Improved Stereophotogrammetric and Multi-View Shape-from-Shading DTMs of Occator Crater and Its Interior Cryovolcanism-Related Bright Spots
, Stephan van Gasselt 2, Ralf Jaumann 1, Julie C. Castillo-Rogez 3, Carol A. Raymond 3, Sebastian H. G. Walter 1 and Frank Postberg 1Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper describes the generation and evaluation of several new regional digital topographic models (DTMs) of Ceres by using stereo- and intensity-based methods in freely available software (ISIS and ASP). The DTMs improve on resolution available in previous such models and their quality has been evaluated by multiple tests. In addition, the methodology is described in detail. This paper thus presents new work (though utilizing existing software tools) and both the data products and the methodology are likely to be of interest to planetary science readers of Remote Sensing. I am therefore recommending that the paper be accepted after minor revision. I have a few overall concerns about the present manuscript, neither of which is substantive in the sense of affecting the validity of the results or conclusions. I also have a large number of suggestions for minor changes and additions to the text and figures. I hope the authors will take these in the constructive spirit in which they are offered, as suggestions to improve the clarity and impact of the paper that they can accept or reject as they see fit.
To cover the essential technical details about the manuscript, it is well organized (apart from some passages that might be improved by minor reordering) and the writing is grammatical and clear (my many quibbles about rewording and additions notwithstanding) so there is no need for copy editing. The tables are all clear and contribute directly to the paper. The figures are well-made but I note some specific issues about both legibility and the necessity of some of the figures and panels below.
Overall issues
The one change I consider essential has to do with terminology and affects the entire paper, starting with the title. I find the use of “stereophotoclinometry” (SPC) to describe the new work reported in this paper inconsistent with usage in the field and thus potentially misleading to readers. In the literature, methods of estimating topography that rely on image intensity have been described as photoclinometry (PC) or shape-from-shading (SfS). This is true even when multiple images are used, and even if the solution process is initialized with and/or constrained to remain close to a DTM from another source such as stereophotogrammetry (SPG). In particular, the creators of the tool in the Ames Stereo Pipeline (ASP) used in this study refer to it as multi-view shape-from-shading (per reference 49 in the manuscript) but it can also be considered to be multi-image photoclinometry.
The term stereophotoclinometry has only been used in the literature to indicate the method developed by Gaskell, which directly combines stereo and intensity cues. Multi-image photoclinometry is used to make “maplets” of local relative elevation. These maplets are then used to align the images they were derived from, resulting in a stereophotogrammetric solution for the absolute location of the maplet. The two techniques are then alternated to achieve convergence in both local detail and absolute positions, so both are equally essential to the overall method.
The published DTMs described in references 58 and 59 were produced by using Gaskell’s methods and should continue to be described as SPC products. The new DTMs that are currently labeled SPG DTMs in the manuscript were made with the ASP tool and should be described as SfS (or PC but not SPC) products. This distinction should be maintained because the ASP and Gaskell methods use multiple images in fundamentally different ways. As a result, the title should be modified to read “Improved Stereophotogrammetric and Photoclinometric DTMs of Occator Crater….”
(As a painful minor footnote, a few early authors used the term “photometric stereo” to refer to multi-image SfS/PC methods. This terminology is highly confusing and little used, so it deserves to be forgotten. I’m glad it doesn’t come up in the present paper. I probably shouldn’t have mentioned it here, but I worried that someone might suggest incorrectly that “photometric stereo” equates to “stereophotoclinometry.”)
A second issue is that I highly recommend that the authors point out clearly and early and often how large the albedo variations in the study areas on Ceres are. This means that the work is a very stringent text of the ASP software’s ability to correct for such variations and that confidence in the method on other bodies with more sutble albedo variations (e.g., Moon and Mars) is bolstered. This is an important “selling” point of the paper. The value of the large albedo variations examined could be mentioned in the Abstract and Introduction and revisited in the Discussion or Conclusions. Quantitative information about the magnitude of the albedo variations on Ceres (and comparison to other bodies for reference) could be given in the text associated with Figures 1 and 2.
