Developing a Semi-Automated Near-Coastal, Water Quality-Retrieval Process from Global Multi-Spectral Data: South-Eastern Australia
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsL25: Need to note in the abstract that the method ultimately relies on algorithm calibration and verification via in situ measurements. Having lots of satellite data is great, but it needs to be demonstrably anchored to reality.
L42: ...are mostly found in the open ocean but can occur in coastal waters.
L51: ...remote sensing process observation....
L56: ....will primarily occupy the space in the middle lower portion of the ternary diagram.
L66-67: may potentially offer potential solutions
L70-71: Can you include references to the satellite systems mentioned? The public S2 and Landsats are easy, but what about the Super Doves as commercial systems?
L93: Reference 12: There is insufficient information in the reference entry to locate the publication.
L96: "fully corrected" - what do you mean; atmospherically corrected, geometrically corrected, calibrated?
L107: Reference 28: Insufficient information; this is a chapter within a larger book, neither the book nor the editor are mentioned.
L110: Is there a reference for the assertion that a prominent absorption peak is observed around 685-700nm?
L118-122: Logic here is very confused. If the waters are turbid, there is no benthic reflectance, but there may be many additional optically active constituents present. If the waters are clear, then the benthic reflectance will become important. Maybe say something like "Close to shore, additional bio-optically active constituents are often present, and even when they are not, in optically shallow waters, bottom reflectance is significant."
L124-127: in water sensors CAN provide accurate and temporally dense streams; but do they not require specialised personnel and techniques? If your argument is that satellites are more cost effective in the longer run, then you need to emphasise the point about the cost of ongoing in situ obs.
L129-130: ...we can develop specify an appropriate capability to measure...
L137: It is a big claim to suggest that the processing and infrastructure capabilities are ideal. I suggest: "..infrastructure and processing capabilities are can make up for it to an extent."
L146: Are you aware of Cherukuru et al, 2014 "Influence of River Discharge and Ocean Currents on Coastal Optical Properties" https://doi.org/10.1016/j.csr.2014.04.022 ? (and the related https://www.researchgate.net/profile/Thomas-Schroeder-9/publication/260106801_Remote_sensing_of_apparent_and_inherent_optical_properties_of_Tasmanian_coastal_waters_application_to_MODIS_data/links/5d2fc774a6fdcc2462e877db/Remote-sensing-of-apparent-and-inherent-optical-properties-of-Tasmanian-coastal-waters-application-to-MODIS-data.pdf). These report detailed bio-optical measurements in your study regions, and note the dominance of riverine-sourced CDOM as the major light absorbing component in Tasmanian coastal waters. If you want to go ahead and just model Chlorophyll in the bio-optics, you need explain how you can ignore CDOM, or at least what impact ignoring CDOM will have (eg added noise after rain events).
L148: replace "extraction" with "estimation".
L150-157: I can not follow the logic of the argument here which seems to confuse absorption and reflectance. What is the "elastic component"?
L165: Can you comment on whether such an algorithm is appropriate to use in CDOM dominated waters? I believe these algorithms were developed for Case-1 waters. Do you need to do something using multiple spectral bands to try and remove the effect of CDOM first?
L181-182: NOMAD overlaps with a significant proportion of the Landsat archive. Why does the fact that it stops in 2008 matter? Are you implicitly asserting that there has been change over that period? Be clear about what you are trying to say.
L193-195: You say why the satellite data are good (appropriate bands, pixel sizes, frequency) but not why they aren't. Because they weren't designed for ocean colour work, how do they fall short? [eg signal to noise ratio or gain, radiometric calibration, ability to be accurately atmospherically corrected]. Also, if you are only getting a look once every 14-16 days, is that really a suitable observation frequency?
L203: Chlorophyll-a
L205: We need to know what ARD means - is there a reference that explains what processing has been applied, including atmospheric correction?
L221: Repeating mention of ARD, but still not saying what it means.
L227: Figure 3 does NOT outline the process. It indicates the locations of the selected sampling sites.
