the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Evaluating and developing a model of specific degradation using geospatial analysis for sediment erosion management in South Korea
Abstract. The South Korean Peninsula is subject to hydrological extremes, and 70 % of its terrain is mountainous, with sharp ridges and steep valley flanks. Recently, rapid urbanization has created an emerging demand for largescale water resources, such as dams and reservoirs. Accordingly, complicated sedimentrelated problems have become an issue, with abundant soil loss during typhoons transported to the reservoirs, and downstream, riverbed degradation is caused by intercepting sediment. Thus, a reliable approach is required for predicting sediment yields of soil erosion and sedimentation. In this study, the specific degradation (SD) of 62 streamriver watersheds and 14 reservoir watersheds were calculated from field measurements of sediment concentration and deposition. Estimated SD ranged between 10 and 1,500 tons·km^{−2}·yr^{−1}. Furthermore, existing empirical models of sediment yield are insufficient for predicting specific degradation upstream of the reservoirs; therefore, a new model was developed based on multiple regression analysis and model tree data mining of 47 watersheds (~75 % national land cover). Accuracy of the developed model was enhanced with the following significant parameters: (1) drainage area, (2) mean annual precipitation, (3) percent urbanized area, (4) percent water, (5) percent wetland and water, (6) percent sand at effective soil depths of 0–10 cm, (7) slope of the hypsometric curve, and (8) watershed minimum elevation. Additionally, erosion maps from the revised universal soil loss equation (RUSLE) were generated to validate model variables and further understand the sediment regime in South Korea. The gross erosion results for 16 ungauged watersheds were used to validate the empirical model by comparing sediment delivery ratios of other references. The modeled meaningful parameters were examined via remote sensing analyses of satellite and aerial imagery and revealed the features affecting erosion and sedimentation with an erosion loss map at 5m resolution. Vulnerable areas of soil loss, including construction sites, and croplands, as well as sedimentation features, such as wetlands and agricultural reservoirs, were highlighted.
 Preprint
(2352 KB)  Metadata XML
 BibTeX
 EndNote
Status: closed

RC1: 'Comment on hess2021225', Anonymous Referee #1, 01 Jul 2021
General comments
This study proposed a method to predict sediment yields of soil erosion and sedimentation in South Korea. The authors first developed an empirical prediction model of specific degradation based on multiple regression analysis and tree data mining. In addition the also predicted gross erosion based on RUSLE and derived sediment delivery ratios by dividing gross erosion prediction estimated from RUSLE by sediment yield estimated with the empirical prediction model (Eq. 12) in order to validate the results of the empirical model.
I found the paper very difficult to read. To my opinion, it is mainly due to the fact that the objective of the paper is not clearly stated. As it stands, this paper gives the feeling of a confused compilation of the previous papers (Kang et al., 2019 and 2021) without the added value of this new paper being really defined and discussed. This feeling is reinforced by the fact that more than half of the discussion focuses on the effect of spatial resolution on the RUSLE results, although this is not a central objective of the paper.
Another major concern is the use of SDR to validate the empirical model (cf. chapter 3.3 Model validation using the sediment delivery ratio) when the authors do not have measured SDR data (nor gross erosion references). To my opinion, the use of SDR data from the literature does not allow any conclusion to be drawn on the validation of the sediment yield prediction (or specific degradation) model, especially as these SDR values are known to be strongly linked to hydrosedimentary connectivity of the considered watershed, and therefore very sitedependent (see for instance De Vente et al., 2007, DOI: 10.1177/0309133307076485). As a consequence, there is little justification for the paper's general approach of articulating an empirical catchment model and a RUSLE approach to derive SDR values...
In addition, the discussion has to be completely rewritten in relation to the objective of the paper.
Specific comments
Line 2021. You use "percent water" and "percent wetland and water" as distinct parameters of the empirical regression analysis. What about the collinearity of these 2 parameters ?
Line 22. Please clarify the sentence "Additionally, erosion maps from the revised universal soil loss equation (RUSLE) were generated to validate model variables". Which model variables are supposed to be validated by the RUSLE approach ?
