Next Article in Journal
L-Band SAR Co-Polarized Phase Difference Modeling for Corn Fields
Next Article in Special Issue
A Review of Quantifying pCO2 in Inland Waters with a Global Perspective: Challenges and Prospects of Implementing Remote Sensing Technology
Previous Article in Journal
Vegetation Resilience under Increasing Drought Conditions in Northern Tanzania
Previous Article in Special Issue
Sentinel-2 and Landsat-8 Observations for Harmful Algae Blooms in a Small Eutrophic Lake
 
 
Article
Peer-Review Record

Spectral and Spatial Feature Integrated Ensemble Learning Method for Grading Urban River Network Water Quality

Remote Sens. 2021, 13(22), 4591; https://doi.org/10.3390/rs13224591
by Xiaoteng Zhou 1, Chun Liu 1,*, Akram Akbar 1, Yun Xue 1 and Yuan Zhou 2
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Remote Sens. 2021, 13(22), 4591; https://doi.org/10.3390/rs13224591
Submission received: 3 September 2021 / Revised: 9 November 2021 / Accepted: 12 November 2021 / Published: 15 November 2021

Round 1

Reviewer 1 Report

=============================================================
GENERAL COMMENTS
The authors have described an interesting new method to determine spatial and temporal dynamics of water quality indices from UAV mounted sensors. They use an approach that places the waterway within the larger spatial context of “water functional zones” as a means to associate land usages with changes in WQ detected by the drone mounted sensor. The use of drone mounted sensors promises to improve the observation and evaluation of small-scale waterways commonly overlooked in urban water quality monitoring programs.

However, despite the efforts of the authors to simplify the description of the work, large sections of the text remain difficult to follow because of imprecise terms, excessive word repetition, awkward placement of adjectives  and a tendency to overuse provocative language (“urgent need”, “prove”). 


Indeed, the  sections 2 and 3 pose the most important problems of writing that render very difficult for me to give a clear opinion of the quality of the work performed. For instance, the methods section cites standard equations (Eq 4 and 5) unnecessarily, while leaving more important ones (e.g. 7 and 8) without adequate explanation of the functions or the units associated with each parameter. The methodology of  ML must be exposed more clearly because I cannot evaluate the authors’ choices with the actual text (ln 312-314). There is also no mention of whether a software package is used or if the functions are coded by the authors.  The sections concerned by the water quality analyses is inadequately described (how were TP and ammonium N measured?) and the sampling plan is not clearly demonstrated, while one seems to have existed if I look closely at the figures. 

Thus, parts of the results and their quality are not possible to fully evaluate because of both language and the organisation of the information. 

I therefore suggest a major revision of the sections 2 and 3 is necessary. There is also an excessive number of figures (19!) that must be reduced. 


Important deletions are indicated in the PDF. The majority are because of redundancies.


I have provided detailed remarks line-by-line below for the sections that were the most problematic.

=============================================================
DETAILED COMMENTS

Title
Spectral and Spatial Feature Integrated Ensemble Learning Method for [Grading] Urban River Network Water Quality 

——————————
——————————

Abstract
10-11    remove repetition of “complex”, awkward construction of opening phrase
12    replace “appropriate” with “suitable”    
13    “Aiming to accurately invert the water quality” ?unclear English usage for statement of purpose
14    replace “proposed” with “describe”

——————————
——————————

Introduction
32    no sense to first phrase, delete
38-39    replace with, “A comprehensive, fine-scale and high-frequency monitoring method is needed to improve urban river  water quality monitoring programs.“ 
44-45    delete “researchers have developed methods from qualitative monitoring to quantitative monitoring and have made great achievements.”
46    “case 1 waters and case 2 waters”, replace with “Case-1” and “Case-2” because they are proper nouns
57    replace “spectrum” with “spectral features”    
57    add “water quality” in front of “indexes”
58    unclear usage of “selected”
58 - 63    too wordy and redundancies, rewrite and shorten
64    replace “Remote sensing has the advantage of accurate recognition of spatial properties …” with “Remote sensing imagery has the advantage of capturing spatial properties at fine scales … ”
66    replace “space” with “environment”
68    Insert “For example,” different usages …
77    Insert “spectral” in front of features.
90    New paragraph at “Machine learning inversion models …
95    Replace “proven” with “shown”
120-137    This paragraph should be cut in half- see next comment

——————————
——————————

Methodology

There are no measurement units in the text - these are only mentioned on the figures themselves. These should be mentioned for example in section 2.1.1 with the naming of the parameters below the equations.


