Using Saildrones to Validate Arctic Sea-Surface Salinity from the SMAP Satellite and from Ocean Models

Round 1

Reviewer 1 Report

The study by Vazquez-Cuervo et al. examines the accuracy of the satellite-derived SSS products using saildrone measurements along the coast of Alaska. The topic is interesting and important as the coastal areas, particularly in the Arctic, are difficult to evaluate because, on the one hand, there is lack of appropriate in-situ (ground truth) data, and, on the other hand, satellite SSS retrievals are prone to large errors and uncertainties due to land/ice contamination, cold water, etc. The paper is generally well-written, the results are solid and worth publishing in Remote Sensing. Yet, there are some issues with the clarity of the presentation, for which I request a minor revision.    

 

Specific comments

 

Line 39, put (SSS) after “sea-surface salinity”

 

Sec 2.1.1. Please indicate which maps are used – daily (8-day running mean), monthly?

 

Sec 2.1.3. Delete “LLC270” from the title as two versions of the model, LLC270 and LLC4320, are used.  

 

Sec 2.2. The description of the collocation method is unclear. E.g., “all Saildrone data was averaged within that grid cell…” – what is the grid cell in this case? All SMAP SSS products are on the same 0.25-deg grid. Why are the grid cells different then? Does the grid cell include time, too, e.g. 0.25-long x 0.25-lat x 8-day?

 

Sec 3 results, lines 150-154: An appropriate place for this paragraph would be somewhere in sec. 2.1.2, Saildrone measurements

 

Line 160, It is not clear how we can see freshening at “day 150” if we are still looking at Fig. 1 - time averaged maps

 

Line 169, The fact that JPL provides data closer to the coast does not necessarily indicate “the robustness of land correction algorithm…”. The correction can simply be wrong. Can this explain larger RMSD for the JPL product (Table 1)?

 

Figure 1. Figure caption. Please indicate the period of averaging. The white color for the saildrone track 1036 is probably not a good choice as the near coastal areas (missing values) are also shown by white (supposed to be black, line 197).

 

Figure 3, figure caption. Line 202, “by RBR sensor…”. Previously (line 152) we learned that it was decided to use the Seabird CTD data for all the comparisons”

 

Line 228. If it is still “the discussion with the reference to Table 1” (line 225), it is not clear how it can be seen that the primary differences are associated “with minimum values seen between days 160 and 170”.

 

Lines 229-230, The feature resolution is not that different between JPL and RSS70 products (60 km vs 70 km, sec 2.1.1), yet the bias and RMSD are very different (Table 1).

 

Line 246. What is the definition of “excellent” here? Again, there are significant differences between the products in terms of comparison to in situ data, e.g., the bias 0.42-0.66 (JPL) vs 0.0 (RSS).  

 

Line 259. “…and coherences were first calculated individually…” Coherences between what?

 

Line 280, “…the model shows a white spectrum…, while…” Fig 6c, black curve (saildrone) appears to show the same “white noise” tail.

 

Line 295, “… increasing again at 200 km and remaining statistically significant up through 50 km.” Does it mean that satellite maps, with the feature resolution claimed to be 70/60 km, realistically resolve scales smaller than L(wavelength)/4=50km?  Please clarify.

 

Figure 6a-c. Why the saildrone spectra (black curves) are different in panels a, b and c? This probably goes back to my previous question about the collocation method and grid-cell averaging.  

 

Line 308, “The decreased coherence could be due to the RSS70km being the smoothest of the three…” Yet, this product shows the highest coherence (approaching 0.8) for wavelengths 300-100 km. This could be related rather to the level of noise in the product, than simply the smoothness.  

 

Lines 327-330. This should be moved to the beginning of the section, or deleted.

 

Sec 4 “Discussion”. Placing Table 1 and the associated discussion here makes the paper read incoherently. I believe, the right place would be in sec 3.2 when it is first mentioned.

The same is about the spectral slopes. The right place would be at the end of sec 3.3.   

 

Line 370. Slopes over what range of wavelengths?

 

Line 372, RSS4km to RSS40km

 

Line 409, “onboard RBR” - Seabird CTD?

 

Line 430. “…excellent…” Again, there are significant differences between JPL and RSS product (table 1).

 

Line 446. “… and 0 psu biases are promising…” What about the JPL product which shows 0.4-0.6 psu biases? Is it not promising? If you exclude a specific interval between days 160-170 (line 227), would be the bias still significant for the JPL product?     