The third overall issue I want to raise concerns the length of the paper. I want to be clear that I am highly supportive of the paper providing full detail about the data, assumptions, and workflow used and the methods by which the results were checked. I do not want to see any of this detail excluded. Nevertheless, the paper is quite long and there are a few opportunities to shorten it slightly without significant loss. Figures 3 and 4 could be reduced by eliminating the information for Vesta, which is not relevant to the paper. Figure 4 might possibly be replaced by a table providing the same information in less space. Figure 12c is redundant with Figure 13 so could potentially be eliminated. Please consider these figure deletions to be highly optional suggestions, to be considered by the authors in consultation with the editor over the desired length of the paper.
Another type of opportunity to shorten and clarify the paper is presented by the considerable redundancy between the tables and text. I will try to call out specific passages in my minor comments where the text repeats many or most of the numbers in a table. This is unnecessary and the impact of the text would be improved if it were to provide a higher-level summary of the results such as “mean differences are a fraction of a pixel but standard deviations are on the order of a pixel” rather than the specific numbers. This is a clarity and impact issue so I hope the suggestion will be considered even if there is no drive to reduce the length of the paper.
If the authors and editors concur, much of the material in the Appendix could be provided as online supporting material as another means to shorten the paper. I would recommend that the table of acronyms be retained in the main paper, because it is so helpful for understanding the text.
Finally, given that the paper focuses much more on the creation of the DTMs than on the products derived by using them (i.e., photometrically corrected mosaics), the level of detail about the photometric models for the latter in Section 4.3 could be greatly reduced. Specifically, the discussion of the Hapke model used in producing the DTMs could be first, followed by a statement of the different type of model used to correct the images. The reader could be directed to the literature already cited in this section for the details, so Equations 1-6 and Table 3 with the parameters could probably be eliminated if this is judged essential.
Given the focus on the DTMs, perhaps some of the other details about the workflow steps needed to make image products could be abbreviated. I actually recommend a change in the opposite direction, however. It would be reasonable to add a short subsection at the end of the Discussion that briefly summarizes the types of image products made and the assessments of both their quality and the lessons learned in making them. In this case the detail about processing steps to make the image products should be retained.
Minor comments, in order of occurrence (P = page, L = line number)
P 1 L 16 XMO7 is not defined here (but other abbreviations are). First definition is at L 104. Maybe the mission phase abbreviations should be defined in Table A1 and that table could be referenced in the text when the first such abbreviation is mentioned.
P 2 Fig. 1 is beautiful and informative. It would be even better if the study areas 1-4 over which DTMs were compared were shown, perhaps with a distinguishing line color or pattern. If the study areas are the same as some of the DTM extents, this could be indicated in the caption.
P 3 Fig. 2 caption: It would be helpful to indicate here (as was done in the Fig. 1 caption) what the stretch limits for the color bands are. Also, as a minor detail perhaps indicate in the caption whether the photometric correction makes use of the DTMs produced in this study as alluded to at L 67. Or perhaps it’s a preliminary product showing what is achieved without a DTM, to be contrasted with DTM-based results later?
If this figure is an example of a photometrically corrected mosaic made based on the final (largest) DTM, then it could be mentioned again in context of Section 4 (it’s the result of the processing chain described there) and Section 6 (Discussion of the quality of the result and the improvements obtained with the new DTMs).
The impact of the paper could also be increased by explicitly stating in Introduction that the albedo variations studied are very large (as much as a factor of 10), a fact that is currently only revealed later, on p. 8. Thus this paper presents a _much_ more stringent test of SfS with albedo modeling than most other studies, which generally involve fractional or at most factor-of-two differences in albedo. The authors can increase the impact of the paper by stressing this and by giving some quantitative estimates of the effect (see my comments about the Discussion section, p. 33).
P 3 L 59-62. I agree heartily with this sentence that defines “the first two steps” as making DTMs and then making image mosaics, in that order. This ordering should be used as consistently as possible in discussing the work. Some other passages of the current manuscript seem to describe the preparation of images for mosaics first, followed by description of the DTM production.
P 4 L 70 ISIS and ASP “will be made publicly available for reproducibility”. Why “will be”? This should be rephrased to indicate that the software systems are already available from their developers.
Also, this is a good location at which to state that the ASP was used to perform SPG and SfS (equivalently PC, not SPC as currently stated) and that its algorithms are independent of and different from the software used to make the previously archived DTMs.