139-149        Replace all text with the 4 points from 129 - 136 

152 - 160    Section 2.1 - This is not a “framework”. The text in fact describes a standard methodology. 

Suggest replacing text with a section simply labelled as data collection and site description? (see ln 391 below):

“2.1    Data collection and site description

At each sampling point, the spectrum data and water quality samples are collected synchronously. Water samples were collected and preserved until analysis and ranking by an accredited laboratory (section 2.1.2). Spectrum data are collected 5 times at the same sampling point with the UAV. Sampling point selection used the rules given in section (2.1.3). ” 


165    add “,using the following:”
170    replace “sea” with “water” everywhere in this section

172-177    It seems from the text that a theoretical quantity is defined by a measurement (for the Ed(0+)) this is an unusual choice that should be explained better by the authors. 

185     Replace “grading basis” with “ranking”
The absence of information raises other questions:

185    In the section 2.1.2, there is no information on *how* the elements of phosphorus and nitrogen were analysed. Additional information is needed: types of analyses, name of laboratory used, errors, detection limits, how were samples preserved in the field? Indeed this information on the quality of the measurements should be treated within the results as an estimate of precision and accuracy


187-191     Replace text with “According to the  People's Republic of China national "Environmental Quality Standards for Surface Water" (GB3838-2002), there are six grades of water quality (Table 1), with Grade I being the best.”

192-193     Replace text with “The laboratory used for the samples in this article passed the China Metrology Accreditation (CMA).” 


194    Replace title of Table 1 with “Water quality rankings”

196    Then in section 2.1.3 (ln 196-220) where I expected to find a sampling “rule” as a mathematical statement, instead there is a description of functional zones and scale. This entire section need revision. Section 2.1.3 needs to be re-titled. Overall it contains a lot of unsupported statements. It also does not contain the information I expected, which was details of a structured sampling plan instead of generic remarks on climate and stream order. It is not possible from the text to understand the choices made.

197    delete first sentence, start with “The in situ sampling …

204-205    Missing reference for the document cited as “Functional Zoning of Water Environment in Shanghai”

211    need to define what is meant by “self-purification” precisely within the classification mentioned and relevant to the study site (dilution?  biological processes? ) because this determines observation frequency

214    a “climate” is usually determined at a larger scale than this study - do the authors mean “season” (called “quarters of a year” in the text, line 215)? are they referring to “micro-climates”? what supporting data do they have for this?

218    What is meant by “the 4 conditions are combined” with samples? 

221    Section 2.2 -this is also quite classical, but poorly described.

240 and 262 Equations 4and 5 are standard definitions, not sure why these are included and they could be removed and replaced with short text, like “we calculated the correlation between …” etc.

General remark on section 2:  it should be revised strongly. I would suggest that the equations that constitute the authors consider are the nucleus of the method be pulled out into a table that includes the parameters with their units. Then the text only describes the reasoning for the model choices, referring to the table. This should improve the readability for a broader audience.


266    “union” is incorrect here
267-269     REPLACE “The same feature bands that are selected by different methods have the higher priority to be chosen. The result intersection of at least 2 methods is finally chosen to be the high-priority bands that are used on the multispectral sensor.” WITH “Spectral feature bands identified by different methods are assigned a higher priority. If at least 2 methods choose the same feature this then becomes the high-priority band for the multispectral sensor.”


274-280    Reads like it is text from an instruction manual, revise.

373-376        delete, start section with “A fivefold cross validation …
378    “common strategy” needs a reference
379-384        These equations should be in a table, with accompanying explanations

 

391 - 425    I don’t like the data collection section at the end, move it up to before the section 2.1.2/3 will get this information out of the way before the ML sections which really involve data treatment.

477 - 498     Section 3.1.3 This entire section should be up in the materials and methods- this is not a result!

690    Table 3 - the grading precision values are over-reported by 2 decimal points 

700    “effective” could be qualified by the Table 3 information as an “improvement”. Effectiveness depends on the exact regulatory requirements of particular sites and  would be expected to vary strongly between regions and countries


——————————
——————————


Globally, the text after line 500  is better presented than the rest of the article, even if it should benefit from a strong revision for the language and deletion of redundancies.


