A Method for Extracting Some Key Terrain Features from Shaded Relief of Digital Terrain Models
Round 1
Reviewer 1 Report
This approach seems to have a good quality of classification and it seems to be ready to be applied in the production line of geomorphologists.
The presentation is long (43 pages), which is acceptable since this proposal introduces a considerable change in the established practises. If other reviewers suggest major/minor changes, then you could shorten the presentation by moving some of the graphic material to supplementary material. My suggestion is that synthetical examples (graphics) and some of the real world examples (graphics) could be supplementary, and then you add a concise table etc. listing the observations from them. But this is conditional (if major/minor revisions need to be done because of other rerviewers).
I would have liked to see a deeper review of curvature based methods. I understand it would bloat the presentation too much, but your future publications could check if parts of your computations could be vectorized and parallelized e.g. by methods presented in:
A compiler approach to map algebra: automatic parallelization, locality optimization, and GPU acceleration of raster spatial analysis (2017)
Jesús Carabaño, Jan Westerholm, Tapani Sarjakoski
| GeoInformatica |
This kind of partial vector computation would reduce the time complexity closer to curvature computation speeds. (This is a suggestion for your future work only).
Author Response
Response to Reviewer 1 Comments
This approach seems to have a good quality of classification and it seems to be ready to be applied in the production line of geomorphologists
Point 1:
The presentation is long (43 pages), which is acceptable since this proposal introduces a considerable change in the established practises. If other reviewers suggest major/minor changes, then you could shorten the presentation by moving some of the graphic material to supplementary material. My suggestion is that synthetical examples (graphics) and some of the real world examples (graphics) could be supplementary, and then you add a concise table etc. listing the observations from them. But this is conditional (if major/minor revisions need to be done because of other rerviewers).
Response 1:
Thank you for the suggestions. The current length of the paper is 35 pages which later the length will be smaller when we update the size of the images. Since we are using all results in the discussion, we decided to do not move them to the supplementary material section.
Point 2:
I would have liked to see a deeper review of curvature based methods.
Response 2:
A paragraph explaining how curvature-based method identifies ridgeline is added
(see lines 288-96)
Point 3:
I understand it would bloat the presentation too much, but your future publications could check if parts of your computations could be vectorized and parallelized e.g. by methods presented in:
A compiler approach to map algebra: automatic parallelization, locality optimization, and GPU acceleration of raster spatial analysis (2017)
Jesús Carabaño, Jan Westerholm, Tapani Sarjakoski This kind of partial vector computation would reduce the time complexity closer to curvature computation speeds. (This is a suggestion for your future work only).
Response 3:
The conclusion section is updated
(see lines 645-647)
Author Response File: 
Author Response.docx
Reviewer 2 Report
the paper is revised based on the comments of the reviewers.
Author Response
Response to Reviewer 2 Comments
the paper is revised based on the comments of the reviewers.
Author Response File: Author Response.docx
Reviewer 3 Report
Dear Authors,
Your manuscript entitled: "A method for extracting some key terrain features from shaded relief of digital terrain models" is in a good shape.
While the manuscript is quite well written there are some points that need to be addressed. The figures that you present need to be strengthening in order to be consistent with the reader comprehension. Regarding the geo-elements you would like to characterise in your manuscript, you need to provide a small analysis for the reader. There are duplicates that need your attention. In addition, the text needs editing cause there are some errors. There are some sentences underlined which need better definition. There are numerous parts that you are repeating the same text in your manuscript. Same with some figures, as a result I could not understand which is the correct that you are willing to publish. Please avoid it. Some cross-references are mistaken and need to be corrected. There are some underlined parts which are really confusing. Please check the following points:
l.28 - GIS in your keywords but you are not mentioning that you used in your manuscript.
l.31 - there is an error in text form
l.38 - methods was.. change to ..methods has been..
l.80 - you are mentioning geomorphon but you are posing the reference on a later stage. (please check l.329 - 332)
l.133-134 - Not clear what the authors are describing.
l.136-144 - the description of your paper is not needed in the manuscript.
l.226 - Why did you choose 44o?