 

Author Response

Dear Reviewer:

Thank you for your thoughtful and detailed review. It is deeply appreciated as we know this can

be time consuming.

 

This is a good manuscript requiring minor revisions.

 

The study by Vazquez-Cuervo et al. examines the accuracy of the satellite-derived SSS products using saildrone measurements along the coast of Alaska. The topic is interesting and important as the coastal areas, particularly in the Arctic, are difficult to evaluate because, on the one hand, there is lack of appropriate in-situ (ground truth) data, and, on the other hand, satellite SSS retrievals are prone to large errors and uncertainties due to land/ice contamination, cold water, etc. The paper is generally well-written, the results are solid and worth publishing in Remote Sensing. Yet, there are some issues with the clarity of the presentation, for which I request a minor revision.    

 

Specific comments

 

Line 39, put (SSS) after “sea-surface salinity”

Done

 

Sec 2.1.1. Please indicate which maps are used – daily (8-day running mean), monthly?

 Done

Sec 2.1.3. Delete “LLC270” from the title as two versions of the model, LLC270 and LLC4320, are used.  

Done

 

Sec 2.2. The description of the collocation method is unclear. E.g., “all Saildrone data was averaged within that grid cell…” – what is the grid cell in this case? All SMAP SSS products are on the same 0.25-deg grid. Why are the grid cells different then? Does the grid cell include time, too, e.g. 0.25-long x 0.25-lat x 8-day?

 

A statement was added to explain the grid cell size being the same in the SMAP products. Thus the Saildrone matchups will be identical. But missing data, due to differences in flagging between the products, may still lead to differences in the retrieved matchups. Additionally, the models are of different spatial resolutions so will have different matchups with respect to the Saildrone data.

 

Sec 3 results, lines 150-154: An appropriate place for this paragraph would be somewhere in sec. 2.1.2, Saildrone measurements

The paragraph was moved to the section describing Saildrone.

 

Line 160, It is not clear how we can see freshening at “day 150” if we are still looking at Fig. 1 - time averaged maps

The statement was changed to Figure 3

 

Line 169, The fact that JPL provides data closer to the coast does not necessarily indicate “the robustness of land correction algorithm…”. The correction can simply be wrong. Can this explain larger RMSD for the JPL product (Table 1)?

A statement was added to clarify this point. The sentence was added that because the RMSD is larger, further work is necessary to determine the accuracy of SMAP products versus distance from land.

 

Figure 1. Figure caption. Please indicate the period of averaging. The white color for the saildrone track 1036 is probably not a good choice as the near coastal areas (missing values) are also shown by white (supposed to be black, line 197).

The statement was added that the mean was taken over the period of time of the Saildrone deployment.

A new figure one was inserted to replace SD1036 with black.

 

 

Figure 3, figure caption. Line 202, “by RBR sensor…”. Previously (line 152) we learned that it was decided to use the Seabird CTD data for all the comparisons”

Was a mistake. The correct statement should read that the RBR sensor was used. This has now been changed in the manuscript. Thank you.

 

Line 228. If it is still “the discussion with the reference to Table 1” (line 225), it is not clear how it can be seen that the primary differences are associated “with minimum values seen between days 160 and 170”.

 A statement was added to clarify the possible reasons for the large freshening event seen in the JPLSSS products versus the RSS product. The statement emphasized the necessity of future work to quantify the differences between the products versus distance to land.

Lines 229-230, The feature resolution is not that different between JPL and RSS70 products (60 km vs 70 km, sec 2.1.1), yet the bias and RMSD are very different (Table 1).

That points to the differences between the two products most likely due to the large freshening event seen between day 160 and day 170. This has also been clarified in the text.  We want to emphasize that these differences do require further analysis, inclusive of land and ice flagging and contamination. But we wanted to clarify that those issues are beyond the scope of this paper and need to be part of future research.

 

Line 246. What is the definition of “excellent” here? Again, there are significant differences between the products in terms of comparison to in situ data, e.g., the bias 0.42-0.66 (JPL) vs 0.0 (RSS).  

The sentence was rewritten as follows:

for the SMAP products, based on the correlations,  show good agreement with the Saildrone SSS. The JPLSSS product show larger biases of greater than 1 PSU.

 

Line 259. “…and coherences were first calculated individually…” Coherences between what?

 The following sentence was added to clarify the coherence derivation. Coherences were derived between each of the SMAP/model products and the Saildrone matchups.

