the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 16 May 2024
River ice analyses and roughness calculations using underwater drones and photogrammetric approach
Abstract. In the Northern Hemisphere, freshwater ice forms a significant part of the cryosphere during winters. River ice cover strongly affects the hydrology and flow characteristics of northern rivers, and the effect can last for several months a year. The magnitude of this effect is on the other hand dependent on characteristics of the ice, especially on subsurface ice roughness. However, ice-covered areas have commonly remained unexplored due to challenging conditions and difficult access. This study focuses on developing an improved approach in studying river ice by applying cost-efficient underwater drone platform and camera solutions in studying the ice underside. Furthermore, the developed methodology utilises a photogrammetric approach, Structure from Motion. One key result of the study is a workflow for reconstructing a digital elevation model of the ice underside. It was found that applied photogrammetric approach also enables calculating roughness coefficient for the ice underside. The results of this study show that underwater drones enable studying river ice in more comprehensive and detailed way compared to conventional methods. Additionally, it is noted that applying Structure from Motion in mapping the ice underside can offer feasible approach in determining subsurface ice roughness, which has wider application potential in modelling fluvial processes in subarctic rivers under changing environmental conditions.
- Preprint
(1431 KB) - Metadata XML
- BibTeX
- EndNote
Status: open (until 30 Jul 2024)
-
RC1: 'Comment on egusphere-2024-1247', Anonymous Referee #1, 20 Jun 2024
reply
General Comments
The authors have prepared a very interesting manuscript on the topic of using underwater drones equipped with cameras to assess under-ice roughness using structure-from-motion photogrammetry. As far as I am aware this is the only such use of this technique, and is therefore quite novel. It complements recent advances in quantifying the top-of-ice roughness of rivers and sea ice using aerial drones.
The writing within the paper could use some improvement. I found it to be somewhat repetitive, and minor grammatical changes throughout the manuscript would help with the overall flow of the paper. Specific suggestions have not been provided since the issue is so prevalent. That said, the writing is understandable and contains no significant errors that I have found.
I would suggest that the ‘track’ of the drone be provided a bit earlier in the paper. I realize that one can see the track based on the point cloud that is provided later; however, I was still left wondering whether what appears to be one single path is in fact a back and forth path. Perhaps a single red line with small arrows pointing in the direction of travel would be helpful? I was also somewhat surprised that a more grid-like pattern was not used, since this appears to be what is done with aerial drone photogrammetry. Perhaps a very brief discussion on why the type of travel path was chosen would be helpful to others who may wish to employ this technique in the future.
The discussion on how the point cloud was divided into cross sections also left me confused. Since based on the point cloud image it appears that no full cross section (i.e. perpendicular from bank to bank) was acquired, perhaps a change in terminology would be a starting point – transects could work. The orientation of the transects was unclear to me, and could be drawn either on the point cloud or on the more simple schematic of the drone track that I’ve suggested. You could label the transects that are later highlighted in Figure 10.
I was surprised that there was no mention of a separate assessment of Manning’s n calculation based on back-calculating the composite roughness and then working out the ice roughness by assuming that the bed roughness remains unchanged throughout the winter. It appears that you’d have most, if not all of what you need. A discharge measurement was apparently done. You had GNSS survey equipment, so the water surface profile could have been measured. You’ve done work in this stretch of river before, so the bed roughness has likely been pre-determined. It would be just one more way to ensure that your Manning’s n value determined through direct ks measurement is grounded in reality. I’d personally rely on this more than the comparison with data from others’ work.
I’d be curious in reading a more thorough description of limitations for this methodology. I’m left with the impression that your site was somewhat ideal for use of this technology. A thicker ice and/or snow cover might have prevented underside of the ice in the video from being visible at all. How might a layer of slush ice have been observed in the video? Do you expect that the more porous ice would have presented challenges? If this was done on a rough ice cover, drilling an access hole at all would have been difficult. The significantly rough under-ice surface and more non-uniform lighting conditions might provide challenges; but then again, perhaps a more distinct roughness would enable more effective tie-in points?
Specific comments
L29 – I don’t believe you mean vertical velocity. I’d call this the vertical profile of streamwise velocity.
L34 – This statement is not universally true (ie. higher flow rate does not always mean smoother ice). For a given river, a low flow rate may allow the formation of smooth skim ice. A higher flow rate may create frazil ice, which will have an initially greater roughness. An even higher flow rate could cause the ice to consolidate, which would again create a higher roughness.
Once an ice cover has formed, the flow can smoothen the ice cover. At that point, higher flow rates could cause the ice to be smoother.
L58 – perhaps ‘scarce’ rather than ‘scant’?
Table 1 – My impression is that you do not have the full range of ice roughness in this table. The underside of an ice jam can be extremely rough, resulting in a Manning’s n value that is well above 0.03.
L319 – ‘keypoints’ seems like a new term at this point in the paper. Should this be one of the previously defined terms instead? If not, please define.
L350 – in which direction were these profiles drawn? Longitudinally, or across the channel? Would it matter? Did you include both directions of profiles, in which case they would intersect? I see that you discuss this somewhat later, but I am still left being confused.
Table 5 – the calculated ks values are reported to an unreasonable number of decimal places – well beyond the accuracy of this methodology. This is also true for the other two columns.
