the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Surface grain-size mapping of braided channels from SfM photogrammetry
Abstract. Braided channels are known as fluvial systems with a high heterogeneity of physical conditions, resulting from particularly active interacting processes of coarse sediment sorting and transport. This in turn generates a complex mosaic of terrestrial and aquatic habitats, supporting an exceptional biodiversity. However, documenting this physical heterogeneity is challenging, and notably the textural variability of these rivers, which is particularly strong. Distributed and continuous grain-size maps of braided channels are notably of great interest in this regard. In this study, high-resolution imagery obtained from UAV equipped for direct georeferencing were used to produce 3D point clouds (Structure from Motion photogrammetry), from which surface grain-size has been inferred. A set of 12 braided river reaches located in SE of France were used to calibrate a roughness-based grain-size proxy, and this proxy was used for the production of distributed grain-size maps. The calibration curve can be used to determine the surface median grain-size with an independent error of 5 mm (14 % of relative error). Resampling procedure shows a good transferability of the calibration, with a residual prediction error ranging from 5 to 17.5 %. Reach-averaged median grain-sizes extracted from roughness-based grain-size maps were in very good agreement with values collected in the field from intensive grain-size samplings (differences of less than 5 %). Some examples of morpho-sedimentary signatures derived from these maps are provided. They notably show a systematic altimetric gradient of the maximum grain-size of bars, that is interpreted as an hydrological imprint, that should be better integrated into conceptual models of grain-size patchiness developed for these rivers.
- Preprint
(8415 KB) - Metadata XML
-
Supplement
(17877 KB) - BibTeX
- EndNote
Status: open (until 22 Jan 2025)
-
RC1: 'Comment on egusphere-2024-3697', Anonymous Referee #1, 14 Jan 2025
reply
1. GENRAL COMMENTS
This manuscript describes an approach for estimating median sediment sizes over a large spatial scale in multiple sections of French braided rivers. This approached is based on photographic data acquired from UAV platform.The feasibility of such an approach has already been demonstrated in the literature. However, this new study presents some important strengths for the community, such as the large number of rivers studied (12), the study of the transferability of this method and its calibration, the evaluation of performance on different geomorphic units including water channels, and some nice examples of application.However, there are points to be improved/discussed or possibly implemented.- The introduction fails to clearly highlight the scientific gaps that need to be filled.
- There seems to be some information missing from the introduction that would help to really understand your objectives
- A grain size description of the reference data (manual or pseudo ground truthdata or Wolman data) is missing, it should be provided either in the methodology section or in the results section.
- information about distribution of samples surface area/dmax area or grain number per samples could be provided instead of active chanel width in figure 1B
- a few small details may be missing in the methodology and results section. (see below the listed details by lines)
- the errors concerning the estimation of the high percentiles by the photosieving tool (here DGS) and the significant scatter for the prediction of the high percentile with the roughness calibration could be related to a problem with the surface area sampled by photosieving not being large enough (1m² for certain samples being too small), coupled with a surface area for the extraction of the roughness values which is different from that for the extraction of the pseudo ground-truth D50. Each member of the data pair does not cover the same extent
- options to be implemented to give more weight to this study:
- a second photosieving tool of the same type but more recent such as Sedinet (Buscombe 2020) or an oriented object detection tool could have been tested in parallel to DGS to evaluate the importance of the tool used on the subsequent performance of the D50 estimation with roughness on this large dataset composed of 12 rivers.
- In the same vein, a second mesh size to extract roughness could have been tested to assess the weight of this parameter.
- The error of the D50 estimate through the roughness method seems to be accurate compared to the Wolman field data. But it is not certain that these differences in D50 in Table 6 characterise well the performance of your method given that such a large spatial integration may include compensation.
2. SPECIFIC COMMENTSIntroductionThe introduction presents the history of the progress of the techniques available to characterize the grain size distribution of braided rivers. Nevertheless, it seems that some recent publications on the subject are not presented here but appear later in the paper.The introduction fails to clearly highlight the scientific gaps that need to be filled. For sure your work is revelent. You just need to reshape the way you present your objectives..I don't understand exactly what the main objective of this paper is:- To evaluate the performance of large-scale spatial estimation of D50 using an RTK-equipped drone (but GCPs are ultimately used, and the errors are not compared with point clouds constructed without the use of RTK during photo acquisition).
- Evaluation for a sedimentological application? (not clear).
- It is the first examples of application of such a method in this field? (in relation to the results obtained in section 3.3.1 and 3.3.3 ).
I think your first 2 objectives should be reworded. Perhaps there aren't 4 but 3?- The first one mentioned doesn't seem to be an objective in itself. You are evaluating the specific error of the DSG tool for photosieving. Studies already exist on this subject. In reality, I think you want to correct the ‘systematic’ errors of your automatic photosiving to then ensure that you minimise the errors of your D50~roughness calibrations. You do this because you know from your literature reviews that automatic photosiving methods are biased. But the problem is that you never talk specifically about DGS in your introduction, so it's complicated for you to introduce, and for the reader to understand, your 1st objective. Eventually you could have had this objective if you had compared several photosieving tools (at least two). In the end, I would say that either this is a sub-objective of your objective 2, or it could be an objective 1 called ‘to evaluate the automatics phototosieving error and developpement of correction with field-base calibration’.
