Watershed Monitoring in Galicia from UAV Multispectral Imagery Using Advanced Texture Methods

Round 1

Reviewer 1 Report

The objetives of this paper can be resumed as follow:

The main objective of this work is to automate the monitoring process of river basins in Galicia, Spain, using UAV image processing algorithms.

A supervised classification chain based on spatio-spectral features and texture descriptors at the superpixel level is proposed. Since the study is performed at the scale of the fluvial corridor, the selected algorithms have to be fast where different strategies are applied to reduce computational times.

Although I found the summary and introduction very interesting and stimulating, I think that aspects of section 2.4 are missing. For example, there is no information on how the reflectance calibration was performed using the calibration plate of the Micasense camera, and there is no discussion on the generation of the mosaic on which the digital processing analysis was performed. Were control points (GCP) used for the mosaic design? What were the light conditions at the time of acquisition? You should clarify these aspects... 

Overall, I think this paper can be published.

I will try to use your new methodologies! And this is the reason why, I recommend it for publishing.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper is relatively well written and it is indeed a nicely designed research. It is enjoyable to read this paper with its clarity, especially the Introduction and methodology sections.

For abstract: please extract a more concrete and general conclusion for your vegetation and man-made structures identification, as high accuracy is not your overarching aim. It is weak to put it at the end of your abstract.

For the methodology, instead of methods such as graph-based, NCUT, quick-shift, Graph-cut, etc, a more concrete justification of using SLIC for your superpixel computation is needed. It is missing in present form. 

For discussion, you repeated yourself (e.g. line 396 to 397). Please improve it by discussing the pros and cons of your own methodology in-depth (e.g. the paragraph from 393 to 403). From 410 to 418 content repetition with the Introduction again, can you integrate this info. with your own study?

 

Minor comments:

line        comments

23-25    I would suggest citing the very influential report from FAO. http://www.fao.org/3/i8087en/i8087en.pdf to demonstrate/introduce your research from the efficient watershed management perspective.

Figure1 It is very nice but can be better if you refer to the threats of each as the subtitle of each picture as you described in the text (line 30-31).

68    Although LIDAR is well-known. Please give the full name before use its abbreviation directly. 

87-88 for the SURF, SIFT as well (same as the above.)

98-102 repetition. please condense it into one sentence.

from 112 I would suggest inserting one sentence to introduce the entire study and its aims. 

146    They are not typical "tree species" of river banks?

165     please replace global with overall

Section 2.2    for your aims, could you depict one step further? For example, to classify and identify both the vegetation and built-up features for timely and efficient watershed management? It would help enrich your discussion through discussion of the significance of your classification method and the justification of your identified 10 classes. 

176     in Figure 8

204-205    Why do not you use water pixels in the end?

256 I would suggest improving the quality of Figure 5 as you used it frequently to demonstrate your sites. 

Figure 13 I would suggest changing the colors of your classification. To be color-wise, please use blue for water, gradient green for vegetation and so forth. 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

The manuscript presents a promising approach to monitor vegetation and manmade structures that can affect watersheds using UAV mounted multispectral sensor. The major strength lies in the use of fairly cheap equipment to perform a daunting job with high accuracy. The manuscript is well written and details the authors' approach in selecting and evaluating various methods for segmentation, texture assessment and classification. The results section give enough detail as to the performance and adequacy of the selected algorithms.

One issue I have is that the methodology lacks an explicit presentation of the accuracy calculation process including the size of the samples used for accuracy assessment and the sampling approach.

However, the main issue for me is with the discussion section that is written more like a literature review with with much repetition from the introduction. I expected a discussion of the authors' results, including their implications and benefits in addressing the problem at hand, the limitations of the approach and its applicability to larger and diverse regions. Additionally, one of the criteria used by the authors to select algorithms was their performance. Would the authors select other algorithms have they had access to higher performance computing equipment?

Minor edits:

-on line 176 add figure before 8

-figure 8 can be improved visually. It is difficult to distinguish the different signatures with the black background and thin low quality curves

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf