewer, thanks for your valuable comment.

Three sets of ICESat-2 / ATLAS data were selected for this experiment using the strong beams they provided. The research data we selected includes the following screening criteria: 1. The research data is located in a forest environment. 2. ATLAS data is strong beam data. 3. Signal_ Con=4 and Classed_pc_flag=1 is used to determine its confidence and the photon count on the understory ground. 4. Determine the elevation difference threshold, i.e. dem_h and Height_1. A total of 790 photons were retained after screening to coincide with the validation data G-LiHT DTM data.

For different grid DEM data, we utilized the "Extract Multi Values to Points" function in ArcGIS software to extract the digital elevation data corresponding to the grid DEM data. In order to explore the impact of different slope and aspect on the accuracy of understory terrain, this study utilized the ArcGIS software to calculate slope and aspect in different study area.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

4. Ground Photon Selection

In the study, we chose to use GLC_ FCS30-2020 data to ensure that study data was under forest environments. Due to the composition of three pairs of laser beams emitted by ATLAS with a strong-to-weak energy ratio of 4:1, the precision of ground photon capture is higher for the strong beams than for the weak beams [40]. As a result, this experiment used to match the elevation data from the strong beams of ATL03 with the corresponding data from ASTER, SRTM, and GLO-30. In forested areas, the LiDAR data is distributed ir-regularly in three-dimensional space, comprising both ground points and vegetation points. To obtain an accurate DTM, this study filters the ground photons from the ATL03 photon cloud data based on the classification parameter "Classed_pc_flag" from ATL08. The values of "Classed_pc_flag" can be 0, 1, 2, or 3, corresponding to noise photons, ground photons, canopy photons, and canopy top photons, respectively. For this research, "Classed_pc_flag=1" (ground photons) and “Signal_ Con=4" from ATL03 is selected to represent ground photons within the forest understory vegetation.

7. Setting Elevation Difference Threshold

The elevation difference threshold is typically set based on the accuracy of the refer-ence DEM. In this experiment, the threshold for elevation differences was set according to the accuracy of the DTM. Generally, the elevation difference threshold is set to twice the RMSE (Root Mean Square Error) value, and in areas with complex terrain, it can be set to 1.5-2 times the RMSE value. Therefore, in forested areas, the elevation difference thresh-olds are typically set at 20m (approximately twice the RMSE value). Interpolation between "dem_h" and "Height_1" was calculated and data points with elevation difference ex-ceeded the threshold were removed as total error. Among them, "Height_1"proposed the most consistent terrain elevation provided by ATL03 with good accuracy, which can be used as the topographic elevation value representing the understory topography. The dem_h is the reference terrain height. The difference between the two partly reflects the accuracy of the data points, and this step helps to eliminate most of the low-precision data points, improve the accuracy of understory topography data, and increase the reliability of subsequent experiments. A total of 790 data were involved in the evaluation.

6. Extraction of Elevation Data

The basic attributes of DTM, ATL03, and the ASTER DEM, SRTM DEM, GLO-30 DEM products have been unified using Arcgis software. At this stage, the "Extract Multi Values to Points" function provided by Arcgis is utilized to extract the corresponding elevation values for each product from their respective positions along the ground track.

Comment 2:

2. The manuscript would benefit from an explanation of why metrics such as LE90 (vertical accuracy at 90% confidence level) were not employed in the accuracy assessment. Including a brief discussion on the choice of evaluation metrics and their relevance to the study could strengthen the methodological foundation of your research.

Response 2:

Dear reviewer, thanks for your valuable comment.

LE90 (Linear Error at 90% Confidence Level) represents the value of linear vertical error at a 90% confidence level, providing an upper limit for vertical error at a certain confidence level. LE90 focuses on the distribution range of errors, especially the upper limit at a given confidence level. LE90 corresponds to Circular Error at 90% Confidence Level (CE90). In this study, we focused more on the vertical error and did not consider the horizontal error. In contrast, ME (Mean Error) is the average value of elevation errors, providing the central position of errors; RMSE (Root Mean Square Error) is the square root of the mean square of errors between each observed value and the true value, emphasizing error measurement; STD (Standard Deviation) is the square root of the mean square of the differences between observed values and the mean value, measuring the dispersion of values in the dataset, specifically the average distance of data points from the mean value, and it emphasizes the measurement of data dispersion. Therefore, these evaluation metrics chosen in this study, ME, RMSE, and STD, already provide a comprehensive analysis of errors, including average errors and the dispersion of errors. Considering this, LE90 was not used as an evaluation metric.