L230: Salmon is important, but at two of your sites, shellfish are the dominant aquaculture. Why mention only the Salmon, when the results at two sites are irrelevant for salmon?
L238-239: What aquaculture is present at each site?
L238-239: I suggest introducing SH, GOB, BI as abbreviations for the three sites and using throughout from here onwards.
L251: Do you have a reference, or a footnote link, for eagle.io?
L262-265: You need to proofread this sentence and address the grammar - it does not make sense
L265-266: Absorption is controlled by more than just Chlorophyll; what about CDOM and NAP?
L269: You are not extracting the optical properties of the water column. You are estimating the optical properties of the near-surface layer. Anything more than an optical depth below the surface is going to be virtually irrelevant given the relative inaccuracy of the surface reflectance observations.
L312: Is black or grey the cloudmask? I think it is probably the grey, but what actually does black represent? This information should be in the figure caption.
Table 1: You really need to include a second table somewhere here that clearly sets out the field survey dates - both the one off visits and the installation periods for the sensors left at the two sites. Without a clear expression of this info the reader is left grasping at straws to figure out what data was being compared with what and when.
L317: What is the YSI sensor? Not previously mentioned or referred to by this name.
L319: What do you mean by "through the water column" - do you operate it at different depths, does it take a depth integrated measurement?
L322-323: Who verified it? Do you have a reference? Or is that what you are trying to show later in the paper - in which case it is premature to state it now.
L326-330: What disturbances do you mean? Are you referring to shadowing by the boat, or mixing of the water column? These effects are unlikely to impact on the precision of the measurements, but they may well affect the representativeness and accuracy.
L332: do you mean satellite data?
L339-341: What atmospheric correction was performed? Is there a Reference for the ARD, or is it something you did yourself? What "surface reflectance correction" are you referring to? Is this BRDF (eg for land), or something else?
L343: How could you distinguish homogeneous water pixels from inhomogeneous water pixels?
L345-L347: these areas are dominated by CDOM, and near-coastal waters are often more influenced by turbidity than chl.
L347-348: What depths were samples taken at? This should be included in the additional in situ data table that I proposed above.
L357: ...concentrations at 3 locations in Tasmanian waters.
L360: imagery.
L360-363: What local parameters do you mean? What evidence is there that these are influential?
L364-368: "Consolidating data from all three locations revealed no correlation between the in situ and satellite data, indicating that significantly different bio-optically properties apply at each location."
L381: Root Mean Square Error (RMSE) errors,...
L388: What was the end date of the satellite processing?
L391: Can you quote some number that gives an indication of the magnitude of the impact of this correction and its uncertainty on the radiometry? If 90% of the signal is atmospheric origin, and it is removed with uncertainty of 20%, what does that mean for your surface reflectance measurement?
L399-401: Isn't it pixels per field location that matters? eg Does 1344 SH pixels mean 1344 pixels for every site in SH, or is that split between 26 sites? Over what total time period was this - when in 2022 did you stop?
L411: situated in at Bruny Island.
L412: The mysterious undefined YSI sensor appears again.
L413: I don't think a buoy can be diligent. Perhaps try "regularly".
L414-416: Repetition of the fact that BI is cloudy.
L416-419. This section mixes up the limitations of only sampling one time and needing to sample many locations. What about sampling at multiple times? It is extremely unclear what point is being made here and the logic of the argument needs to be clarified.
L422-423: How does the unavailability of historical data highlight the need for robust validation? Robust validation is a requirement for confidence in the model - whether the data is historical or not is only relevant if you want to use the model historically.
L426: You are quoting the sensor location with nanometre precision. Surely you don't mean that you had it that accurately?
L445: ... captured observed...
Table 2: What does Depth >=2m mean? Do you mean locations where the water was > 2m depth, or do you mean in situ samples were taken at more than 2m depth? Should the heading of this column be "sites" rather than "samples"?
Table 2: What are the units of RMSE? mg/l?
Table 2: What do corrected and uncorrected mean?
Figure 5: What is the red line?