Line 280, “…the model shows a white spectrum…, while…” Fig 6c, black curve (saildrone) appears to show the same “white noise” tail.

Thank you. I looked at the figure again, but it does appear to me that the Saildrone does not show the white noise or flattening of the slope, at least not the sam extent that the LLC4320 model does.

 

Line 295, “… increasing again at 200 km and remaining statistically significant up through 50 km.” Does it mean that satellite maps, with the feature resolution claimed to be 70/60 km, realistically resolve scales smaller than L(wavelength)/4=50km?  Please clarify.

This is likely due to the spectral averaging. Also the spectral plot being shown on log scale. But yes the interpretation indicates that energy does exist up to the feature resolution of the data set.

 

Figure 6a-c. Why the saildrone spectra (black curves) are different in panels a, b and c? This probably goes back to my previous question about the collocation method and grid-cell averaging.  

Overall the shape of the Saildrone spectra is similar in figures (a-c). The reason they are the same is due to the fact the three products have gaps in different locations of the time series. The spectra were derived after a linear interpolation was applied to the SMAP data before the calculation of the spectra. Thus identical spectra would not be expected as the matchups won;t match exactly.

 

Line 308, “The decreased coherence could be due to the RSS70km being the smoothest of the three…” Yet, this product shows the highest coherence (approaching 0.8) for wavelengths 300-100 km. This could be related rather to the level of noise in the product, than simply the smoothness.  

Thank you. The sentence has been rewritten.  Coherences peak at about 0.7 for JPLSSS and 0.8 for RSS70km. RSS40km peak lower at about 0.6. The decreased coherences could be due to the RSS70km being the smoothest of the three SMAP products and consistent with RSS40km containing the greater noise.

 

Lines 327-330. This should be moved to the beginning of the section, or deleted.

Done

 

Sec 4 “Discussion”. Placing Table 1 and the associated discussion here makes the paper read incoherently. I believe, the right place would be in sec 3.2 when it is first mentioned.

The same is about the spectral slopes. The right place would be at the end of sec 3.3.

Thank you to the reviewer for the thoughtful comment. After examining this further we decided it was probably best to leave table 1 and the discussion of  spectral slopes in the discussion section. A sentence in section 3.2 points to the discussion section for fruther analysis of the statistics.  We feel this makes for a better Discussion section having these results in the same place as the discussion about the spectral slopes. Removing these would eliminate the discussion section of the manuscript.  Thank you for the comments. 

 

Line 370. Slopes over what range of wavelengths?

The sentence was added to specify  that the slopes were derived over a range from 1000km to 100km.

 

Line 372, RSS4km to RSS40km

Done

 

Line 409, “onboard RBR” - Seabird CTD?

Done. This should be the RBR.Apologies as this was a mistake. The sensor used was the RBR.

 

Line 430. “…excellent…” Again, there are significant differences between JPL and RSS product (table 1).

Changed to “good” agreement

 

Line 446. “… and 0 psu biases are promising…” What about the JPL product which shows 0.4-0.6 psu biases? Is it not promising? If you exclude a specific interval between days 160-170 (line 227), would be the bias still significant for the JPL product?    

Following statement was added:  Biases were larger for the JPLSSS product and due to most likely to a large freshening event between days 160 and 170.

 

 

Submission Date

15 January 2021

Date of this review

27 Jan 2021 02:29:30

© 1996-2021 MDPI (Basel, Switzerland) unless otherwise stated

Disclaimer Terms and Conditions Privacy Policy

 

Author Response File: Author Response.docx

Reviewer 2 Report

This is a good manuscript requiring minor revisions.

My biggest comment/concern surrounds the freshening event near day 170.  From Fig 3, I am unable to see results from the RSS products.  Were retrievals obtained?  This has an influence on the statistics and discussion/conclusions.  This is a period right before a gap in data from the satellite – is there any connection?  Line 211 speaks to a “difference” in the data, but is there a difference if there is no data?  Line 229 also speaks of differences I am unable to see.  Line 250:  were RSS data available?

 

Line 102:  Decreasing light – I know what you mean, but I think this could be made a bit clearer for a reader less familiar

 

Line 104:  Undefined acronyms

 

Line 117:  Clarify here that the insufficient “horizontal resolution” corresponds to the climatology

 

Line 127:  In proceeding discussion it could be beneficial for the reader for you to explicitly clarify which products contain SSS

 

Line 135:  “of course” – this statement isn’t “obvious” to me not knowing the product.

