the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Feb 2021
GNSS and RPAS integration techniques for studying landslide dynamics: Application to the areas of Victoria and Colinas Lojanas, (Loja, Ecuador)
Abstract. This research tests the application of GNSS and RPAS techniques to the spatiotemporal analysis of landslide dynamics. Our method began by establishing non-permanent GNSS networks on the slope surfaces to perform periodic measurements by differential GNSS. Similarly, RPAS flights were made to acquire high-resolution images, which were oriented and georeferenced using ground control points and structure-from-motion algorithms to obtain digital surface models and orthophotos ultimately. Based on GNSS measurements, the direction and velocity of displacements were accurately calculated, and orthophotos and DSMs were used to calculate horizontal and vertical displacements in a set of significant points throughout the study area, reaching accuracies higher than 0.035 m in the GNSS data and 0.10 m in the RPAS data. These values were within the accuracy required for such studies. Based on the field observations and the results from the photogrammetric studies, the two studied landslides were classified as very slow flows.
This preprint has been withdrawn.
-
Withdrawal notice
This preprint has been withdrawn.
-
Preprint
(2157 KB)
Interactive discussion
Status: closed
-
RC1: 'Comment on nhess-2021-32 - Review', Anonymous Referee #1, 23 Feb 2021
General comments
The manuscript presents a study of landslide movement from a series of campaigns obtained with GNSS and UAS surveys at two test sites. The presentation of the manuscript is good in terms of details on data collection, understanding and interpretation of results. The scientific questions are also relevant to the scope of NHESS. While the methodology adopted in the manuscript generally follows the common practice, it does not provide any novel concept or tool and does not directly attract readers. For example, in order to improve the manuscript the authors should think beyond the fundamental applications of GNSS and UAS processing, they could further explore errors associated with those techniques and the impact those errors can often have on landslide estimation. Alternatively, they could also explore other morphological attributes of DEM to identify cracks and scarps in a more automated fashion. These are some examples to further investigate landslide dynamics with a step away from the conventional geomatics techniques (i.e. GNSS and UAS). I would recommend some major revision of the manuscript to include some more interesting concepts that are currently missing and correct other issues that are included in the specific comments. The positive point in this study is that there are a lot of data collected which is not always feasible. The authors should take this as an opportunity to generate something catchy and interesting providing greater value to the geomorphological research community.
While the authors have nicely given good credit to related work there are few publications worth including in the manuscript that I have commented below in the specific comments. Additionally, there are some recommendations to change the structure of the text and explain the error propagation law earlier within the Methodology section. The UAS camera calibration part is also missing, and SfM self-calibrating bundle adjustment should be mentioned in the Methodology. It would be clearer if there are some figures showing the unstable/stable regions and which targets have been used for GCPs and which for CPs. Some more figures of detailed maps over particular scarps would be also good to add. The abstract is currently under presented. The authors should show in a better way the bigger picture in the abstract (why this study is important) and the novel part of the methodology, with some tangible comparable results to attract the general audience.
Specific comments
Line 43: RPAS also known as UAS too.
Line 50: Strictly Speaking, these processes (i.e. SfM and MVS image matching) constitute a pipeline that does not result only in an orthophoto. The actual output is a dense point cloud, then this is converted into an orthophoto, and a DEM via interpolation methods. The DTM is a by-product after ground classification. The DSM is also a by-product after some additional processing to create the surface model without noisy points. As a generic term I would recommend to use DEM throughout the manuscript.
Figure 2. Caption: What do you mean by photograph February 2016? Is this a UAV-based orthophoto of the entire study site or a single image taken by a UAV? Please make sure to specify this in the caption.
Line 82: Figure 2 shows an orthophoto reconstructed with UAV imagery, and detailed views over particular scarps and cracks taken with a compact camera? Please specify accordingly, see also the previous comment.
Figure 3: Same correction as for the caption of Figure 2. This is not a photograph, it seems like a UAV-based orthophoto.
Figure 4: What do you mean by ventral camera? Is this a typo? Could you please state here the GoPro camera that was carried by the quadrotor?
Line 109-110: A reference base station is used from SIRGAS which means that the accuracies mentioned earlier are relative to this station. How far is this station from the study areas (i.e. how long the baselines are)? Please make it clear that relative accuracies have been calculated with Trimble and also mention the length of the baselines.
Table 2: Could you please elaborate more in the text on what is the number of flyovers? Also, it is more interesting to include in the table the number of tie points reconstructed.
Line 136: It should be mentioned that the 2.77 mm focal length is the nominal one and not the calibrated one, since no calibration was performed. It is quite important to be specific with those terms to help the readers.
Line 140-141: This sentence is a bit vague. What do you mean that textured digital 3D models where appropriate? I assume you created a texture of the reconstructed 3D model for the entire study site. Also, I believe that DEM is more correct rather than using DSM as a terminology here see previous comment.
Just for clarity: SfM pipeline constitutes the first step for image alignment which leads to tie points (or sparse point cloud reconstruction based on Agisoft terminology), the densification of point cloud results in a dense point cloud and is performed with the aid of dense image matching multi view stereo algorithms, texturing helps the orthophoto creation. I would suggest to rephrase those sentences and make the description clearer. A good reference to use when it comes to terminology in photogrammetry is: https://doi.org/10.1111/phor.12146. Please correct some terminology throughout the manuscript using this reference: for example “point cloud is reconstructed”
Line 148-149: It is very good that ASPR standards are mentioned here, but we should be quite careful when it comes to accuracies. As only 3 and 4 check points were used, it should be mentioned that the achieved accuracies were based on a low redundancy. Such a small number of check points is not adequate to claim high accuracies in general. I recommend to read this publication (https://doi.org/10.1016/j.geomorph.2016.11.021) and apply some of the concepts presented there to understand how errors are propagated through the SfM process. The error values Agisoft gives are not directly adequate indicators of accuracy.
Figure 5 and 6 a and b do not offer anything interesting as they are similar to Figure 2 and 3. It would be more interesting to differentiate which points were GCPs and which were check points in all maps. Also it would be interesting to show the scarps and cracks over the generated DEMs. I think it is better to show the DEMs only rather than orthophotos. If you really want to have all generated orthophotos you can have them as a separate figure in Appendix (or supplementary material).
In terms of dense matching: it would be interesting to mention in the manuscript what method was set up in Agisoft in terms of depth mode reconstruction (e.g aggressive, mild or moderate?) Each method has different results. Also, have you used high (or ultra) accuracy for point cloud generation? What did you do to check any errors in the point cloud (biases e.g. high uncertainties in the tie points)? Have you performed any clean-up process before generating the DEM and orthophoto? Perhaps some points were erroneously located (e.g. flying points etc.).
Line 173-174: Could you map unstable/stable areas on the previous figures, perhaps categorise the GNSS points that are in the unstable and those in stable with different colours for example. That would help the reader.
Line 176: Why did you use 3 m as a threshold how did this come from, could you please elaborate more on this?
Line 189: The GPS positioning precision equal to 0.03 ± 1ppm refers to a measurement of a single point. This value does not apply for the displacements which are the differences between two measurements.
Table 4 and 5 do not really add anything in the narrative, can be just included as complementary materials. Instead, the displacement uncertainty per point is more valuable to be added in a table. For some reason it is not really clear that you have calculated the uncertainties based on the error propagation law from the beginning of the methodology description. I had to reach the Discussion to actually realise that you have adopted the error propagation law but this is placed wrongly too late in the manuscript. I would suggest to move the accuracies and errors section back in section 3 as a main part of the methodology before presenting the results.
Line 234: Before calculating the displacements at monitoring points from the subsequent DEMs, have you checked how well the DEMs are co-registered with each other? The fact that they have been georeferenced with very few GCPs (5 and 6 are not really many points) does not mean that there might be unresolved co-registration errors. This is a step that should be included in the methodology.
Table 6 and 7: SDs and RMSEs in Tables 6 and 7 are relatively high (e.g. 0.08 in xy). Have you considered to remove features above ground first, before calculating DODs? It would be perhaps better to remove/mask/filter out areas with buildings and trees and perform a point cloud cleaning process and then a ground classification to remove unwanted flying points over the grass that might be picked when extracting their location in DEM. Other questions that arise here is how easily you could identify those concrete targets on the DJI images in Agisoft (there are no black and white markers in the middle for a better recognition aren’t they) and what are the settings you used for the image alignment?
Lines 315-316: All this section should be part of Methodology before Results. I also think that the way the error propagation law is adopted is not fully correct, as a threshold of 90 or 95 % level of confidence is missing. Please see and also cite these two relevant highly-cited publications https://doi.org/10.1002/esp.4125 and https://doi.org/10.1016/j.geomorph.2016.11.021. I think they should be mentioned as they are very relevant to the presented study and because they provide some very good tips for example among others on how to use Agisoft in geomorphological studies.
