Nonparametric estimation method for river cross-sections with point cloud data from UAV photography URiver-X version 1.0 -methodology development

Lee, Taesam; Park, Jaewoo; Hwang, Sunghyun; Singh, Vijay

doi:https://doi.org/10.5194/gmd-2023-57

Preprints

https://doi.org/10.5194/gmd-2023-57

Preprints

Submitted as: model description paper

27 Jun 2023

Submitted as: model description paper |

| 27 Jun 2023

Status: this preprint was under review for the journal GMD but the revision was not accepted.

Nonparametric estimation method for river cross-sections with point cloud data from UAV photography URiver-X version 1.0 -methodology development

Taesam Lee, Jaewoo Park, Sunghyun Hwang, and Vijay Singh

Abstract. Aerial surveying with unmanned aerial vehicles (UAVs) has been popularly employed in river management and flood monitoring. One of the major processes in UAV aerial surveying for river applications is to demarcate the cross-section of a river. From the photo images of aerial surveying, a point cloud dataset can be abstracted with the structure from motion technique. To accurately demarcate the cross-section from the cloud points, an appropriate delineation technique is required to reproduce the characteristics of natural and manmade channels, including abrupt changes, bumps, and lined shapes. Therefore, a nonparametric estimation technique, called the K-nearest neighbor local linear regression (KLR) model, was tested in the current study to demarcate the cross-section of a river with a point cloud dataset from aerial surveying. The proposed technique was tested with synthetically simulated trapezoidal, U-shape, and V-shape channels. In addition, the proposed KLR model was compared with the traditional polynomial regression model and another nonparametric technique, locally weighted scatterplot smoothing (LOWESS). The experimental study was performed with the river experiment center in Andong, South Korea. Furthermore, the KLR model was applied to two real case studies in the Migok-cheon stream on Hapcheon-gun and Pori-cheon stream on Yecheon-gun and compared to the other models. With the extensive applications to the feasible river channels, the results indicated that the proposed KLR model can be a suitable alternative for demarcating the cross-section of a river with point cloud data from UAV aerial surveying by reproducing the critical characteristics of natural and manmade channels, including abrupt changes and small bumps as well as different shapes. Finally, the limitation of the UAV-driven demarcation approach was also discussed due to the penetrability of RGB sensors to water.

Received: 09 Mar 2023 – Discussion started: 27 Jun 2023

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.

Download & links

Preprint (PDF, 6251 KB)

Supplement (4962 KB)

Download & links

Taesam Lee, Jaewoo Park, Sunghyun Hwang, and Vijay Singh

Status: closed

RC1:
'Comment on gmd-2023-57', Anonymous Referee #1, 31 Jul 2023
This paper concerns the nonparametric estimation of river cross-sections with point cloud data from UAV photography. Advances in structure from motion (SfM) methods have led to widespread use of photogrammetry in many environmental applications. In this application, the goal of the method is to to reproduce the characteristics of natural and manmade channels, including abrupt changes, bumps, and lined shapes. The authors have proposed the use of a nonparametric estimation technique, called the K-nearest neighbor local linear regression (KLR) model, and compared its results to more traditional locally weighted scatterplot smoothing (LOWESS) model for two test cases in Korea. They found that the KLR model perform well to derive river cross section from point cloud data.

While I find the findings of this research relevant and interesting, I find this work reads more like a research paper than a ‘model development’ paper. While it is listed as a model description paper, it is not 'comprehensive descriptions of numerical models'. It has not been describing how the code is developed and can be used. So perhaps it is more suitable for a journal other than GMD, maybe in a quaternary science journal. I also feel the paper can be greatly shortened to be a methods and techniques paper or a research letter.

Major comments:

There are a total of 27 figures in the main text. I wonder whether any of them can be consolidated to make the same points. Also, the results of KLR/LOESS/polyfit can simply be put in the same figure but with different line styles.

It is not clear from the introduction why it is important to get the river cross section right. What are the issues if wrong cross section is imported to models (e.g. Neal et al 2015)? Is 2D cross-section useful enough? Why not produce 3D DEM of channels from the point clouds?

As can be observed in the point clouds, in most cases the point cloud captures the channel cross-section quite well. It begs the question why not use a hand-drawn best-fit line or simple parametric methods?