Flowchart is not clear for the reader
l. 235 - not clear the equation for q and p, please define
l. 280-289 - not clear. please rephrase
Figure 5,6 - a and b looks exactly the same please correct
regarding the altitude is it correct to pose it as altitude or sun (light) angle?
figure analysis should be followed right after from the exact figures.
Your illustrations need better analysis in the text.
Evaluation metrics numbering is not cross-referenced in the text.
It should be nice to see the manual dataset in order to have an overview of the reference dataset.
In your evaluation process there are some negative values, what are representing and what does it mean? I am quite sceptic with your current evaluation values cause there are really low for a valid procedure.
Legend in figures should be follow the same pattern. Please use the same amount of decimals to all.
In the discussion part you mentioning that your proposed method outperformed the other three which in your evaluation is not true. Please provide a more realistic approach.
Table 5 does not put any further knowledge for the reader. It needs some corrections and I suggest to transform it as specific points rather than in the current format.
l.587 - How cab you quantify the sensitivity you are suggesting?
l. 639-640 - In which sense the approach is efficient and effective?
l. 683-684 - Please avoid this kind of sentences cause they are too vague.
Please check the references. Some of them need modifications. (eg l.760)
Author Response
Response to Reviewer 3 Comments
Your manuscript entitled: "A method for extracting some key terrain features from shaded relief of digital terrain models" is in a good shape.
While the manuscript is quite well written there are some points that need to be addressed. The figures that you present need to be strengthening in order to be consistent with the reader comprehension. Regarding the geo-elements you would like to characterise in your manuscript, you need to provide a small analysis for the reader. There are duplicates that need your attention. In addition, the text needs editing cause there are some errors. There are some sentences underlined which need better definition. There are numerous parts that you are repeating the same text in your manuscript. Same with some figures, as a result I could not understand which is the correct that you are willing to publish. Please avoid it. Some cross-references are mistaken and need to be corrected. There are some underlined parts which are really confusing. Please check the following points:
Point 1:
GIS in your keywords but you are not mentioning that you used in your manuscript
Response 1:
See line (117, 127).
Also, in all methods we used GIS tool
(see line 301-304)
Point 2:
l.31 - there is an error in text form
Response 2:
Fixed
(see lines 31-32)
Point 3:
l.38 - methods was.. change to ..methods has been..
Response 3:
Updated
(see line 35)
Point 4:
l.80 - you are mentioning geomorphon but you are posing the reference on a later stage. (please check l.329 - 332)
Response 4:
References related to geomorphon were fixed
Point 5:
l.133-134 - Not clear what the authors are describing.
Response 5:
Updated
(see lines 99-101)
Point 6:
l.136-144 - the description of your paper is not needed in the manuscript.
Response 6:
We included this structure as it is required by most journal papers.
Point 7:
l.226 - Why did you choose 44o?
Response 7:
Explained.
(see lines 228-231)
Point 8:
Flowchart is not clear for the reader
Response 8:
Flowchart is updated
(see figure 2)
Point 9:
- 235 - not clear the equation for q and p, please define
Response 9:
Clarified
(see lines 181-188)
Point 10:
- 280-289 - not clear. please rephrase
Response 10:
Updated
(see lines 228-231)
Point 11:
Figure 5,6 - a and b looks exactly the same please correct
Response 11:
Figure 6 is updated, detected pixels were converted onto lines. In figure 5 a and b are different.
(see Figures 5 and 6)
Point 12:
regarding the altitude is it correct to pose it as altitude or sun (light) angle?
Response 12:
The word altitude referred to altitude of illumination source (sun (light) angle). Paper is updated.
Point 13:
figure analysis should be followed right after from the exact figures
Response 13:
The position of the figures was changed.
Point 14:
Your illustrations need better analysis in the text.
Response 14:
Updated
(see lines 166-176)
Point 15:
Evaluation metrics numbering is not cross-referenced in the text.
Response 15:
The result section with Evaluation metrics were updated
(see section 5)
Point 16:
It should be nice to see the manual dataset in order to have an overview of the reference dataset.
Response 16:
Map with manual dataset will be added
(see Figure 12)
Point 17:
In your evaluation process there are some negative values, what are representing and what does it mean? I am quite sceptic with your current evaluation values cause there are really low for a valid procedure.
Response 17:
Negative Cohen's Kappa coefficient values indicate no agreement (McHugh, 2012).