 

Fig 1:  The graphics could be a bit larger in the final version to improve readability.  The caption refers to “black” missing data areas, but how does tis compare to the white regions and line 166 that says white is land contamination.

 

Fig 2:  Would the satellite graphic have looked significantly different if were for the period of the Saildrone passage only instead of averaged over the whole period?

 

Line 197/205:  Here and other places in the text, the text is interrupted with an inappropriate period and indentation when figures have been inserted in the middle of a paragraph.

 

Line 219:  Wasn’t the subject of this paragraph opening already discussed above?

 

Fig 5:  panels b and c are flipped in sense of caption and legends

 

Line 259:  perhaps could clarify that the Saildrone spectra are computed specifically for the collocated data to explain the differences between the spectra in the different comparisons.

 

Line 307:  I don’t see 0.4 as peak – values look higher to me in the figure.

 

Line 327:  This paragraph seems out of place – meant for the start of the subsection?

 

Line 343:  SD1038 -> 1037

 

Line 348/9 vis 270:  Here (later) it is said not adding noise, whereas on line 270 it implies more noise – please clarify for consistency.

 

Line 377:  the -2.27 appears to be -2.26 elsewhere

 

Line 377 vs 385:  What slope is really “best associated”?  There appears to be reference to different numbers

 

Line 451:  a conclusion is that retrievals should be improved, but what specific aspect did this paper imply needs to be improved?  This connection is seemingly missing to me and would be a very beneficial aspect to close the paper.

Author Response

Dear reviewer:

Thank you for your thoughtful and detailed review. It is deeply appreciated as we know this can

be time consuming.

 

This paper focus on the evaluation of SMAP satellite products and a couple of models to represent the SSS observed by 2 saildrones deployed several month in the Bering Sea, where the main SSS signal occurs off the Yukon River. It provides useful and elaborate statistics on the satellite/model/in situ comparison, following an approach already used by the authors in other regions (Vazquez-Cuervo et al, 2019). The results are significant as the high latitude and coastal regions are a critical zone for the satellite SSS products. Therefore I recommend publication, however I have two main concerns about this paper that I would like the authors to address, as well as a list of minor revisions to improve the clarity of presentation.

 

I understand that NASA funding incites to focus primarily on SMAP but science requires more objectivity and would benefit from a comparison between SMAP and SMOS. The authors obviously have all the tools in hand to apply the same analysis conducted with SMAP to SMOS. Moreover the paper title « SSS from SatelliteS » is misleading when only one satellite is really used. Therefore I strongly recommend to add SMOS products in the comparison with saildrone observations, specifically these two:

- The SMOS product that has been specially developed for the Arctic region (https://www.seanoe.org/data/00607/71909/, Supply et al., 2020), with specific correction for low SST and sea ice detection. It covers the northern 2/3 of the domain presented in Figure 1 and all the Y-K delta region presented in Figure 2.

- The SMOS global product with coastal bias correction to improve representation of river plumes (https://www.seanoe.org/data/00417/52804/#79565, Boutin et al., 2018), which would be very appropriate in the Y-K delta region.

We thank the reviewer for the recommendation about the SMOS data. It is appreciated as we realized we had not clarified the issue.  The intent of the manuscript was not co conclude which data set is best, but to present a representative study showing the utility of comparisons with Saildrone and the ability of satellite derived sea surface salinity to resolve coastal signals in the Arctic associated with river discharge. To avoid any confusion and replyging the comment we have specifically added SMAP to the title of the manuscript. One of the intentions of the manuscript was to also highlight possible future research topics. We will explicitly mention comparisons and analysis with improved SMAP products as well as SMOS. A sentence was added at line 502 of the manuscript. Thank you!

 

 

 The strongest freshening signal in the SMAP JPL product occurs at a time when the authors indicate that SMAP suffers from malfunction, and the fact that SMAP RSS product are not available at that time support this assessment, but this point must be clarified at is has potentially an heavy weight on the statistical differences between the 2 products. More details are given in the following minor comments.

Thank you. Sorry about the confusion about and description of the freshening at day 150. The freshening at day 150 did not occur when SMAP suffered from the malfunction. Additionally, that freshening is also seen in the Saildrone data. The freshening seen at day 170 is more difficult to assess as it is seen only in the JPLSSS product. However, this could be due to several factors, including the different land flagging in the JPL and RSS products,  as the data were quality controlled during  the malfunction, it is difficult to attribute the freshening to the SMAP malfunction.  A sentence has been added at line 230 that explicitly relates that day 150 occurs bedore the SMAP data gap happened. The text has now been rewritten to make this points more clear.