P 4 L 86 (and throughout the paper): It’s fine to speak of the image resolution changing with orbital distance (as at L 77, 83, 85, for example) but the figures in meters/pixel are not resolutions in the strict sense of the resolving power of the images, but the size of the pixels. This distinction can be maintained by using a phrase such as “pixel scale,” “pixel spacing” or “ground sample distance” (GSD) for such quantities. The authors clearly understand this distinction, as evidenced by multiple passages later where resolution and GSD are clearly distinguished. Thus I’m just suggesting the text be checked for any inadvertent uses of one term where the other was meant.
P 5 Fig. 3 The grey lines and text appear quite faint on my screen. It should be possible to use thicker lines and larger type (and/or a darker shade of grey) to make them more visible. It is not clear to me why black can’t simply be used, since there doesn’t seem to be an important distinction between the information in grey and that already presented in black. The Vesta portion of the figure is not relevant to the paper so could be omitted.
P 6 L 119, 122. The footnote numbers following units (m) unfortunately look a lot like exponents. The use of color helps somewhat with this. Perhaps consider adding space between the “m” and the number? If there is a citation number as well, the footnote number could be moved after it (e.g., L 122 but unfortunately not L 119). This issue may occur in additional places in the manuscript but I won’t comment on it further.
P 6 L 122. The DTM in references 58-59 is “really” a SPC DTM, produced with the stereophotoclinometry algorithm developed by Gaskell. I’ve argued above that the method used in this paper should not be called SPC in order to distinguish it from the Gaskell method. If that recommendation is taken, there will be no confusion here (except that the abbreviation SPC will need to be explained). If it is not taken, then it should be noted at this location that the DTM is not produced by the same methodology as the “SPC” DTMs created for this paper.
P 7 Fig. 4. This is a pretty figure, but not essential. The Vesta portion is not relevant to the paper. The Ceres portion could be eliminated (or the information presented more concisely in a table) if it becomes desirable to shorten the paper. If the figure is retained, the grey elements could be made darker (even black) and the type could be enlarged slightly to improve readability, similar to what was suggested for Fig. 3.
PP 7-8 Kudos for distinguishing here photoclinometry/shape-from-shading from stereophotoclinometry, and for distinguishing resolution from ground sample distance. Hopefully these distinctions can be made consistently throughout the paper.
P 8 L 158-171. The factors listed can indeed result in points (or correct points, if some are mismatches) being obtained at spacings larger than the image pixel. There’s another effect that’s as or more important that isn’t mentioned. Stereo matching depends on recognizing multi-pixel patterns (whether patches or features) so even if points are obtained at each pixel they will be based partly on shared pixels. Therefore the resolution will still be multiple pixels even if the point cloud is dense. This is just a more nuanced explanation; it doesn’t change the conclusions cited in the following paragraph.
Also, at L 162, reference 72 can be omitted. Reference 73 supersedes it and contains all its information and more. On the other hand, the authors might consider citing Kirk et al. (2022), which builds on the earlier papers and contains a much more extensive investigation of the results from the ASP SfS module and the optimal settings for it. It could be referenced where the ASP SfS module is first mentioned (alongside reference 49) and/or alongside reference 73 to provide some information about the expected resolution of photoclinometry DTMs to compliment that about stereo DTMs in [73]. The full reference is
Kirk, R.L., Mayer, D.P., Dundas, C.M., Wheeler, B., Beyer, R.A., and Alexandrov, O., 2022, Comparison of digital terrain models from two photoclinometry methods, International Archives of Photogrammetry, Remote Sensing, and Spatial Information Sciences, XLIII-B3-2022, 1059–1067, doi:10.5194/isprs-archives-XLIII-B3-2022-1059-2022.
P 8 L 168-187. This is a key passage where multi-view shape-from -shading (MV-SFS) as implemented in ASP needs to be distinguished from SPC.
P 8 L 204. The values given are major and minor axes, so should be divided by 2 to get the semi-axis values that correspond to the mean radius 470 km given at L 200.