(see line 403)
Point 18:
Legend in figures should be follow the same pattern. Please use the same amount of decimals to all.
Response 18:
All maps were updated
Point 19:
In the discussion part you mentioning that your proposed method outperformed the other three which in your evaluation is not true. Please provide a more realistic approach
Response 19:
The discussion section describes evaluation of synthetic dataset where proposed method outperformed other methods
(see lines 593-596)
Point 20:
Table 5 does not put any further knowledge for the reader. It needs some corrections and I suggest to transform it as specific points rather than in the current format.
Response 20:
Table 5 is updated
Point 21:
l.587 - How cab you quantify the sensitivity you are suggesting?
Response 21:
Explained
(see lines 568-570)
Point 22:
- 683-684 - Please avoid this kind of sentences cause they are too vague.
Response 22:
The sentence was deleted
Point 23:
Please check the references. Some of them need modifications. (eg l.760)
Response 23:
The references are updated.
References:
McHugh, M. L. (2012) ‘Interrater reliability: The kappa statistic’, Biochemia Medica. doi: 10.11613/bm.2012.031.
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
Dear Authors,
Please remodify and correct some minor parts in your manuscript regarding the syntax. Your manuscript has been updated successfully but minor changes are needed.
Author Response
Point 1:
Please remodify and correct some minor parts in your manuscript regarding the syntax. Your manuscript has been updated successfully but minor changes are needed.
Response 1:
The manuscript has been updated the issues regarding the syntax are fixed.
Author Response File: Author Response.docx
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
Review of the paper “A method for extraction of key terrain features from shaded relief” byy Syzdykbayev et al. submitted to Remote Sensing.
This paper presents and novel method to extract some specific terrain features from DEM and shaded relief derived rasters. In my opinion the paper is interesting and I enjoyed reading however I have important doubts about supporting the publication of this work. My main concerns are introduced below:
-Your method is limited to terrain features that are assumed as lines which is not always true in real world features. In fact your method works in a very nice way for synthetic data with lines clearly defined, but is not the best when you use real world data. This limits the usefulness of the proposed methodology.
-On the other hand, your method just detect three features against other methods that outperform this. For example, geomorphon gives a much richer information. The processing time is also an interesting point that is not mentioned in the paper. A deeper discussion is needed in the key drawbacks of your method: processing time and the number of features detected against other methods.
-The results you show in Tables 1 and 2 and figures 8 and 9 are influence by the way you or your two operators digitize or understand landforms or features, if you digitize features as lines, your method works better for these features, however if you allow your operator to delineate ridges, for example, as polygons, you'll see that, for example, the probabilistic or the curvature method works better. On the other hand, curvature method is based on the selection of a specific threshold but no sensitivity analysis about the influence of the threshold is shown.
-I do not see very useful to use not real features to compare the performance of your method, I would just keep this as a part of the methods section to show the performance of you proposed methodology or, if you prefer keep this analysis as supplementary material. You should support your results in the real world data.
-When you use real world data your method is not the most accurate so what is the reason to use your methodology instead of the existing methods? this should be the basis for your discussion section.
-Some figures should be improved.
-I attach a pdf file with some other minor corrections or suggestions.
Comments for author File: 
Comments.pdf
Author Response
Response to Reviewer 1 Comments
This paper presents and novel method to extract some specific terrain features from DEM and shaded relief derived rasters. In my opinion the paper is interesting and I enjoyed reading however I have important doubts about supporting the publication of this work. My main concerns are introduced below:
Point 1:
-Your method is limited to terrain features that are assumed as lines which is not always true in real world features. In fact your method works in a very nice way for synthetic data with lines clearly defined, but is not the best when you use real world data. This limits the usefulness of the proposed methodology.
Response 1:
Our experimentation results show that the performance of the proposed method is similar to the geomorphon method, one of the state-of-the-art methods. We have explained the reasons why the curvature and probabilistic methods result in a higher accuracy than the proposed and geomorphon methods.
(see lines 540-560)
Point 2:
-On the other hand, your method just detect three features against other methods that outperform this. For example, geomorphon gives a much richer information.
Response 2:
The output of the geomorphon method has been discussed in the experimental setup section.