Lines 9293. Finally which values of the trap efficiencies were used in your studies ? Why don't you use Heinemann’s formula or a similar formula to estimate them (Heinemann, H. G. 1981. A new sediment trap efficiency curve for small reservoirs. Water Resources Bulletin, 17, 825830.)
Line 99102. A more detailed presentation of the SD assessment procedure is required in this paper. For future works, I particularly suggest to provide a evaluation of uncertainties in your SD estimations, because this evaluation is necessary if you want to show that your new empirical model is better than the previous ones. In a scarcedata context as yours, it can be assumed that the uncertainties on the SD values will be large (as seen in Fig. 2a for the SY/discharge relationship) and that therefore a variety of models can give similar results when considering the uncertainties in the calibration/validation datasets.
Line 109. I am a bit surprised that some gauging stations were removed from this study whereas many of them were used in the previous ones... To what extent could this partly explain the improved quality of the regression?
Line 111 (section 2.2). Please consider dividing this section in several subsections to make this section easier to read and understand (At least on subsection for the empirical approach and another for the RUSLE approach). Please also consider to provide more details on the regression model procedure and the parameters tested both for the regression model and the RUSLE approach.
Line 121. What is the signification of SWATK here ?
Line 124. Please clarify what you mean by "based on the RUSLE structure" in a more explicit way. The link is not obvious as the RUSLE structure was developed on a plot based scale and the regression model on a watershedbased scale.Line 200. What is the signification of "W" in eq. 13 ? idem for "Sa" and "Hyp"...
Lines 200201. As far as I can see, the main difference between the previous model (eq. 13) and the proposed new model (eq. 14) lies in the value of the exponent associated with the Hyps parameter (positive in Eq. 13 and negative in Eq. 14). How do you interpret this difference ?
Line 216. What is the signification of SA010 in Eq. 16 ?
Lines 248249. "Thus, the results suggest that stream watersheds carry more sediment, and alluvial rivers provided more opportunities for deposition." Do you think this sentence provides useful new information ?
Line 254. Madiment, 1993 (or Maidment ?) has to be added in the final reference list.
Line 255. In the sentence "...as well as results of other studies", please provide the references of these other studies and discussed how their context is similar, or not, to yours...
Page 12. Please consider enlearging Fig. 1 as it is difficult to read as it stands .
Page 15. In legend of Fig.7, your aerial images (in 7 bd) looks to be terrestrial (or ground) images
Page 16. Change ungagued into ungauged in Fig. 8 and explain what means WS, MR and MT
Page 18. Please consider adding a legend to Figure 11, or removing Figure 11.
Page 19. In Table 1, Have you an explanation for the very high dry mass density 1 for SR1 (value of 2.1).
Page 20. In Table 2, please explain why the SD values are not similar from those shown in your previous paper (e.g., Kang et al., 2019). And why some of the stations (e.g., NU4 and NU5) are marked as "discarded and/or unreasonable results) but still considered in Table 7.
Page 21. Several variables in Table 3 are not defined. What is the significance of "Line:total", "Main" ... What is the difference between "Middle relative height at middle relative area", and "elevation at middle relative area", and "middle elevation" ?
Page 22. Please explain why the conservation practice factors (Pvalue) for each land use type increase with slopes (Table 5). In general, the conservation practice factors do not have a direct systematic link with slopes. Morover, the influence of slopes in RUSLE is already taken into account through the LS factor.
Citation: https://doi.org/10.5194/hess2021225RC1  AC1: 'Reply on RC1', Woochul Kang, 03 Sep 2021
 AC2: 'Reply on RC1', Woochul Kang, 03 Sep 2021

RC2: 'Comment on hess2021225', Anonymous Referee #2, 07 Jul 2021
The research deals about evaluating specific degradation and sediment yield in South Korea. The procedure is a composed by different stages that considers: 1) analysing measurements of sediment yield from 62 streams/rivers and 14 reservoirs, 2) developing a regression / tree mining model for sediment yield in 47 upstreams catchments, 3) using RUSLE with mapping variables to validate the model, 4) using 16 ungauged watersheds to validate empirical data, and, 5) remote sensing is used for spatial variables.