P 8 L 205. It’s a subtle point of language, but “projection” has two different meanings. As a result, there is a distinction between “projecting the data onto a reference ellipsoid” and “basing the map projection on a reference ellipsoid.” The first means estimating 3D point locations by intersecting pixels with the ellipsoid. That is, the ellipsoid is used as a crude (better than a sphere but not as detailed as a DTM) shape model. The second means that 3D points (however they are determined) are dropped perpendicularly onto the ellipsoid and then the map projection from that ellipsoid to the plane is used. For example, a version of the stereographic projection that is conformal for a biaxial ellipsoid rather than for a sphere. The phrasing used in the text corresponds to the first of these meanings, but I think it’s unlikely, since the authors are making and using DTMs. The second meaning is probably meant, and would be better expressed by saying that “to enable minimally distorted cartographic representations and measurements, Figures … utilize conformal projections of this reference ellipsoid.” The sentence about profiles that follows is OK the way it is.
P 9 L 228 “Table ??”. I don’t see any tables that list image numbers. If such a table isn’t added, the reference to it should be removed. Maybe it’s intended to go in supplemental material. “Table ??” is referenced at 4 other locations in the manuscript.
P 10 L 260 “Table ??”
P 10 L 277-278. The 3x3 filter does smoothing, not downsampling. The “projection” to different GSD can accurately be described as downsampling.
P 11 Fig. 5 caption, penultimate sentence: “four erroneous pixel accumulations” should be five. Also, in the text immediately following the figure (L 310) “catchment area” and “accumulations” are terms with specific geologic meanings that don’t seem very relevant here. Maybe just write “The clusters of pixels affected are outlined by dashed lines in the bottom row of Figure 5.”
P 12 L 329 The term “kernels” has not been defined up to this point and may not be familiar to all readers. I suggest expanding on the existing sentence to say “…also provided updated camera position and pointing information in the form of NAIF SPICE kernel files” with a reference to Acton, C. H. (1996). Ancillary data services of NASA's navigation and ancillary information facility. Planetary and Space Science, 44(1), 65–70. https://doi.org/10.1016/0032‐0633(95)00107‐7.
P 12 L 367 “iterative corrections of the Z-vector in the ground control point networks” is unclear. If this means that the elevation (often called the Z value) to which ground points were constrained was updated based on the DTM and latest horizontal positioning, it could be stated a little more explicitly.
P 12 L 368-369 “sigma0” is not defined, and should be at this first appearance. I’m guessing it’s an overall root-sum-squared measure of the image residuals, maybe weighted. The subscript 0 is reminiscent of radar terminology for backscatter cross-section, but I’m sure that’s a coincidence. The expression is used only 5 times in the paper, so maybe use an easily understood phrase like “RMS image residuals” instead, to avoid confusion entirely. Also, units (presumably pixels) should be given for the value of 0.1 that is mentioned.
P 13 Section 4.3. This section involves two different uses of photometric models, with different types of model used for each. That’s fine, but the section is potentially confusing because this is not laid out explicitly. A few added words would greatly help readers follow what was done. The two uses are correction (normalization) of images to produce uniform mosaics and allow slope-corrected comparisons of albedo and color. The second is the photoclinometric modeling used to infer topography from images. The current first sentence of the section makes it seem that correction of images will be the only topic, so I suggest an introductory sentence like what I wrote above stating the two applications of photometric models, followed by the description of what a photometric model entails (current L 372-379). Then at L 380 change “In this study” to “For correction of images to be used in mosaics and color products”. At P 14 L 403 Insert a sentence before “Currently, the ASP…” as follows: “Software limitations made it necessary to use a different photometric formulation for the production of DTMs by photoclinometry.”
Per my earlier comments it would also make sense to reorder this section to discuss the Hapke model for SfS first and the model used for image correction second, with the introductory sentence changed to reflect this order.
P 13 L 302 It’s a minor point, but this leaves me curious what sort of “good results” are meant. Uniformity of image mosaics? Or consistency of the photoclinometry results? This could be made clear by a few words added in the text.
P 13 L 385 For “uniform” I would suggest “universal” is closer to the likely intended meaning.
P 14 L 409. The parameters of the photoclinometry algorithm are entirely distinct from the parameters of the photometric model. I would suggest moving the discussion of how values for the two weighting parameters were chosen to Section 4.6 where those weights (there called mu and lambda) are defined. If this is not done, I would suggest at a minimum to start a new paragraph before “Additionally, two values….”