 

L41 where low salinity waters exits from

Thank you. That has now been changed to “exit”

L116 it would be better to put both resolutions in km for easier comparison.

That was now changed to 37km.

L135 increased

Done

L136 Given the region of study, it would be more appropriate to indicate the resolution in the Arctic rather than in the Antartica, which is presumably similar ?

The statement was changed to “high latitudes”.

L150-154 this paragraph would better fit in the 2.1.2 section than in the results section.

The paragraph was changed to section 2.1.2

L152 decided to use

Done

L160 « day 150 » has not been defined at this stage (from the beginning of the year, from the day of Saildrone deployment?), also no precise temporal information can be taken from Figure 1, which is a mean map over several months. I therefore suggest to suppress reference to this date here.

”of 2019 was added

L162 the freshening is minimal in model LLC270 but comparable in amplitude to satellite data for LLC4320 near the Yukon outflow. Please correct accordingly.

The sentence was rewritten: “The magnitude of the freshening depends on the SMAP product, with the LLC270 model showing little or no freshening, while the higher resolution LLC4320 shows similar freshening as the SMAP products.”

L166 On the opposite, the legend of Figure 1 states that land contamination is in black, please correct.

Done

L170 RSS40km has indeed data closer to the coast but it is noisy and not consistent with JPL so I suspect land contamination and would advice to use a similar land mask as RSS70km. Can you comment on that ?

A statement has been added at line 183  to clarify the point. Thank you for pointing this out to us as this is an issue that needs to be addressed in future research. The primary issue is quantifying the differences of the products versus distance from land.

L177 The SMOS sea ice correction methodology, presumably different, should be cited too (Supply et al, 2020).

We appreciate the comment, but want to maintain the focus of the paper on SMAP comparisons. However, we now mention SMOS comparisons as a future study. The intent of this paper was to give an overall assessment of the SMAP products when compared to Saildrone. We hope it opens up future research. The issue of quality of salinity measurements close to land needs to be one of those areas studies. To clarify this point we changed the title to reflect the focus on SMAP salinity.

L188 Then how does it come that, while SMAP RSS products are indeed not available on this period, but SMAP JPL is available on part of it ? This is problematic as the strongest freshening in SMAP JPL occurs at this time (L212) so can we believe it if SMAP is not functioning right ? And what is the point in comparing SMAS products here (L211) when only one is available ? Please clarify.

There are different flagging and quality control procedures that go into the processing of the SMAP products. Additionally, the products have different issues with respect to land correction. Thus it is not surprising that there is not a one to one correspondence. We wanted the focus to be on the comparisons with Saildrone and not each other. Saildrone has no gaps so the co-location methodology would match up with all available SMAP pixels.

L205 This is apparently the continuation of the sentence in L190, although the later seemed finished. Please correct.

Done

Figure 3 The first freshening events occur simultaneously in the two saildrone time series, at the location of the southernmost diamond in Figure 1 where both saildrones sail through, so I guess they were there at the same time, but it is never said clearly. On the contrary, while both saildrones went through the northermost diamond as shown in Figure 1, only SD1037 records the freshening. Were they here at the same time, maybe because one is closer to the coast, which may explain the difference, but again it is never said clearly. Can you add this important information in section 2.1.2 so that the reader does not necessarily has to refer to the full cruise report ? Also can you relate this second coastal freshening to a particular river runoff ? And can you shift your time axis to month and days, and use this format everywhere in the text rather than day X ?

We added a statement that “Both deployments were launched from Dutch Harbor at approximately the same time, but did not necessarily follow identical tracks. This is clearly seen in Figure 1 where the grey line indicates the 1037 deployment and the white indicates the 1036 deployment. Thus, it is not expected that both deployments will have identical values” inserted at Line 104. We have added a statement that the second coastal freshening could be due to the JPLSSS product resolving features closer to land. Thank you for the suggestion about switching to month and day for the x-axis. We probably will stick the “day of year label”. Because the deployments were over a time range of less than 200 days, we feel the year day labeling is better for identifying  the major freshening events.

L222 The RSS40km and RSS70km products can obviously not reproduce the freshening observed by the JPL produt as they are not available at that time. This may be what you mean by « contain » but it is ambiguous, please clarify.

Apologies for the confusion. The following sentences were added: “The RSS40km and RSS70km products do not contain the freshening seen at approximately day 170. This could be due to several factors, including differences in the land mask and  in the application of quality flags. This requires future research. “ The RSS products do not contain data for day 170. This has been explained in the text.

 

L225 I would not reference Table 1 here as it is only analysed in the disc

Thank you for the suggestion! I would prefer to keep the following statement:

“A more quantitative discussion of the agreement between the products is undertaken in the discussion with reference to Table 1.”  My fear is that many of those reading the manuscript might not read the whole article, thus missing the summary of all the statistics. The rationale for adding that sentence is to make the reader aware sooner of the statistics in Table1, and thus encourage reading the manuscript further.

L236 which clearly do not

Done

Figures 5 & 6, 7 & 8 You should merge the satellite and model analyses in a single figure as you did previously.

Thank you very much for your suggestion. One reason we decided to keep the figures separate for the spectra and coherence is that they are derived over a different range for observations and model. Additionally, because of the differences in location of gaps between the SMAP products, this will also cause the Saildrone spectra to be different. Thus combining them all in one figure would make it difficult to clearly identify the different spectra and easily identify the differences.

L270 This is probably due to the coastal noise in RSS40 that could be removed by extending the land mask (that would be a good sensitivity test), as I previously mentioned. This should be discussed in part 4, does it fully support the statement of L349 ?

The following sentence has been added to the discussion to clarify the point. “Based on the spectra, the RSS40km product has increased noise at higher wavelengths. This would be consistent with values closer to the coast and increased land contamination. In terms of extending the land mask that is beyond the scope of the paper. The SMAP data is provided by Remote Sensing Systems  and JPL. The data is directly downloaded from the Physical Oceanography Distributed Active Archive Center (PO.DAAC).

L300 Move this analysis to the text and stick to factual in the legend. Just indicate that the right end of the spectrum corresonds to the product resolution.

Done

L295 & 303 The sentence is split between 2 paragraphs.

Done

L304 Rather 50 and 70 km, there is a minimum at 60 km in Figure 7a.

Thank you. Was changed to reflect minima at 60km.

L307 Coherence peaks at 0,8 around 200 km in Figure 7b, and it is overall not much lower than for JPLSSS. Please correct.

This was corrected.

L327-330 This paragraph should be moved farther in the discussion.

The paragraph was completely removed as it was redundant with the same words repeated in the first paragraph of the Discussion.

Table 1 This table is first referenced in L225, which is disturbing, made me try to relate the following lines to numbers presented in the table, and made me think that you could insert these numbers in the corner of each panel in figure 4 for more synthetic reading. I later found it is properly discussed later, so it should be referenced in the discussion only, and I suggest to replace the table with a plot bar for more intuitive reading, as the main statistical numbers are repeated in the text.

We do refer to the table (line 249) once earlier. The statement referring to the table is included earlier to motivate the reader to continue reading the manuscript.

L366 Could you materialise these slopes in Figure 6 ?

Thank you for the suggestion. We feel it is better to keep Figure 6 as is. The rationale is to make them easy to compare. Do not feel that adding two lines with the given slopes would add much as the slopes basically just fit a line to the spectra which are already close to being linear.

Table 2 Again, I suggest to replace the table with a graphical representation for more intuitive reading, as the main statistical numbers are repeated in the text.

Thank you for your comment. The goal of the paper was to present the statistics in as clear and concise way as possible. We wanted to not present the results in any way that would indicate the best data set. That is a major reason we presented the results in a table format rather than graph. Additionally, the table presents the actual numbers, which for comparison sake we felt it is important. As there are also only 5 data sets, a table format would not be difficult to interpret and compare the correlations, biases, RMSD simultaneously,

L446 Negligible rather than « 0 PSU », it’s all relative.

Changed to negligible

 

 

 

 

 

Submission Date

15 January 2021

Date of this review

29 Jan 2021 00:18:41

© 1996-2021 MDPI (Basel, Switzerland) unless otherwise stated

Disclaimer Terms and Conditions Privacy Policy

 

Author Response File: Author Response.docx

Reviewer 3 Report

This paper focus on the evaluation of SMAP satellite products and a couple of models to represent the SSS observed by 2 saildrones deployed several month in the Bering Sea, where the main SSS signal occurs off the Yukon River. It provides useful and elaborate statistics on the satellite/model/in situ comparison, following an approach already used by the authors in other regions (Vazquez-Cuervo et al, 2019). The results are significant as the high latitude and coastal regions are a critical zone for the satellite SSS products. Therefore I recommend publication, however I have two main concerns about this paper that I would like the authors to address, as well as a list of minor revisions to improve the clarity of presentation.