We founded the subject of the article interesting, but in general the manuscript is a little confusing on how the procedures are adopted. We had a more clear understanding looking back to the article Kang et al., 2019 where a similar research is conducted and explained in a more fluid behave. We think that the use of methodologies adopted should be follow a more linear explication of the procedures, we are a bit perplexed about the question of spatial resolution that can be interesting but that is to much in evidence in discussions. Some concerns are on the material and methods not really clear or reported from previous article where better explained (e.g. the use of TE, trap efficiency in in defining SD, specific degradation, questionable dealing on both the catchment and the reservoir; use and description of the Modified Einstein Procedure  MEP).
In general, our opinion is that the manuscript could be reconsidered for the publication in HEES Journal after a new submission.ABSTRACT
L19. “significant parameters:”: the term significant, should have a statistical meaning.
We think the abstract it is a bit confusing when showing the procedures and the results should be more concise.INTRODUCTION
L3241 We are not sure that a pedagogic description of the erosion terms is necessary. In general, the introduction must give a larger spectrum of the state of the art that is much wide than that here showed.MATERIAL AND METHODS
L7981.“When water enters a reservoir, the flow velocity decreases, flow depth increases, and sedimentation occurs as a result of the overall decreased transport capacity of the stream.”: it is true but, over a pedagogic approach, here not demanded, scientifically thinking, it is a bit more complex.
L84 sediment deposition rate (ðð, m3·km2·yr1): in this form is not appropriate to call it like this, it is more an erosion/denudation rate in the end (m1y1) (i.e., metres over the catchment surface in a period)
L85: field measurements: this is maybe more a continuous monitoring of the dams
L89 impoundment: not sure it is the right term for that
Equations 1 and 2: x: not adapted mathematical notation
Equation 2 (Specific Degradation): ðð¸ is the trap efficiency (%): we suppose this is more a fraction than a percentage. If using the term TE, the term Specific Degradation is questionable because is the part of sediment captured by the dam while the degradation of the surface catchment refers to the whole sediment eroded.
Caption Figure2: maybe add that are average values
L100. The Modified Einstein Procedure (MEP) should be at least briefly described and why the author choose this procedure (maybe suitable to this kind of data or study site etc.)
L111190. We suggest to better structure this part to give a more clear presentation of the models, it is a little bit confusing and sometimes not enough well in details for equations. Additionally, some terms are not clear, W versus WW for instance (Eq. 13, 14, 16, 17) or the eq. 15 itself.Results
Is a bit confusing the part of models, showing a sort of evolution of previous models; we think a better structured explication should help.L250255. The use of SDR from the literature as in figure 8 is in general allowed, but should pay attention to the specific condition of the reservoirs, being the SDR very dependent on the specific vegetation of the soil (as also observed in this research) and connectivity condition of the reservoirs.
Discussion
Here, we talk about processes (eroding land surface) and a methodological issue, spatial resolution. We are not sure a methodological issue is important, as it is not the core of this paper, while the process has a limited discussion.Conclusion
All the main finding are evoked, maybe a more synthetic or better structured presentation should help.Citation: https://doi.org/10.5194/hess2021225RC2  AC3: 'Reply on RC2', Woochul Kang, 03 Sep 2021

RC3: 'Comment on hess2021225', Anonymous Referee #3, 16 Aug 2021
There are a number of significant problems with this paper that would need to be addressed before it could be considered for publication, so I think it should be rejected in its current form.
Even if one believes that the sedimentdelivery ratio is physically meaningful, rather than an artefact of the ways erosion rates have been analyzed historically (Parsons et al. doi: 10.1002/esp.1395), the fact that the paper considers them to be a good test of the modelled rates of erosion is highly problematic. There are common factors in the numerator and denominator of equation (12) that will lead to issues of spurious correlation. The “test” seems to be a comparison of whether the new model can fall somewhere within the bounds of the SDR estimated elsewhere, which is a target of over an order of magnitude. This target is missed in a nonnegligible number of cases, and the text then turns to special pleading of why specific datasets are problematic. Either you believe your data or you don’t!