Lines 326-327: Displacements of points over stable areas should be null. Table 6 and 7 show high SD and RMSEs with a cm-level mean displacement. So small mean value but high errors, how do you explain this? Do you think that if you had applied point cloud cleaning and ground classification before calculating the displacements would have reduced those errors? Also, what about co-registration errors (see previous comment).
Lines 332-333: I would suggest to clean the point clouds and check any unresolved co-registration errors (Cloud Compare is a good free tool for that) and re calculate the uncertainty threshold, because I think some noise can be reduced at earlier stages of the methodology.
Line 446: Even though it is well acknowledged here in the Conclusions that the methodology need to be improved with vegetation filtering, I still believe that point cloud noise filtering and error checking should be undertaken at this stage and not in the future to further enhance the quality of the results.
Technical corrections
Lines28-29: instead of “including in the” it is better to write “such as”.
Line 40: Better written: DTMs or DSMs are constructed using automatic image correlation techniques.
Line 46: The computer vision algorithms included in SfM are not really new, they have just become popular and have been improved to handle unordered imagery from UAV and big volume of data.
Line 49: It is better to use e.g. for the software in the parenthesis, as those are not the only ones, there are other software such as Context Capture, MicMac that provide very reliable outputs. It should be nice to mention.
Line 65: English correction: It is better to say “The present research was conducted at two study sites: a) ….; and b)…” And you should name those sites in the first sentence as well after a) and b).
Line 78: Typo: The predominant vegetation cover in the study area is grass”
Figure 1 caption: Please correct the citation style. You should cite as follows: “Adapted from Soto et al., (2017)”
Line 110: This is not an antenna, it is a GNSS station named LJEC. Better to rephrase the sentence as: The LJEC station from the Ecuadorian network of the Geocentric Reference System for the Americas (SIRGAS) was used as a reference base station.
Line 112: Better and clearer to say: Six GNSS surveys were conducted in each study site, as outlined…… Please use either study area or study site and not sector for your study areas throughout the manuscript for consistency.
Line 123: Instead of “GNSS measurements were taken”, you can write, “GNSS were surveyed”
Citation: https://doi.org/10.5194/nhess-2021-32-RC1 -
AC3: 'Reply on RC1', Belizario Zarate, 21 Mar 2021
The manuscript presents a study of landslide movement from a series of campaigns obtained with GNSS and UAS surveys at two test sites. The presentation of the manuscript is good in terms of details on data collection, understanding and interpretation of results. The scientific questions are also relevant to the scope of NHESS. While the methodology adopted in the manuscript generally follows the common practice, it does not provide any novel concept or tool and does not directly attract readers. For example, in order to improve the manuscript the authors should think beyond the fundamental applications of GNSS and UAS processing, they could further explore errors associated with those techniques and the impact those errors can often have on landslide estimation. Alternatively, they could also explore other morphological attributes of DEM to identify cracks and scarps in a more automated fashion. These are some examples to further investigate landslide dynamics with a step away from the conventional geomatics techniques (i.e. GNSS and UAS). I would recommend some major revision of the manuscript to include some more interesting concepts that are currently missing and correct other issues that are included in the specific comments. The positive point in this study is that there are a lot of data collected which is not always feasible. The authors should take this as an opportunity to generate something catchy and interesting providing greater value to the geomorphological research community.
While the authors have nicely given good credit to related work there are few publications worth including in the manuscript that I have commented below in the specific comments. Additionally, there are some recommendations to change the structure of the text and explain the error propagation law earlier within the Methodology section. The UAS camera calibration part is also missing, and SfM self-calibrating bundle adjustment should be mentioned in the Methodology. It would be clearer if there are some figures showing the unstable/stable regions and which targets have been used for GCPs and which for CPs. Some more figures of detailed maps over particular scarps would be also good to add. The abstract is currently under-presented. The authors should show in a better way the bigger picture in the abstract (why this study is important) and the novel part of the methodology, with some tangible comparable results to attract the general audience.
Firstly, we would like to sincerely appreciate your comments and suggestions. This an advance of the response we are preparing after the closure of the open discussion.
We are rewriting the abstract and restructuring the sections to make the paper more easily for reading and understanding. We also rewrite some parts of the methodology especially those related with the SfM and MVS pipeline for image orientation and DEM/orthophoto generation, and also to the uncertainties estimation. We attend also to specific suggestions.
Specific comments
Line 43: RPAS also known as UAS too.
Answer: Suggested change has been made in the text.
Line 50: Strictly Speaking, these processes (i.e. SfM and MVS image matching) constitute a pipeline that does not result only in an orthophoto. The actual output is a dense point cloud, then this is converted into an orthophoto, and a DEM via interpolation methods. The DTM is a by-product after ground classification. The DSM is also a by-product after some additional processing to create the surface model without noisy points. As a generic term I would recommend to use DEM throughout the manuscript.
Answer: Your suggestion is appropriate and will be taken into account for the revision of the whole article
Figure 2. Caption: What do you mean by photograph February 2016? Is this a UAV-based orthophoto of the entire study site or a single image taken by a UAV? Please make sure to specify this in the caption.
Line 82: Figure 2 shows an orthophoto reconstructed with UAV imagery, and detailed views over particular scarps and cracks taken with a compact camera? Please specify accordingly, see also the previous comment.
Figure 3: Same correction as for the caption of Figure 2. This is not a photograph, it seems like a UAV-based orthophoto.
Answer. The images were captured by UAV procedures. The supporting field photographs of cracks and other surface phenomena were recorded with a conventional camera.
Figure 4: What do you mean by ventral camera? Is this a typo? Could you please state here the GoPro camera that was carried by the quadrotor?
Answer: The ventral camera is the one used to capture aerial images and was located at the bottom of the UAV. The term ventral refers to what is underneath the UAV.
Line 109-110: A reference base station is used from SIRGAS which means that the accuracies mentioned earlier are relative to this station. How far is this station from the study areas (i.e. how long the baselines are)? Please make it clear that relative accuracies have been calculated with Trimble and also mention the length of the baselines.
Answer: It is indeed a reference station in the SIRGAS network. The length of the baseline will be inserted in the document.
Table 2: Could you please elaborate more in the text on what is the number of flyovers? Also, it is more interesting to include in the table the number of tie points reconstructed.
Line 136: It should be mentioned that the 2.77 mm focal length is the nominal one and not the calibrated one, since no calibration was performed. It is quite important to be specific with those terms to help the readers.
Answer: The camera has a focal length of 2.77 mm and corresponds to the nominal focal length. The clarification will be made in the document.
Line 140-141: This sentence is a bit vague. What do you mean that textured digital 3D models where appropriate? I assume you created a texture of the reconstructed 3D model for the entire study site. Also, I believe that DEM is more correct rather than using DSM as a terminology here see previous comment.
Just for clarity: SfM pipeline constitutes the first step for image alignment which leads to tie points (or sparse point cloud reconstruction based on Agisoft terminology), the densification of point cloud results in a dense point cloud and is performed with the aid of dense image matching multi view stereo algorithms, texturing helps the orthophoto creation. I would suggest to rephrase those sentences and make the description clearer. A good reference to use when it comes to terminology in photogrammetry is: https://doi.org/10.1111/phor.12146. Please correct some terminology throughout the manuscript using this reference: for example “point cloud is reconstructed”
Answer: The paragraph has been rewritten according to the suggestions and the terminology recommended. Besides, the term DEM has been used in a coherent way through the whole text.
Figure 5 and 6 a and b do not offer anything interesting as they are similar to Figure 2 and 3. It would be more interesting to differentiate which points were GCPs and which were check points in all maps. Also it would be interesting to show the scarps and cracks over the generated DEMs. I think it is better to show the DEMs only rather than orthophotos. If you really want to have all generated orthophotos you can have them as a separate figure in Appendix (or supplementary material).
Answer: We have preferred to keep all the figures. Thus, in Figures 2 and 3 we present the morphological features and the GNSS network over the orthoimage. We thick that cracks and scarps are better visible on orthophotos at the figure resolution. Then, in figures 4 and 5, we present both the orthoimages and the DSMs, with the GCPs and checkpoints.
In terms of dense matching: it would be interesting to mention in the manuscript what method was set up in Agisoft in terms of depth mode reconstruction (e.g aggressive, mild, or moderate?) Each method has different results. Also, have you used high (or ultra) accuracy for point cloud generation? What did you do to check any errors in the point cloud (biases e.g. high uncertainties in the tie points)? Have you performed any clean-up process before generating the DEM and orthophoto? Perhaps some points were erroneously located (e.g. flying points etc.).