The paper shows a different use case of KLR, which was originally proposed in Lee et al. (2017) for climate dynamics. The paper needs to show more clearly the specific advances it makes in the delineation of channel from UAV photogrammetry, rather than a generic application of KLR.
Citation: https://doi.org/10.5194/gmd-2023-57-RC1
- AC1: 'Reply on RC1', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC1
RC2:
'Comment on gmd-2023-57', Anonymous Referee #2, 04 Aug 2023
The paper aims to use the KLR technique to demarcate river cross-sections from point cloud data obtained from UAV photography. The authors apply the KLR technique to two case studies in Korea and compare it with other regression approaches, such as polynomial regression and LOWESS. They propose KLR as an alternative method for demarcating river cross-sections with point cloud data.
General comment:
I agree with the comment of reviewer 1. In addition to his comment, here are my comments as well.
The paper structure could be improved, and the length could be reduced. The paper should focus more on the main contribution of the study, which is the KLR technique, and describe other regression methods briefly and only for comparison purposes.

Is the heuristic approach based on Eq. (10) better than cross-validation or other methods for selecting K value?

The authors should justify the need for using the KLR technique to demarcate river cross-sections from point cloud data. From line 80-82, it is stated that "the dense cloud point dataset obtained from UAV aerial surveying and the SfM technique mostly contains errors and does not provide direct cross-sectional information". However, from the results, it seems that the point cloud data already describes the cross-section very well. What is the advantage of using KLR overusing the mean or median of the point cloud data?

I recommend revising and restructuring the paper and resubmitting it.
Citation: https://doi.org/10.5194/gmd-2023-57-RC2
- AC2: 'Reply on RC2', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC2
RC3:
'Comment on gmd-2023-57', Anonymous Referee #3, 05 Aug 2023
The authors propose a K-nearest-neighbor-based approach for estimating river cross-section from point cloud data gathered from UAV. It seems this is the second time that the authors have tried to submit their manuscript to GMD. The first submission in 2021, which can be found at https://gmd.copernicus.org/preprints/gmd-2021-309/, was rejected since the authors didn't address the comments provided by the referees. In this submission, the authors made some improvements, but, in my opinion, they still haven't addressed some of the major comments given in the first revision. Moreover, I agree with the comments provided by the two other referees in this submission. In addition to the comments given so far, I have some additional notes.
The introduction focuses too much on UAVs instead of point cloud data analysis and river cross-section estimation methods. I recommend the authors give a more in-depth literature review on river bathymetry estimation methods, their implications, and applications. Currently, the introduction does not present the novelty of the manuscript sufficiently.

One of the major hurdles in river bathymetry estimation from data gathered by instruments that do not penetrate water has been estimating the "wet" part underwater surface during the survey. Although this manuscript mentions it as a limitation of their methodology, this shortcoming significantly limits the applicability and novelty of their methodology, for river cross-section applications.

When comparing with observation, I think a comparison with existing DEM data can be informative as well, so we can measure the improvements over using DEMs.

Regarding reproducibility, maybe I am missing something, but I couldn't find the point cloud data that the authors generated using WebODM. If the data is public, providing the point cloud data is necessary for running the scripts.
Citation: https://doi.org/10.5194/gmd-2023-57-RC3
- AC3: 'Reply on RC3', Taesam Lee, 04 Sep 2023
  
  The comment was uploaded in the form of a supplement: https://gmd.copernicus.org/preprints/gmd-2023-57/gmd-2023-57-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC3
- AC4: 'Reply on RC3', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC4
RC4:
'Comment on gmd-2023-57', Anonymous Referee #4, 10 Aug 2023

The author use UAV to obtain the point cloud data of river cross section, and propose the nonparametric estimation method to realize the fitting of river cross section. In manuscript KLR algorithm can better fit various key features of different types of river cross sections, but it cannot compare with other methods for fitting river cross-section, such as SfM, DEM, etc. I have the following suggestions and questions to further improve the current manuscript.
1.For this title, the nonparametric estimation method covers too much, and this article only focuses on the KLR algorithm.
2.In Introduction section, too many drone applications were introduced, but there was a lack of introduction to methods related to river cross-section fitting.
3.page 4, line 79, 'How, the dense cloud point dataset obtained from UAV aerial survey and the SfM technology mostly contains errors', lacking relevant experiments or literature to prove the existence of errors here
4.line 103 of page 5, "a fixed function of the multilateral regression with a feed parameters is limited to the highly variable shape of the cross section." It expresses that Polynomial regression is not applicable to river cross section, and this method is still used in subsequent comparative experiments
5.page 8, line 165 introduces the improvement of KLR algorithm on KNN algorithm. It is recommended to include KNN algorithm in comparative experiments to ensure the effectiveness of this improvement.
6.page 13, line 236, there is a lack of relevant experiments or references regarding the suggestion of overlapping parts.
7.All images are at the end of the article, which is not conducive to reading.