I do not see the rationale for the structure of the regression model in equation (3). There are many critiques in the literature of the structure of (R)USLE. Furthermore, this is not the same structure, as it is the product of powers of the original variables.
Although the “proposed model” has a better RMSE, it seems to have more bias than the other models, overpredicting lower values, and underpredicting higher ones.
The description of “data mining” to produce alternative model structures is minimal and wouldn’t allow the approach to be replicated. In the results, “meaningful parameters” are mentioned, but it is not clear what meaning they have. In particular, what is the physical meaning of “lowest elevation”?
The overall aim and rationale of the paper is vague. There seems to be an invaluable dataset underlying the paper that could be much better employed in estimated sediment fluxes in different locations.
Citation: https://doi.org/10.5194/hess2021225RC3  AC4: 'Reply on RC3', Woochul Kang, 03 Sep 2021
Status: closed

RC1: 'Comment on hess2021225', Anonymous Referee #1, 01 Jul 2021
General comments
This study proposed a method to predict sediment yields of soil erosion and sedimentation in South Korea. The authors first developed an empirical prediction model of specific degradation based on multiple regression analysis and tree data mining. In addition the also predicted gross erosion based on RUSLE and derived sediment delivery ratios by dividing gross erosion prediction estimated from RUSLE by sediment yield estimated with the empirical prediction model (Eq. 12) in order to validate the results of the empirical model.
I found the paper very difficult to read. To my opinion, it is mainly due to the fact that the objective of the paper is not clearly stated. As it stands, this paper gives the feeling of a confused compilation of the previous papers (Kang et al., 2019 and 2021) without the added value of this new paper being really defined and discussed. This feeling is reinforced by the fact that more than half of the discussion focuses on the effect of spatial resolution on the RUSLE results, although this is not a central objective of the paper.
Another major concern is the use of SDR to validate the empirical model (cf. chapter 3.3 Model validation using the sediment delivery ratio) when the authors do not have measured SDR data (nor gross erosion references). To my opinion, the use of SDR data from the literature does not allow any conclusion to be drawn on the validation of the sediment yield prediction (or specific degradation) model, especially as these SDR values are known to be strongly linked to hydrosedimentary connectivity of the considered watershed, and therefore very sitedependent (see for instance De Vente et al., 2007, DOI: 10.1177/0309133307076485). As a consequence, there is little justification for the paper's general approach of articulating an empirical catchment model and a RUSLE approach to derive SDR values...
In addition, the discussion has to be completely rewritten in relation to the objective of the paper.
Specific comments
Line 2021. You use "percent water" and "percent wetland and water" as distinct parameters of the empirical regression analysis. What about the collinearity of these 2 parameters ?
Line 22. Please clarify the sentence "Additionally, erosion maps from the revised universal soil loss equation (RUSLE) were generated to validate model variables". Which model variables are supposed to be validated by the RUSLE approach ?
Lines 9293. Finally which values of the trap efficiencies were used in your studies ? Why don't you use Heinemann’s formula or a similar formula to estimate them (Heinemann, H. G. 1981. A new sediment trap efficiency curve for small reservoirs. Water Resources Bulletin, 17, 825830.)
Line 99102. A more detailed presentation of the SD assessment procedure is required in this paper. For future works, I particularly suggest to provide a evaluation of uncertainties in your SD estimations, because this evaluation is necessary if you want to show that your new empirical model is better than the previous ones. In a scarcedata context as yours, it can be assumed that the uncertainties on the SD values will be large (as seen in Fig. 2a for the SY/discharge relationship) and that therefore a variety of models can give similar results when considering the uncertainties in the calibration/validation datasets.
Line 109. I am a bit surprised that some gauging stations were removed from this study whereas many of them were used in the previous ones... To what extent could this partly explain the improved quality of the regression?
Line 111 (section 2.2). Please consider dividing this section in several subsections to make this section easier to read and understand (At least on subsection for the empirical approach and another for the RUSLE approach). Please also consider to provide more details on the regression model procedure and the parameters tested both for the regression model and the RUSLE approach.
Line 121. What is the signification of SWATK here ?