Comment 3:

3. The manuscript mentions an "elevation difference threshold" in line 478. It would be helpful to elaborate on how this threshold was defined and its significance in the context of your study. Additionally, providing a reference or justification for the chosen threshold value would add credibility to this aspect of the methodology.

Response 3:

Dear reviewer, thanks for your valuable comment. The text has been modified in detail.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

The elevation difference threshold is typically set based on the accuracy of the reference DEM. In this experiment, the threshold for elevation differences was set according to the accuracy of the DTM. Generally, the elevation difference threshold is set to twice the RMSE (Root Mean Square Error) value, and in areas with complex terrain, it can be set to 1.5-2 times the RMSE value. Therefore, in forested areas, the elevation difference thresholds are typically set at 20m (approximately twice the RMSE value). Interpolation between "dem _ h" and "Height_1" was calculated, and data points whose elevation difference exceeded the threshold were removed as the total error. Among them, Height_1 proposed the most consistent terrain elevation provided by ATL 03 with good accuracy, which can be used as the topographic elevation value representing the understory topography. The dem _ h is the ATL08 reference terrain height. The difference between the two partly reflects the accuracy of the data points, and this step helps to eliminate most of the low-precision data points, improve the accuracy of understory topography data, and increase the reliability of subsequent experiments. A total of 790 data were involved in the evaluation.

Reviewer 2 Report

Comments and Suggestions for Authors

The objective of this paper is to assess the precision of ICESat-2/ATLAS data, ASTER, SRTM, and GLO-30 sets in estimation of understory terrain (in other words the location of the ground), thereby offering robust data support for forestry applications.

There is no explanation as to why these datasets were chosen, nor a description of how the reference DEM and DTMs were developed.  G-LiHT information is brought in as a reference standard, without clear explanation.  A concise clear table summarizing key information from each of the different instruments by their acronyms, and a clear explanation of the reference DEM and DTM, is essential in this type of study, and including all the different products that may come from the same instrument.  Understanding these datasets and then hypothesizing which one or ones would be most useful before running the analyses, and then writing up the results, including only the evidence needed to prove the hypotheses would be a research paper.  By using a table and referencing it, repetitive text can be deleted.  The paper appears to be at least three times longer than it needs to be.  Lines 517-525 seem like a concluding paragraph and yet there are many more pages to the paper.  

Also, summarizing the information now listed out in the conclusions, in a table, would make for a more understandable paper.  If the information was presented well enough, it might be worth publishing though as a more standard work.    

Lines 15-17 in the abstract states that ‘The study reveals that ICESat-2 ATL03 at the footprint scale exhibits the highest accuracy within forested regions, with a correlation coefficient (R²) close to 1 when compared to the reference dataset DTM.’   But we should suspect this given that NASA supports a reference dataset DTM that appears to be the ICESat-2 ATL03 layer.  Going to  https://nsidc.org/data/atl03/versions/3, National Snow and Ice Data Center, to find out what this reference DTM is, the text describes the ATL03 product.

Some of the reasons to chose certain products over others is cost of processing, or timing of the information, geographic availability, and these factors are not brought into the discussion of performance.

There is text that says the information being presented is novel, however lines 552-555 discusses how results of this study aligns with that or two other previously published studies. If similar results have already been published several times, then it seems inaccurate to say the work is novel.  And Line 550 lists another publication study that is similar, but applied to more broad land types than in this study.  However, this section 4.3 in this study is on DEM precision under different non-forested land cover, such as bushes and grassland.  Why is this information included if this study is focused on forested environments? L23-25 in abstract mentions performance by different land cover, but this study is on forested environments, so why is the different land cover mentioned?