P 14 Section 4.4. I wonder if the section title “Pre-processing” is the most appropriate description of this section. The early steps are clearly needed as pre-processing before making DTMs by either the stereo or photoclinometry routes. The full set of steps would be needed to prepare the images for incorporation into derived products such as image mosaics and color composites. The most important distinction would be that one certainly shouldn’t perform photometric correction on an image before using it for photoclinometry! Images might or might not be projected to map coordinates before stereo or photoclinometry processing, depending on the software. Certainly one wouldn’t normally project onto a DTM before stereo analysis for the same reason that one wouldn’t photometrically correct before photoclinometric analysis. (Projecting onto the reference ellipsoid is workable because the calculation is easy to invert during the stereo computation, and it has advantages over using the raw image geometry, so it’s not surprising to read later in the paper that this approach was used. Being clear about it the distinction in Section 4.4 would be helpful, though.) Maybe a few additional introductory sentences could make it clearer that this workflow was followed to different endpoints for different uses of the images. In particular, it might help to add to Figure 6 some “output” arrows where images are sent to photoclinometry before the photometric correction, to mosaicking after the final step, and so on.
P 14 L 421 and 423 “Figure ??” should presumably be “Figure 6”.
P 15 L 451 What does “+ Unix*” in the section title signify? The text of the section doesn’t mention Unix, though the mention of a script may be relevant. The meaning of the asterisk is completely unclear. Maybe the Unix/script aspect doesn’t need to be mentioned in the title as long as it’s explained in the text.
P 19 L 617 “However, we did not encounter this problem” is a bit confusing wording, because the antecedent of “this problem” is unclear. It could be read to suggest that albedo variations were not encountered, but the rest of the sentence makes it clear that they were. I suggest rewording the sentence along the following lines: “Multiple images are needed to separate the effects of albedo and surface orientation. Fortunately, our study areas are covered by images obtained under varying lighting and observation conditions.” I would also suggest “better modeled” should be “modeled” on L 620 because, as noted, the modeling can’t be done at all with a single image.
P 19 L 625. I suggest “…photometric model” be changed to “…photometric model, along with weighted terms that penalize excessive roughness and excessive departures from a prior model of the surface.” This will allow the reader to see immediately why there are 3 terms in Equation 7.
P 19 Section 5. My first reaction was that the introductory text of Section 5 needs to say something about how the statistical comparisons were made between DTMs that were produced with different grid spacings. Was the coarser DTM enlarged to the finer one? Or was the fine DTM downsampled to the coarse one? If the later, was it subsampled by extracting values or was some sort of smoothing or averaging of pixels done? These details are important because the difference between any two DTMs can come from differences in what features are resolved, as well as from errors in either or both DTMs.
That said, my reaction after reading further is that the comparisons in 5.1 and 5.2 focus almost entirely on the long-wavelength differences between DTMs. If the local variations are not examined then the details of how the data were resampled for comparison don’t matter for these sections. Instead, it would be good to include a clearer statement in the Section 5 introduction that the comparisons in 5.1 and 5.2 focus on long-wavelength trends whereas 5.3 compares the local artifacts and true details in the DTMs.
In keeping with this emphasis, it would be useful to add a column to Table 5 with another statistic that measures the amplitude of long-wavelength differences. This could be the standard deviation of the difference after smoothing at some spatial scale of choice (maybe 2x or 3x the true resolution of the finer scale DTM, so as to eliminate features that are represented in one dataset but not the other). The raw SD is dominated by the local variations that aren’t being discussed in the first subsections. It would even be reasonable to use profiles of the smoothed rather than raw differences in the figures.
PP 19-20. The lat-lon extents of the areas are stated in the Table 4 title and again in the text. It would be clearer (as well as shorter) to provide them in table form. Table 2 could be expanded to include the study area extents as well as the DTM extents and Fig. 1 could show them graphically.
P 20 L 667 Symbols d and D need to be defined at this first use of d/D.
P 21 L 677-678. The triangulation error described in Section 4.5.3 is not a vertical error. It’s the perpendicular miss distance of the stereo rays. The two types of error do go somewhat hand in hand, but the geometry of the image set (how strongly do the rays converge on the ground point?) also plays a factor. Triangulation errors and vertical errors shouldn’t be directly compared, but the fact that the triangulation errors are kept to a subpixel level does place a limit on the vertical errors. Thus, I agree with the choice of the SPG DTMs as reference. Not only are some images not ideal for photoclinometry (as stated), the photoclinometry calculation lacks the geometric rigor of the SPG one.
