The Methodological Aspects of Constructing a High-Resolution DEM of Large Territories Using Low-Cost UAVs on the Example of the Sarycum Aeolian Complex, Dagestan, Russia
Round 1
Reviewer 1 Report
Although the use of fixed-wing UAVs is still preferable for lar-ge areas, the study of how to carry out these surveys with multi-rotor UAVs is interesting. Therefore, the paper may be useful to those who intend to address similar issues.
On the other hand, the data needs to be supplemented with more information.
Considering an altitude difference of 197 meters, a well-distributed set of Ground control points is necessary. It is essential to add a diagram showing the position of the ground control points and the overlap between the different adjacent areas.
Were the flights performed at a constant height or following the slope of the terrain to obtain a constant ground sample distance?
Was the average flight height 150 meters, as indicated in 203?
It is important to add information about the achieved GNSS survey accuracy.
It does not appear that well-distributed control points were used, the use of only ground control points does not allow quantification of model deformation, especially in a case where reliable ground truth is not available.
At 219-221 lines, it is not clear how the camera angle was increased.
The designed and actual Ground sample distance are not indicated. Is the "resolution of the cloud" (line 231) the ground sample distance? Considering the available data on DJI Phantom 4 (pixel size 0.002412mm, focal lenght 8.8mm), at 150m Ground Sample Distance should be about 41mm.
Some acronyms are not explained at first appearing.
Author Response
Dear Reviewer.
Thank you for your careful reading of the manuscript and your valuable comments.
Q1: Although the use of fixed-wing UAVs is still preferable for large areas, the study of how to carry out these surveys with multi-rotor UAVs is interesting. Therefore, the paper may be useful to those who intend to address similar issues.
Response 1: Yes, the practice of surveying large areas with fixed-wing UAVs is indeed common, but due to the introduction of restrictions on the flight of aircraft of this type, as well as for safety reasons, it is becoming difficult to carry out such work. The multirotor UAVs are always (for the most part) in sight and in case of danger of collision with something, immediate action can be taken.
Q2: Considering an altitude difference of 197 meters, a well-distributed set of Ground control points is necessary. It is essential to add a diagram showing the position of the ground control points and the overlap between the different adjacent areas.
Response 2: The scheme of ground control point placement was created in beforehand and was adjusted on site to take into account the terrain conditions. Lines 205-218 describe the principle of the arrangement of ground control and ground control points. Geopositioning and verification points locations now presented in Figure 5.
Q3: Were the flights performed at a constant height or following the slope of the terrain to obtain a constant ground sample distance?
Response 3: Since the terrain conditions are difficult and the slope in some areas could reach 33%, to maintain a constant height above the terrain at the time of the work in 2017 was not possible. All popular UAV flight planners now allow to load an altitude model and adjust the flight altitude during the flight mission. In our case, 150 m is the flight altitude over the takeoff point, which was chosen based on the possibility to start and finish the flight without problems, as well as to cover the flight area without affecting the percentage of longitudinal and lateral overlap.
Q4: Was the average flight height 150 meters, as indicated in 203?
Response 4: As already noted, 150 m is the altitude above the takeoff point. The terrain in the study area is variable, so in some areas the height of the UAV above the surface can reach >150 m, and somewhere <150 m. On average over the entire study area, the UAV flight altitude above the surface was 127 m with 60% longitudinal and 55% transverse overlap.
Q5: It is important to add information about the achieved GNSS survey accuracy.
Response 5: The achieved accuracy is 0.1 m in every measured points and explained by the fact that the internal GNSS antenna of the Trimble Geoexlorer 6000 receiver was used for ground control point surveying, using which an accuracy of 0.1 m can be achieved. Taking into account the specifics of the object of research and permissible accuracies for this kind of work (15-20 cm horizontally and 15-25 cm vertically according to the national standard in the field of creating models based on the results of airborne survey, valid on the territory of the Russian Federation), these limitations can be considered acceptable. (added to Lines 239-243)
Q6: It does not appear that well-distributed control points were used, the use of only ground control points does not allow quantification of model deformation, especially in a case where reliable ground truth is not available.
Response 6: In the process of performing post-processing of the obtained results, additional work was carried out to analyze the deformation of the model. Validation points were used to evaluate it. Also in the revised version of the manuscript, information about estimation of absolute errors in comparison with the topographic map of scale 1:10000 was added. (lines 323-358).
Unfortunately, due to the law on state secrets, we cannot present a fragment of the topographic map of this scale in the text of the manuscript.
Q7: At 219-221 lines, it is not clear how the camera angle was increased.
Response 7: For preliminary equalization of UAV survey results, the photogrammetric software starts the process of searching for common points on the photos. As a rule, 3-4 neighboring photos are taken, the neighborhood is controlled by GPS data in the EXIF information of the photos. Depending on the settings the operator sets, the number of common points may vary. Running the floating window, photogrammetric software bypasses the entire study area and gradually adds common points of photos with a set of so-called key points. The floating window is fixed, usually using an area of 10 degrees of the field of view of the central image, relative to which the alignment takes place, for the alignment of neighboring images. This is done to optimize and save computational resources. Increasing the search angle, we increase the number of potentially common points on neighboring images, but decrease the accuracy of their alignment to each other, because common objects on neighboring images are not always captured without distortion (thanks to fisheye lens). The more precisely are coordinates of image centers and the greater is overlap between neighboring images, the smaller, in theory, should be the angle of search for common points to create a high-precision model. In our case, the accuracy of positioning of the image centers, to put it mildly, is not ideal, and the percentage of overlap is variable due to elevation variability.