Answer: In terms of depth mode reconstruction we used aggressive mode and high accuracy. The DEMs were subjected to a process of manual removal of vegetation such as grasses and small shrubs. High vegetation areas were not eliminated. The control points were located on the ground with the utmost care in order to avoid location errors.
Line 176: Why did you use 3 m as a threshold how did this come from, could you please elaborate more on this?
Landslide evidence in the study area are related to subtle changes in the terrain elevation, usually lower than 3 m and even 1m. Thus to detect visually these changes, we saturate the palette in values higher 3 m that is enough to observe the landslides evidence. The text has been to express better this idea.
Table 4 and 5 do not really add anything in the narrative, can be just included as complementary materials. Instead, the displacement uncertainty per point is more valuable to be added in a table. For some reason it is not really clear that you have calculated the uncertainties based on the error propagation law from the beginning of the methodology description. I had to reach the Discussion to actually realise that you have adopted the error propagation law but this is placed wrongly too late in the manuscript. I would suggest to move the accuracies and errors section back in section 3 as a main part of the methodology before presenting the results.
The suggested change has been made, and the uncertainties will be included in the methodology section.
Line 234: Before calculating the displacements at monitoring points from the subsequent DEMs, have you checked how well the DEMs are co-registered with each other? The fact that they have been georeferenced with very few GCPs (5 and 6 are not really many points) does not mean that there might be unresolved co-registration errors. This is a step that should be included in the methodology.
Answer: Responding to Table 4 and 5 and line 234, the suggested changes have been made
Table 6 and 7: SDs and RMSEs in Tables 6 and 7 are relatively high (e.g. 0.08 in xy). Have you considered to remove features above ground first, before calculating DODs? It would be perhaps better to remove/mask/filter out areas with buildings and trees and perform a point cloud cleaning process and then a ground classification to remove unwanted flying points over the grass that might be picked when extracting their location in DEM. Other questions that arise here is how easily you could identify those concrete targets on the DJI images in Agisoft (there are no black and white markers in the middle for a better recognition aren’t they) and what are the settings you used for the image alignment?
Answer: For the generation of the DoD, manual filtering of the arboreal areas and pastures was carried out using the Agisoft software tools. For the orientation of the models, the control points and checkpoints were used, and they are found on the ground in the form of crosses of rocks painted white, in the maps they are not possible to visualize them due to the assembly of the used symbols. Considering the suggestion, this topic will be supplemented in the article to clarify the adjustment work.
Lines 326-327: Displacements of points over stable areas should be null. Table 6 and 7 show high SD and RMSEs with a cm-level mean displacement. So small mean value but high errors, how do you explain this? Do you think that if you had applied point cloud cleaning and ground classification before calculating the displacements would have reduced those errors? Also, what about co-registration errors (see previous comment).
Answer: The main photogrammetric products obtained using the UAV technique are orthophotos and MDS. If an automatic or manual filtering of vegetation or other elements on the surface is carried out, a DTM could be generated, however, this could be obtained with the use of LIDAR techniques that allow generating a more real terrain.
This difference causes errors to be generated in the measurements that can be generated mainly by manual filtering.
Lines 332-333: I would suggest to clean the point clouds and check any unresolved co-registration errors (Cloud Compare is a good free tool for that) and re calculate the uncertainty threshold, because I think some noise can be reduced at earlier stages of the methodology.
Answer: The dense point clouds generated in the Agisoft software were entered into the Cloundcompare software. Vegetation filtering was performed, generating DEMs and DoDs were generated. The results generated the same results. The variation of the errors generated is + - 0.012 m, which when compared with precision thresholds do not affect the initial results. It should be noted that the area of the slope movement lacks mostly vegetation greater than one meter, so in this work, the most significant errors occur in the margins where there is vegetation greater than 1 m.
Line 446: Even though it is well acknowledged here in the Conclusions that the methodology need to be improved with vegetation filtering, I still believe that point cloud noise filtering and error checking should be undertaken at this stage and not in the future to further enhance the quality of the results.
Answer: The comment is valid, this section will be analyzed and written
Technical corrections
All technical corrections have been made according to the suggestions given.
Citation: https://doi.org/10.5194/nhess-2021-32-AC3
-
AC3: 'Reply on RC1', Belizario Zarate, 21 Mar 2021
-
RC2: 'Comment on nhess-2021-32', Anonymous Referee #2, 01 Mar 2021
This study describes the use of GNSS and RPAS for studying two landslide and their dynamics. It is a well written and structured manuscript. Even if the level of innovation is limited, it is a nice application of existing technologies which – in my opinion – is fine for a paper. However, there are a few things I recommend to adapt. First, the descriptions of the displacements, etc., with all the numbers that are provided both in the tables and text, are a bit lengthy. As a reader, I tend to go through these sections very quickly. However, I understand that a detailed description of the results is worth it, but maybe it could be a bit condensed and some redundancies could be checked. Second, while technical details and results are described in detail, I am missing any descriptions of the practical implications of this study. Providing more information on that would add much more value to the manuscript. For example, how do the landslides affect the infrastructure, in particular the buildings in the Victoria study site and the large road in the Colinas Lojanas study site, what can be expected in future in terms of displacement, are there intentions to implement measures to counteract the movement, etc. Currently, the discussion focuses mainly on technological aspects, but I think a separate section should focus on the practical aspects and impacts of the.
Please find a few detailed comments in the following:
Line 24-25: I doubt that the given year (1994) for the reference "Malet et al., 1994" is correct. Please check.
Line 79: How was the landslide area identified/delineated? I am wondering if it is possible to provide the area with up to a single m² accuracy and to delineate the boundary that accurately?
Line 89: See the previous comment. I doubt that it is possible to give such an accurate number. When looking at the rather coarse delineation in Figure 3 it becomes obvious that this is hardly possible to exactly delineate the landslide (assuming that the area matches with the dashed black line shown there). I suggest using a rounded number.
Line 124: Is there any reason why the flights were made independently of the GNSS measurements and does this have any influence on the further analysis?
Figure 5: The green GNSS network dots are hardly visible. Better use another colour and increase the size.
Figure 6: See the previous comment. Also, the currently used tone of green for the GNSS network is different from Figure 5.
Figure 7 and 8: Also here the visibility of the symbols could be increased.
Line 346 ff: Please check for redundancies in this section in comparison to the previous sections. Some descriptions might be shortened in this section.
Line 437-438: Only here a very short note on the affected road is given. More details on the practical implications of the findings and potential risks for the infrastructure should be provided.
Citation: https://doi.org/10.5194/nhess-2021-32-RC2 -
AC2: 'Reply on RC2', Belizario Zarate, 21 Mar 2021
This study describes the use of GNSS and RPAS for studying two landslide and their dynamics. It is a well written and structured manuscript. Even if the level of innovation is limited, it is a nice application of existing technologies which – in my opinion – is fine for a paper. However, there are a few things I recommend to adapt. First, the descriptions of the displacements, etc., with all the numbers that are provided both in the tables and text, are a bit lengthy. As a reader, I tend to go through these sections very quickly. However, I understand that a detailed description of the results is worth it, but maybe it could be a bit condensed and some redundancies could be checked. Second, while technical details and results are described in detail, I am missing any descriptions of the practical implications of this study. Providing more information on that would add much more value to the manuscript. For example, how do the landslides affect the infrastructure, in particular the buildings in the Victoria study site and the large road in the Colinas Lojanas study site, what can be expected in future in terms of displacement, are there intentions to implement measures to counteract the movement, etc. Currently, the discussion focuses mainly on technological aspects, but I think a separate section should focus on the practical aspects and impacts of the.
Firstly, we would like to sincerely appreciate your comments and suggestions. This an advance of the response we are preparing after the closure of the open discussion.
We are rewriting the abstract and the whole text, and restructuring the sections to make the paper more easily for reading and understanding. We also simplify some tables as you suggest.
Please find a few detailed comments in the following:
Line 24-25: I doubt that the given year (1994) for the reference "Malet et al., 1994" is correct. Please check.
Answer: The reference has been checked and the error was detected
Line 79: How was the landslide area identified/delineated? I am wondering if it is possible to provide the area with up to a single m² accuracy and to delineate the boundary that accurately?
Answer: The delimited area of the lanslide was established based on field recognition where the presence or absence of surface geomorphic changes was evidenced. The area was determined with the ArcGIS Area tool.
Line 89: See the previous comment. I doubt that it is possible to give such an accurate number. When looking at the rather coarse delineation in Figure 3 it becomes obvious that this is hardly possible to exactly delineate the landslide (assuming that the area matches with the dashed black line shown there). I suggest using a rounded number.