——————————
——————————

Figures
There are way too many figures - My suggested changes are below:

Figure 1 - delete, understood from text

Figure 2 - delete, understood from text

Figure 3 - OK but why so much repetition? It seems to be the same everywhere

Figure 4 - delete, understood from text

Figure 5 - OK

Figure 6 - place in supplementary data

Figure 7 - OK

Figure 8 - OK

Figure 9 - OK

Figure 10 - place in supplementary data, or possible to combine with Figure 13? as and a/b plot?

Figure 11 - OK

Figure 12 - OK

Figure 13 - OK

Figure 14 - Delete the “b” part of the figure, add the information to the legend, or the caption

Figure 15 - convert to table of data instead

Figure 16 - improve organisation - Labels for Total P and NH3-N should be on the plot for clarity. Label top row a-c  and d-f for the bottom row. All the Y-scales should be the same (0 - 0.8). Change to “case number” on X-axes. Change the plot colors to indicate whether they concern P or N (e.g. leave top row blue-red, change bottom row to purple-orange). Could also indicate on axis when the spatial features are included (case 5 and above) - this is the obvious explanation for the jump in precision

Figure 17 - improve organisation -Labels for Total P and NH3-N should also be on the plot for clarity. Classically, the presentation of the matrix would be reversed: the prediction would be on the vertical (“y”) and the actual (true) label value would be on the (“x”)

Figure 18 - place in supplementary data, unless there is something specifically interesting to note

Figure 19 - OK

 

Author Response

Dear reviewer:

We thank you very much for the comments and suggestions. The comments and suggestions are valuable and very helpful for revising and improving our manuscript. We have made revisions according to the comments and suggestions. Please see the attachment.

Yours sincerely,

Xiaoteng Zhou

Author Response File: Author Response.docx

Reviewer 2 Report

General comments:

In my opinion this paper is an interesting study and authors have presented new application for water quality assessment using spectral and spatial feature integrated ensemble learning method. Because water environment protection is one of the most important problems due to climate changes, it is necessary to find new methods and techniques to monitor water quality changes. The paper is generally well written and structured. My main concern regarding to the discussion is lack of information about advantages and limitations of proposed method on the background of other methods. Additionally, there is no discussion with results obtained from other scientists regarding water quality measurements. All shortcomings were presented below.

 

Specific comments:

  • Abstract is too long. Lines 10-14 are not necessary, it’s good sentence to introduction or discussion chapter, not for abstract. Please add main aim of the study.
  • Line 443. In Composition of the Dataset subsection authors describe location of measurements point on the map so before it should be characterization of Study site. Please change order of sub-sections. First should be Study site, then Experimental Dataset.
  • Please add information about advantages and limitation of applied method on the background of other studies. Additionally, I agree that UAV’s have many advantages but also have limitations comparing to different remote sensing sources.  It is cost-intensive, not allow to monitor a large target area over time, so we can observe changes of water quality only in selected part of river during one flight. If we look at satellite data there is around 290 km in one scene (Sentinel-2). Additionally, in my opinion, satellites can be used to monitoring of small rivers (Pleiades or SuperView-1 satellite 50 cm spatial resolution).
  • I don’t see author contributions section.

Author Response

Dear reviewer:
We thank you very much for the comments and suggestions. The comments and suggestions are valuable and very helpful for revising and improving our manuscript. We have made revisions according to the comments and suggestions. Please see the attachment.
Yours sincerely,
Xiaoteng Zhou

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Compared with the earlier version, the authors have improved greatly aspects of the the presentation. However, there are still many difficult to read sections, despite the improvements.

I have listed some below, but I encourage the authors to revise strongly their text for clarity, going beyond the specific remarks I have made here to create a coherent presentation of their work.

----------------

Please reduce the number of times "moreover" and "constraint" are used. 

Abstract
17 replace "proved" with "showed" OR "demonstrated"

Introduction

31 remove repetition of "small- and medium-sized", should read "these rivers" ...

33 change "(P) that may" to "(P), which may"

75 change "indexes" for "indices"

86-92 Numerous small vocabulary problems. Flow rate is awkward here, better to use "discharge" to avoid confusion, also talking about river bed morphological properties which have spatial dimensions of course, but need to use hydrological terms instead of putting everything under "spatial". 

Replace:
" Different discharges cause different pollution situations. Moreover, in hydrodynamics, the self-purification ability influences river water quality [22]. Streamflow and flow rate are two main factors that are closely related to self-purification ability. Among the river spatial properties, river width is a key factor that reflects streamflow and flow rate [23]. Thus, in addition to spectral features, spatial fea- tures also need to be used in water quality remote sensing. Furthermore, the surrounding environment should be recorded during sampling."