- Perhaps objective 2 should be reformulated. It seems to be strongly linked to the use of RTK, but then neither the results nor the discussions refer to RTK.
In L 35 to 49 or L 87 to 10: you should clearly introduce the fact that it may be necessary to correct the data obtained by photo-sieving with a calibration.L39: The list of photosieving solutions could be extended to include more recent solutions (ex: Buscomb 2020), the latest of which is already 11 years old.L42: “with typical measurement errors of ...”. Could you specify the error of what, b-axis? number of grains? what type of error, bias? RMSE?L90: rather strange to introduce Basegrain when you don't use it at all in your workflow. In addition, Basegrain is more of an object detection type, whereas DSG is not. Perhaps part of the methodology section 2.2.2 should be reworked in the introduction.L106/107: “but, it has never been tested…”. I recommend that authors consult the publication by Mair et al (2022) (https://doi.org/10.5194/esurf-10-953-2022) which evaluated the variations in performance when using UAVs equipped with RTK or GCP for sedimentological purposes.MethodologyThe methodology section is very well structured. However, the description of the active width of the channel, although obviously very interesting, does not seem to have a place in this paper since it is never included in the results or the discussion.It might be interesting to use this space to describe the particle size distribution of the study sites. It seems unthinkable to carry out this study concerning the estimation of the D50 without giving any information on the particle size in the presentation of the study site, not even in one column of a table.L75/76: “Obtained images..”. How were these images scaled? Do they come from SfM processing (orthophotos)? Or do they come from simple photographs taken with the drone, then scaled using the meter placed next to the frame? is this step performed in a GIS environment? How many georeferencing points were used? 2? 4 points?L180: “Digital Grain-size (DGS) code developed by Buscombe (2013)," It is curious that this DGS tool is not referenced in the introduction, whereas BASEGRAIN is while not been utilized in this study. Additionally, Sedinet, developed by the same author as DGS (Buscomb, 2020), but a more recent tool, is not referenced in the introduction either. I guess the reason is that Chardon et al (2020) have shown that DGS performs better than Sedinet. This could be added somehow in the introduction.L189: “images were corrected...”. Can you indicate how, with which program you are carrying out this phase, which seems to be important for the rest of your study?L199: “At least one hundred ..” What is the distance between the nodes? The number of nodes will be very small to avoid sampling the same grain several times. Could you give us some details? It might be useful to quickly mention the range of the number of grains sampled and inter distance node.L203/204: “several studies demonstrated that this systematic error is low for grid-by-number sampling”.As if the systematic error in grid-by-number sampling were small compared to something else. But what is that other thing?L259: “The radius…”. Why extract the roughness in a sphere rather than in a ‘box’ ? A box covering exactly the same area as the photosieving data could be more suitable for accurately assessing the deviation between the SfM data and the photosieving data? Why use a radius of 0.5m on the SfM data if the calibrations are obtained from photosieving data covering at least 1m² ?L262/263: “all theses predictors..”. It is not clear whether the D50 used in these regression analyses is the raw photosieving D50 or the D50 corrected by the previous calibrations from section 2.2.3 . This should be clearly stated.L301: “No major morphological changes occurred..”. Totally understandable that no campaign was carried out in 2024. Having said that, it would have been interesting to have a new data acquisition for the Arigéol site to assess the differences between the two sets of data (2020/2024) for a similar expected result (the precision, the reproductibility of the method), in comparison to Drac where D50 estimate are expected to be significatively different.ResultsThe section is well structured.L324/325: “Median of the computed...". error compaire to manual data may simply due to sampling area being too small (1 m² for these patches?) to assess the entire distribution correctly, as there are too few coarse grains to characterize them correctly. In addition, using only the grains present on grid nodes and not all the grains present in the photo amplifies this phenomenon. The error may not be so important, but may apear large due to the method used to acquire and process the reference data.L336: “Although the number of grains…”. Another recommendation is the use of a sampling area equal to 100 times the area of the largest grain (Petrie and Diplas, 2000). Instead of chanel width, the distribution of the ratio sample area/dmax area or the grain number per samples could be provided. After an explanation concerning acceptable area sampling or grain number samples, I think you could simply present the error results and establish roughness calibration without these 2 outliers, These errors don't seem to be linked to photosieving, but to the reference data obtantion..Thinking about it, it might be appropriate to use a a roughness extraction area 100 time greater than the expected Dmax to extract the roughness.L360/361: “The high data scatter”. This continues to make me think that perhaps these errors for the high percentiles are linked to the too small size of the surface area photographed for some samples and the size of the mesh used to extract the roughness data, which are also not the same..L379: “The residuals prediction..”