Retrieving Total and Inorganic Suspended Sediments in Amazon Floodplain Lakes: A Multisensor Approach

Round 1

Reviewer 1 Report

After reviewing the revision and in particular the detailed comments, I am satisfied that the authors have addressed my initial criticisms. While this analysis still overlaps with similar papers, I agree that the authors have addressed a unique enough situation to warrant publication.

Regarding the Nechad evaluation, I am glad to see that the authors tested this quite extensively. I am OK with not including the results but would still suggest that they include a few sentences saying that they evaluated that method and that it wasn't nearly as good, since it will be an obvious question for people reading this paper. Perhaps include the summary points about the need to calibrate the Nechad algorithm for this specific region, so it's clear why the global method fails here. 

Author Response

The authors would like to thanks to the reviewer and consider that all of our comments were essentialy for manuscript improvement. 

Regarding the Nechad evaluation, I am glad to see that the authors tested this quite extensively. I am OK with not including the results but would still suggest that they include a few sentences saying that they evaluated that method and that it wasn't nearly as good, since it will be an obvious question for people reading this paper. Perhaps include the summary points about the need to calibrate the Nechad algorithm for this specific region, so it's clear why the global method fails here. 

Regarding Nechad algorithm, authors added a sentence at page 10, lines 312-315 in the revised manuscript, as follows:

"The semi-analytical approach proposed by Nechad et al. [62] was also evaluated. However, it didn’t present sound results for our study area and required a re-calibration. Then, Nechad algorithm results were not included in the paper."

Reviewer 2 Report

Authors have addressed my comments and the revised version is much improved. Therefore, I recommend its publication in Remote Sensing.

Author Response

The authors would like to thanks to the reviewer and consider that all of our comments were essentialy for manuscript improvement. 

Reviewer 3 Report

I acknowledge the efforts done by the Authors to address my previous comments. I think that the manuscript could be considered for publication.

Author Response

The authors would like to thanks to the reviewer and consider that all of our comments were essentialy for manuscript improvement. 

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

This is generally a well-written manuscript that describes the application of regionally tuned TSS algorithms to the Amazon floodplain. The authors make a reasonable argument for the importance of the study site and the need for higher resolution data. The methods are generally well described (but see below—it is not clear to me exactly what they are using Monte Carlo simulations for), and the results, while limited to only a few satellite images, are reasonable. 

I have three major comments that I think need to be addressed before recommending this for publication. First, the authors argue that 250 m resolution MODIS data is not sufficient, but they are using 3x3 pixel averages for OLI which is 90 m resolution, versus 30-60 m resolution for MSI. I would like to see some discussion of the relative importance of pixel size for the region, since this was a primary incentive for switching to higher resolution sensors. Is 90m resolution adequate? What do you get with 30-60m resolution?

Second, I think they need to be a bit more careful about their conclusions regarding the need, or lack of need, for glint corrections. MSI has a serious glint issue because of the orbit, while OLI is better, but it’s VERY dependent on the time of year and scene, etc. They may rather broad conclusions based on a very few number of images. 

Finally and this also touches on the glint issue, this paper is very similar to Novoa et al., Remote Sensing 2017. Those authors came up with very similar conclusions (atmospheric correction is important, you can merge multiple sensors, etc.) except that they ALSO evaluated the semi-analytical models of Nechad. I personally don’t think we need yet another set of empirical algorithms developed for a specific region (even though it was done very well in this case), and I would strongly encourage the authors to test the Nechad method, which was deliberately developed as a generic algorithm that could be applied to any sensor. This is much more attractive than building empirical algorithms for each possible sensor and region, since (a) it’s based on a theoretical underpinning, and (b) it allows the same algorithm to be applied to different regions. I therefore think it’s important to at least evaluate that method and show that it’s not as good (if that’s true) before developing more empirical algorithms that may or may not be applicable to other regions 

Some specific comments are below:

Line 184: is there a reason the timespan isn’t given for State 3? It can be inferred from the other periods but it seems odd to leave it out.

