Water Quality Chl-a Inversion Based on Spatio-Temporal Fusion and Convolutional Neural Network

Round 1

Reviewer 1 Report

Review manuscript

remotesensing-1596927-peer-review-v1

Title: 
Water quality Chl-a inversion based on spatio-temporal fusion and convolutional neural network

Authors: 
Haibo Yang, Yao Du, Hongling Zhao, Fei Chen

Summary: 
The manuscript describes and tests a new CNN method to evaluate spatio-temporal changes in Chl-a concentrations from time series of satellite images. The advantage of their approach is that they develop a means to convert between different sources of satellite images in order to extend and fill-in gaps of observational time series. This represents an important contribution to the field.

GENERAL REMARKS:

Overall, this article is well-organized and the figures, even if I was doubtful of their number, end up to compliment the text well. 

However, parts of the text are inadequately developed and presented. For instance, the reasoning underlying decisions made in the methodological development (Sections 2) are often not explained or referenced. In the results section, the descriptions of the figures are often too short, leaving the reader to have to puzzle over what each plot and figure actually show. Each figure needs a proper description of what is plotted, then what the figure is supposed to show or demonstrate in terms of a result. For instance, Figures 15 and 16 don't even mention that they are time series. The discussion is surprisingly underdeveloped. Instead of a considered reflection of the work and progress accomplished and, importantly, what is not yet accomplished, it reads like a protocol test report. This article is however presented as a research article. The discussion needs to place the results in the context of published work, recalling both the advantages and disadvantages offered with respect to other authors. For instance, what spatial and temporal resolutions are actually necessary to optimize decision-making procedures using satellite imagery? What are the range of sizes of water features that would be targeted for monitoring? What kinds of precision are acceptable or unacceptable? What kinds of errors could occur and where would these interfere with the NN results?

Finally, apart from the need to expand and develop further the results and discussion with respect to other workers, I have made a long series of minor changes and suggestions that either correct misspellings, small grammatical problems or aim to improve the accessibility of the text. 

--REMARKS--

GLOBAL CHANGES:

To avoid: word repetition in the same sentence !

Add "Lake" after BYD throughout

Change "GF-2 like" to "GF2-like" to clarify this is being used as a label (and already used like this on Figure 16)

"Nir" should be "NIR" throughout

"Pixel" should be "pixel" throughout

FIGURES:
Overall the graphic quality of many figures is poor, with differing file resolutions, subplots not properly aligned and extraneous borders showing (e.g. Figures 7, 8, 15) and repeated information in the labels. The figures should be reviewed in detail to have a more consistent presentation quality.

Figure 1 - " parts by the dotted line", replace with "regions indicated by the dashed lines" 

after "two detail display boxes" add "; see Figures 6, 10 and 12."

change "details display box" to "Area of detail"

change dashed rectangle in legend to an oval instead

Figure 2 - add "Example of an image matching ..." to the caption. Add the distances represented by the sides of the blown-up detail areas shown on the right-hand side of the figure.

Figures 3 and 4 could be combined into one

Figure 6 - add reference back to Figure 1 in caption - "August 16, 2019 (region I, see Figure 1)." Delete title from above the images, the same information is already in the caption. Add distances on detailed areas - like for Figure 2

Figure 7 - on x-axis should read "NIR", not "Nir", must add descriptive title for Y-Axis. Also the caption should be revised to indicate what is plotted (?mode ?mean ? range??).

Figure 8 - information on the fits are repeated on the plots and in the text - need to revise to remove unnecessary repetition

Figure 10 - add reference back to Figure 1 in caption 

"Retrieved Chl-a results for BYD Lake on August 21, 2019, after pixel correction (a). For insets results of region II (same as Figure 1) are shown for: Chl-a based on Sentinel-2 image (b), Chl-a based on GF-2-like image (c), and Chl-a after pixel correction (d)." 

Figure 11 - could be removed and data reported on the fits in the text only. 

Figure 12 - Add labels above each column of images 
Sentinel-2    GF-2-like     Sentinel-2    GF-2-like

Invert placement of N arrow and the colored Chl-a scale bar (which should appear directly under last two columns)

There is a missing English units description for scale bars on plots a4 and b4

And need to clarify description
Images a1 and b1 indicate Sentinel-2 image, a2 and b2 indicate GF-2 like image, a3 and 311 b3 indicate Chl-a result from Sentinel-2, a4 and b4 indicate Chl-a result from GF-2 like image after 312 pixel correction)

Figure 14 - The different patterns are difficult to distinguish - wouldn't this be better if converted to colors instead? 