Citation: https://doi.org/10.5194/gmd-2023-57-RC4
- AC5: 'Reply on RC4', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC5

Status: closed

RC1:
'Comment on gmd-2023-57', Anonymous Referee #1, 31 Jul 2023
This paper concerns the nonparametric estimation of river cross-sections with point cloud data from UAV photography. Advances in structure from motion (SfM) methods have led to widespread use of photogrammetry in many environmental applications. In this application, the goal of the method is to to reproduce the characteristics of natural and manmade channels, including abrupt changes, bumps, and lined shapes. The authors have proposed the use of a nonparametric estimation technique, called the K-nearest neighbor local linear regression (KLR) model, and compared its results to more traditional locally weighted scatterplot smoothing (LOWESS) model for two test cases in Korea. They found that the KLR model perform well to derive river cross section from point cloud data.

While I find the findings of this research relevant and interesting, I find this work reads more like a research paper than a ‘model development’ paper. While it is listed as a model description paper, it is not 'comprehensive descriptions of numerical models'. It has not been describing how the code is developed and can be used. So perhaps it is more suitable for a journal other than GMD, maybe in a quaternary science journal. I also feel the paper can be greatly shortened to be a methods and techniques paper or a research letter.

Major comments:

There are a total of 27 figures in the main text. I wonder whether any of them can be consolidated to make the same points. Also, the results of KLR/LOESS/polyfit can simply be put in the same figure but with different line styles.

It is not clear from the introduction why it is important to get the river cross section right. What are the issues if wrong cross section is imported to models (e.g. Neal et al 2015)? Is 2D cross-section useful enough? Why not produce 3D DEM of channels from the point clouds?

As can be observed in the point clouds, in most cases the point cloud captures the channel cross-section quite well. It begs the question why not use a hand-drawn best-fit line or simple parametric methods?

The paper shows a different use case of KLR, which was originally proposed in Lee et al. (2017) for climate dynamics. The paper needs to show more clearly the specific advances it makes in the delineation of channel from UAV photogrammetry, rather than a generic application of KLR.
Citation: https://doi.org/10.5194/gmd-2023-57-RC1
- AC1: 'Reply on RC1', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC1
RC2:
'Comment on gmd-2023-57', Anonymous Referee #2, 04 Aug 2023
The paper aims to use the KLR technique to demarcate river cross-sections from point cloud data obtained from UAV photography. The authors apply the KLR technique to two case studies in Korea and compare it with other regression approaches, such as polynomial regression and LOWESS. They propose KLR as an alternative method for demarcating river cross-sections with point cloud data.
General comment:
I agree with the comment of reviewer 1. In addition to his comment, here are my comments as well.
The paper structure could be improved, and the length could be reduced. The paper should focus more on the main contribution of the study, which is the KLR technique, and describe other regression methods briefly and only for comparison purposes.

Is the heuristic approach based on Eq. (10) better than cross-validation or other methods for selecting K value?

The authors should justify the need for using the KLR technique to demarcate river cross-sections from point cloud data. From line 80-82, it is stated that "the dense cloud point dataset obtained from UAV aerial surveying and the SfM technique mostly contains errors and does not provide direct cross-sectional information". However, from the results, it seems that the point cloud data already describes the cross-section very well. What is the advantage of using KLR overusing the mean or median of the point cloud data?

I recommend revising and restructuring the paper and resubmitting it.
Citation: https://doi.org/10.5194/gmd-2023-57-RC2
- AC2: 'Reply on RC2', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC2
RC3:
'Comment on gmd-2023-57', Anonymous Referee #3, 05 Aug 2023
The authors propose a K-nearest-neighbor-based approach for estimating river cross-section from point cloud data gathered from UAV. It seems this is the second time that the authors have tried to submit their manuscript to GMD. The first submission in 2021, which can be found at https://gmd.copernicus.org/preprints/gmd-2021-309/, was rejected since the authors didn't address the comments provided by the referees. In this submission, the authors made some improvements, but, in my opinion, they still haven't addressed some of the major comments given in the first revision. Moreover, I agree with the comments provided by the two other referees in this submission. In addition to the comments given so far, I have some additional notes.
The introduction focuses too much on UAVs instead of point cloud data analysis and river cross-section estimation methods. I recommend the authors give a more in-depth literature review on river bathymetry estimation methods, their implications, and applications. Currently, the introduction does not present the novelty of the manuscript sufficiently.

One of the major hurdles in river bathymetry estimation from data gathered by instruments that do not penetrate water has been estimating the "wet" part underwater surface during the survey. Although this manuscript mentions it as a limitation of their methodology, this shortcoming significantly limits the applicability and novelty of their methodology, for river cross-section applications.

When comparing with observation, I think a comparison with existing DEM data can be informative as well, so we can measure the improvements over using DEMs.