. We don't know where to look. Could be changed by: “The standard error of the prediction residuals varies from 5 to 17.5%.L354: “The D50 calibration curve…”. I suppose that the calibration used to map D50 spatially is: D50 = 1.9 Rh + 12 (according to table 4) for all the sites. Could you specify this in the text ?L396: Change Figure 11 by Figure 8Figure 8: very nice output from your workflowL446, 448, 449: change Figure 8 by Figure 9Figure 9: Very interesting result. I wonder how it can be done quickly on more sites. I have the feeling that a limiting factor is the precise delimitation of the area to be mapped in the end, the exclusion of vegetated areas, woods...L454: Change Figure 9 by Figure 10.DiscussionsThe discussions section is well structured. The general comments at the beginning and the points listed below (sometimes linked) can at least be discussed.L476: “shift for small percentiles”. Do you have any explanation?L480: I wonder what the errors might have been if you had removed the 2 outlier patches and also used another photosieving tool as Sedinet for examples instead of DSG. This is just a question that doesn't negate your study, of course.L485: "a very good performance” Perhaps ‘Very’ is excessive, as there is still a 24% error. How many % of error must we accept to consider a result as just ‘good’ (50% error)?L495: “Further studies are needed …” I still think it would have been useful to use at least one other tool to assess whether better calibration with roughness is possible. And also use another radius or "box area" for extracting roughness value.This kind of analyses could have given clues to indicate limiting factors and theire weight.Perhaps using one or more other tools would have given you D50 estimation results similar to your current DGS results, Then you could have conclude that there's no point in using more sophisticated tools. I see this as an option for improvement (not mandatory) which would add interest to your paper for the community. Your work seems perfectible.L506: "and whith shapes dominated by spherical”, Is this your opinion or concrete facts? How did you quantify this ( imbrication, grain shape). Did you measure a sample to confirm this?L512 to 518: "Another limitation…" Again, error on the highest percentile may initially be due to a difficulty in correctly characterising the end of the distribution because the calibration area is too small. To refute this hypothesis, no data is available. No presentation of the particle size distribution of the study reaches is given in the material and methods section, even though this is precisely what this study seeks to predict. It would also be interesting to know the distribution of the ratio area photographed/ Dmax area. Furthermore, you explain these errors by mentioning a high degree of imbrication variability when 8 lines above you talk about homogeneous patches with a low degree of imbrication. It's a bit contradictory, or maybe you need to rework this paragraph.L538/541: You are presenting a method for estimating the D50 with a high resolution (1m²) but the performance (or error) is assessed at an extremely low resolution (Wolman over a distance of at least 800m). How is the average D50 obtained by Wolman representative of the reality on sites? What is the standard deviation, for example, along the different transects? Your quantification of the errors aggregates so many factors that it seems difficult to really isolate the quality of the D50 estimates based on roughness and may results from compensation. Wouldn't it have been better to characterize each of the sedimentary unit in detail with 2/3 field samples per unit, and then compare per unit the D50 estimated using the roughness extracted from these same locations? Of course, this doesn't seem feasible at this stage.L541/547: "Interestingly …underestimation of the median grain-size has been observed”. You mention a "very good" estimate of the D50 in submerged areas, with a "substantial" underestimate. That may be the case for the Arigéol site (around 5%) but it's absolutely not the case for DRAC, with at least a 17% difference. I don't know what you define by the terms ‘substantial’, ‘good’ and ‘very good’ in terms of percentage, but I think they should be avoided.L538/556: This paragraph is a bit messy, it could be reworded with something like this:Comparisons of SfM-based D50 with field-based values obtained from intensive Wolman pebble counts revealed that our roughness-based grain-size maps can be used for extracting a reach-averaged median grain-size of high-quality (less than 5% of error) along several kilometers of river channels, not only at the scale of the whole active channel, but also at the scale of different geomorphic units (unvegetated bars and low-flow channels). Nevertheless, the median grain-size computed in the submerged portion of the active channel present more contrasting results, respectively -5.5 and -17.5% for the Argéol and Dracs site. If low-flow channels are considered, underestimation of the median grain-size has been observed and may increase when water depth increase. Water-depth have been evaluated by subtracting the minimum elevation of the channel obtained at a 10-meter interval from the elevation of the external limit of the channel (water/bar limit). The averaged water depth was 9 cm greater in the Drac (mean value of 0.35 m) than in the Arigéol (mean value of 0.26 m) at the time of drone flights. The refraction effect on SfM topography must then be greater on the Drac, thereby introducing more error into the positioning of the photogrammetric points which probably biases the prediction of the submerged grain-size This means that the submerged roughness is not always a good proxy of the grain-size, notably when water depth increases.Citation: https://doi.org/10.5194/egusphere-2024-3697-RC1
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
170 | 48 | 5 | 223 | 14 | 1 | 1 |
- HTML: 170
- PDF: 48
- XML: 5
- Total: 223
- Supplement: 14
- BibTeX: 1
- EndNote: 1
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1