Change "in different regions" to "in the eight regions indicated on the map of Figure 1"

Figure 15 and 16 could be exchanged with Figure 13 and 14. This would make a better narrative to end the results section on.

Figure 17 - correct the misspelling of "iteration" on the plot c legend

LINE-BY-LINE COMMENTS AND CHANGES

33-34    "Water quality conditions directly affect national survival and production"
suggest change to: 
"Water quality conditions directly affect public health and economic productivity of nations."

35    delete "traditional"

36    "they are time-consuming,"
clarification - add "also": they are also time-consuming, 

37 "complete timely and large-scale monitoring"
clarification - change to "complete timely and large-scale monitoring"

38-40    "Remote sensing technology has intuitive and macroscopic characteristics, which can reflect the spectral characteristics of water quality in large-scale [7-9]"

revise for word repetition:    "Remote sensing technology has macroscopic qualities that are well-suited to the spectral characteristics of water quality monitoring at large-scales[7-9]"

40-43    "Various remote sensing data and continuously updated water quality inversion models make water quality inversion a real-time and convenient path, which facilitated the development of remote sensing applications for water quality monitoring[10-12]."

revise for clarity and word repetition:    
"Various remote sensing data and continuously updated water quality inversion models make water quality inversion a convenient means of achieving real-time monitoring; this has facilitated the emergence of new applications [10-12]." 

46    delete "mainly" and add ":" after include

48-49    "It is worth mentioning that a lot of Chl-a content inversion researches 48 are more valuable and reliable in application[15]"

revise for clarity:

It is worth mentioning that much of Chl-a content inversion research is used in applications[15]

50    change "New" to "new"

53    change "qualities" to "qualities"

54-55    "Due to the constraints of the existing sensors, no remote sensing data can take into account high resolution in temporal, spatial and spectral dimension" 

revise for clarity:
"Due to the constraints of existing sensors, no remote sensing data is acquired at high resolution data in temporal, spatial and spectral dimensions" 

56 delete "a lot of"

60 "neighbors. Fu .." - delete the period and replace with a comma

60-61 "proposed ESTARFM" - need to define "E" in acronym, as have done for ESTDFM

69    "It is difficult" 

suggest:  "It can be difficult"

78    "Poyang Lake , etc. "

Remove "etc" and replace with other concrete examples, OR delete if don't wish to add more and replace "including" with "such as"

78-79 "Whether the neural network ..." this should be the first sentence of next paragraph

82    "The water quality grade of BYD Lake was usually higher 82 than V in a slightly eutrophic state[53]. "

Unclear what the scale is - needs additional explanation

84-85    " ... and convenient by using remote sensing data because of lower temporal resolution or cloud coverage"

Unclear sense, needs reworking

85-87    "And also, due to the 85 complicate process of the water quality inversion, there are great needs to establish intel- 86 ligent models to retrieve Chl-a. "

Revise to something like this:
"Also, due to the complicated water quality inversion process, there is an important need to establish robust models for Chl-a. "

87    add ":" after "to"

90    add Lake after "Baiyangdian"

91-93    "Through this method, we can realize multi-period continuous Chl-a monitoring with high frequency in time and fine resolution in space. It provides scientific basis for inversion of multi-spectral Chl-a concentration and quick diagnosis of eutrophication in Baiyangdian Lake."

Revise to:
"We use this method to do a multi-period, continuous Chl-a monitoring at high frequency and with fine-scale spatial resolution. The aim is to provide a quantitative basis for inversion of multi-spectral Chl-a concentrations and enable rapid diagnosis of eutrophication in Baiyangdian Lake."

99    to avoid redundancies in phrase, delete "fan-edge" and add "formed" after "depression"

100-104    "We collected 92 Chl-a concentration data in BYD, selected eight oval regions to calculate the changes of Chl-a concentration in different regions with time, and selected two small rectangular regions to verify the details of STF results, the details of water quality correction algorithm, the changes of water quality with time, and the details of STF and water quality results."

Simplify and correct:
"We collected 92 water samples from BYD Lake, within eight oval-shaped regions (Figure 1) to calculate changes in Chl-a concentration over time. We also selected two smaller rectangular regions (Figure 1, detail areas I and II) to verify the STF results at fine-scale, including the water quality correction algorithm, changes of water quality with time."

109     Missing subject, insert at start of phrase, "Water samples from ..."

110    Change: " using EXO multiparameter and RTK from" TO "... using an EXO multiparameter sonde and real-time kinematic (RTK) positioning from ..."

111    Insert: "sensor" after "multi-spectral" 

111    Replace "69 days of return period." WITH "a return period of 69 days."

112    Replace "Because of coarser temporal resolution," WITH "This means that ..."

113    "preprocess " should be "preprocessing "

115    "imager" should be "camera" OR "sensor"

116    "5-days of temporal resolution" should be "a temporal resolution of 5-days."