(see lines 272-286)
Point 3:
The processing time is also an interesting point that is not mentioned in the paper. A deeper discussion is needed in the key drawbacks of your method: processing time and the number of features detected against other methods.
Response 3:
A deeper discussion about drawbacks of the proposed method and a new table with processing time have been added to the discussion and the experimental setup sections.
(see lines 531-581)
Point 4:
-The results you show in Tables 1 and 2 and figures 8 and 9 are influence by the way you or your two operators digitize or understand landforms or features, if you digitize features as lines, your method works better for these features, however if you allow your operator to delineate ridges, for example, as polygons, you'll see that, for example, the probabilistic or the curvature method works better.
Response 4:
Ridges and spurs are lines. The output of each method is raster and the thickness of the detected lines varies based on the method. For this reason, we do not convert the output of each method into line polygons. The comparison has been conducted in a raster form.
(see lines 270-286)
Point 5:
- On the other hand, curvature method is based on the selection of a specific threshold but no sensitivity analysis about the influence of the threshold is shown.
Response 5:
The discussion about sensitivity analysis and how it influences the threshold value has been added.
(see lines 520-526)
Point 6:
-I do not see very useful to use not real features to compare the performance of your method, I would just keep this as a part of the methods section to show the performance of you proposed methodology or, if you prefer keep this analysis as supplementary material. You should support your results in the real world data.
Response 6:
The discussion section has been updated.
We conducted an experiment on both synthetic and real-world data. In addition, some of the synthetic data were generated to be close to the real-world data.
(see Figure 10).
Point 7:
-When you use real world data your method is not the most accurate so what is the reason to use your methodology instead of the existing methods? this should be the basis for your discussion section.
Response 7:
We show that the performance of the proposed method is similar to the performance of the geomorphon method, one of the state-of-the-art methods.
(see lines 553-560).
We also explained why the manual detection has a higher accuracy than the proposed and geomorphon methods.
(see lines 540-546).
Point 8:
-Some figures should be improved.
Response 8:
Maps and figures have been updated.
Point 9:
-I attach a pdf file with some other minor corrections or suggestions.
Response 9:
The corrections and suggestions from the attached PDF file have addressed.
(see table below).
Corrections and suggestions from the attached pdf document
|
Location (on the attached pdf document) |
Comments |
Response |
|
Line 2 |
who says that these are key and the other aren't, why key, I do not agree with this word in the title, to be honest you'll need to specify what features are you extracting, because you are only detecting 3 types of features. |
The title has been updated. |
|
Line 11 |
In my opinion the abstract takes many words to set the scene, however it should be concise and use just two or three introductory sentences, then, you should explain your method because this is the core of your paper. Finally, it is mandatory that you show the most important results showing some figures or statistics. |
The abstract has been updated. |
|
Line 12 |
better "digital terrain analysis" |
The abstract has been updated. |
|
Line 15 |
do you mean "landscape or surface evolution"? |
The abstract was updated |
|
Line 17 |
seems that you are forgeting the use of OBIA
|
The abstract has been updated and OBIA has been discussed in the introduction and related works section. (see lines 61-88).
|
|
Line 41 |
when you say sources it seems that you are going to present techniques such as photogrammetry, radar, laser (LIDAR or TLS), etc., however you introduce platforms instead of techniques. |
The abstract has been updated.
(see lines 40-46). |
|
Line 48 |
you may also resample a point cloud, here you only show the good part of DEM against point clouds but you forget that DEMs are 2.5D surfaces and the inability of DEMs to show walls, cliffs or vertical surfaces where you have more than one Z value for a XY location.
|
Additional discussion about DEM and point clouds has been included. (see lines 51-54). |
|
Line 56 |
vertical cliffs? I am expecting to see how do you handle that with a DEM, if you are not showing the case of vertical cliffs you should specify here what kind of cliff.
|
Yes, vertical cliffs. (see line 61).