So much of the text in this paper would go better in a supplement, or perhaps omitted.    

Focusing the text on the exact study, being clear about hypotheses based on understanding of the strengths and design of these instruments, and summarizing information in tables would help make it clear what this study is focused on, and then also adding in terms of research results about these instruments that is not already known.

Author Response

Comment 1:

1. There is no explanation as to why these datasets were chosen, nor a description of how the reference DEM and DTMs were developed.  G-LiHT information is brought in as a reference standard, without clear explanation. A concise clear table summarizing key information from each of the different instruments by their acronyms, and a clear explanation of the reference DEM and DTM, is essential in this type of study, and including all the different products that may come from the same instrument. 

Response 1:

Dear reviewer, thanks for your valuable comment.

This experiment utilized G-LiHT DTM data as validation data to compare the differences between ICESat-2/ATLAS ATL03 and DEM products including ASTER DEM V3, SRTM DEM V3, and the GLO-30 DEM in terms of understory terrain data. The choice of G-LiHT DTM data (with a spatial resolution of 1 m) as the reference standard for comparative analysis was based on its higher point cloud density compared to ICESat-2's spaceborne LiDAR data, allowing for more precise capture of subtle terrain changes and higher measurement accuracy. Additionally, ICESat-2/ATLAS ATL03 was chosen alongside the other three DEM products for this study because ICESat-2/ATLAS provides data in the form of spot clouds, while ASTER DEM V3, SRTM DEM V3, and the GLO-30 DEM are widely used grid-form DEM products. The above data are all obtained for free, and they have shown good estimation ability in different aspects in other studies. The most important thing is that the three types of DEM data selected in the study used two common methods for generating DEM data. Among them, the ASTER DEM is a data product that utilizes optical stereo photogrammetry to acquire global elevation data, SRTM DEM and GLO-30 DEM used InSAR technology to acquire elevation data.

G-LiHT (Goddard’s LiDAR, Hyperspectral, and Thermal Imager) is a relatively cost-effective and highly rugged portable airborne imaging system equipped with a VQ-480 miniaturized LiDAR for providing vertical and horizontal distribution infor-mation of tree canopies. G-LiHT can emit discrete laser pulses and it provides 1 m spatial resolution data products including Canopy Height Models (CHM), DTM, point clouds, and hyperspectral images [38]. These data products are used for measuring forest under-story terrain and canopy height. The G-LiHT official provides a set of algorithms for pro-cessing raw data. This algorithm employs a grid-based Delaunay triangulation morpho-logical filter that creates an irregular Triangulated Irregular Network (TIN) using the low-est elevation points (ground echoes). The TIN is then interpolated to generate a Digital Terrain Model by adjusting grid sizes in proportion to flight altitude, laser repetition rate, surface roughness, and reflectance to ensure sufficient point cloud coverage within each grid.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

Evaluating the accuracy of these four products in generating DEMs for forested areas allows for a comprehensive consideration in selecting the most suitable product for understory DEM estimation. Table 1 showed the study data in the study.

Table 1. Basic parameters of the study data.

2.2.5. G-LiHT data

G-LiHT (Goddard’s LiDAR, Hyperspectral, and Thermal Imager) is a relatively cost-effective and highly rugged portable airborne imaging system equipped with a VQ-480 miniaturized LiDAR for providing vertical and horizontal distribution infor-mation of tree canopies. G-LiHT can emit discrete laser pulses and it provides 1 m spatial resolution data products including Canopy Height Models (CHM), DTM, point clouds, and hyperspectral images [38]. These data products are used for measuring forest under-story terrain and canopy height. The G-LiHT official provides a set of algorithms for pro-cessing raw data. This algorithm employs a grid-based Delaunay triangulation morpho-logical filter that creates an irregular Triangulated Irregular Network (TIN) using the low-est elevation points (ground echoes). The TIN is then interpolated to generate a Digital Terrain Model by adjusting grid sizes in proportion to flight altitude, laser repetition rate, surface roughness, and reflectance to ensure sufficient point cloud coverage within each grid.

