the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Sediment source and sink identification using Sentinel-2 and (kayak-based) lagrangian river turbidity profiles on the Vjosa River
Srividya Hariharan Sudha
Gabriel Singer
Peter Molnar
Abstract. Measurement of SSC at a basin outlet yields a basin-integrated picture of sediment fluxes, however it does not give a full spatial perspective on possible sediment sources, sinks, and pathways within the catchment. More effortsome spatially resolved estimates of suspended sediment concentrations (SSC) can be used to identify sediment sources, track erosion gradients in river basins, and quantify anthropogenic effects on catchment-scale sediment production, e.g. by dam construction or erosion control. Here we explore the use of high-resolution Sentinel-2 satellite images for this purpose in narrow and morphologically complex mountain rivers, combined with ground station turbidity sensing for calibration, and supported by a lagrangian kayak-derived river profile measurement. The study is carried out on the Vjosa River in Albania, which is one of the last intact large river systems in Europe. We developed a workflow to estimate river turbidity profiles from Sentinel-2 images including atmospheric, cloud cover, and deep water corrections, for the period May 2019 to July 2021 (106 images). In-situ turbidity measurements from four turbidity sensors located along the Vjosa River provided ground truthing. A multivariate linear regression model between turbidity and reflectance was fitted to this data. The extracted longitudinal river turbidity profiles were qualitatively validated with two descents of the river with a turbidity sensor attached to a kayak. The satellite-derived river profiles revealed variability in turbidity along the main stem with a strong seasonal signal, with the highest mean turbidity in winter along the entire length of the river. Most importantly, sediment sources and sinks could be identified and quantified from the river turbidity profiles, both for tributaries and within the reaches of the Vjosa. The river basin and network acted as a sediment source most of the time and significant sediment sinks were rare. Sediment sources were mostly tributaries following basin-wide rainfall, but also within-reach sources in river beds and banks were possible. Finally, we used the data to estimate the mean annual fine sediment yield at Dorez at ~2.5 ± 0.63 Mt/y in line with previous studies, which reveals the importance of the Vjosa River as an important sediment source into the Adriatic Sea. This work presents a proof of concept that open-access high-resolution satellite data has the potential for suspended sediment quantification not only in large water bodies but also in smaller rivers. The potential applications are many, from identifying sediment sources, activation processes, local point sources, and glacial sediment inputs, to sediment fluxes in river deltas, with a necessary future focus on improving accuracy and reducing uncertainty in such analyses.
Jessica Droujko et al.
Status: open (until 05 May 2023)
-
RC1: 'Comment on egusphere-2023-156', Anonymous Referee #1, 13 Mar 2023
reply
The manuscript (MS) entitled “Sediment source and sink identification using Sentinel-2 and (kayak-based) lagrangian river turbidity profiles on the Vjosa River”. Overall, the manuscript is well written. However, I have a few specific questions which have to be addressed for a better understanding of the concept. The following comments may help the authors to improve the manuscript.
- Page 3, Paragraph 2: Although authors have mentioned literature for remote sensing-based SSC estimation methods, please categorized the existing methods, viz., semi-analytical approach and empirical approach. Subsequently, describe the pros and cons of approaches to motivate the present study.
- Page 3, Paragraph 2: For narrow-width rivers, there are several existing research studies are available using Landsat, Sentinel-2, Synthetic Landsat (derived from MODIS), and other commercial satellites (e.g., CubeSat, Planet lab satellites, etc.). Please highlight those studies in this section.
- Figure 1: Please provide the elevation of the study site and demark the tributaries, main river, and catchment area distinctly.
- Page 6, Paragraph 2, Section 2.2: While highlighting the methodology, the authors mentioned that they have first established the regression model in the section where in situ data is available and then, the models are used in the entire mainstream. I agree with this hypothesis is true for lakes and estuaries where the water is nearly stagnant or less affected by lateral flows; however, adopting the same for dynamic river systems is still questionable.
- Page 6, Paragraph 2, Section 2.2: What is the significance of the citation “Talluto (2023)” in this statement?
- Page 6, Section 2.2.2: Why the ACOLITE correction was adopted in this study? Although there are several other atmospheric correction algorithms are available for water colour inversion studies in the literature, why its only ACOLITE? It may be SNAP (specially developed for Sentinel) or SeaDAS etc.
- Page 6, Paragraph 2, Section 2.2: What is the average depth of water for deep water? Is there any threshold defined for deep water?
- Page 6-7, Paragraph 1, Section 2.2: When the authors considered the rivers are narrow, especially the tributary, the water pixels are more pronounced to the land adjacency effect. How to address those effects while considering this framework?