The detection of cliffs has been added to section 3.5. (see lines 239-248). |
|
Figure 1 |
I would recommend to use a layer with the sea to put over the hillshade, in this way you avoid the triangles in the hillshade. I also recommend you the use of elements in the map: scale, north arrow, location map or geographic coordinates.
|
Figure 1 has been updated. |
|
Line 96 |
are you anticipating results in the introduction section? please delete or move this sentence to the proper section.
|
The introduction section has been updated. |
|
Line 102 |
are you anticipating results in the introduction section? please delete or move this sentence to the proper section.
|
The introduction section has been updated. |
|
Line 108 |
discussed |
Done. (see line 112). |
|
Line 114 |
DEMs have or DEM has |
Done. (see lines 131-132). |
|
Line 151 |
DEM or DTM? |
When we use DEM and when we use DTM have been explained. (see lines 115-117). |
|
Line 154 |
what are false terrain features? have you introduced this term previously in the paper?
|
The false terrain features are introduced and have been explained in Section 3.2. (see lines 217 -232). |
|
Line 191 |
to give more weigth or to delete false features?
|
The false terrain features are introduced and have been explained. (see lines 226-232). |
|
Line 194 |
providing these experiments as supplementary material would be nice.
|
The results of the experiment have been added. (see figures 16-17). |
|
Line 268 |
this is methododology, I do not understand why is in results section? |
This paragraph was moved to the experimental setup section. (see lines 291-299). |
|
Line 285 |
as curvature method is highly dependent on threshold selection I do not see fair comparison. |
The threshold value in curvature has been discussed. (see lines 520-526). |
|
Line 353 |
Many of your results depend on the way you digitize features, if you use lines, your method outperform the others, however if you digitize features as polygons the probabilistic method works better. |
The comparison was conducted in a raster data type, not lines or polygons. (see lines 270-286). |
|
Line 376 |
not true, avoiding Geomorphon (which is not comparable in my opinion) you method is almost always the less accurate. |
Lower accuracy has been explained in the discussion section. (see lines 540-560). |
|
Figure 10 |
This figure is not easy to understand, explain why curvature and probablistic show the three features at the same time. The correspondence between real world features is not possible with the current figure. -Your scales in the left part are different, why? please use uniform scales if you are showing the same portion of terrain. Please use more standard divisions, i.e. 250 m, 500 m, 750 m -Would be possible to change vertical scale in the 3D blocks? I guess it also would be nice to show the viewshed together with the DEM to provide a better view of terrain morphology. |
Maps and figures have been updated. |
|
Figure 11 |
It is very difficult to see with this kind of figure where every method and manual match, probably you could use one color (for example red) where manual and proposed match and green where we only have proposed and yellow where we only have manual. In the current way it is hard to get some conclusions from your figure, you may also apply this strategy to the other zoomed sub-figures. Please see also some advices given in the previous figure that may be also helpful in this one.
|
Maps and figures have been updated. |
|
Line 410 |
I wouldn't use synthetic data to support results |
Done. The discussion section has been updated. |
|
Line 425 |
this is the key you are using lines instead of polygons |
The comparison was conducted in a raster data type, not lines or polygons. (see lines 270-286). |
|
Line 428 |
I do not agree, just for not real world features
|
We conducted an experiment on both synthetic and real-world data. In addition, some of the synthetic data were generated to be close to the real-world data. (see Figure 10). |
Reviewer 2 Report
this paper proposed a method to extract terrain features from the shaded relief map. overall, the paper is well written and the method is usefull. However, the following problems should be solved before its acceptance.
- Background and Related Works should be further expanded. the present contents are too simple.
- Give a diagram to show the terrain features including ridges, spurs, cliffs and peaks. What is the spur?
- How to derive the results in Figure 4?
- P9 Line246: Figures 8 and 9?
- P364: Table 1? Table 2?
- Figures 10-14: 2D figures are more proper. The features with the line format are more popular than the others. hence, I suggest that the extracted features should be transformed into line format to better show the performance of the proposed methods and the others.
- How the proposed method can deal with both ridge and valley lines?
Author Response
Response to Reviewer 2 Comments
this paper proposed a method to extract terrain features from the shaded relief map. overall, the paper is well written and the method is usefull. However, the following problems should be solved before its acceptance.
Point 1:
Background and Related Works should be further expanded. the present contents are too simple.
Response 1:
The background and related works section have been expanded.
(see lines 114-171).
Point 2:
Give a diagram to show the terrain features including ridges, spurs, cliffs and peaks.