P 21 L 685-690. There are a couple of problems with this paragraph. First, the difference statistics are presented in Table 5 not Table 4. Second, why repeat in the text what is in the table? It would be more concise and more insightful to summarize that the mean differences are <2 m for 68 m/pixel images and the standard deviations are ~30 m. (Similar remarks apply to the next section, where L 720-724 repeat information in Table 5.) Third, the text seems to give means but no standard deviations for areas 2 and 3 but these areas are not listed in the table. If the statistics are available, they should be in the table and the standard deviations should be given along with the means.
P 21 L 608. I suggest “spatial dependence” -> “long-wavelength spatial dependence”. This, also describable as overall distortions of the DTM surface, is an important aspect to check. Averaging over one coordinate at a time is one way to do this. Another that might be even more helpful would be to present the difference between DTMs in grayscale form, and possibly smoothed to remove the local fluctuations so the long wavelength variations can be seen clearly. In any case, these distortions are a small fraction of the pixel size, as noted (L 711-712) which is gratifying.
It would also be of interest to examine the local fluctuations of the difference to see if they can be attributed to the resolution of the DTMs or to artifacts, and whether the artifacts can be attributed to one or the other DTM being compared.
P 21 L 714-715 “SPC DTMs differ from the baseline SPG DTMs by only a few meters” is an over-claim. This needs to be qualified by saying “apart from local variations, which are on the order of 30 m RMS.”
P 22 L 733 “Significant discrepancies” -> “Significant long-wavelength discrepancies”. Indeed, these long-wavelength differences are much more striking than one would expect from the larger standard deviation alone.
P 22 L 738 “no long-wavelength discrepancies” is an overstatement. They are ≤ something like ±2 m (based on my reading of the figure) but not zero.
P 24 L 788 I suggest “the crater” -> “the small crater” to distinguish from Occator Crater, which was the subject of the previous sentence.
P 25 Figure 11. First of all, this is a lovely and very informative figure! I am puzzled by two things in this figure. Explanations could be added to the text if they are understood. First, why does the resolution of the F1CLEAR image vary from one row to the next, when it’s the same image projected on different surfaces? Is each orthoimage prepared at the grid spacing of the DTM used and then enlarged to appear the same size in the figure? Second, why are the artifacts on the east side of the crater so much stronger in some of the RGB photomet images than in the corresponding clear images? Does this have to do with the color images being taken under different conditions such that this side of the crater was more shadowed?
P 26 Figure 12c could be eliminated if it is necessary to shorten the paper. Figure 13 shows the profiles more clearly, though this figure does allow the reader to see the “envelope” of overall variation between results. Parts a) and b) are needed to explain how the profiles were averaged, so they should not be eliminated. Retaining Fig. 12c and eliminating Fig. 13 would save even more space, but a table would have to be added to contain the depth and diameter measurements, so this is not an attractive option.
P 26 L 810 Should “lower multi-temporal coverage” be “lower resolution multi-temporal coverage”? The essence of the better performance seems more likely to me to be the inclusion of some images in which these areas are out of shadow, rather than the better resolution (relative to image resolution) of photoclinometry.
P 27 L 846-848. This sentence strikes me as slightly odd. Why mention statistical significance? Not only is it unlikely that readers will worry about statistical significance with a sample of one, I would think that the significance of a systematic depth error would depend on the precision expected for the profile as much as on the number of craters examined. Perhaps a less prejudicial way to summarize the result would be that the JPL SPC DTM departs visibly from the other DTMs in crater depth but the discrepancy is only 3.6%. This suggests (without proving conclusively) that the depths of well-resolved craters are not substantially over- or under-estimated in this dataset. (If there is a published suggestion that they are, it could be cited here, with the indication that the effect wasn’t observed in this limited dataset.) Of course, such vertical scaling error, if it exists, would only affect features in a middle range of sizes. Large features will be constrained to stay near the initial (SPG) DTM used. Smaller features may be defined mainly by the photoclinometry process but if they are very small they will be smoothed regardless of the accuracy of that process.
P 31 L 900-905 This text is largely redundant with the figure caption. It could be reduced to a statement that Figure 15 shows the Cerealea Tholus area in detail along with the location of profiles shown in Figure 16.