 

I understand that NASA funding incites to focus primarily on SMAP but science requires more objectivity and would benefit from a comparison between SMAP and SMOS. The authors obviously have all the tools in hand to apply the same analysis conducted with SMAP to SMOS. Moreover the paper title « SSS from SatelliteS » is misleading when only one satellite is really used. Therefore I strongly recommend to add SMOS products in the comparison with saildrone observations, specifically these two:

- The SMOS product that has been specially developed for the Arctic region (https://www.seanoe.org/data/00607/71909/, Supply et al., 2020), with specific correction for low SST and sea ice detection. It covers the northern 2/3 of the domain presented in Figure 1 and all the Y-K delta region presented in Figure 2.

- The SMOS global product with coastal bias correction to improve representation of river plumes (https://www.seanoe.org/data/00417/52804/#79565, Boutin et al., 2018), which would be very appropriate in the Y-K delta region.

 

The strongest freshening signal in the SMAP JPL product occurs at a time when the authors indicate that SMAP suffers from malfunction, and the fact that SMAP RSS product are not available at that time support this assessment, but this point must be clarified at is has potentially an heavy weight on the statistical differences between the 2 products. More details are given in the following minor comments.

 

L41 where low salinity waters exits from

L116 it would be better to put both resolutions in km for easier comparison

L135 increased

L136 Given the region of study, it would be more appropriate to indicate the resolution in the Arctic rather than in the Antartica, which is presumably similar ?

L150-154 this paragraph would better fit in the 2.1.2 section than in the results section.

L152 decided to use

L160 « day 150 » has not been defined at this stage (from the beginning of the year, from the day of Saildrone deployment?), also no precise temporal information can be taken from Figure 1, which is a mean map over several months. I therefore suggest to suppress reference to this date here.

L162 the freshening is minimal in model LLC270 but comparable in amplitude to satellite data for LLC4320 near the Yukon outflow. Please correct accordingly.

L166 On the opposite, the legend of Figure 1 states that land contamination is in black, please correct.

L170 RSS40km has indeed data closer to the coast but it is noisy and not consistent with JPL so I suspect land contamination and would advice to use a similar land mask as RSS70km. Can you comment on that ?

L177 The SMOS sea ice correction methodology, presumably different, should be cited too (Supply et al, 2020)

L188 Then how does it come that, while SMAP RSS products are indeed not available on this period, but SMAP JPL is available on part of it ? This is problematic as the strongest freshening in SMAP JPL occurs at this time (L212) so can we believe it if SMAP is not functioning right ? And what is the point in comparing SMAS products here (L211) when only one is available ? Please clarify.

L205 This is apparently the continuation of the sentence in L190, although the later seemed finished. Please correct.

Figure 3 The first freshening events occur simultaneously in the two saildrone time series, at the location of the southernmost diamond in Figure 1 where both saildrones sail through, so I guess they were there at the same time, but it is never said clearly. On the contrary, while both saildrones went through the northermost diamond as shown in Figure 1, only SD1037 records the freshening. Were they here at the same time, maybe because one is closer to the coast, which may explain the difference, but again it is never said clearly. Can you add this important information in section 2.1.2 so that the reader does not necessarily has to refer to the full cruise report ? Also can you relate this second coastal freshening to a particular river runoff ? And can you shift your time axis to month and days, and use this format everywhere in the text rather than day X ?

L222 The RSS40km and RSS70km products can obviously not reproduce the freshening observed by the JPL produt as they are not available at that time. This may be what you mean by « contain » but it is ambiguous, please clarify.

L225 I would not reference Table 1 here as it is only analused in the disc

L236 which clearly do not

Figures 5 & 6, 7 & 8 You should merge the satellite and model analyses in a single figure as you did previously.

L270 This is probably due to the coastal noise in RSS40 that could be removed by extending the land mask (that would be a good sensitivity test), as I previously mentioned. This should be discussed in part 4, does it fully support the statement of L349 ?

L300 Move this analysis to the text and stick to factual in the legend. Just indicate that the right end of the spectrum corresonds to the product resolution.

L295 & 303 The sentence is split between 2 paragraphs.

L304 Rather 50 and 70 km, there is a minimum at 60 km in Figure 7a.

L307 Coherence peaks at 0,8 around 200 km in Figure 7b, and it is overall not much lower than for JPLSSS. Please correct.