Response 2:
The full diagram of the workflow has been added.
(see Figure 2)
Point 3:
What is the spur?
Response 3:
The definition of the spur has been added.
(see lines 69-74).
Point 4:
How to derive the results in Figure 4?
Response 4:
The extraction of the cliffs has been explained (section 3.4) and is shown in the diagram.
(see Figure 2 and lines 238-247).
Point 5:
P9 Line246: Figures 8 and 9?
Response 5:
Done.
(see line 304).
Point 6:
P364: Table 1? Table 2?
Response 6:
Done.
(see lines 435-438).
Point 7:
Figures 10-14: 2D figures are more proper. The features with the line format are more popular than the others. hence, I suggest that the extracted features should be transformed into line format to better show the performance of the proposed methods and the others.
Response 7:
Maps and figures have been updated.
Point 8:
How the proposed method can deal with both ridge and valley lines?
Response 8:
The proposed method can only detect some terrain features such as ridges, spurs, peaks, and cliffs. This is a drawback of the proposed method in comparison with the geomorphon method.
(see lines 272-286)
Reviewer 3 Report
The basic idea is simple and interesting, and this method surely is worth to implement in the GIS software. I tried to find corresponding "complete" curvature and slope based methods, but did not find anything worth adding as a reference.
Limitation to 8 orientations x 2 illuminaton angles seems to be enough, and that is the main result of the article. Of course, there should be more applications, new and old, to actually judge this method, since it is very close to some of the existing methods.
There should have been a short presentation of the computational complexity: how fast is the new method when compared to existing methods (I assume it is comparable or better in most cases). Include either complexity analysis or timed results of one of the examples! This is conditional though: if other reviewers do not require new revisions, then this requirement can be omitted.
Author Response
Response to Reviewer 3 Comments
The basic idea is simple and interesting, and this method surely is worth to implement in the GIS software. I tried to find corresponding "complete" curvature and slope based methods, but did not find anything worth adding as a reference. Limitation to 8 orientations x 2 illuminaton angles seems to be enough, and that is the main result of the article. Of course, there should be more applications, new and old, to actually judge this method, since it is very close to some of the existing methods.
Point 1:
There should have been a short presentation of the computational complexity: how fast is the new method when compared to existing methods (I assume it is comparable or better in most cases). Include either complexity analysis or timed results of one of the examples! This is conditional though: if other reviewers do not require new revisions, then this requirement can be omitted.
Response 1:
A table showing processing times has been added to the discussion.
(see lines 568-582).
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
To be honest, I always support the first round of reviews in a paper because I think that everybody has the right to discuss about their works and to answer reviewers. I think this producedure is a fairer one. In this case I think that questions and concerns I set out in the first round were not fulfilled. Authors responses are vague and far from a response to reviewers proper of a journal like Remote Sensing. Let's take the example of my points 1 and 2. I said that your method is similar in accuracy to other existing methods, like geomorphon, however your method only detects three features against geomorphon that detects many more. I wanted you to justify in the discussion why I should use your method instead geomorphon and your answers were: the performance is similar and the performance of geomorphon has been discussed. This is very poor in my opion. With this circunstances I have to support the rejection of the paper.
The new table with processing times is a good step in my opinion to justify why your method is important, however, I see that is faster than geomorphon, but again, you only classify a few features, probably if you take time per features your method is slower. I would recommend the authors for future submission that they should justify why your new methodd is relevant against state-of-the-art existing methods.
Reviewer 2 Report
the following questions should be solved:
- DEM is used as an umbrella name to describe both DTMs and Digital Surface Models (DSMs）--this is wrong. DEM or DTM is different from DSM.
2. the five steps should be denoted in Figure 2.
3. Eq（1）8d or 6d? Check
4. Eq(2) what is the mean of R(p,q)
5 why the values of p=1/sqrt(2) and q=-1/sqrt(2) correspond to the light source in standard cartographic position with 315 azimuth and 45 altitude angle?
6. give a figure to show the azimuth and altitude angle.
7 Eq(3), f is R?
8 each time with a different azimuth orientation-- a figure with sun could clearly explain the different orientations.
9 how the 3*3 filters work?
10 Figure 5: a-c, line overlaid with shaded relief map is more clear