116-118    Replace "Sentinel-2 data was open- 116 source and free of charge from the ESA Copernicus Data Center, it was widely used in 117 many aspects. "

with
"Sentinel-2 data is open-source, and available from the ESA Copernicus Data Center (GIVE WEB PAGE ADDRESS USED). "

118    "preprocess " should be "preprocessing "

120/121    Replace "selected" with "used" on both lines

121    change "landsat" to "Landsat"

124    "The remote sensing image query 124 results are shown in Table 1." 
Change to: "The dates and sources for remote sensing imagery analysed in this study are given in Table 1."

TABLE 1 - remove last row with "STF" - this is unnecessary and confusing information. 

129    Replace "Prediction and generation" with "The production"

131    need to explain the abbreviations "T1" "T2" 

146-151    need to revise, eq. 2 is described first - equations should be described in order of their appearance

169    instead of "a pair of data", use "paired data values"

171-172    "This fusion calculation was performed on the computer 171 with Intel Core i7-10700K plus 64G memories, and took a total of 204 hours."

Revise to:
This calculation required about 204 hours of computing time using a computer with an Intel Core i7-10700K and 64G memory.

175    Suggest replace "In this study" with "To achieve this"

180    Replace "which" with "and"

181    Replace "to" with "at "

179 - 194    The letter "O" and number "0" look almost the same in the manuscript font, could another letter, be substituted for legibility?

199    "Image" should be with lowercase "i"

201    Unclear meaning "we used landsat8 conversion ", and Landsat should be capitalized

202    Change "filling time series" to "filling-in the time series"

204-205    "We choose 92 sampling points and corresponding grid pixels from GF-2 like image  on August 21, 2019 as training input to design Chl-a inversion model based on CNN."

correct to: 
"We chose 92 sampling points and their corresponding grid pixels from the GF-2-like image  taken on August 21, 2019 as training input to design a Chl-a inversion model based on CNN."

206    Change " each 5 × 5 " to " 5 × 5 "

207    Changing "was corresponding to each sampling point position" to "corresponds to each sampling point position"

225-226    Unclear "determine the best combination of parameters by traversal to improve the accuracy of the model"

228    "R2" should use an exponential 

240    "Pixel" should be in lowercase throughout the text

243    Change "in formula indicates " to "is" 

259    Change "Figure 7. shows the box line plots of 1000 sampling points, respectively."
to
"Figure 7 shows the distribution of the 1000 sampled points for the the GF-2 and GF-2-like images for each spectral band. "

271    Information on fit is repeated in text and on plots, need to revise

280    deleted " kind"

281-282    Not clear usage of "unexpanded/expanded/extended" in this part

282    should avoid use of "believe" in the results section - this is better for the discussion

287    "with the core distribution in 3.53-19.47 μg/L, "- unclear what is meant by core distribution 

289    delete "It can be found that " 

305    "region" takes an "s": regions

307-308    "both could well express the spatial distribution, while the per- 307 formance of the result from GF-2 like image were richer details."

change to
"both express well the spatial distribution of Chl-a values, while the GF-2-like images (Figure 12-a2, -b2) are richer in details."

317 change "a trained neural network model" to "the trained neural network model"

322-324    ?? unclear, needs revision : "the change trend of the inversion results of sentinel-2 data generated by 322 Landsat is consistent with that of real sentinel-2 data, indicating that the data supplement 323 is feasible."

329    Change  "in different areas of BYD." to "the eight areas of BYD Lake (Figure 1)."

330    Change "We collected 1000 points from each of these areas and plotted them as box plots"

to something more informative, like:
"We plotted the mean and range of 1000 points from each area, for 8 consecutive dates."

332     Capitalize "during"

329-337    The description of the trends observed needs to be revised. For example it is incorrect to state that "box shapes" become larger. It is the spread, or range of values in the region becomes wider. In addition, since this is the results, the last line (335-337) which makes an inference belongs in the discussion.

356+406    Missing from the discussion section is any perspective on the work presented relative to other research groups, or even progress made in the application of NN to water quality monitoring with remote sensing imagery. Only three articles are cited in the discussion! 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

This paper has a great findings. However, The authors need to add an illustration of the overall research framework in the method section. Surprisingly, there was no citation in the method section including the equation (why?). Inappropriate figure caption is everywhere. The explanation of the figures in the result section is not adequate. The discussion section looks like the result section since there was no discussion. Please check the further comment in the attached file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The manuscript is greatly improved