From the Moon to Mercury: Release of Global Crater Catalogs Using Multimodal Deep Learning for Crater Detection and Morphometric Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Global crater catalogs is a foundation dataset for lunar exploration missions. The detection of crater is of vital importance for ensuring the safety of lunar exploration missions and enhancing their scientific value. This manuscript introduces an enhanced VOLOLens model, establishing a more comprehensive crater catalog with smaller dimensions, increased quantity, and global coverage.  This model has also been applied to the detection of carter in the surface of Mercury. The following issues are suggested to be considered:

1. Additional ablation experiments should be designed to elucidate the methodological advantages.

2. The second paragraph in the introduction requires further enhancement. It should clarify the classification and limitations of existing crater detection methods.

3. Given the multitude of AI-based crater detection methods, it is recommended to compare the proposed method not only with existing databases but also with other approaches to highlight its superior features.

4. How are different datasets (LOLA, WAC) fusion? For instance, in terms of registration and cross-scale processing?

5. Can this method be effectively applied to identify craters of varying morphologies, such as secondary impact craters?

6. Is the model capable of processing higher-resolution data (e.g., NAC) to obtain a database of smaller-diameter craters?

7. In Table 4, why does ME6M300TGT extract fewer craters with diameters between 80-100 km compared to Herrick?

8. Equations require numbering, such as the formula mentioned on Page 6, line 187.

Author Response

See attached pdf

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Craters record the impact history of solid celestial bodies and play an important role in the formation and evolution of the Moon and Mercury. But how to detect craters is a key and hot topic. Although hundreds of people have been studying identification methods and geomorphological characteristics of the crater, data and methods always affect the accuracy and efficiency of identification. In this article, the authors utilized a wider range of data sources to more accurately detect craters on the Moon and Mercury, and obtained the spatial density and distribution of these craters. The analysis results sound nice. However, I suggest the authors make some improvements.

 

Comments 1: The abstract reported the diameters of the first impact-crater dataset for the Mercury in this article are greater than 400 meters. The resolutions (250m/p,650m/p) can support the result? I didn’t find any evidence in this article. Could the authors provide the calculation process.

 

Comments 2: In the abstract, one result shows that those "meters-scale secondary impact craters dominate the crater population in certain regions of Mercury’s surface". In 3.3.2, there is insufficient evidence to support the above result. Perhaps the meters-scale is kilometers-scale. However, generally speaking, there are more little secondary caters in on the solid celestial bodies. So, this result is necessary?

 

Comments 3: in section 2, the grid and tile are the same concept? If not, could the authors give some further information and introduce how to get the grid or tile?

 

Comments 4: in line 92,101 and so on, the distortion is made by the projection or reprojection. So, I suggest using the projection distortion.

 

Comments 5: in line 109-114, “This multiscale strategy ensures that no significant craters are discarded”. In this strategy, the craters at the middle scale can been detected, but those craters on the boundary could been discarded at the max ( for lager craters) or min ( for smaller craters) scale. As shown in the Figure5 and Fig 9, many craters haven’t been detected on the W90，E90, and 60N, where is a dark circle gap. If amplify the map with Mollweide projection, maybe more discarded craters can be found. So, some people made an overlap between the grids or tiles at the same scale to avoid discarding crater identification. So, I suggest the author to validate the strategy.

 

Comments 6: in133-134, “Simulates how the terrain would appear when illuminated from a specific angle”, what is the specific angle? why? Would the authors give some descriptions?

 

Comments 7: In Tab 1, the FoV is more like resolution, because the unit is m/px, not angle or range.

 

Comments 8: in 198, the rimr, driml, drimt and drimb represent the right, left, top and bottom of the rim. For reading easily, I suggest use the r, l, t and b to replace the Rim R, N, E and S respectively in the Fig 2.

 

Comments9: in Tab 3, why the confidences are not the same in the Southern Highlands and Northern Highlands? Would the author give some description?

 

Comments 10: in section 3.2, could the authors give a  comparison of size-frequency distributions (SFDs) between the Robbins catalog, LU5M812TGT catalog and LU6M371TGT catalog，which can provide some supports for the identification accuracy rising of the new model.

 

Comments11: in Fig4 and Fig5, the maps show the density and distribution of crater catalog LU6M371TGT and ME6M300TGT. Could the author give some further information or findings?

 

Comments12: in Tab 5, why is the difference between the YOLOlens Counting and Manually Counting so big. Could give some reasons?

 

Comments13: in line 390, there is no information about the architecture of the YOLOLens. Right?

 

Comments14: in Fig 12, I suggest reseting the central meridian of the map projection to rotate the map, which can save the page space.

 

Author Response

see attached pdf

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

General aspects

The manuscript is about automatized crater analysis for Mercury and comparison to the Moon. The topic is relevant and up to date, both considering the target (Mercury) and the method (AI based). The topic is important for many readers. The methods are moderately well described, however some adjustment is needed. The results are presented, but only part of the potential is described, here improvement is needed (as expected to have more results at the authors). The language is good, the illustrations are also good, the cited references are relevant, however the illustrations and references should be improved – thus the referee suggests major revise and encourages the authors to improve their work. Main points:

 

Please discuss the separation method of secondaries.

 

Some discussion is needed on the role of solar elevation and related change of craters’ appearance on automatic identification (for example around 326-334 lines)

 

Please give more information of the findings related to d/D ratio, for example around 360-368 lines

 

In general, the referee thinks there should be more specific results using the produced crater dataset – do not the authors intend to share their results?

 

 

Specific aspects

 

lines 36-84

this is a one page long paragraph, suggest separating toat least 2 or 3 paragraphs

 

around 114-115 or 215 or 393

also cite the recent work on the testing statistical impact crater analysis in permanently shadowed lunar polar regions

 

98

„composed of location areas”

sounds strange, suggest to reformulate

 

Figure 1

„in light green rectangle”

not visible in black and white print

write out „evaluation” in the right middle

 

242

„than the previous catalog”

not absolutely clear which is this catalogue

 

251

„Each Crater”

need not capitalize „crater” here

 

Table 3

Please indicate in the caption what is the brackets for

 

Table 4

Was the crater counting global?

 

284

„Ground-Truth data are partial”

in what sense?

 

Fig 6

also not useful in black and white print, suggest to improve

 

288

„106km2”

is it 106 km 10^6 km?

same for 298 line

 

304

„rounded”

not clear what the authors mean, all carters are round shaped

 

338

„fulfilled only of the primary craters,”

why? explain

 

Author Response

see attached pdf

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

1. If this YOLOLens model has already been published, the authors should introduce it in a concise way. Pages 5/6 are still elaborating on the model extensively, but there is no innovation. The authors should not emphasize the model's performance (Section 3.2) anymore, but rather focus on the advantages and characteristics of the two new datasets. 

2. What do the colors in Figures 16 and 19 represent? The same as Figure 11?

Author Response

see attach

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The author revised the manuscript according to the suggestions. There are few very small issues, what might be realized in the proof stage, so acceptance is suggested, with the following small modifications:

440 line
“causing noticeable seams”
is it properly formulated?

451
“enhances the terrain consistency”
short explanation of the term would be useful

518-520
“When the solar altitude angle is not within this range, factors such as underexposure or overexposure of the image, may lead to additional reconstruction errors.”
OK, and what bout emerging absolute shadowed areas?

532
„places of the moon”
capitalize ”Moon”
534, 539 also

533
“(e.g., south pole) „
suggest top put a citation for the readers to understand the importance, citing the work Polar Ice on the Moon by Springer published recently.

 

Author Response

see attach

Author Response File: Author Response.pdf