essd-2023-94
Multitemporal characterisation of a proglacial system: a multidisciplinary approach
I carefully read the authors' response to my comments and looked at the manuscript with the tracked changes. The authors have made a great effort to respond to my comments, and the article has benefited greatly from this work. As I pointed out in my previous paper, I consider this work a very good contribution and it demonstrates the importance of studying the functioning of proglacial margins with a multidisciplinary approach. The results presented here will allow further investigation of the Rutor Glacier and its proglacial margin, and the papers that follow will certainly be of interest to the geomorphological community.
The paper is now stronger, more informative, and better structured. However, I still have some concerns about the error assessment in the DSMs, but it may be that I am too picky.
The authors present a first DSM assessment in Table 4. They calculate the vertical and horizontal RMS of the GCPs and checkpoints, but it would have been more informative to include the mean error and the standard deviation of error as well (this would allow characterizing systematic and random errors in the DSMs). Then, the authors calculate such statistics in Table 5 by using the delta of Z of a number of stable points, however these results are not informative since those stable points are not well scattered across the study zone.
By looking at the DoD in Figure 9a, it appears that the DSMs are of good quality, and no apparent systematic deformation is occurring. However, without a deeper error assessment some doubts might arise. Building on my own experience in error assessment in DSMs and DoDs, I would suggest two paths to assess the DSMs.
1. Assess the DSMs by using your independent checkpoints, which are widely distributed across the study area, although their number is low (but this is OK in such extreme and dangerous environments). This analysis would inform if your DSMs are impacted by random errors (std of error) and systematic errors (mean error). In the presence of systematic errors, check if those errors have some spatial structure (e.g., systematic tilt, which I formerly named datum shift), and if any correct it in respect to your checkpoint network (if possible). Once the presence of systematic errors is assessed (and corrected), use the std of error to calculate your Limit of Detection (LoD), either at 68 or 95 confidence limits. In this way, you do not violate the assumption of the error propagation theory which states that the errors in your DSM should be random, independent and gaussian. I suspect that this approach would provide a large LoD, but this is fine.
2. Assess your reference DSM (i.e., 2021) by using your independent checkpoints, and check for systematic errors. In the presence of systematic errors, check if those errors have some spatial structure (e.g., systematic tilt), and if any correct it in respect to your checkpoint network (if possible). This would inform about the quality of your reference. Then create a new set of checkpoints based on a number of widely distributed stable points (e.g., bed-rock, etc.). Calculate the shift between the reference (2021) and the DSM of 2020, and assess the presence of random and systematic errors. In the presence of systematic deformations such as tilt, correct them. Once the presence of systematic errors is assessed (and corrected), use the std of error of your stable points to calculate the LoD.
Both ways are widely accepted, although (1) is probably more robust although your checkpoint population is somehow little. I would suggest discussing this with the editors, and get to know what their expectations in terms of error assessment are. I want to emphasize that my suggestions should not be seen as criticism, but as genuine suggestions to improve the quality of the article, which - as I have already said - in my opinion is already good. I am of the opinion that the authors can respond to my concerns fairly quickly and I don't think another round of review by the referees is needed.
I know the challenges of collecting (and processing) this amount of data in these extreme environments, therefore I compliment the authors for having produced and discussed this remarkable dataset.
Some specific comments are listed down below.
Line 192: "During the 2021 field activities, a total of 32 artificial photogrammetric markers"
Following Table 2, you deployed 33 markers. Either the text or the table are wrong.
---
Lines 223 – 225: "Outwash plains, which may have been affected by geomorphological changes (e.g. due to erosion and water deposits) between the time of the surveys, were also considered stable areas."
I do not see the point of using such points if you want to assess the error in your datasets. Outwash plains are extremely unstable as you pointed out. What is the rationale of choosing points within the outwash plain? You bring in this point later in the paragraph, stating that using such points may have worsen your statistics, which again questions the rationale of choosing them. Could you please clarify? How many stable points did you select?
---
Lines 352 – 354: "A standard Structure-from-Motion (SfM) photogrammetric approach was adopted, following a consolidated workflow (i.e. interior and exterior orientation, camera calibration, dense point cloud generation, DSM and orthomosaic generation) using the software Agistoft Metashpe."
I would move this sentence to methods. I am a bit picky here, but could you please clarify what consolidated workflow means. I guess you mean the estimation of the internal/external orientation parameters through the alignment in Metashape, then markers, re-fitting through a bundle adjustment (camera calibration), and so on. Is this the workflow? Perhaps, it would be useful for future readers to specify the parameters you used within the bundle adjustment (e.g., focal length, principal point offset, etc.).
---
Table 4: See my general comments about including the mean error and std of error.
---
Line 365: "about 0.2 m"
Consider using 20 cm instead of 0.2 m.
---
Lines 377 – 379: "Glacier surface elevation differences were estimated by subtracting 2021 DSM to 2020 one, to quantify ablation and displacement (Table 5)"
I suspect that the reference to Table 5 is wrong here.