Answer: Suggested setting has been made
Figure 5: The green GNSS network dots are hardly visible. Better use another colour and increase the size.
Figure 6: See the previous comment. Also, the currently used tone of green for the GNSS network is different from Figure 5.
Figure 7 and 8: Also here the visibility of the symbols could be increased.
Answer: Suggested setting has been made and will be show in the new text.
Line 346: Please check for redundancies in this section in comparison to the previous sections. Some descriptions might be shortened in this section.
Answer: The reduction of certain parts of the manuscript will be reviewed
Line 437-438: Only here a very short note on the affected road is given. More details on the practical implications of the findings and potential risks for the infrastructure should be provided.
Answer: In fact, details of the effects of the results obtained in the existing infrastructure in the two study areas have been inserted.
Technical corrections
All technical corrections will be made according to the suggestions given.
Citation: https://doi.org/10.5194/nhess-2021-32-AC2
-
AC2: 'Reply on RC2', Belizario Zarate, 21 Mar 2021
-
RC3: 'Comment on nhess-2021-32 Review', Anonymous Referee #3, 01 Mar 2021
General comments:
The paper by Belizario et al. entitled “GNSS and RPAS integration techniques for studying landslide dynamics: Application to the areas of Victoria and Colinas Lojanas, (Loja, Ecuador)” presents the application of GNSS and RPAS techniques to study the landslide dynamics of two landslides that affected the Victoria and Colinas Lojanas sector of the city of Loja, Ecuador. The direction and velocity of landslide displacements were calculated by GNSS measurements, while orthophotos and DSMs were used to calculate horizontal and vertical displacements in a set of significant points throughout the study area.
Overall, this is an appropriate subject area for NHESS journal, and the amount of data collected is very important from a risk monitoring and prevention perspective. However, this work should try to better develop the application of topographic techniques to the case of study where some innovative aspects or tools are missing. In addition, the methodology used is not effectively illustrated and lacks some aspects related to an accurate assessment of errors. I believe that this paper has great potential and interesting aspects that could be improved to make it more appealing to a reader but in its current state is not ready now to be published. It requires a substantial upgrading (major review), maybe assessing the limits and errors associated with the used topographic techniques and the comparison with other technologies in terms of landslide dynamics estimation. The text is unclear in some sections (requires English revision and often the terminology used is incorrect, especially regarding the photogrammetric process), difficult to follow for a reader and the structure of the text must be revised because some parts are not in their optimal location (see specific comments below). With some important improvements, this work can be interesting and useful for the scientific community.
Specific comments
- Abstract: I suggest rewriting it to make it more attractive to the reader perhaps emphasizing the innovative aspect of this work and the usefulness of these results in terms of the mitigation of natural hazard problems.
- Introduction: this part should be underlined the innovative aspects of the work, motivated the choice of technologies used for the surveys, and highlighted the usefulness of the data obtained.
- Methods:
- A GoPro 3+ camera was used to carry out the SfM surveys, but it was not shown how the problems related to image distortion were solved given the use of a fisheye lens with a flight altitude very high.
- The study areas are quite large (around 6 and 2 ha) and 6 and 5 GCPs were used respectively, by what criteria were this number and GCP/CP ratio chosen? Is this number sufficient? (look at this paper: https://doi.org/10.3390/rs10101606 and https://doi.org/10.1016/j.geomorph.2016.11.021). Where are they located in the study area (a figure could be added about this)? Are the errors related to GCP and CP referred to the point cloud? and the errors related to DSM? Because for the displacements monitoring, the points are extracted from DSMs in stable areas so the interpolation error should also be considered.
- Were the DoDs thresholded to account for the errors or do they represent raw differences? Line 180: why was a 3 m threshold chosen?
- Has the problem of co-registration of point clouds been considered in making multi-temporal DSMs?
- Results: should be presented more effectively by trying not to repeat some of the data already summarised in the various tables.
- Discussion: This part is a kind of repetition of the results and is lacking in comparisons with other case studies or similar work, especially about the methodologies used. The discussion misses an in-depth analysis of the problems and errors caused by the technologies used, how to improve these aspects, and a comparison with other works using the same techniques. It would also be interesting to analyze how the results and the information obtained could be used and exploited for risk prevention purposes.
Technical corrections
- Abstract: acronyms should always be made explicit here as well. In the first sentence, a verb is missing after the "to".
- UAV is more known than RPAS
- Line 41 and 409: “interpolated” is not the correct term to use, better generated or realized.
- Line 49-50: this sentence is unclear, and I suggest rewriting it.
- Line 181-182: this sentence is unclear, and I suggest rewriting it and explained which ArcGis tool was used, please.
- Figure 9 b and c: I think it is useful to improve the presentation of these figures since the legend covers the image itself.
- Line 333: what is meant by errors of the orientation process?
- Equations 1 and 2: should be moved to the methods section.
- Lines 332-333 and 338-341: should be moved to the methods section.
Citation: https://doi.org/10.5194/nhess-2021-32-RC3 -
AC1: 'Reply on RC3', Belizario Zarate, 21 Mar 2021
General comments:
The paper by Belizario et al. entitled “GNSS and RPAS integration techniques for studying landslide dynamics: Application to the areas of Victoria and Colinas Lojanas, (Loja, Ecuador)” presents the application of GNSS and RPAS techniques to study the landslide dynamics of two landslides that affected the Victoria and Colinas Lojanas sector of the city of Loja, Ecuador. The direction and velocity of landslide displacements were calculated by GNSS measurements, while orthophotos and DSMs were used to calculate horizontal and vertical displacements in a set of significant points throughout the study area.
Overall, this is an appropriate subject area for the NHESS journal, and the amount of data collected is very important from a risk monitoring and prevention perspective. However, this work should try to better develop the application of topographic techniques to the case of study where some innovative aspects or tools are missing. In addition, the methodology used is not effectively illustrated and lacks some aspects related to an accurate assessment of errors. I believe that this paper has great potential and interesting aspects that could be improved to make it more appealing to a reader but in its current state is not ready now to be published. It requires a substantial upgrading (major review), maybe assessing the limits and errors associated with the used topographic techniques and the comparison with other technologies in terms of landslide dynamics estimation. The text is unclear in some sections (requires English revision and often the terminology used is incorrect, especially regarding the photogrammetric process), difficult to follow for a reader and the structure of the text must be revised because some parts are not in their optimal location (see specific comments below). With some important improvements, this work can be interesting and useful for the scientific community.
Firstly, we would like to sincerely appreciate your comments and suggestions. This an advance of the response we are preparing after the closure of the open discussion.
We are rewriting the abstract and restructuring the sections to make the paper more easily for reading and understanding. We are also rewriting some parts of the methodology and the error assessment, as well as revising some tables and figures as you suggest.
Specific comments
Abstract: I suggest rewriting it to make it more attractive to the reader perhaps emphasizing the innovative aspect of this work and the usefulness of these results in terms of the mitigation of natural hazard problems.
Answer: Modifications are being made to the abstract based on suggestions
Introduction: this part should be underlined the innovative aspects of the work, motivated the choice of technologies used for the surveys, and highlighted the usefulness of the data obtained.
Answer: Suggestion will be added in the manuscript introduction
Methods:
A GoPro 3+ camera was used to carry out the SfM surveys, but it was not shown how the problems related to image distortion were solved given the use of a fisheye lens with a flight altitude very high.
Answer: The fisheye function was not used for the present study, the wide mode was used. Capture mode specified in the manuscript.
The study areas are quite large (around 6 and 2 ha) and 6 and 5 GCPs were used respectively, by what criteria were this number and GCP/CP ratio chosen? Is this number sufficient? (look at this paper: https://doi.org/10.3390/rs10101606
and https://doi.org/10.1016/j.geomorph.2016.11.021).
Answer: Pepa et al (2016, 2019) carried out a study to evaluate the precision of a UAV-based monitoring system, in which a total of 5 GCPs are considered to monitor an area of 6.6 ha, the accuracies obtained are similar to those of the present study. Other studies cited in the text present similar errors and uncertainties. The geographical conditions given in said study are similar in the two study areas. Although studies suggest having a greater number of GCPs, the results obtained were considered relevant.
Where are they located in the study area (a figure could be added about this)? Are the errors related to GCP and CP referred to the point cloud? and the errors related to DSM? Because for the displacements monitoring, the points are extracted from DSMs in stable areas so the interpolation error should also be considered.
We are considering your suggestions, and revising the figures and the tables to include this information. The control points were located on the ground with the utmost care in order to avoid location errors.