L672 – hyphen not needed in the word ‘combines’
Citation: https://doi.org/10.5194/egusphere-2024-1247-RC1 -
RC2: 'Comment on egusphere-2024-1247', Anonymous Referee #2, 28 Jun 2024
reply
The authors describe an approach to map the underside surface of inland water ice as a Digital Elevation Model (DEM) using Structure from Motion(SfM). The images are captured with a Remotly Operated Underwater Vehicle (ROV). The novelty lies in the combination of using ROVs for SfM for mapping the ise underside in arctic inland waters, while SfM as such has already been used by other others for the same application. The authors tested different viewing geometries (camera angles) to determine the optimal configuration (image quality, geometric details). The authors further derive the Manning coefficient from the captured data and, thus, present a method to directly measure ice-induced roughness rather than back calculating it from hydrodynamic-numerical modelling.
It is an interesting approach. The English reads well and the paper is well structured.
The main points of criticism are listed below.
- The manuscript is overly long. Please try to streamline the paper and restrict yourself to the relevant information. Parts of the methods section, for example, readl like an SfM best-practice tutorial, which should not be the focus of a scientific article.
- The main criticism is the lack of photogrammetric expertize in the entire work. I highly recommend to team up with a specialist in underwater photogrammetry.
- While the description of capturing and modelling the ice underside is cleare an concise, it dod not get clear, how ice thickness was determined. I understand that the focus lies on the maping the underside, specifically the roughness of the underside's surface, but I assume that ice thickness plays a role as well. Please comment on that and adapt the manuscript accordingly.
- I disagree with using low resolution images and high frame rates as the best parameter setting. I assume that the ROV's velocity was slow (< 1m/s)). Thus, high frame rates will result in overly high overlap and reducing the overlap to a reasonable degree but using higher resolution imagery would be better to my strong believe. I even recommend to use still frame images instead of video (if this is an option for the GoPro 10). Please comment and adapt the manuscript accordingly. At least the vendor's recommendation for using high frame rates for capturing fast action is not applicable in this slow-capturing-motion setting. The ROV's velocity should be reported in the paper.
- Georeferencing: The underwater targets are installed through drill holes and are measured with RTK GNSS. While the location above the ice is accurate to the RTK accuracy, it is not stated if the inclination of the poles (where the GCP targets are mounted) was considered. This clearly has an effect on the accuracy of the underwater target accuracy. Please comment. Please also find ideas for the GCP network in the attached commented PDF (i.e. levelled/vertical poles anchored on the ground with multiple targets on it).
- GCP accuracy: A setting of 1cm is overly optimistic. I assume the accuracy rather to be at 5cm level (RTK accuracy, potential errors due to inclination of poles/angles). Please consider and comment.
- Dense Matching strategy (line 322-339): The entire paragraph reads as a best-practice Metashape tutorial. In the paragraph georeferencing, dense point cloud generation, and strategies for handling blocks which broke into >1 chunks. In general, the clear photogrammetric workflow of relative and absolute image orientation, followed by dense matching, followed by DSM generation is not adherered to. Instead, in this paragraph georeferencing follows dense matching, although GCPs have already been measured previously (thus, the model should already be georeferenced). I suggest that the authors contact photogrammetry experts to make the entire image-to-model pipeline more sound.
- Roughness calculation: You describe that your block (i.e. point cloud) suffered from a spherical/cylindrical deformation. To calculate roughness, you segmented the data and horizontally aligned the piecewise deformed segments. This entire process could have been automated and optimized by calculating the roughness in a sliding-window approach based on the de-trended height coordinates within the local neighbourhood. If you are only interested in the roughness, and not so much in the absolute shape itself, then this would have speared you the segmentation and "horizontal alignment".
- Ground Sampling Distance(GSD): This important value (i.e. the size of an image pixel on the ground/ice surface) of every photogrammetric block is not reported. The (local) height accuracy mainly depends on the GSD (rule of thumb: sigma_z=2 x GSD) and, thus, knowing the GSD would enable to draw conclusions on the detectablility of the smallest height deviations. In other words, the GSD limits the detectablilty of small-scale roughness.
- Methods section: The section is overly long and often contains discussion related material (justifications) to a too high degree. Please separate methods and discussion more clearly and streamline the workflow description (short and concise instead of describing every detour).
- Results: Accuracy vs precision: You reported the RMS at the three GCPs. Due to the lack of redundancy, the reported values cannot be seen as representative for the absolute block accuracy. In addition, the more interesting measure would be precision, i.e. local errors. These could be derived by comparing the (detrended) relative coordinate differences for the individual markes mounted on the poles (Fig 4). Thus would yield an estimate of the aibility to derive small-scale roughness.
- Figure 9: It is hard to understand the correspondence (left colum - image; right column - DEM). You do have a georeferenced block, thus, it is possible to exactly locate the images. Please improve the Figure in this respect. The ROV positions are well known via the exterior orientation of the images.
- I see it as one of the weak points of the paper that there is no independent reference data. If roughness estimation and shape reconstruction are the primary goals, then it would have been possible to saw out a block of ice after capturing it with the ROV and to measure the ice underside from above (full frame cameras, terrestrial laser scanning, total station). That way, the ability to capture the ice underside with a ROV/camera system could be verified and quantified.
More comments can be found in the attached PDF.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 195 | 48 | 17 | 260 | 11 | 12 |
- HTML: 195
- PDF: 48
- XML: 17
- Total: 260
- BibTeX: 11
- EndNote: 12
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1