---
Lines 380 – 385: The introduction of a LoD is indeed very good, however there is no reference to the works of Brasington et al. (2000) and Lane et al. (2003), both published in ESPL, that are the foundation of the LoD theory in DoDs (in my opinion). Furthermore, you use the RMS of Table 4 that was previously noted as accuracy and here as precision. When calculating the LoD, one should use precision i.e. the standard deviation of error as illustrated in both Brasington et al. (2000) and Lane et al. (2003) papers.
---
Table 5: I appreciate the choice of including Table 5 in the results, however I think that the way it is presented now does not provide an informative assessment since the stable points cover a fraction of the study area and they are not widely distributed across the study zone. I acknowledge the difficulty of selecting stable points in such unstable environments, but you might find stable zones even in close proximity of the glacier (e.g., bed rock) so that your assessment becomes more informative. Please refer to my general comments at the beginning of this report, and if possible calculate the statistics again to leave no room for any doubt.
---
Line 386: "Figure 9 (a) shows the differences between the 2021 and 2020 DSMs adopting a LoD threshold of 95% = 40 cm"
Since you used the 95% LoD in your DoD (Figure 9a) I would remove the description of the 68%. I do not see any point of including the calculation of the 68% if you do not use it in further analysis.
---
Figure 9a: The DoD is now clearer with the bivariate scale and the LoD.
I would suggest using the color white for changes that fall within the LoD or even use the orthomosaic as background (with LoD range in transparent), so that significant changes are easier to see. Furthermore, consider using more classes to be more informative, and perhaps use better contrasting colors. I would also suggest reversing the scale, since now it shows that the glacier termini gained in Z between 2020 and 2021. This results from subtracting the DSM of 2021 from the DSM of 2020 instead of doing the opposite, that is subtracting 2020 from 2021. |
The authors present a multi-temporal and multi-disciplinary characterization of the pro-glacial margin of the Rutor Glacier. The dataset that comes out from this characterization is remarkable, and shows the importance of using a multi-disciplinary approach to appreciate the functioning of pro-glacial systems in full. The paper is well written, and I enjoyed reading it. The paper fits within the aims and scope of ESSD, and it is worth publishing but I would like to suggest some modifications that may improve the overall quality of the manuscript.
---------
General comments:
Section 2.2.1 Geomatic survey: The section does not provide a detailed explanation of the photogrammetric processing, and it is not supported by the relevant literature. In my opinion, this section needs to be improved. For instance, the name "SfM-MVS photogrammetry" is never mentioned throughout the article and no effort is made to explain the steps needed in a rigorous photogrammetric study. A few examples: the authors do not mention how the images were collected (e.g., drone flight geometries), how many ground control points they used, or which software or freeware they used to process the images. Finally, the author do not provide any information on the quality of the DSMs, therefore questioning if their results are reliable or not. This important weakness needs to be addressed.
Section 2.2.4 Bedload monitoring: As per the geomatics, I believe there is the need of providing a more detailed explanation of the bedload monitoring since the use of seismometers in bedload studies is relatively recent. It would be very useful to know in more detail how the data were processed and the steps required to go from the raw signal to the results presented here.
Data availability: The 2020 orthophoto and DSM are not available on Zenodo (https://zenodo.org/record/7713299). Are you going to include them in future?
---------
Detailed comments:
1. Lines 26 – 28: “Alpine glacier retreat is leading to increased exposure of formerly glaciated terrain, entailing the colonization of plants and animals, and changes in morphodynamics and sediment transfer.”
Consider adding one or more citations here.
2. Lines 29 – 30: “Little Ice Age (LIA)”
You already defined the acronym; perhaps just use LIA instead of “Little Ice Age (LIA)”.
3. Lines 35 – 36: “On the one hand, plant colonization stabilizes glacial sediment and reduces sediment fluxes; on the other hand, geomorphic processes disturb and limit vegetation succession.”
Consider adding one or more citations here.
4. Lines 54 – 55: “Sediment yield depends on water discharge and sediment availability which are both highly variable in space and time.”
Consider adding one or more citations here.
5. Line 148: “manned photogrammetric flights”
I would move away from “manned” and describe those as crewed or airborne.
6. Lines 154 – 159:
How many targets did you use in total? Did you deploy the targets only in 2021? Why did you not consider collecting independent checkpoints for quality assessment?
7. Line 160: “Unlike drone flights which were oriented exploiting a direct georeferencing approach”
What do you mean with “direct georeferencing”? Did you use the camera positions alone? If yes, why? The GPS onboard of the DJI P4 is of poor quality for high-precision photogrammetric surveys, and it is a standard practice to use ground targets in SfM-MVS studies (particularly to reduce the occurrence of systematic deformations in DEMs).
8. Lines 163 – 166: “Due to a large number of well-distributed ground control points, the 2021 aerial survey was considered the reference model (referred to as ’Model” Zero’) to be used for multitemporal analyses. The 2020 survey was, therefore, co-registered (i.e., georeferenced in the same reference system, enabling the overlap of all the derivative products) with the 2021 survey.”
This suggests that you did not use any target in 2020 (see comment 6), am I right? How did you co-register the 2020 survey? Could you please explain the co-registration procedure? Could you provide statistics on the quality of the co-registration?