P 32 L 912-913 The wording “we used our highest-resolution CSL/CXL/XMO7 ASP SPC DTM as a reference and plotted deviations from the remaining six DTMs” agrees with the legends in the plot, which are of the form [reference DTM] – [other DTM]. However, the difference curves are low where the other DTM profile is below the reference profile in the top row. This is “deviation of” rather than “deviation from” and corresponds to [other DTM] – [reference DTM]. I definitely find it more natural to subtract the reference from the dataset being compared, so don’t change the plots, change the legends and text to agree with what is shown.
P 33 L 909-973 I would (again) probably use a phrase like “not necessarily representative” rather than “not statistically significant” for results from looking at a single crater. I’m not sure how you might do a statistical test even if you looked at depths of more craters. This passage does bring in the idea that results may depend on crater diameter. I agree strongly, but I think some known effects could be described here explicitly and would improve the discussion. Craters that are small in relation to the true resolution of the DTM are going to be poorly represented and too shallow. In addition to this direct effect of resolution, the curvature term in Eq. 7 will act to suppress smaller craters more than larger ones. For the same d/D (and thus the same slopes) small craters will have higher curvature and be penalized more. At the other extreme of crater size, the third term of Eq. 7 ensures that very large craters will have depths that are determined mainly by the starting DTM (phi_0 in the equation). Therefore any errors that suppress depths of smaller features (wrong photometric function, incomplete convergence, or too much smoothing, to name three) will not affect the largest craters as much. It would be great if these ideas could be expressed in this paragraph.
I don’t understand that idea that is stated on L 971-973, that underestimation of depth caused by shadowing will be greater in larger craters. Above a threshold size, larger craters generally have smaller d/D so less of their floor would be expected to be in shadow. It seems to me that this should decrease any related error.
P 33 L 973-976 I agree strongly with this concluding sentence.
P 33 L 978- 981 Again, I agree with the claim that good results were obtained in this area of albedo contrasts and agree strongly that this is important. However, the discussion misses an opportunity to quantify the effect and emphasize that the SfS algorithm has been shown to perform well in the face of much more dramatic albedo contrasts than are encountered on most bodies (or in past tests like the Dawn “wall” test). I suggested above that the introduction emphasize that Vinalia and Cerealia are respectively about 4x and 10x higher albedo than the background material. This could be contrasted quantitatively with the typical fractional contrast from topographic features. For example, the authors could calculate and describe the fractional contrast of the photometric law for a 1° slope at typical incidence angles. Or they could just measure the relative contrast in an area outside the faculae (i.e., ratio of standard deviation to mean in an image area with topographic features). This is likely to be <1 whereas the albedo ratio is >1, making it a strong contrast. A third possible comparison would be with typical albedo ranges for Mars or the Moon that might be quoted from the literature. I believe these are only about a factor of 2 (e.g., highlands/maria or Tharsis/Syrtis) and local albedo features may be even more subtle. So the albedo variations in this study’s area are (a) much greater than the brightness variations from which slopes are estimated, so they would lead to severe distortions if neglected, and (b) much greater than typical on other bodies. Thus, this is a powerful demonstration of the utility of multi-view SfS.
P 33 L 1001 DTM resolution is _at least_ 3-5x coarser than image GSD; reference 73 indicates the ratio is often larger.
P 34 L 1015-1035 This is another passage in which the values in the tables could be summarized more broadly to bring out the main conclusions, e.g., standard deviations of 30-80 m but mean differences of only a few m.
P 34 Section 6.3 describes the issue that there can be effects from the choice of photometric model but doesn’t contain any quantitative results. The section could be omitted for brevity or some effort could be made to draw some quantitative conclusions. For example, the seminal paper by McEwen (1991, Photometric functions for photoclinometry and other applications, Icarus, 92, 298-311) describes how closely the lunar-Lambert model (with correctly chosen parameters) agrees with the Hapke model. Perhaps some conclusion can be drawn from the residual artifacts in the corrected images in this study about how closely the Hapke and Shkuratov-Akimov models agree?