Despite the acquisition time of G-LiHT data is different from the ASTER, SRTM, GLO-30, and ATLAS data used in this experiment, there have been no geological disasters in the terrain of the research area over the past 20 years. The focus of the study on the ac-curacy of the terrain under the canopy makes the time difference irrelevant. Consequently, this study employs 1 m resolution DTM products generated using the official algorithm from data collected over three study areas as validation data. The relevant data can be freely downloaded from the G-LiHT official website (https://glihtdata.gsfc.nasa.gov).

Comment 2:

2. The paper appears to be at least three times longer than it needs to be.  Lines 517-525 seem like a concluding paragraph and yet there are many more pages to the paper.

Response 2:

Dear reviewer, thanks for your valuable comment.

Thanks for your comment and valuable advice. We have deleted some content according to your advice. We carefully organized the entire text and removed similar descriptions.

Comment 3:

3. Lines 15-17 in the abstract states that ‘The study reveals that ICESat-2 ATL03 at the footprint scale exhibits the highest accuracy within forested regions, with a correlation coefficient (R²) close to 1 when compared to the reference dataset DTM.’   But we should suspect this given that NASA supports a reference dataset DTM that appears to be the ICESat-2 ATL03 layer.  Going to  https://nsidc.org/data/atl03/versions/3, National Snow and Ice Data Center, to find out what this reference DTM is, the text describes the ATL03 product.

Response 3:

Dear reviewer, thanks for your valuable comment.

The reference DTM chosen for validation in this study is the G-LiHT DTM. The experimental findings indicate that, in forested areas, ICESat-2 ATL03 exhibits the highest accuracy at the footprint scale. Compared to the reference data provided by G-LiHT, the correlation coefficient (R²) approaches 1. The Root Mean Square Error (RMSE) is 1.96 m, and the Standard Deviation (STD) is 1.93 m. I apologize for any lack of clarity in the abstract.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

We assessed the vertical accuracy of ASTER, SRTM, GLO-30, and ATLAS in the forest study areas of the United States compared to the reference dataset DTM provided by G-LiHT and we will further discuss the influence of different ground altitude, forest type, slope and aspect on vertical accuracy. The study reveals that in forested environment, ICESat-2 ATL03 exhibits the highest accuracy at the footprint scale, with a correlation coefficient (R²) close to 1 and Root Mean Square Error (RMSE) = 1.96 m. SRTM exhibits the highest accuracy at the regional scale, with a R² close to 0.99, RMSE=11.09 m. A significant decrease in accuracy was observed with increasing slope, especially for slopes above 15°. With a sudden increase in altitude, such as in mountainous situations, the accuracy of vertical estimation will significantly decrease. Our results show that ICESat-2 and SRTM data might be sufficient and stable vertical accuracy in forested environment.

Comment 4:

4. Some of the reasons to chose certain products over others is cost of processing, or timing of the information, geographic availability, and these factors are not brought into the discussion of performance.

Response 4:

Dear reviewer, thanks for your valuable comment.

Reasons for choosing G-LiHT DTM as validation data: As is well known, airborne data is high-precision data, but it is also very expensive. Therefore, we chose to use G-LiHT data, which is highly accurate and can be downloaded for free. Therefore, G-LiHT is used as our validation data, and our research area is also selected based on G-LiHT as the main reference. Moreover, as our research focuses on understory terrain, it will not change unless major geological disasters occur. Therefore, we believe that time factors will not have a significant impact on our research results. G-LiHT airborne LiDAR point cloud data, with higher density compared to ICESat-2 satellite-borne photon cloud data, can be considered as actual measurement data. Therefore, this study utilizes the DTM products generated by the official algorithms using data collected over the three study areas with a 1-meter resolution as validation data.

Reasons for choosing ICESat-2/ATLAS: ICESat-2/ATLAS was the first spaceborne LiDAR to adopt photon counting format. ICESat-2/ATLAS is characterized by strong penetration ability, high sensitivity and low power consumption. In forested areas, some of the photons from ICESat-2 penetrate the canopy and reach the ground and understory vegetation, providing accurate information on ground elevation and understory vegetation growth, contributing to the high accuracy of understory DEM.