Figure 5 caption: Why mention the sites - you've labelled the top row of panels, just label the bottom row as well. You do not make clear what the difference is between the top and bottom row.
L457 - what studies - give a reference, or else just state that "optically shallow sites are affected by bottom reflectance".
L462: What do you mean by "the bottom influence may exceed two metres"? Is bottom influence measured in metres? Are you attempting to say that you applied a threshold of 2m depth, and excluded sites that did not meet that threshold?
L480: ... in terms of such as sediments...
Figure 6: In panel (a) the chl estimates are clearly correlated with the channels in the north east part of the image. Are these artefacts? In panel (c), there are pretty obvious artefacts in the chl estimates adjacent to some of the clouds, for example in the north west of the image and in the south east in great bay. If you do not comment on these, then what else are you not noticing?
L481: Where did these 57 sample points come from - one of the 3 sites, or all of them, and were they from different in situ sites within each of the larger sites, or just one?
L482: How did you ensure they were free from atmospheric disturbances? By what criteria did you do this?
L483: Why only seagrass? What about other benthic types?
L484-487: Repetition of the point previously made about the impact of bottom reflectance.
L493: what does "low-high energy" mean?
L496: Why only macroalgae?
L500-501: Most case-1 waters are the open ocean where bottom reflectance is irrelevant. Your study areas are not case-1, so why is this point being made?
L506-507: This would be a good point to mention that in situ data is needed to develop a location based model.
L507-509: But isn't this exactly what you have done? Taken a model for case-1 waters and applied it directly to turbid CDOM-dominated waters?
L509: Usually rivers have a lot of CDOM too - not just sediment.
L511-512: I can't figure out which in situ data you have selected to perform this comparison.
L523: Figure 7 is NOT a time series. Neither axis is time.
L526: replace "recently" with "only". A table with time ranges for the relevant in situ sampling would really help with understanding what you've done.
L529: "study period" - same comment as for L526.
L536: ...cloud-free satellite pixels for both in situ sensors...
Figure 7: Is it really coincidence that both r^2 values are the same? You need to mention in the caption that the red dots are excluded outliers, and what the dotted and solid lines are. In panel (b) I find it hard to understand the logic that the point at the top right is an outlier while the one at the left (in situ=1, modelled=1.6) is not.
L546: Units of RMSE?
L552: What depth? Is it near the surface?
L553: Is there any evidence for sub-30m spatial variability? Did you sample at multiple in situ points along a 30m line to see if there was any consistency or variability? Is it possible that the presence of unmodelled CDOM is responsible for biasing the modelled estimate?
L562: What do you mean by "instrumental"? Related to the sensor instrument, or to do with the algorithm?
L562-563: What is an exceptionally high or concerning value? How do you make that value judgement?
L565: "delve deeper" is a poor choice of words - do you mean make sub-surface measurements or something else? Ambiguous.
L577: Because I can't figure out from the description earlier what actual in situ data you compared, I can't be confident that this conclusion (about accuracy) is correct.
L579: Bruny Island site is hardly open-coast; it is a channel subject to modest tides in the western part but relatively slack and undisturbed in the eastern part of the channel. It would have been helpful to describe the tidal ranges at each of the 3 sites when they were introduced to give readers an idea of the extend of tidal mixing that might be expected, not to mention whether a 2m bottom depth threshold was sufficient.
L583: Are these estimates from satellite data or in situ measurements?
L593: Do you mean more satellite or in situ data?
L594: A satellite measurement doesn't tell you much about the water column, only the (optically) near surface layer. Unless there is significant vertical mixing present - eg if there are strong tidal currents.
L599: What unique trends do you mean - are you trying to say that the models were different in the two locations?
L615: Why will longer data collection improve the model reliability? If the model is inadequate because it does not represent the optical complexity, additional data will likely just make any correlation gradually disappear.
L616-618: Regarding the starting point of the study; what point are you making here?
Figure 8: Are these plots showing estimates for each of the locations in Figure 2? If so the choice of symbols (different shades of grey) makes it impossible to figure what is going on. Unless you use different symbols (eg cross, box, plus etc) for each location within a site, you might as well just show one location at each site.