PP 35-36 Section 6.4 is much more impressive than 6.3. This comparison clearly and quantitatively shows how the present results improve on those from SPG that were previously published and avoid major distortions by albedo variations that are still quite substantial. As a minor quibble, I think that it would be OK to omit any discussion of the average DTM differences, since SfS/PC methods don’t provide any independent information about absolute elevation. Can you quantify the conclusions about possible albedo-related slope errors? The long wavelength deviations are ±10 m and ±140 m over ~4 km, so slopes are in error by 0.3° or 4° in the respective datasets. How much fractional difference in model brightness does this correspond to, for typical incidence angles? Rather than putting any effort into getting the right photometric model, I’ll evaluate for a Lambertian model at 30° incidence (typical in Table A1). The fractional brightness differences are 0.3% and 4% respectively; other photometric functions generally have less contrast than the Lambertian, so the realistic brightness differences will be smaller. It’s clear the current SfS model does much better than the published SPC model, but even the latter is excluding most of the albedo variation (factor of several) and only allowing a few percent variation to influence the topography. The authors could redo this calculation carefully and with the right photometric law and quote the impressive results in this section.
P 36 Fig. 17a A minor point, but in the legend at the bottom of this panel the JPL SPC curve should be described as published, not “this study.”
P 36 Maybe there should be a brief Section 6.5 added, discussing that the DTMs produced were used to make photometrically corrected color mosaics (Fig. 1?) and summarizing how the improved resolution and registration accuracy reduced the visible errors (referencing Figs. 11 and 14).
P 37 Section 7 The Conclusion is excellent. To strengthen it further, maybe add a sentence at L 1151 reminding readers that the albedo variations corrected for were very large compared to those in past studies and on other (rocky) bodies. Thus, this study not only provides confidence in the new DTMs reported, it suggests even more strongly that SfS with albedo modeling witll produce useful results for other planetary bodies. Likewise, at L 1158 you could justifiably add a sentence to the effect that “We have provided a detailed description of our methodology for using these tools to aid future researchers who may apply them in other contexts.”
P 39 There are several ordering problems in the Appendix. The title (Appendix A) should appear before the tables. The table of acronyms probably deserves to be labeled as a separate Table A2 rather than being appended to A1. The remaining Appendix tables appear after the References for some reason (PP 45-49) so the ordering of these materials needs to be corrected.
Author Response
Thank you for taking the time to read and review our manuscript. We have done our best to address the comments and recommendations in most cases. For practical reasons, we have created a PDF in which we respond to the comments of all three reviewers collectively.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript constitutes a complete overview of methodologies applied to data obtained for Ceres by the Dawn mission, as well as a complete description of the implementation of freely available software for scientific handling and use of these data. The manuscript is well written and organized, and it is fully in the scope of Remote Sensing. Because the manuscript is focused on an specific feature (Occator crater and associated structures), a bit more discussion of the scientific problematic and aplications would improve it significantly.
Author Response
Thank you for taking the time to read and review our manuscript. We have done our best to address the comments and recommendations in most cases. For practical reasons, we have created a PDF in which we respond to the comments of all three reviewers collectively.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThe accurate topography and orthorectified image data are significant to plan and prepare the potential landing sites in Occator crater. In this work, the authors demonstrate the capabilities of the freely available and open source USGS ISIS and ASP in creating high-quality image data products as well as SPG and SPC digital terrain models of the region with spectroscopically challenging features. The results provided the improved stereophotogrammetric and stereophotoclinometric DTMs of Occator crater and its interior cryovolcanism-related bright spots. This is a good work to reconstruct the topography of the planetary surface and the description of the methods is detailed.
My main comments are listed as follows.
1. Please provide the reference ellipsoid to generate the DTMs.
2. Section 4 is lengthy. The expression of the contents is more like a technical process but not a scientific introduction. I suggest to express the content using one technical flowchart and provide the necessary equations to the key methods.
3. Section 5.4 is lengthy. (1) I could not understand what the authors want to express in this section. After all, Sections 5.1 to 5.3 have demonstrated the effectiveness of the methods used in this paper. (2) The authors should clarify the improvement in scientific views by using the new datasets.
4. The usage of the referred Figures and tables in the paper is problematic in several. E.g., “… are listed in Table ??.” in Line 269.
5. Section 4.1: There occurs “Bad/Warm Pixels” in the original dataset. Please give the scientific way to look for and remove the bad/warm pixels.
Author Response
Thank you for taking the time to read and review our manuscript. We have done our best to address the comments and recommendations in most cases. For practical reasons, we have created a PDF in which we respond to the comments of all three reviewers collectively.