Line 197: where

Figure 1: the text is difficult to read on the graphics (too small), and the image source, etc. should be included in the legend, not embedded into the graphic (so for example what sensor was used? It looks like the Brazilian sensor (WFI), when it would probably be more appropriate to use an L8 or S2 image to match the analysis. 

Line 215: three (not tree)

Line 231: where

Line 238: please specify where the SRFs were obtained, or which ones were used

Table 1: only S2A is listed, but the revisit time is given as 5 days. Does that mean you are using S2A and S2B interchangeably? Each satellite has an individual revisit time of 10 days.

Line 287: what package?

Section 2.4: it’s not clear to me how the Monte Carlo simulations were used. There are a finite number of observations, and they were split 70:30. So for the 70% of observations, how do you generate 10,000 Monte Carlo runs? That implies that you are selecting variables from a probability distribution function or similar, but from what I read, you only have simulated Rrs and matched TSS, etc. So what is being iterated?

Line 517: section (not session)

Reviewer 2 Report

Comments to the Author

Reviewer recommendation and comments for Manuscript ID: remotesensing-411198, entitled "Retrieving Total and Inorganic suspended sediments in Amazon Floodplain Lakes: A multisensor approach" for Remote Sensing.

General Comments:

This study presents a validation exercise for the different atmospheric correction methods for Landsat-8 and Sentinel-2 sensors. It is hard for me to find the contribution by the authors, i.e. (i) authors have validated the Remote Sensing Reflectance (Rrs) estimation methods only for the month of August, (ii) they have used the existing empirical methods for the estimation of water quality parameters and (iii) they claim to combine datasets from the Landsat-8 OLI and Sentinel-2 MSI sensors. I would like to clarify that (i) there has already been an extensive exercise by Doxani et al., (2018) for the validation of different atmospheric correction methods over different geographical location, around the world, so what are contributions by authors? (ii) have authors developed their own empirical method for the estimation of the water quality parameters? (iii) there has already been a Landsat-8 OLI and Sentinel-2 MSI combined data product named as Harmonized Landsat Sentinel (HLS) product (https://hls.gsfc.nasa.gov/), why authors have not used it? My further comments are as follows

Major Comments:

1.     Page 1 lines 34 and 35, the sun-glint correction was validated with the in situ measurements of sun-glint?

2.     Page 2 line 41, “MAPE < 21%” for both, OLI (561 nm) and MSI red band. Do authors think that a value of 21% of MAPE is a good value?

3.     Page 2 line 43, have you validated the virtual constellation of OLI and MSI, or authors are just extrapolating their results?

4.     Page 2 lines 43 to 45, I don’t agree with the statement of being the first study for calibration/validation of empirical algorithms for Amazon region. Please read the article by Montanher et al. (2014) and many others.

5.     Page 3 lines 100 to 106, such merged data from Landsat 8 and Sentinel-2 sensors already exist in the form of Harmonized Landsat Sentinel product.

6.     Page 4 lines 151 to 154, such exercise has already been conducted over different geographical location around the globe by Doxani et al, 2018.

7.     Page 8 line 235, “interpolated from 1 to 1 nm” could not understand? 1 to 1 nm?

8.     Page 9 lines 274 to 277, the usage of the satellite data from only the month of August makes the results very limited.

References Cited:

Doxani, G., Vermote, E., Roger, J. C., Gascon, F., Adriaensen, S., Frantz, D., ... & Louis, J. (2018). Atmospheric correction inter-comparison exercise. Remote Sensing, 10(2), 352.

Montanher, O. C., Novo, E. M., Barbosa, C. C., Rennó, C. D., & Silva, T. S. (2014). Empirical models for estimating the suspended sediment concentration in Amazonian white water rivers using Landsat 5/TM. International Journal of Applied Earth Observation and Geoinformation, 29, 67-77.