L626-628: This is a superfluous point.
L628-629: In what sense are they relatively stable? They seem to very widely between 0 and 2?
L636-637: errors and biases are only minimised if the automating model is representative of the relevant processes. Otherwise it just introduces a new set of errors and biases.
L640: What do you mean by "more" Landsat data? More locations, longer time series? How would adding data increase the scope and applicability?
L646: ...likely to be required to be incorporated....
L651-658: This section seems like uneccessary repetition of previous material.
L658-659: which studies - references required.
L667-668: You mean the Australian Ocean Data Network (AODN) collection?
L667-679: You could just say that "although there is an extensive network or in situ sensors of relevant parameters operated by the AODN, very few are at the surface which makes them of limited use for validation of satellite-based models." The map is not needed.
L679-695: Similar statement for NOMAD and GLORIA would be sufficient. Also, why do you not mention NASA SeaBASS (https://www.earthdata.nasa.gov/learn/articles/seabass-in-situ-aquatic-data)
L702: replace "integrating" with "including"
L705-706: Again - how would you identify (presumably spatially) homogeneous pixels?
Table 3: What does "elevation" mean. Do you mean depth? Consistency of terminology goes a long way to eliminating ambiguity and making your arguments clearer.
L718: Surely a nationwide (and I think you mean continental-wide) model is inconceivable?
L721: Which inherent characteristics of water do you mean? Or do you mean the inherent optical properties (absorption and reflectance) of water including its optically active constituents?
L724-726: It would be good to mention that the presence of these elements could confound a method that is based solely upon an assumption of chl-only, not just that they contribute to water quality.
L729-730: replace "integration" with "inclusion"
L730: Which commercial satellites? I've heard there are a lot.
L731: replace "persistent" with "frequent". If it is persistent you'll never get a clear pixel.
L731-732: Which additional variables? Do you mean more satellites or something like the tides or river inflows?
L733-735: I don't follow the logic here. Why do you need to check and adjust periodically? Surely the challenge is just getting some observations when it isn't cloudy, and then you are set?
L739: replace "insights" by "observations"
Conclusions: Would probably be good to mention the potential afforded by the PACE and other hyperspectral missions to cross calibrating the higher resolution satellite models (landsat, S2) in representative coastal waters. Also, Aqua-watch mission proposal is potentially very relevant for Australian coastal waters (although it may yet be vapourware).
Comments on the Quality of English Language
Many of the descriptions, discussions and conclusions are excessively wordy and/or repetitive, and use language that is imprecise or unclear. Many specific examples have been provided in the detailed feedback. A significant effort is required to provide more succinct, precise and unambiguous descriptions of the methodologies and discussions. Without that effort, readers will be unable to discern what actual methodology was followed and thus whether the results are valid.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsIn this study, the authors proposed a method for extracting chlorophyll-a specificity from surface water in the near coastal environment of eastern Australia based on multispectral sensor Landsat-8. Overall, the paper has a good structure, a large amount of work, and outstanding results.
The following are the limitations and suggestions:
1. The abstract description is somewhat redundant, and more quantitative results should be added to this section.
2. Figure 2 and Figure 9 lack compass.
3. The scale format in the paper should be consistent.
4. In section 4, "Bruny Island shows the opposite temporal trend" can explain the cause.
5. The process of automatically building the GEE model should be elaborated in detail.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsMany improvements have been made to the paper since the first review.
However in many places where I asked what was meant have been explained in the cover note without changing the text of the paper. My asking what was meant was intended to signal to the authors that their explanations were unclear or unsatisfactory, and where no changes have been made beyond elaboration in the cover letter - that remains true.
I have noted a number of detailed further comments below:
L42 I suggest: Case-1 waters are characterised by suspended….biomass and primarily occur in open oceans but also in some coastal waters.
L55-56 – “…highly spatially and temporally dynamic..” is just repeating yourself from L51. Try to be succinct
L125 In shallow (depth < 20m) coastal near-shore areas, where…..; otherwise it looks like you mean within 20m of the shore!