To something like this:
"The hydrodynamic properties of a river or stream affect it's capacity to self-clean [22]. Streamflow and discharge rates are two of the factors that are important. The channel morphology also affects this capacity, because it is affected by streamflow and discharge patterns [23]. Thus, in addition to spectral features of the water, water quality remote sensing should also include spatial features of the stream and surrounding environment."

100 replace "waters" with "water bodies"

102 delete "basic"

106 replace "be established" with "establish"

110 add "s" to network

111 replace "robust ability" with "capacity"

116 replace "proven" with "shown"

119 replace "acquisition" with "acquired" - This is an important point - the data must be of good quality to begin with

142 replace " the bands of satellite" with "satellite bands"

143 Need new paragraph at "Unmanned ..."

146 replace "fill in the gap" with "fill the gap"

147 replace "groundwork" with "ground"

158 change "fine" to "fine-scale"
etc.


Methods

166-167 Need to revise: "We propose an urban river in situ sampling dataset constraint for the large quantity and complex urban rivers."  I don't understand this phrase because of the non-standard usage of "constraint"

295-298 As far as I know, the Hach DR2800 is a spectrophotometer and not a multi-parameter probe ?? 


304 it's not climate but "season" here. 

352 replace "brook" with "stream"

355 This study does not concern climate. Replace with "season". 

392 Need to add web page for source of Pix4D.

394 delete "Using the ensemble learning model, we can finally obtain the water quality grading results."

399 typo "reasearch" should be "research"

407 delete "from life and industry"

563 "paramegamma" must be a typo?

576 replace "usually show" with "have shown"


Experiments and Results

Experiments should be separated from Results. 

Not necessary to repeat the same formulas between table 2 and Figure 2 and the text. Take out the formulas from Figure 2, and gather all the equations into a table, for example

627 appears to be a typo on this line

629 Section 3.1.1 and 3.1.2 belong in or after the Methodology section, not with results

835 replace "of the" with "at"

847 replace "43 thousand rivers." with "43 thousand waterways"

847 can delete "Rivers have a large water flowrate, and the water network is dense."

849 delete "condition"

853 replace "in different places in Shanghai, which are displayed in Figure 3.", with "throughout Shanghai (Figure 3)."

Figure 3 : the blue on the legend and the blue on the maps are not the same shade


Results
1034-1037 : Delete "In Table 3, for TP, the correlation coefficient of the original dataset was 0.66 and the other correlation coefficients were 0.71, 0.68, 0.76, and 0.68. for NH3-N, the correlation coefficient of the original dataset was 0.68 and the other correlation coefficients were 0.75, 0.85, 0.83, and 0.79. ", this information is in the Table !

1104 "still performed more stable" ??? no sense

1108-1109 "Nevertheless, all the models using only spectral features cannot be applied for urban water quality grading." ??

1134 Figure 16? This should be corrected

Figure 10 caption seems to be mislabeled beta5, beta6 etc. don't appear on plots

 

Discussion
There is no appreciable discussion of the limits, or other advantages of the approach. For example, what other elements of water quality could be detected by this approach, or not? 

There is also little discussion of when and where ML may fail to produce exploitable results. The only example given is for when shadows fall across the water surface. 

ML itself also has weaknesses that should at least be mentioned. If a water quality surveillance system fails, this can have big consequences. It seems worthwhile to consider where these weaknesses may exist in the discussion, under section 4.4 for instance. 

1248-1249  "Furthermore, other environmental features will be considered. Thus, the dataset can be expanded and more reasonable.",  should revise to mention specific features instead of resting on generalities

1252 "curse of", remove, personal opinion that is out of place in a method article
 
1282 delete "We proved that ensemble learning has higher accuracy for water quality inversion."

1291 delete "The processes of these studies are of great significance for reference. " No meaning as written

 

Author Response

Dear reviewer:

We are very grateful for your comments and suggestions in the round 2 review. The comments and suggestions are valuable and very helpful for revising and improving our manuscript. We have made revisions according to the comments and suggestions as described in the corresponding response.

In addition, we also think it is better to submit this manuscript for English editing. However, there was not enough time to Polish English. Thus, we decided to submit the revised manuscript first. We will submit this manuscript for MDPI English editing.

Best regards,

Xiaoteng Zhou

Author Response File: Author Response.docx

Back to TopTop