Were the DoDs thresholded to account for the errors or do they represent raw differences? Line 180: why was a 3 m threshold chosen?
Landslide evidence in the study area is related to subtle changes in the terrain elevation, usually lower than 3 m and even 1m. Thus to detect visually these changes, we saturate the palette in values higher than 3 m that is enough to observe the landslides evidence. The text has been to express better this idea.
Technical corrections
All technical corrections have been made according to the suggestions given.
Citation: https://doi.org/10.5194/nhess-2021-32-AC1
Interactive discussion
Status: closed
-
RC1: 'Comment on nhess-2021-32 - Review', Anonymous Referee #1, 23 Feb 2021
General comments
The manuscript presents a study of landslide movement from a series of campaigns obtained with GNSS and UAS surveys at two test sites. The presentation of the manuscript is good in terms of details on data collection, understanding and interpretation of results. The scientific questions are also relevant to the scope of NHESS. While the methodology adopted in the manuscript generally follows the common practice, it does not provide any novel concept or tool and does not directly attract readers. For example, in order to improve the manuscript the authors should think beyond the fundamental applications of GNSS and UAS processing, they could further explore errors associated with those techniques and the impact those errors can often have on landslide estimation. Alternatively, they could also explore other morphological attributes of DEM to identify cracks and scarps in a more automated fashion. These are some examples to further investigate landslide dynamics with a step away from the conventional geomatics techniques (i.e. GNSS and UAS). I would recommend some major revision of the manuscript to include some more interesting concepts that are currently missing and correct other issues that are included in the specific comments. The positive point in this study is that there are a lot of data collected which is not always feasible. The authors should take this as an opportunity to generate something catchy and interesting providing greater value to the geomorphological research community.
While the authors have nicely given good credit to related work there are few publications worth including in the manuscript that I have commented below in the specific comments. Additionally, there are some recommendations to change the structure of the text and explain the error propagation law earlier within the Methodology section. The UAS camera calibration part is also missing, and SfM self-calibrating bundle adjustment should be mentioned in the Methodology. It would be clearer if there are some figures showing the unstable/stable regions and which targets have been used for GCPs and which for CPs. Some more figures of detailed maps over particular scarps would be also good to add. The abstract is currently under presented. The authors should show in a better way the bigger picture in the abstract (why this study is important) and the novel part of the methodology, with some tangible comparable results to attract the general audience.
Specific comments
Line 43: RPAS also known as UAS too.
Line 50: Strictly Speaking, these processes (i.e. SfM and MVS image matching) constitute a pipeline that does not result only in an orthophoto. The actual output is a dense point cloud, then this is converted into an orthophoto, and a DEM via interpolation methods. The DTM is a by-product after ground classification. The DSM is also a by-product after some additional processing to create the surface model without noisy points. As a generic term I would recommend to use DEM throughout the manuscript.
Figure 2. Caption: What do you mean by photograph February 2016? Is this a UAV-based orthophoto of the entire study site or a single image taken by a UAV? Please make sure to specify this in the caption.
Line 82: Figure 2 shows an orthophoto reconstructed with UAV imagery, and detailed views over particular scarps and cracks taken with a compact camera? Please specify accordingly, see also the previous comment.
Figure 3: Same correction as for the caption of Figure 2. This is not a photograph, it seems like a UAV-based orthophoto.
Figure 4: What do you mean by ventral camera? Is this a typo? Could you please state here the GoPro camera that was carried by the quadrotor?
Line 109-110: A reference base station is used from SIRGAS which means that the accuracies mentioned earlier are relative to this station. How far is this station from the study areas (i.e. how long the baselines are)? Please make it clear that relative accuracies have been calculated with Trimble and also mention the length of the baselines.
Table 2: Could you please elaborate more in the text on what is the number of flyovers? Also, it is more interesting to include in the table the number of tie points reconstructed.
Line 136: It should be mentioned that the 2.77 mm focal length is the nominal one and not the calibrated one, since no calibration was performed. It is quite important to be specific with those terms to help the readers.
Line 140-141: This sentence is a bit vague. What do you mean that textured digital 3D models where appropriate? I assume you created a texture of the reconstructed 3D model for the entire study site. Also, I believe that DEM is more correct rather than using DSM as a terminology here see previous comment.
Just for clarity: SfM pipeline constitutes the first step for image alignment which leads to tie points (or sparse point cloud reconstruction based on Agisoft terminology), the densification of point cloud results in a dense point cloud and is performed with the aid of dense image matching multi view stereo algorithms, texturing helps the orthophoto creation. I would suggest to rephrase those sentences and make the description clearer. A good reference to use when it comes to terminology in photogrammetry is: https://doi.org/10.1111/phor.12146. Please correct some terminology throughout the manuscript using this reference: for example “point cloud is reconstructed”
Line 148-149: It is very good that ASPR standards are mentioned here, but we should be quite careful when it comes to accuracies. As only 3 and 4 check points were used, it should be mentioned that the achieved accuracies were based on a low redundancy. Such a small number of check points is not adequate to claim high accuracies in general. I recommend to read this publication (https://doi.org/10.1016/j.geomorph.2016.11.021) and apply some of the concepts presented there to understand how errors are propagated through the SfM process. The error values Agisoft gives are not directly adequate indicators of accuracy.
Figure 5 and 6 a and b do not offer anything interesting as they are similar to Figure 2 and 3. It would be more interesting to differentiate which points were GCPs and which were check points in all maps. Also it would be interesting to show the scarps and cracks over the generated DEMs. I think it is better to show the DEMs only rather than orthophotos. If you really want to have all generated orthophotos you can have them as a separate figure in Appendix (or supplementary material).
In terms of dense matching: it would be interesting to mention in the manuscript what method was set up in Agisoft in terms of depth mode reconstruction (e.g aggressive, mild or moderate?) Each method has different results. Also, have you used high (or ultra) accuracy for point cloud generation? What did you do to check any errors in the point cloud (biases e.g. high uncertainties in the tie points)? Have you performed any clean-up process before generating the DEM and orthophoto? Perhaps some points were erroneously located (e.g. flying points etc.).
Line 173-174: Could you map unstable/stable areas on the previous figures, perhaps categorise the GNSS points that are in the unstable and those in stable with different colours for example. That would help the reader.
Line 176: Why did you use 3 m as a threshold how did this come from, could you please elaborate more on this?
Line 189: The GPS positioning precision equal to 0.03 ± 1ppm refers to a measurement of a single point. This value does not apply for the displacements which are the differences between two measurements.
Table 4 and 5 do not really add anything in the narrative, can be just included as complementary materials. Instead, the displacement uncertainty per point is more valuable to be added in a table. For some reason it is not really clear that you have calculated the uncertainties based on the error propagation law from the beginning of the methodology description. I had to reach the Discussion to actually realise that you have adopted the error propagation law but this is placed wrongly too late in the manuscript. I would suggest to move the accuracies and errors section back in section 3 as a main part of the methodology before presenting the results.
Line 234: Before calculating the displacements at monitoring points from the subsequent DEMs, have you checked how well the DEMs are co-registered with each other? The fact that they have been georeferenced with very few GCPs (5 and 6 are not really many points) does not mean that there might be unresolved co-registration errors. This is a step that should be included in the methodology.
Table 6 and 7: SDs and RMSEs in Tables 6 and 7 are relatively high (e.g. 0.08 in xy). Have you considered to remove features above ground first, before calculating DODs? It would be perhaps better to remove/mask/filter out areas with buildings and trees and perform a point cloud cleaning process and then a ground classification to remove unwanted flying points over the grass that might be picked when extracting their location in DEM. Other questions that arise here is how easily you could identify those concrete targets on the DJI images in Agisoft (there are no black and white markers in the middle for a better recognition aren’t they) and what are the settings you used for the image alignment?
Lines 315-316: All this section should be part of Methodology before Results. I also think that the way the error propagation law is adopted is not fully correct, as a threshold of 90 or 95 % level of confidence is missing. Please see and also cite these two relevant highly-cited publications https://doi.org/10.1002/esp.4125 and https://doi.org/10.1016/j.geomorph.2016.11.021. I think they should be mentioned as they are very relevant to the presented study and because they provide some very good tips for example among others on how to use Agisoft in geomorphological studies.
Lines 326-327: Displacements of points over stable areas should be null. Table 6 and 7 show high SD and RMSEs with a cm-level mean displacement. So small mean value but high errors, how do you explain this? Do you think that if you had applied point cloud cleaning and ground classification before calculating the displacements would have reduced those errors? Also, what about co-registration errors (see previous comment).
Lines 332-333: I would suggest to clean the point clouds and check any unresolved co-registration errors (Cloud Compare is a good free tool for that) and re calculate the uncertainty threshold, because I think some noise can be reduced at earlier stages of the methodology.