Q8: The designed and actual Ground sample distance are not indicated. Is the "resolution of the cloud" (line 231) the ground sample distance? Considering the available data on DJI Phantom 4 (pixel size 0.002412mm, focal lenght 8.8mm), at 150m Ground Sample Distance should be about 41mm.
Response 8: We agree that "resolution of the cloud" is not a good term. Replaced by "point spacing" as it meant the distance between neighboring points of the cloud by analogy with the resolution of the DEM, which is the distance between neighboring cells of a regular grid.
Q9: Some acronyms are not explained at first appearing.
Response 9: Corrected
Reviewer 2 Report
Overall comments
The manuscript entitled "The methodological aspects of constructing a high-resolution DEM of large territories using low-cost UAVs on the example of the Sarycum aeolian complex, Dagestan, Russia" discusses the construction of a DEM made through UAV photogrammetry. The construction of 3D models described by the authors covers one of the topics of the Journal. At present it seems more like a Technical Note than a research paper. The innovative aspects are very limited. In addition, there are several aspects that leave me doubtful. The resideui on GCPs are rather high. How do the authors explain these high values? Probably the camera network is weak. I suggest, therefore, that the values reported in Table 2 be reviewed. Also, to evaluate the quality of the model, it is strictly necessary to consider points outside the model (Check Points).
Finally, to increase the level of quality of the paper it is necessary to make other considerations, such as comparing the DEM obtained by UAV photogrammetry with other DEMs in the study area, evaluating the effect of image compression (JPEG, TIFF, RAW), the impact of interpolation techniques in the construction of the model etc.
Minor comments
specify acronyms:
LS (line 22), d-GPS (line 31), UAV (line 67), GNSS (line 83), ICP (line 94)
What is the projection system used in figure 5? Why don't you report the coordinates in a single reference system in Table 2 as well?
Were there any problems or needs to homogenize texture across blocks?
I suggest to rewrite the sentence from line 199 to 201 because it is unclear
Author Response
Dear Reviewer.
Thank you for your careful reading of the manuscript and your valuable comments.
Q1: At present it seems more like a Technical Note than a research paper. The innovative aspects are very limited.
Response 1: We are aware of this. In the revised version of the manuscript, the research portion has been expanded. They are highlighted in yellow for easy tracking of changes.
Q2: In addition, there are several aspects that leave me doubtful. The residues on GCPs are rather high. How do the authors explain these high values? Probably the camera network is weak.
Response 2: It's not just the number of cameras that cover the control point. The terrain conditions sometime make complete absence of cellular communication, which is necessary to obtain real-time kinematic corrections. In addition, it should be kept in mind that the passport accuracy of used GNSS receiver is 10 cm. Such accuracy of the receiver, as well as the achieved error in determining coordinates of the model, including at validation points, is permissible error when conducting works in the field of aerial survey according to the current national standards of the Russian Federation.
Q3: I suggest, therefore, that the values reported in Table 2 be reviewed. Also, to evaluate the quality of the model, it is strictly necessary to consider points outside the model (Check Points).
Response 3: The values of the reference points have been revised and the model has been recalculated. Table 3 was added with the values of the validation points not involved in the geopositioning of the model.
Q4: Finally, to increase the level of quality of the paper it is necessary to make other considerations, such as comparing the DEM obtained by UAV photogrammetry with other DEMs in the study area, evaluating the effect of image compression (JPEG, TIFF, RAW), the impact of interpolation techniques in the construction of the model etc.
Response 4: A comparison with an independent source of relief information has been added. Unfortunately, the available information for this area is a 1:10000 scale topographic map, which differs greatly in the level of detail and relevance from the results obtained from the UAV survey. Concluding the comparison, we can note that the results of the UAV survey satisfy the requirements of permissible error for this kind of survey.
In addition, we added information comparing different methods of interpolation of UAV survey results. We have done such work before, but the results were not given. In the revised version of the manuscript, we explain why the interpolation method used was chosen and give the results of the comparison with other algorithms, see L268-324.
Q5: LS (line 22), d-GPS (line 31), UAV (line 67), GNSS (line 83), ICP (line 94)
Response 5: Corrected
Q6: What is the projection system used in figure 5? Why don't you report the coordinates in a single reference system in Table 2 as well?
Response 6: The coordinate system in the figure is UTM 38N. We have slightly updated the figure to take into account the recalculation of the whole project according to the comments, and to display the model in a different color scheme.
Q7: Were there any problems or needs to homogenize texture across blocks?
Response 7: Since the work on the survey area was carried out in approximately the same weather conditions with approximately the same level of light, no additional work on texture color correction was required. When the textured model was constructed, the texture contrast was adjusted. The only problem we encountered was the homogeneous texture in areas of sand. To compensate this, a ground reference point, which was a 70*70 cm black-and-white panel was placed on such areas and in a number of cases this solution helped.
Q8: I suggest to rewrite the sentence from line 199 to 201 because it is unclear
Response 8: Corrected
Round 2
Reviewer 1 Report
The paper is more clear and improved after the revision and can be published in the current form.
About flying in a slope following mode, I suggest to try UgCS flight planner.
Reviewer 2 Report
The manuscript is much improved over the first version. The author responded to all my doubts and/or requests for clarification. Therefore, I no further requests and, consequently, in my opinion the manuscript can be accepted in the present form.