- Page 7, Paragraph 1, Section 2.2: Please mentioned the type of interpolation followed during the processing of the images.
- Page 8, Section 3.1: From the results, it was found that the best correlation of the in situ turbidity was found in the blue band of electromagnetic radiation also represented by Eqn. 1. However, it is completely different from the findings of other studies as the majority of the literature tested and proved the sensitive bands are red and NIR bands (as elaborated in Page 9, Section 3.1) which is also obvious due to the peak occurred in that region only. So, on what basis, the blue band will be considered?
- Page 8-9, Section 3.1: What are the correlation values in the red and NIR region when the reflectances of ROI (region of influence) are compared against the in situ SSC or Turbidity? Please provide scattered plots for individual stations (like region-specific).
- Page 8-9, Section 3.1: Band ratio techniques are a very popular technique for both SSC and turbidity estimation along the river claimed in many literatures. Include observations while comparing the developed model with the existing models.
- Page 8-9, Section 3.1: As this study emphasized narrow-width rivers, please include the studies in this section that were exclusively conducted along rivers using the finer resolution satellite images (Sentinel, Landsat, or any derived satellite products).
- Page 8-10, Section 3.1: If the particular methodology will shift to another river system worldwide, how to implement this? Please add a comprehensive discussion regarding this.
- Page 10, Section 3.2: How the land adjacency effect was addressed while mapping the turbidity along the narrow-width river section?
- Page 14, Section 3.3: With reference to #Comment 15, while averaging the 106 number of image time series river pixels, the uncertainty of Turbidity estimation is still existing and due to the land adjacency effect, the estimation may be quite high near the bank. In this aspect, it will be very much difficult to judge the concept provided in this section. Please provide substantial evidence regarding the claim.
Citation: https://doi.org/10.5194/egusphere-2023-156-RC1 -
AC1: 'Comment on egusphere-2023-156', Jessica Droujko, 21 Mar 2023
reply
-
Page 3, Paragraph 2: Although authors have mentioned literature for remote sensing-based SSC estimation methods, please categorized the existing methods, viz., semi-analytical approach and empirical approach. Subsequently, describe the pros and cons of approaches to motivate the present study.
We will categorize these approaches as mostly empirical and we will briefly describe the pros/cons of this approach (e.g. this method has been successfully implemented/validated by other studies however, the work is not directly physically-based). We are not sure if the reviewer has seen that in a later section (3.1), we have further categorized that various empirical equations (band values) used. -
Page 3, Paragraph 2: For narrow-width rivers, there are several existing research studies are available using Landsat, Sentinel-2, Synthetic Landsat (derived from MODIS), and other commercial satellites (e.g., CubeSat, Planet lab satellites, etc.). Please highlight those studies in this section.Thank you, we will include references to these additional studies.
-
Figure 1: Please provide the elevation of the study site and demark the tributaries, main river, and catchment area distinctly.
We will incorporate the suggested changes. -
Page 6, Paragraph 2, Section 2.2: While highlighting the methodology, the authors mentioned that they have first established the regression model in the section where in situ data is available and then, the models are used in the entire mainstream. I agree with this hypothesis is true for lakes and estuaries where the water is nearly stagnant or less affected by lateral flows; however, adopting the same for dynamic river systems is still questionable.Although it is true that lateral flows certainly affect the turbidity of the river in the lateral direction, the degree to which lateral flows affect the turbidity differs from reach to reach and riffle to riffle. Here we wanted to see the limits of applying a single regression to a dynamic system. We know that larger errors are introduced from the application of one general regression to the entire stem (as opposed to individual regressions for the reaches around the individual in-situ stations - further explained in section 3.1) and we quantified these errors and compared the R2 of the regressions (overall vs. single stations) in section 3.1.
-
Page 6, Paragraph 2, Section 2.2: What is the significance of the citation “Talluto (2023)” in this statement?We apologize. This seems to have been an oversight. The citation will be removed.
-
Page 6, Section 2.2.2: Why the ACOLITE correction was adopted in this study? Although there are several other atmospheric correction algorithms are available for water colour inversion studies in the literature, why its only ACOLITE? It may be SNAP (specially developed for Sentinel) or SeaDAS etc.Although there are many atmospheric correction algorithms for water, ACOLITE was specifically developed for atmospheric correction over inland waterbodies.
-
Page 6, Paragraph 2, Section 2.2: What is the average depth of water for deep water? Is there any threshold defined for deep water?The average depth of the water for deep water pixels is unknown. An explanation of how the threshold (in sr-1) was selected is explained in section 2.2.2, but it was mostly done through trial and error through visual inspection of the Sentinel images in addition to personal observations during the kayak sampling.