Line 446: Even though it is well acknowledged here in the Conclusions that the methodology need to be improved with vegetation filtering, I still believe that point cloud noise filtering and error checking should be undertaken at this stage and not in the future to further enhance the quality of the results.
Technical corrections
Lines28-29: instead of “including in the” it is better to write “such as”.
Line 40: Better written: DTMs or DSMs are constructed using automatic image correlation techniques.
Line 46: The computer vision algorithms included in SfM are not really new, they have just become popular and have been improved to handle unordered imagery from UAV and big volume of data.
Line 49: It is better to use e.g. for the software in the parenthesis, as those are not the only ones, there are other software such as Context Capture, MicMac that provide very reliable outputs. It should be nice to mention.
Line 65: English correction: It is better to say “The present research was conducted at two study sites: a) ….; and b)…” And you should name those sites in the first sentence as well after a) and b).
Line 78: Typo: The predominant vegetation cover in the study area is grass”
Figure 1 caption: Please correct the citation style. You should cite as follows: “Adapted from Soto et al., (2017)”
Line 110: This is not an antenna, it is a GNSS station named LJEC. Better to rephrase the sentence as: The LJEC station from the Ecuadorian network of the Geocentric Reference System for the Americas (SIRGAS) was used as a reference base station.
Line 112: Better and clearer to say: Six GNSS surveys were conducted in each study site, as outlined…… Please use either study area or study site and not sector for your study areas throughout the manuscript for consistency.
Line 123: Instead of “GNSS measurements were taken”, you can write, “GNSS were surveyed”
Citation: https://doi.org/10.5194/nhess-2021-32-RC1 -
AC3: 'Reply on RC1', Belizario Zarate, 21 Mar 2021
The manuscript presents a study of landslide movement from a series of campaigns obtained with GNSS and UAS surveys at two test sites. The presentation of the manuscript is good in terms of details on data collection, understanding and interpretation of results. The scientific questions are also relevant to the scope of NHESS. While the methodology adopted in the manuscript generally follows the common practice, it does not provide any novel concept or tool and does not directly attract readers. For example, in order to improve the manuscript the authors should think beyond the fundamental applications of GNSS and UAS processing, they could further explore errors associated with those techniques and the impact those errors can often have on landslide estimation. Alternatively, they could also explore other morphological attributes of DEM to identify cracks and scarps in a more automated fashion. These are some examples to further investigate landslide dynamics with a step away from the conventional geomatics techniques (i.e. GNSS and UAS). I would recommend some major revision of the manuscript to include some more interesting concepts that are currently missing and correct other issues that are included in the specific comments. The positive point in this study is that there are a lot of data collected which is not always feasible. The authors should take this as an opportunity to generate something catchy and interesting providing greater value to the geomorphological research community.
While the authors have nicely given good credit to related work there are few publications worth including in the manuscript that I have commented below in the specific comments. Additionally, there are some recommendations to change the structure of the text and explain the error propagation law earlier within the Methodology section. The UAS camera calibration part is also missing, and SfM self-calibrating bundle adjustment should be mentioned in the Methodology. It would be clearer if there are some figures showing the unstable/stable regions and which targets have been used for GCPs and which for CPs. Some more figures of detailed maps over particular scarps would be also good to add. The abstract is currently under-presented. The authors should show in a better way the bigger picture in the abstract (why this study is important) and the novel part of the methodology, with some tangible comparable results to attract the general audience.
Firstly, we would like to sincerely appreciate your comments and suggestions. This an advance of the response we are preparing after the closure of the open discussion.
We are rewriting the abstract and restructuring the sections to make the paper more easily for reading and understanding. We also rewrite some parts of the methodology especially those related with the SfM and MVS pipeline for image orientation and DEM/orthophoto generation, and also to the uncertainties estimation. We attend also to specific suggestions.
Specific comments
Line 43: RPAS also known as UAS too.
Answer: Suggested change has been made in the text.
Line 50: Strictly Speaking, these processes (i.e. SfM and MVS image matching) constitute a pipeline that does not result only in an orthophoto. The actual output is a dense point cloud, then this is converted into an orthophoto, and a DEM via interpolation methods. The DTM is a by-product after ground classification. The DSM is also a by-product after some additional processing to create the surface model without noisy points. As a generic term I would recommend to use DEM throughout the manuscript.
Answer: Your suggestion is appropriate and will be taken into account for the revision of the whole article
Figure 2. Caption: What do you mean by photograph February 2016? Is this a UAV-based orthophoto of the entire study site or a single image taken by a UAV? Please make sure to specify this in the caption.
Line 82: Figure 2 shows an orthophoto reconstructed with UAV imagery, and detailed views over particular scarps and cracks taken with a compact camera? Please specify accordingly, see also the previous comment.
Figure 3: Same correction as for the caption of Figure 2. This is not a photograph, it seems like a UAV-based orthophoto.
Answer. The images were captured by UAV procedures. The supporting field photographs of cracks and other surface phenomena were recorded with a conventional camera.
Figure 4: What do you mean by ventral camera? Is this a typo? Could you please state here the GoPro camera that was carried by the quadrotor?
Answer: The ventral camera is the one used to capture aerial images and was located at the bottom of the UAV. The term ventral refers to what is underneath the UAV.
Line 109-110: A reference base station is used from SIRGAS which means that the accuracies mentioned earlier are relative to this station. How far is this station from the study areas (i.e. how long the baselines are)? Please make it clear that relative accuracies have been calculated with Trimble and also mention the length of the baselines.
Answer: It is indeed a reference station in the SIRGAS network. The length of the baseline will be inserted in the document.
Table 2: Could you please elaborate more in the text on what is the number of flyovers? Also, it is more interesting to include in the table the number of tie points reconstructed.
Line 136: It should be mentioned that the 2.77 mm focal length is the nominal one and not the calibrated one, since no calibration was performed. It is quite important to be specific with those terms to help the readers.
Answer: The camera has a focal length of 2.77 mm and corresponds to the nominal focal length. The clarification will be made in the document.
Line 140-141: This sentence is a bit vague. What do you mean that textured digital 3D models where appropriate? I assume you created a texture of the reconstructed 3D model for the entire study site. Also, I believe that DEM is more correct rather than using DSM as a terminology here see previous comment.
Just for clarity: SfM pipeline constitutes the first step for image alignment which leads to tie points (or sparse point cloud reconstruction based on Agisoft terminology), the densification of point cloud results in a dense point cloud and is performed with the aid of dense image matching multi view stereo algorithms, texturing helps the orthophoto creation. I would suggest to rephrase those sentences and make the description clearer. A good reference to use when it comes to terminology in photogrammetry is: https://doi.org/10.1111/phor.12146. Please correct some terminology throughout the manuscript using this reference: for example “point cloud is reconstructed”
Answer: The paragraph has been rewritten according to the suggestions and the terminology recommended. Besides, the term DEM has been used in a coherent way through the whole text.
Figure 5 and 6 a and b do not offer anything interesting as they are similar to Figure 2 and 3. It would be more interesting to differentiate which points were GCPs and which were check points in all maps. Also it would be interesting to show the scarps and cracks over the generated DEMs. I think it is better to show the DEMs only rather than orthophotos. If you really want to have all generated orthophotos you can have them as a separate figure in Appendix (or supplementary material).
Answer: We have preferred to keep all the figures. Thus, in Figures 2 and 3 we present the morphological features and the GNSS network over the orthoimage. We thick that cracks and scarps are better visible on orthophotos at the figure resolution. Then, in figures 4 and 5, we present both the orthoimages and the DSMs, with the GCPs and checkpoints.
In terms of dense matching: it would be interesting to mention in the manuscript what method was set up in Agisoft in terms of depth mode reconstruction (e.g aggressive, mild, or moderate?) Each method has different results. Also, have you used high (or ultra) accuracy for point cloud generation? What did you do to check any errors in the point cloud (biases e.g. high uncertainties in the tie points)? Have you performed any clean-up process before generating the DEM and orthophoto? Perhaps some points were erroneously located (e.g. flying points etc.).
Answer: In terms of depth mode reconstruction we used aggressive mode and high accuracy. The DEMs were subjected to a process of manual removal of vegetation such as grasses and small shrubs. High vegetation areas were not eliminated. The control points were located on the ground with the utmost care in order to avoid location errors.
Line 176: Why did you use 3 m as a threshold how did this come from, could you please elaborate more on this?
Landslide evidence in the study area are related to subtle changes in the terrain elevation, usually lower than 3 m and even 1m. Thus to detect visually these changes, we saturate the palette in values higher 3 m that is enough to observe the landslides evidence. The text has been to express better this idea.
Table 4 and 5 do not really add anything in the narrative, can be just included as complementary materials. Instead, the displacement uncertainty per point is more valuable to be added in a table. For some reason it is not really clear that you have calculated the uncertainties based on the error propagation law from the beginning of the methodology description. I had to reach the Discussion to actually realise that you have adopted the error propagation law but this is placed wrongly too late in the manuscript. I would suggest to move the accuracies and errors section back in section 3 as a main part of the methodology before presenting the results.
The suggested change has been made, and the uncertainties will be included in the methodology section.
Line 234: Before calculating the displacements at monitoring points from the subsequent DEMs, have you checked how well the DEMs are co-registered with each other? The fact that they have been georeferenced with very few GCPs (5 and 6 are not really many points) does not mean that there might be unresolved co-registration errors. This is a step that should be included in the methodology.
Answer: Responding to Table 4 and 5 and line 234, the suggested changes have been made
Table 6 and 7: SDs and RMSEs in Tables 6 and 7 are relatively high (e.g. 0.08 in xy). Have you considered to remove features above ground first, before calculating DODs? It would be perhaps better to remove/mask/filter out areas with buildings and trees and perform a point cloud cleaning process and then a ground classification to remove unwanted flying points over the grass that might be picked when extracting their location in DEM. Other questions that arise here is how easily you could identify those concrete targets on the DJI images in Agisoft (there are no black and white markers in the middle for a better recognition aren’t they) and what are the settings you used for the image alignment?
Answer: For the generation of the DoD, manual filtering of the arboreal areas and pastures was carried out using the Agisoft software tools. For the orientation of the models, the control points and checkpoints were used, and they are found on the ground in the form of crosses of rocks painted white, in the maps they are not possible to visualize them due to the assembly of the used symbols. Considering the suggestion, this topic will be supplemented in the article to clarify the adjustment work.
Lines 326-327: Displacements of points over stable areas should be null. Table 6 and 7 show high SD and RMSEs with a cm-level mean displacement. So small mean value but high errors, how do you explain this? Do you think that if you had applied point cloud cleaning and ground classification before calculating the displacements would have reduced those errors? Also, what about co-registration errors (see previous comment).
Answer: The main photogrammetric products obtained using the UAV technique are orthophotos and MDS. If an automatic or manual filtering of vegetation or other elements on the surface is carried out, a DTM could be generated, however, this could be obtained with the use of LIDAR techniques that allow generating a more real terrain.
This difference causes errors to be generated in the measurements that can be generated mainly by manual filtering.
Lines 332-333: I would suggest to clean the point clouds and check any unresolved co-registration errors (Cloud Compare is a good free tool for that) and re calculate the uncertainty threshold, because I think some noise can be reduced at earlier stages of the methodology.
Answer: The dense point clouds generated in the Agisoft software were entered into the Cloundcompare software. Vegetation filtering was performed, generating DEMs and DoDs were generated. The results generated the same results. The variation of the errors generated is + - 0.012 m, which when compared with precision thresholds do not affect the initial results. It should be noted that the area of the slope movement lacks mostly vegetation greater than one meter, so in this work, the most significant errors occur in the margins where there is vegetation greater than 1 m.
Line 446: Even though it is well acknowledged here in the Conclusions that the methodology need to be improved with vegetation filtering, I still believe that point cloud noise filtering and error checking should be undertaken at this stage and not in the future to further enhance the quality of the results.
Answer: The comment is valid, this section will be analyzed and written
Technical corrections
All technical corrections have been made according to the suggestions given.
Citation: https://doi.org/10.5194/nhess-2021-32-AC3
-
AC3: 'Reply on RC1', Belizario Zarate, 21 Mar 2021
-
RC2: 'Comment on nhess-2021-32', Anonymous Referee #2, 01 Mar 2021
This study describes the use of GNSS and RPAS for studying two landslide and their dynamics. It is a well written and structured manuscript. Even if the level of innovation is limited, it is a nice application of existing technologies which – in my opinion – is fine for a paper. However, there are a few things I recommend to adapt. First, the descriptions of the displacements, etc., with all the numbers that are provided both in the tables and text, are a bit lengthy. As a reader, I tend to go through these sections very quickly. However, I understand that a detailed description of the results is worth it, but maybe it could be a bit condensed and some redundancies could be checked. Second, while technical details and results are described in detail, I am missing any descriptions of the practical implications of this study. Providing more information on that would add much more value to the manuscript. For example, how do the landslides affect the infrastructure, in particular the buildings in the Victoria study site and the large road in the Colinas Lojanas study site, what can be expected in future in terms of displacement, are there intentions to implement measures to counteract the movement, etc. Currently, the discussion focuses mainly on technological aspects, but I think a separate section should focus on the practical aspects and impacts of the.
Please find a few detailed comments in the following:
Line 24-25: I doubt that the given year (1994) for the reference "Malet et al., 1994" is correct. Please check.
Line 79: How was the landslide area identified/delineated? I am wondering if it is possible to provide the area with up to a single m² accuracy and to delineate the boundary that accurately?
Line 89: See the previous comment. I doubt that it is possible to give such an accurate number. When looking at the rather coarse delineation in Figure 3 it becomes obvious that this is hardly possible to exactly delineate the landslide (assuming that the area matches with the dashed black line shown there). I suggest using a rounded number.
Line 124: Is there any reason why the flights were made independently of the GNSS measurements and does this have any influence on the further analysis?
Figure 5: The green GNSS network dots are hardly visible. Better use another colour and increase the size.
Figure 6: See the previous comment. Also, the currently used tone of green for the GNSS network is different from Figure 5.
Figure 7 and 8: Also here the visibility of the symbols could be increased.
Line 346 ff: Please check for redundancies in this section in comparison to the previous sections. Some descriptions might be shortened in this section.
Line 437-438: Only here a very short note on the affected road is given. More details on the practical implications of the findings and potential risks for the infrastructure should be provided.
Citation: https://doi.org/10.5194/nhess-2021-32-RC2 -
AC2: 'Reply on RC2', Belizario Zarate, 21 Mar 2021
This study describes the use of GNSS and RPAS for studying two landslide and their dynamics. It is a well written and structured manuscript. Even if the level of innovation is limited, it is a nice application of existing technologies which – in my opinion – is fine for a paper. However, there are a few things I recommend to adapt. First, the descriptions of the displacements, etc., with all the numbers that are provided both in the tables and text, are a bit lengthy. As a reader, I tend to go through these sections very quickly. However, I understand that a detailed description of the results is worth it, but maybe it could be a bit condensed and some redundancies could be checked. Second, while technical details and results are described in detail, I am missing any descriptions of the practical implications of this study. Providing more information on that would add much more value to the manuscript. For example, how do the landslides affect the infrastructure, in particular the buildings in the Victoria study site and the large road in the Colinas Lojanas study site, what can be expected in future in terms of displacement, are there intentions to implement measures to counteract the movement, etc. Currently, the discussion focuses mainly on technological aspects, but I think a separate section should focus on the practical aspects and impacts of the.
Firstly, we would like to sincerely appreciate your comments and suggestions. This an advance of the response we are preparing after the closure of the open discussion.
We are rewriting the abstract and the whole text, and restructuring the sections to make the paper more easily for reading and understanding. We also simplify some tables as you suggest.
Please find a few detailed comments in the following:
Line 24-25: I doubt that the given year (1994) for the reference "Malet et al., 1994" is correct. Please check.
Answer: The reference has been checked and the error was detected
Line 79: How was the landslide area identified/delineated? I am wondering if it is possible to provide the area with up to a single m² accuracy and to delineate the boundary that accurately?
Answer: The delimited area of the lanslide was established based on field recognition where the presence or absence of surface geomorphic changes was evidenced. The area was determined with the ArcGIS Area tool.
Line 89: See the previous comment. I doubt that it is possible to give such an accurate number. When looking at the rather coarse delineation in Figure 3 it becomes obvious that this is hardly possible to exactly delineate the landslide (assuming that the area matches with the dashed black line shown there). I suggest using a rounded number.
Answer: Suggested setting has been made
Figure 5: The green GNSS network dots are hardly visible. Better use another colour and increase the size.
Figure 6: See the previous comment. Also, the currently used tone of green for the GNSS network is different from Figure 5.
Figure 7 and 8: Also here the visibility of the symbols could be increased.
Answer: Suggested setting has been made and will be show in the new text.
Line 346: Please check for redundancies in this section in comparison to the previous sections. Some descriptions might be shortened in this section.
Answer: The reduction of certain parts of the manuscript will be reviewed
Line 437-438: Only here a very short note on the affected road is given. More details on the practical implications of the findings and potential risks for the infrastructure should be provided.