L153: …image pixel at wavelength \{lambda} is as follows….
L191: do you mean minimise or estimate? If you can estimate the CDOM from the whole spectrum you could account for it in the CHl measurement.
L221 – “often enough” - you’ve previously said that these waters are highly temporally dynamic; so you should qualify what “often enough” means – every 16 days doesn’t really cut it. Maybe with a constellation you can do better – what time scale is “often enough”?
L249 – having introduced the GEE abbreviation in L245, perhaps you could use it here?, also L261.
L268 – Who says “Tasmania boasts”? Use objective/neutral succinct expression. “In 2022 44 farmers collectively produced oysters valued at AU$39 million [60]”. Also – in :266 you should give some idea of the $ value of the salmonid industry for consistency and comparison.
L348-350 The sentence “This meticulous approach…” simply makes an assertion without support, adding nothing to the previous statement. Remove it.
L388 – no need to define GEE a second time
L390 – you have said in your cover letter what atmos+surface correction means – but you need to include the information for the readers of the paper, not just me! Can you include a reference, or even the statement as a footnote?
L420: These factors…
L447-450: Applying scale factor and offset is NOT radiometric correction, it is simply converting the data stored as integers to their floating point values. Radiometric correction is to do with the calibration of the sensor to units of radiance which is done as part of the creation of ARD. That said, (Surface) Reflectance is a ratio of out radiance to in radiance so is dimensionless. I suggest replacing the whole section L445-L450 “A total of 389 scenes (STH – 130, GOB – 140, BNI – 119) were selected between the beginning of January 2019 and the end of March 2023 and processed the GEE.”
L475: do you mean “calibrate the model”
L485: do you mean Table 2?
Table 3: You need to clarify/explain Depth >=2m in the paper – not for me in the cover letter. I asked because it is not clear what you mean. Maybe change “All” to “All depths”.
Table 3: Change the second last heading to “RMSE (mg/m3)”
Figure 5: Again – the explanation about what the red line is, that you in the cover letter, is no use to readers of the paper – the information should be in the caption where it is useful to readers.
Figure 5: What are the coloured contours? Unless you explain what they are and they add interpretive value, simplify the figure and reduce the cognitive load for the reader and leave them out.
Figure 6: Again – you have addressed my comment about artefacts in the cover letter but not improved the description in the paper. Any experienced remote sensing person would look at these images and note the possible artefacts. That you do not even comment on them suggests an unquestioning incurious approach which will lead readers to doubt whether you are paying attention. You need to at least acknowledge the obvious potential shortcomings.
L541: So, from your explanation in the cover letter, what you mean is not “free from atmospheric disturbances” but “the data were flagged using the quality information provided in the metadata.”
L542-543: …bottom reflectance including seagrass, a potential confounding source of Chl-a pigment signal.
L548: Make clearer: “Applying a minimum water depth threshold of 2m across all three …..”
L550-552: This does not make sense. If the bottom effects are less important why would you need to investigate further? What are you trying to say here?
L557 – If the Tasmanian waters are dominated by CDOM, are they really predominantly clear waters? Highly turbid waters are characterised by high sediment concentrations. Perhaps you mean to say that these waters are predominantly absorbing due to CDOM and Chl-a, rather than backscattering due to turbidity?
L574. I have no idea what you mean by a temporal correlation plot. Isn’t figure 7 plotting the same thing as figure 5 – modelled chl-a versus in situ chl-a, just with data selected differently. Are you trying to say that each plot is for a single location for which data is available at multiple times? In that case, try something like: ”Figure 7 compares modelled (satellite) chl-a against in situ measurements for the two locations where data were available at multiple dates.”
Figure 7: I think you need to rename this plot as per my previous comment. “Comparison of modelled (satellite) chl-a against in situ measurements for the two locations where data were available at multiple dates.” You don’t need the caption to say it is great oyster bay and bruny island because the plots are already labelled as such.
Figure 7: Once again the plots have coloured contours without explanation. If you don’t explain what they are and they don’t convey information relevant to your argument – simplify by removing.