9. Lines 282 – 284: “The aerial DSMs were preliminary compared to the LiDAR DSM as of 2008 available on Valle d’Aosta Geoportal to verify the consistency of the produced model, checking the stability of the periglacial rocky areas. Subsequently, 2021 and 2020 DSMs were subtracted to quantify glacier ablation and displacement”
I would move this section into the methods, and explain how you compared the DSMs. In the results, it would be more informative to provide the statistics of such a comparison (e.g., mean error, std of error – not the RMSE) in order to demonstrate that your DSMs were free of systematic (mean error close to 0) and random (std of error close to 0) errors.
10. Figure 8a:
I am a little concerned about the way you presented the DSM of difference. First, why did you not use a bivariate scale (from –X to +X)? A bivariate scale would help a lot in my opinion. Second, why did you not apply a Limit of Detection? The use of a Limit of Detection is a common practice in DoDs, and allows showing changes that are statistically significant (e.g., at 68 or 95% confidence limits). Lastly, from the DoD presented in Fig. 8a it seems that the whole study area experienced at least some movements in Z, is that really possible? Did you check for systematic deformations (e.g., doming, datum shift) in your DSMs? There is the need of providing statistics that illustrate the quality of your DSMs, e.g. the mean error in Z (i.e., systematic errors) and the std of error in Z (i.e., random errors) in respect to reference point altitudes or independent check points.
11. Line 289: “Additionally, a comparison with the 2008 DSM shows a lowering of glacier surface up to 50 meters in glacial front areas”
Where is this result presented?
12. Lines 290 – 293: “As far as very high-resolution satellite stereo pairs are concerned, they enable the extraction of 3D information with a lower vertical accuracy (metric level) with respect to aerial and drone data. Nevertheless, the coverage of a much larger area (in the range of hundreds of square kilometres) enables a multiscale and multiplatform approach to identify the most critical areas where to focus the monitoring activities in the field”
You do not present these results nor discuss them later, what is the point of including such a thing?
13. Lines 300 – 303: “The x-y-z locations of the first interface, representing the lake bottom, detected in all the GPR sections, were interpolated to produce a bathymetry map (Figure 10, which also displays the sediment thickness distribution and the electrical conductivity measurements). The perimeter of the lake, retrieved from the 6-cm-resolution orthophoto, was useful to fix the 0-depth in the interpolation process.”
This section reads like methods. I would move it to the methods.
14. Lines 326 – 334: “The ecoLog1000 and CTDs instruments were first installed in July 2021 and June 2022, respectively. The measuring periods of each sensor are shown in a time: measured-quantity diagram in Table 1. At the L4 gauging station, a set of velocity-based discharge measurements (Q) taken in the summer of 2021 and 2022 were related to the corresponding water depth measured at the gauge (h), in order to plot the stage-discharge diagram (Fig. 11(a); details of the procedure followed to determine the stage-discharge relationship are given in Appendix A). Discharge measurements were also used to calibrate the lake outflow curve, i.e., the relationship between the hydraulic head (H) in the lake and the flowing discharge (see Fig. 11(c)). For this purpose, a linear fitting between the water depth at the gauge (h) and the Hydraulic head in the lake (H) was also calibrated (Fig. 11(b), R2 ∼ 0.98), since the water levels in the lake and in the control cross-section in the stream are strictly related but not equal, due to the head-dependant outflow process and water speed.”
This section reads like methods, I would therefore re-arrange this part.
14. Line 357: “permits the identification of time intervals characterized by intense transport”
Although bed load would easily occur at high discharges, is the signal you see necessarily related to intense transport events? Or, could the peaks be related to flow turbulence instead?
15. Lines 358 – 359: “Raw seismic signals were filtered in the band 5-95 Hz and then the envelope was calculated as the average of the absolute value of the filtered signal over a time window of 1 min”
The sentence reads like methods, consider moving it in the appropriate section.
16. Lines 363 – 364: “In 2021, we directly observed the absence of bedload transport in three days (10 July, 20 July and 13 September).”
What does “directly observed” mean here? Could you be more precise?
17. Lines 364 – 371: “During the 2022 season, we performed direct measurements of bedload transport at the glacier mouth by means of portable samplers on the occasion of one day of intense glacier melt (14 July) and at the end of the monitoring season (16 September). Bedload traps (4 mm mesh size, 20 × 30 cm opening, (Bunte et al., 2004)) were deployed simultaneously at 2 positions. Measured unit bedload rates feature a large variability ranging from 0.02 to 16.2 kg/m/min in a few hours, as already observed in glacierized basins (Coviello et al., 2022). Bedload samples were sieved and weighed to obtain the grain size distribution. The total bedload transport rate Qs (kg/min above 4 mm) for each sampling period (ranging 370 from 2 to 30 min) was estimated as width-weighted averages based on the available positions sampled.”
This section really is about methods, I would therefore move into the methods.
18. Lines 416 – 417: “It is important to stress that the accurate georeferencing of all the acquired data with respect to the same Datum plays a crucial role in the data integration phase and in enabling the multitemporal analyses.”
This is right, but what about systematic deformations that could well lead to erroneous multitemporal analysis? See comment 10.