Answer: In fact, details of the effects of the results obtained in the existing infrastructure in the two study areas have been inserted.
Technical corrections
All technical corrections will be made according to the suggestions given.
Citation: https://doi.org/10.5194/nhess-2021-32-AC2
-
AC2: 'Reply on RC2', Belizario Zarate, 21 Mar 2021
-
RC3: 'Comment on nhess-2021-32 Review', Anonymous Referee #3, 01 Mar 2021
General comments:
The paper by Belizario et al. entitled “GNSS and RPAS integration techniques for studying landslide dynamics: Application to the areas of Victoria and Colinas Lojanas, (Loja, Ecuador)” presents the application of GNSS and RPAS techniques to study the landslide dynamics of two landslides that affected the Victoria and Colinas Lojanas sector of the city of Loja, Ecuador. The direction and velocity of landslide displacements were calculated by GNSS measurements, while orthophotos and DSMs were used to calculate horizontal and vertical displacements in a set of significant points throughout the study area.
Overall, this is an appropriate subject area for NHESS journal, and the amount of data collected is very important from a risk monitoring and prevention perspective. However, this work should try to better develop the application of topographic techniques to the case of study where some innovative aspects or tools are missing. In addition, the methodology used is not effectively illustrated and lacks some aspects related to an accurate assessment of errors. I believe that this paper has great potential and interesting aspects that could be improved to make it more appealing to a reader but in its current state is not ready now to be published. It requires a substantial upgrading (major review), maybe assessing the limits and errors associated with the used topographic techniques and the comparison with other technologies in terms of landslide dynamics estimation. The text is unclear in some sections (requires English revision and often the terminology used is incorrect, especially regarding the photogrammetric process), difficult to follow for a reader and the structure of the text must be revised because some parts are not in their optimal location (see specific comments below). With some important improvements, this work can be interesting and useful for the scientific community.
Specific comments
- Abstract: I suggest rewriting it to make it more attractive to the reader perhaps emphasizing the innovative aspect of this work and the usefulness of these results in terms of the mitigation of natural hazard problems.
- Introduction: this part should be underlined the innovative aspects of the work, motivated the choice of technologies used for the surveys, and highlighted the usefulness of the data obtained.
- Methods:
- A GoPro 3+ camera was used to carry out the SfM surveys, but it was not shown how the problems related to image distortion were solved given the use of a fisheye lens with a flight altitude very high.
- The study areas are quite large (around 6 and 2 ha) and 6 and 5 GCPs were used respectively, by what criteria were this number and GCP/CP ratio chosen? Is this number sufficient? (look at this paper: https://doi.org/10.3390/rs10101606 and https://doi.org/10.1016/j.geomorph.2016.11.021). Where are they located in the study area (a figure could be added about this)? Are the errors related to GCP and CP referred to the point cloud? and the errors related to DSM? Because for the displacements monitoring, the points are extracted from DSMs in stable areas so the interpolation error should also be considered.
- Were the DoDs thresholded to account for the errors or do they represent raw differences? Line 180: why was a 3 m threshold chosen?
- Has the problem of co-registration of point clouds been considered in making multi-temporal DSMs?
- Results: should be presented more effectively by trying not to repeat some of the data already summarised in the various tables.
- Discussion: This part is a kind of repetition of the results and is lacking in comparisons with other case studies or similar work, especially about the methodologies used. The discussion misses an in-depth analysis of the problems and errors caused by the technologies used, how to improve these aspects, and a comparison with other works using the same techniques. It would also be interesting to analyze how the results and the information obtained could be used and exploited for risk prevention purposes.
Technical corrections
- Abstract: acronyms should always be made explicit here as well. In the first sentence, a verb is missing after the "to".
- UAV is more known than RPAS
- Line 41 and 409: “interpolated” is not the correct term to use, better generated or realized.
- Line 49-50: this sentence is unclear, and I suggest rewriting it.
- Line 181-182: this sentence is unclear, and I suggest rewriting it and explained which ArcGis tool was used, please.
- Figure 9 b and c: I think it is useful to improve the presentation of these figures since the legend covers the image itself.
- Line 333: what is meant by errors of the orientation process?
- Equations 1 and 2: should be moved to the methods section.
- Lines 332-333 and 338-341: should be moved to the methods section.
Citation: https://doi.org/10.5194/nhess-2021-32-RC3 -
AC1: 'Reply on RC3', Belizario Zarate, 21 Mar 2021
General comments:
The paper by Belizario et al. entitled “GNSS and RPAS integration techniques for studying landslide dynamics: Application to the areas of Victoria and Colinas Lojanas, (Loja, Ecuador)” presents the application of GNSS and RPAS techniques to study the landslide dynamics of two landslides that affected the Victoria and Colinas Lojanas sector of the city of Loja, Ecuador. The direction and velocity of landslide displacements were calculated by GNSS measurements, while orthophotos and DSMs were used to calculate horizontal and vertical displacements in a set of significant points throughout the study area.
Overall, this is an appropriate subject area for the NHESS journal, and the amount of data collected is very important from a risk monitoring and prevention perspective. However, this work should try to better develop the application of topographic techniques to the case of study where some innovative aspects or tools are missing. In addition, the methodology used is not effectively illustrated and lacks some aspects related to an accurate assessment of errors. I believe that this paper has great potential and interesting aspects that could be improved to make it more appealing to a reader but in its current state is not ready now to be published. It requires a substantial upgrading (major review), maybe assessing the limits and errors associated with the used topographic techniques and the comparison with other technologies in terms of landslide dynamics estimation. The text is unclear in some sections (requires English revision and often the terminology used is incorrect, especially regarding the photogrammetric process), difficult to follow for a reader and the structure of the text must be revised because some parts are not in their optimal location (see specific comments below). With some important improvements, this work can be interesting and useful for the scientific community.
Firstly, we would like to sincerely appreciate your comments and suggestions. This an advance of the response we are preparing after the closure of the open discussion.
We are rewriting the abstract and restructuring the sections to make the paper more easily for reading and understanding. We are also rewriting some parts of the methodology and the error assessment, as well as revising some tables and figures as you suggest.
Specific comments
Abstract: I suggest rewriting it to make it more attractive to the reader perhaps emphasizing the innovative aspect of this work and the usefulness of these results in terms of the mitigation of natural hazard problems.
Answer: Modifications are being made to the abstract based on suggestions
Introduction: this part should be underlined the innovative aspects of the work, motivated the choice of technologies used for the surveys, and highlighted the usefulness of the data obtained.
Answer: Suggestion will be added in the manuscript introduction
Methods:
A GoPro 3+ camera was used to carry out the SfM surveys, but it was not shown how the problems related to image distortion were solved given the use of a fisheye lens with a flight altitude very high.
Answer: The fisheye function was not used for the present study, the wide mode was used. Capture mode specified in the manuscript.
The study areas are quite large (around 6 and 2 ha) and 6 and 5 GCPs were used respectively, by what criteria were this number and GCP/CP ratio chosen? Is this number sufficient? (look at this paper: https://doi.org/10.3390/rs10101606
and https://doi.org/10.1016/j.geomorph.2016.11.021).
Answer: Pepa et al (2016, 2019) carried out a study to evaluate the precision of a UAV-based monitoring system, in which a total of 5 GCPs are considered to monitor an area of 6.6 ha, the accuracies obtained are similar to those of the present study. Other studies cited in the text present similar errors and uncertainties. The geographical conditions given in said study are similar in the two study areas. Although studies suggest having a greater number of GCPs, the results obtained were considered relevant.
Where are they located in the study area (a figure could be added about this)? Are the errors related to GCP and CP referred to the point cloud? and the errors related to DSM? Because for the displacements monitoring, the points are extracted from DSMs in stable areas so the interpolation error should also be considered.
We are considering your suggestions, and revising the figures and the tables to include this information. The control points were located on the ground with the utmost care in order to avoid location errors.
Were the DoDs thresholded to account for the errors or do they represent raw differences? Line 180: why was a 3 m threshold chosen?
Landslide evidence in the study area is related to subtle changes in the terrain elevation, usually lower than 3 m and even 1m. Thus to detect visually these changes, we saturate the palette in values higher than 3 m that is enough to observe the landslides evidence. The text has been to express better this idea.
Technical corrections
All technical corrections have been made according to the suggestions given.
Citation: https://doi.org/10.5194/nhess-2021-32-AC1
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 810 | 553 | 51 | 1,414 | 40 | 41 |
- HTML: 810
- PDF: 553
- XML: 51
- Total: 1,414
- BibTeX: 40
- EndNote: 41
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