Line 124. Please clarify what you mean by "based on the RUSLE structure" in a more explicit way. The link is not obvious as the RUSLE structure was developed on a plot based scale and the regression model on a watershedbased scale.Line 200. What is the signification of "W" in eq. 13 ? idem for "Sa" and "Hyp"...
Lines 200201. As far as I can see, the main difference between the previous model (eq. 13) and the proposed new model (eq. 14) lies in the value of the exponent associated with the Hyps parameter (positive in Eq. 13 and negative in Eq. 14). How do you interpret this difference ?
Line 216. What is the signification of SA010 in Eq. 16 ?
Lines 248249. "Thus, the results suggest that stream watersheds carry more sediment, and alluvial rivers provided more opportunities for deposition." Do you think this sentence provides useful new information ?
Line 254. Madiment, 1993 (or Maidment ?) has to be added in the final reference list.
Line 255. In the sentence "...as well as results of other studies", please provide the references of these other studies and discussed how their context is similar, or not, to yours...
Page 12. Please consider enlearging Fig. 1 as it is difficult to read as it stands .
Page 15. In legend of Fig.7, your aerial images (in 7 bd) looks to be terrestrial (or ground) images
Page 16. Change ungagued into ungauged in Fig. 8 and explain what means WS, MR and MT
Page 18. Please consider adding a legend to Figure 11, or removing Figure 11.
Page 19. In Table 1, Have you an explanation for the very high dry mass density 1 for SR1 (value of 2.1).
Page 20. In Table 2, please explain why the SD values are not similar from those shown in your previous paper (e.g., Kang et al., 2019). And why some of the stations (e.g., NU4 and NU5) are marked as "discarded and/or unreasonable results) but still considered in Table 7.
Page 21. Several variables in Table 3 are not defined. What is the significance of "Line:total", "Main" ... What is the difference between "Middle relative height at middle relative area", and "elevation at middle relative area", and "middle elevation" ?
Page 22. Please explain why the conservation practice factors (Pvalue) for each land use type increase with slopes (Table 5). In general, the conservation practice factors do not have a direct systematic link with slopes. Morover, the influence of slopes in RUSLE is already taken into account through the LS factor.
Citation: https://doi.org/10.5194/hess2021225RC1  AC1: 'Reply on RC1', Woochul Kang, 03 Sep 2021
 AC2: 'Reply on RC1', Woochul Kang, 03 Sep 2021

RC2: 'Comment on hess2021225', Anonymous Referee #2, 07 Jul 2021
The research deals about evaluating specific degradation and sediment yield in South Korea. The procedure is a composed by different stages that considers: 1) analysing measurements of sediment yield from 62 streams/rivers and 14 reservoirs, 2) developing a regression / tree mining model for sediment yield in 47 upstreams catchments, 3) using RUSLE with mapping variables to validate the model, 4) using 16 ungauged watersheds to validate empirical data, and, 5) remote sensing is used for spatial variables.
We founded the subject of the article interesting, but in general the manuscript is a little confusing on how the procedures are adopted. We had a more clear understanding looking back to the article Kang et al., 2019 where a similar research is conducted and explained in a more fluid behave. We think that the use of methodologies adopted should be follow a more linear explication of the procedures, we are a bit perplexed about the question of spatial resolution that can be interesting but that is to much in evidence in discussions. Some concerns are on the material and methods not really clear or reported from previous article where better explained (e.g. the use of TE, trap efficiency in in defining SD, specific degradation, questionable dealing on both the catchment and the reservoir; use and description of the Modified Einstein Procedure  MEP).
In general, our opinion is that the manuscript could be reconsidered for the publication in HEES Journal after a new submission.ABSTRACT
L19. “significant parameters:”: the term significant, should have a statistical meaning.
We think the abstract it is a bit confusing when showing the procedures and the results should be more concise.INTRODUCTION
L3241 We are not sure that a pedagogic description of the erosion terms is necessary. In general, the introduction must give a larger spectrum of the state of the art that is much wide than that here showed.MATERIAL AND METHODS
L7981.“When water enters a reservoir, the flow velocity decreases, flow depth increases, and sedimentation occurs as a result of the overall decreased transport capacity of the stream.”: it is true but, over a pedagogic approach, here not demanded, scientifically thinking, it is a bit more complex.