-
Page 6-7, Paragraph 1, Section 2.2: When the authors considered the rivers are narrow, especially the tributary, the water pixels are more pronounced to the land adjacency effect. How to address those effects while considering this framework?Since the entire width of the river, from headwaters to sea for the main stem and tributaries, is narrow, we assume that land adjacency has an effect everywhere (on all processed pixels in this study). To investigate the effect of land adjacency would go beyond the scope of this work, but adjacency effects are important and should be investigated on narrow (and large) rivers.
-
Page 7, Paragraph 1, Section 2.2: Please mentioned the type of interpolation followed during the processing of the images.We simply split each pixel into four pixels (to fit into the 10x10m grid). We will clarify this.
-
Page 8, Section 3.1: From the results, it was found that the best correlation of the in situ turbidity was found in the blue band of electromagnetic radiation also represented by Eqn. 1. However, it is completely different from the findings of other studies as the majority of the literature tested and proved the sensitive bands are red and NIR bands (as elaborated in Page 9, Section 3.1) which is also obvious due to the peak occurred in that region only. So, on what basis, the blue band will be considered?Some of the outlined studies in section 3.1 also used the green band, not only red and NIR. However, since there are many factors that affect the optical properties of inland water bodies, and since these empirical equations are not physically-based, we decided to investigate all of the available bands and fit models to several combinations of bands. Using the Bayesian Information Criterion (BIC), we were able to select an empirical model using only those bands that act as the best predictors for turbidity (in our catchment).
-
Page 8-9, Section 3.1: What are the correlation values in the red and NIR region when the reflectances of ROI (region of influence) are compared against the in situ SSC or Turbidity? Please provide scattered plots for individual stations (like region-specific).We see the benefit of the referee's suggestion to investigate the effect that turbidity has on the red and/or NIR bands (since these bands were discussed already in section 3.1 and used in several past studies). We will provide some additional information on the relationship between red+turbidity and NIR+turbidity within each ROI.
-
Page 8-9, Section 3.1: Band ratio techniques are a very popular technique for both SSC and turbidity estimation along the river claimed in many literatures. Include observations while comparing the developed model with the existing models.We are unsure what the author means by "include observations." During our BIC analysis, we had also investigated empirical relationships using band ratios, however eqn. (1) was the best predictor of turbidity in our catchment.
-
Page 8-9, Section 3.1: As this study emphasized narrow-width rivers, please include the studies in this section that were exclusively conducted along rivers using the finer resolution satellite images (Sentinel, Landsat, or any derived satellite products).Here, we wanted to give an impression of the performance range (R2) in similar studies and there were very few studies conducted on narrow rivers. For this reason, we included studies from wide rivers and other inland water bodies.
-
Page 8-10, Section 3.1: If the particular methodology will shift to another river system worldwide, how to implement this? Please add a comprehensive discussion regarding this.Thank you for this comment. This would certainly add to our paper, we will engage with this question.
-
Page 10, Section 3.2: How the land adjacency effect was addressed while mapping the turbidity along the narrow-width river section?Unfortunately, addressing the effects of adjacency goes beyond the scope of this work (see reply to comment #8).
-
Page 14, Section 3.3: With reference to #Comment 15, while averaging the 106 number of image time series river pixels, the uncertainty of Turbidity estimation is still existing and due to the land adjacency effect, the estimation may be quite high near the bank. In this aspect, it will be very much difficult to judge the concept provided in this section. Please provide substantial evidence regarding the claim.We assume that land adjacency affects all of the river pixels. It would be interesting to see a future study that investigates at which point (e.g. at which river width) does land adjacency not affect the pixels in the middle of the river. With this information, we could determine where and when to remove the effects of land adjacency. Land has a general effect on the river pixels in our methodology, one effect of which is the "land adjacency effect." We will add a small section in the outlook addressing all issues stemming from land adjacency.
Citation: https://doi.org/10.5194/egusphere-2023-156-AC1 -
Jessica Droujko et al.
Data sets
Data used to produce this work Jessica Droujko https://doi.org/10.5281/zenodo.7590129
Model code and software
Code used to produce this work Jessica Droujko https://doi.org/10.5281/zenodo.7590129
Jessica Droujko et al.
Viewed
HTML | XML | Total | BibTeX | EndNote | |
---|---|---|---|---|---|
273 | 78 | 9 | 360 | 3 | 2 |
- HTML: 273
- PDF: 78
- XML: 9
- Total: 360
- BibTeX: 3
- EndNote: 2
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1