Regarding reproducibility, maybe I am missing something, but I couldn't find the point cloud data that the authors generated using WebODM. If the data is public, providing the point cloud data is necessary for running the scripts.
Citation: https://doi.org/10.5194/gmd-2023-57-RC3
- AC3: 'Reply on RC3', Taesam Lee, 04 Sep 2023
  
  The comment was uploaded in the form of a supplement: https://gmd.copernicus.org/preprints/gmd-2023-57/gmd-2023-57-AC3-supplement.pdf
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC3
- AC4: 'Reply on RC3', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC4
RC4:
'Comment on gmd-2023-57', Anonymous Referee #4, 10 Aug 2023

The author use UAV to obtain the point cloud data of river cross section, and propose the nonparametric estimation method to realize the fitting of river cross section. In manuscript KLR algorithm can better fit various key features of different types of river cross sections, but it cannot compare with other methods for fitting river cross-section, such as SfM, DEM, etc. I have the following suggestions and questions to further improve the current manuscript.
1.For this title, the nonparametric estimation method covers too much, and this article only focuses on the KLR algorithm.
2.In Introduction section, too many drone applications were introduced, but there was a lack of introduction to methods related to river cross-section fitting.
3.page 4, line 79, 'How, the dense cloud point dataset obtained from UAV aerial survey and the SfM technology mostly contains errors', lacking relevant experiments or literature to prove the existence of errors here
4.line 103 of page 5, "a fixed function of the multilateral regression with a feed parameters is limited to the highly variable shape of the cross section." It expresses that Polynomial regression is not applicable to river cross section, and this method is still used in subsequent comparative experiments
5.page 8, line 165 introduces the improvement of KLR algorithm on KNN algorithm. It is recommended to include KNN algorithm in comparative experiments to ensure the effectiveness of this improvement.
6.page 13, line 236, there is a lack of relevant experiments or references regarding the suggestion of overlapping parts.
7.All images are at the end of the article, which is not conducive to reading.

Citation: https://doi.org/10.5194/gmd-2023-57-RC4
- AC5: 'Reply on RC4', Taesam Lee, 04 Sep 2023
  
  Reply is attached
  
  Citation: https://doi.org/10.5194/gmd-2023-57-AC5

Taesam Lee, Jaewoo Park, Sunghyun Hwang, and Vijay Singh

Supplement

https://doi.org/10.5194/gmd-2023-57-supplement

Taesam Lee, Jaewoo Park, Sunghyun Hwang, and Vijay Singh

Viewed

Total article views: 699 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
486	156	57	699	56	34	54

HTML: 486
PDF: 156
XML: 57
Total: 699
Supplement: 56
BibTeX: 34
EndNote: 54

Views and downloads (calculated since 27 Jun 2023)

Month	HTML	PDF	XML	Total
Jun 2023	48	12	3	63
Jul 2023	79	14	1	94
Aug 2023	73	13	8	94
Sep 2023	57	26	10	93
Oct 2023	21	6	0	27
Nov 2023	11	1	0	12
Dec 2023	16	7	1	24
Jan 2024	10	5	1	16
Feb 2024	13	22	1	36
Mar 2024	20	15	6	41
Apr 2024	18	7	5	30
May 2024	14	6	6	26
Jun 2024	27	5	3	35
Jul 2024	11	5	2	18
Aug 2024	20	2	5	27
Sep 2024	9	4	4	17
Oct 2024	8	1	0	9
Nov 2024	27	4	1	32
Dec 2024	4	1	0	5

Cumulative views and downloads (calculated since 27 Jun 2023)

Month	HTML	PDF	XML	Total
Jun 2023	48	12	3	63
Jul 2023	79	14	1	94
Aug 2023	73	13	8	94
Sep 2023	57	26	10	93
Oct 2023	21	6	0	27
Nov 2023	11	1	0	12
Dec 2023	16	7	1	24
Jan 2024	10	5	1	16
Feb 2024	13	22	1	36
Mar 2024	20	15	6	41
Apr 2024	18	7	5	30
May 2024	14	6	6	26
Jun 2024	27	5	3	35
Jul 2024	11	5	2	18
Aug 2024	20	2	5	27
Sep 2024	9	4	4	17
Oct 2024	8	1	0	9
Nov 2024	27	4	1	32
Dec 2024	4	1	0	5

Viewed (geographical distribution)

Total article views: 647 (including HTML, PDF, and XML) Thereof 647 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 13 Dec 2024

Short summary

The current study presents a novel method to demarcate the cross-section of a river channel using a very flexible regression model, called K-nearest neighbor local linear regression (KLR). The proposed method draws the cross-section automatically based on the point cloud data taken from unmanned aerial vehicles (UAVs). The proposed model can provide a further development of 4th industy innovation by employding the UAV-based photogrammetry.


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