L84 sediment deposition rate (ðð, m3·km2·yr1): in this form is not appropriate to call it like this, it is more an erosion/denudation rate in the end (m1y1) (i.e., metres over the catchment surface in a period)
L85: field measurements: this is maybe more a continuous monitoring of the dams
L89 impoundment: not sure it is the right term for that
Equations 1 and 2: x: not adapted mathematical notation
Equation 2 (Specific Degradation): ðð¸ is the trap efficiency (%): we suppose this is more a fraction than a percentage. If using the term TE, the term Specific Degradation is questionable because is the part of sediment captured by the dam while the degradation of the surface catchment refers to the whole sediment eroded.
Caption Figure2: maybe add that are average values
L100. The Modified Einstein Procedure (MEP) should be at least briefly described and why the author choose this procedure (maybe suitable to this kind of data or study site etc.)
L111190. We suggest to better structure this part to give a more clear presentation of the models, it is a little bit confusing and sometimes not enough well in details for equations. Additionally, some terms are not clear, W versus WW for instance (Eq. 13, 14, 16, 17) or the eq. 15 itself.Results
Is a bit confusing the part of models, showing a sort of evolution of previous models; we think a better structured explication should help.L250255. The use of SDR from the literature as in figure 8 is in general allowed, but should pay attention to the specific condition of the reservoirs, being the SDR very dependent on the specific vegetation of the soil (as also observed in this research) and connectivity condition of the reservoirs.
Discussion
Here, we talk about processes (eroding land surface) and a methodological issue, spatial resolution. We are not sure a methodological issue is important, as it is not the core of this paper, while the process has a limited discussion.Conclusion
All the main finding are evoked, maybe a more synthetic or better structured presentation should help.Citation: https://doi.org/10.5194/hess2021225RC2  AC3: 'Reply on RC2', Woochul Kang, 03 Sep 2021

RC3: 'Comment on hess2021225', Anonymous Referee #3, 16 Aug 2021
There are a number of significant problems with this paper that would need to be addressed before it could be considered for publication, so I think it should be rejected in its current form.
Even if one believes that the sedimentdelivery ratio is physically meaningful, rather than an artefact of the ways erosion rates have been analyzed historically (Parsons et al. doi: 10.1002/esp.1395), the fact that the paper considers them to be a good test of the modelled rates of erosion is highly problematic. There are common factors in the numerator and denominator of equation (12) that will lead to issues of spurious correlation. The “test” seems to be a comparison of whether the new model can fall somewhere within the bounds of the SDR estimated elsewhere, which is a target of over an order of magnitude. This target is missed in a nonnegligible number of cases, and the text then turns to special pleading of why specific datasets are problematic. Either you believe your data or you don’t!
I do not see the rationale for the structure of the regression model in equation (3). There are many critiques in the literature of the structure of (R)USLE. Furthermore, this is not the same structure, as it is the product of powers of the original variables.
Although the “proposed model” has a better RMSE, it seems to have more bias than the other models, overpredicting lower values, and underpredicting higher ones.
The description of “data mining” to produce alternative model structures is minimal and wouldn’t allow the approach to be replicated. In the results, “meaningful parameters” are mentioned, but it is not clear what meaning they have. In particular, what is the physical meaning of “lowest elevation”?
The overall aim and rationale of the paper is vague. There seems to be an invaluable dataset underlying the paper that could be much better employed in estimated sediment fluxes in different locations.
Citation: https://doi.org/10.5194/hess2021225RC3  AC4: 'Reply on RC3', Woochul Kang, 03 Sep 2021
Viewed
HTML  XML  Total  BibTeX  EndNote  

704  354  45  1,103  27  26 
 HTML: 704
 PDF: 354
 XML: 45
 Total: 1,103
 BibTeX: 27
 EndNote: 26
Viewed (geographical distribution)
Country  #  Views  % 

Total:  0 
HTML:  0 
PDF:  0 
XML:  0 
 1