Reasons for choosing ASTER, SRTM, and GLO-30: These three are commonly used DEM products with wide coverage, and they are available for free download. The most important thing is that the three types of DEM data selected in the study used two common methods for generating DEM data. Among them, the ASTER DEM is a data product that utilizes optical stereo photogrammetry to acquire global elevation data, SRTM DEM and GLO-30 DEM used InSAR technology to acquire elevation data.

Reasons for choosing GLC_FCS30 as land cover data: It is the world's first dataset providing 30 m classification accuracy in a fine-grained forest cover and is available for free download.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

ICESat-2/ATLAS, ASTER, SRTM and GLO-30 can be obtained for free and represent different data production methods (LiDAR, SAR and stereophotogrammetry). Therefore, we chose ICESat-2/ATLAS, ASTER, SRTM, and GLO-30 data as our study data. This paper aims to assess the accuracy of ICESat-2/ATLAS data with used DEM datasets (ASTER, SRTM, and GLO-30) in estimating understory terrain, providing robust data support for forestry applications.

Comment 5:

5. There is text that says the information being presented is novel, however lines 552-555 discusses how results of this study aligns with that or two other previously published studies. If similar results have already been published several times, then it seems inaccurate to say the work is novel.  And Line 550 lists another publication study that is similar, but applied to more broad land types than in this study. 

Response 5:

Dear reviewer, thanks for your valuable comment.

In lines 552-555, the conclusions of this study align with those of Gao et al. [41] and Zhu et al. [27], indicating that within vegetation-covered areas, the accuracy of SRTM surpasses that of ASTER. However, this experiment, building on previous research, concludes that GLO-30 performs less effectively than ASTER in vegetation-covered regions, demonstrating the absolute advantage of ATL03 data provided in footprint form over gridded DEM data.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

In line 550, the findings of this study are consistent with Neuenschwander et al.'s research [43], revealing that ICESat-2 data exhibit smaller elevation variations due to geographical inaccuracies in flat regions, with smaller average differences and RMSE values. However, this paper focuses more specifically on the accuracy of DEMs in forested regions. In summary, this experiment not only confirms the limited impact of slope on ATL03 within forested areas but also suggests that GLO-30 may be more suitable for gentle slope regions, while SRTM may be better suited for obtaining understory DEM in larger area coverage with slope >6°.

Comment 6 :

6. This section 4.3 in this study is on DEM precision under different non-forested land cover, such as bushes and grassland.  Why is this information included if this study is focused on forested environments? L23-25 in abstract mentions performance by different land cover, but this study is on forested environments, so why is the different land cover mentioned?

Response 6:

Dear reviewer, thanks for your valuable comment.

At your suggestion, I removed experiments under different non-forest land covers, such as shrubs and grassland, and focused on the different forest cover. We have reorganized the description of non-forest types in the text.

To make it easier for you to observe the changes, we also made a copy here below.

In the revised manuscript:

We assessed the vertical accuracy of ASTER, SRTM, GLO-30, and ATLAS in the forest study areas of the United States compared to the reference dataset DTM provided by G-LiHT and we will further discuss the influence of different ground altitude, forest type, slope and aspect on vertical accuracy. The study reveals that in forested environment, ICESat-2 ATL03 exhibits the highest accuracy at the footprint scale, with a correlation coefficient (R²) close to 1 and Root Mean Square Error (RMSE) = 1.96 m. SRTM exhibits the highest accuracy at the regional scale, with a R² close to 0.99, RMSE=11.09 m. A significant decrease in accuracy was observed with increasing slope, especially for slopes above 15°. With a sudden increase in altitude, such as in mountainous situations, the accuracy of vertical estimation will significantly decrease. Our results show that ICESat-2 and SRTM data might be sufficient and stable vertical accuracy in forested environment.

Thank you again for your valuable advices! We hope these modifications could make our paper better!