L613-616: This sentance effectively says “Low A versus high B means A is lower than B” – which is so obvious that it is superfluous – you will improve the text simplify by removing it.
L618-620: There is a problem with the logic here – you only flag the data because there is an anomalous discrepancy between in situ and modelled, however the method you are proposing in this paper is supposed to useful because you can estimate conditions using the satellite data only. If you only have satellite data, how will you identify anomalous values that are so important to aquaculture managers? You need to explain this better in the text.
L668 – July and August are winter months, not spring. As such there is a likelihood that there is increased runoff from the Georges River due to seasonal rainfall – so your peaks could be due to CDOM or other inflow related bio-optical substances. That’s fine – your method could be detecting that, which is useful – so you should say so!
Figure 8: Vastly improved – well done. You can probably leave out the last sentence in the caption – it is pretty obvious that the line joins the dots.
L702: The acronym ADON is still wrong. Also in L726
L746: SeaBASS is an in situ data collection that was compiled as part of the SeaWifS mission – but it doesn’t use Seawifs data – the unsuitability of seawifs imagery for your application is immaterial to the utility of the in situ data in the seabass collection; you can still compare that in situ data with landsat imagery. The point of this whole section is surely to illustrate that there are some existing in situ data sets but that they are spares, offshore, or historical, and thus validation of the method you are proposing (hi-res sats with cloud infrastructure) will require more contemporary and relevant in situ data collection.
L771-772: The spatial resolution of PACE may indeed not meet the requirements of your study areas – but you might want to point out that the spectral resolution and sensitivity may be valuable for understanding the bio-optics in a larger scene (eg slightly offshore) that you can also observe with Landsat and therefore compare the performance of your algorithms.
L773-774: Through its use of Landsat 8, this study addresses….
L837: ADON acronym wrong, and explanation incorrect as well.
Figures A1, A2, A3: It remains unclear to me what purpose is achieved by including the maps that cannot be taken care of by saying “the data is sparse and often not in locations of interest ot aquaculture” or words to that effect.
L860 – insufficient information to identify reference; is it a book, with a publisher or editor, or a journal?
Comments on the Quality of English Language
As a general principle we would expect succinct descriptions of work in a journal. Repetition, apart from making the paper longer to read, introduces opportunities for inconsistency and ambiguity, both of which reduce the clarity of description. A ruthless proof-reading to eliminate superfluous words, and especially subjective descriptions, would be very beneficial.
For example: "The handheld Exo 2 multi-parameter sensor played a pivotal role in the data collection process at each survey location site. It’s crucial to emphasize that the sensor captures data at a specific location on the water surface and through the water column, providing measurements in micrograms per litre (μg/l). The sensor captured chlorophyll-a values at various depths, which were later averaged across the top 1-meter of the water column for comparison and contrast with the satellite’s multi-spectral data."
could become
"The Exo 2 sensor measures concentrations in the water in micrograms per litre. Observations were made and averaged across the top 1-m of the water column for comparison with the concentrations estimated from the satellite data."
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 3
Reviewer 1 Report
Comments and Suggestions for AuthorsL386 – you should probably make a comment that “the accuracy of an atmospheric correction developed for terrestrial environments, which are generally higher brightness, may not be as applicable for ocean scenes where the reflectance is much lower. Nevertheless it allows a correction to be made to enable a proof-of-concept demonstration such as this.” Just show that you are aware that terrestrial atmoco is not sufficiently stringent for ocean colour work but that it is expedient for the present purposes.
L507 – By “linear trend-line” do you mean a “least squares linear fit”?
L581: This temporal correlation plot is utterly misleadingly named. There is no correlation taking place in the time domain, which is what such a name implies. I strongly recommend that you reword this and call it “a comparison plot of modelled satellite values compared with in situ observations across the study period. Only cloud-free satellite data acquired within X hours of an in situ observation are shown.”
Figure 7: Is ‘potential trendline’ in the caption another “linear least squares fit”?
Reference 76: ADON is wrong.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