L327-330 This paragraph should be moved farther in the discussion.

Table 1 This table is first referenced in L225, which is disturbing, made me try to relate the following lines to numbers presented in the table, and made me think that you could insert these numbers in the corner of each panel in figure 4 for more synthetic reading. I later found it is properly discussed later, so it should be referenced in the discussion only, and I suggest to replace the table with a plot bar for more intuitive reading, as the main statistical numbers are repeated in the text.

L366 Could you materialise these slopes in Figure 6 ?

Table 2 Again, I suggest to replace the table with a graphical representation for more intuitive reading, as the main statistical numbers are repeated in the text.

L446 Negligible rather than « 0 PSU », it’s all relative.

 

 

 

 

Author Response

Dear reviewer:

Thank you for your thoughtful and detailed review. It is deeply appreciated as we know this can

be time consuming.

This is a good manuscript requiring minor revisions.

Thank you

My biggest comment/concern surrounds the freshening event near day 170.  From Fig 3, I am unable to see results from the RSS products.  Were retrievals obtained?  This has an influence on the statistics and discussion/conclusions.  This is a period right before a gap in data from the satellite – is there any connection?  Line 211 speaks to a “difference” in the data, but is there a difference if there is no data?  Line 229 also speaks of differences I am unable to see.  Line 250:  were RSS data is available?

Thank you for comments. Yes there are no retrievals

 

The following line was added in the text. “ The freshening is not seen in the RSS products. This also occurs before SMAP experienced a data gap.” The two products have different quality mechanisms as well as flagging for land contamination. This is now explicitly stated in the text. A key point is that there was Saildrone data, thus still making it possible to compare directly with Saildrone.

 

At approximately  Line 249 the following statement was also added. “The RSS40km and RSS70km products do not contain the freshening seen at approximately day 170. This could be due to several factors, including differences in the land mask and  in the application of quality flags. This requires future research as both the RSS40km and RSS70km products do not contain data on those dates”.

“

 

 

Line 102:  Decreasing light – I know what you mean, but I think this could be made a bit clearer for a reader less familiar.

The following was explicitly added in parentheses in the same sentence:

(Saildrones are solar powered)

 

 

Line 104:  Undefined acronyms

Conducitivity, Temperature, Depth was added to define the acronym CTD. RBR and Seabird are the two names of the companies.

 

Line 117:  Clarify here that the insufficient “horizontal resolution” corresponds to the climatology

The following was added to the sentence: “of the climatology”

 

Line 127:  In proceeding discussion it could be beneficial for the reader for you to explicitly clarify which products contain SSS

Thank you for the recommendation. The following was added, “All these products measure SSS.

“

 

Line 135:  “of course” – this statement isn’t “obvious” to me not knowing the product.

 Obvious was removed

Fig 1:  The graphics could be a bit larger in the final version to improve readability.  The caption refers to “black” missing data areas, but how does tis compare to the white regions and line 166 that says white is land contamination.

Thsnk you!! Apologies as that was leftover from a previous version. “Black” was a mistake and changed to “white”

 

Fig 2:  Would the satellite graphic have looked significantly different if were for the period of the Saildrone passage only instead of averaged over the whole period?

Apologies. Thank you for pointing this out. We have now explicitly added that the map was created as the mean over the time period of the Saildrone deployment. This was added to the figure caption. In terms of whether it would have looked different, the answer is probably yes. The main reason is that the Y-K delta discharge is most likely seasonal after snow and ice melt. Averaging over a longer period of time would most likely have smoothed out the signal.

 

Line 197/205:  Here and other places in the text, the text is interrupted with an inappropriate period and indentation when figures have been inserted in the middle of a paragraph.

Thank you so much for pointing this out. We will correct for the next version to be sent.

 

Line 219:  Wasn’t the subject of this paragraph opening already discussed above?

 This has now been fixed. that paragraph was removed.

Fig 5:  panels b and c are flipped in sense of caption and legends

Thank you. The figure caption was mistyped and is now fixed to accurately reflect the figure.

 

Line 259:  perhaps could clarify that the Saildrone spectra are computed specifically for the collocated data to explain the differences between the spectra in the different comparisons.

 Thank you! This has now been added.

Line 307:  I don’t see 0.4 as peak – values look higher to me in the figure.

Apologies and thank you. This has now been rewritten to accurately reflect the peak.

 

Line 327:  This paragraph seems out of place – meant for the start of the subsection?

 Thank you. That paragraph was removed and was replaced with a updated paragraph in the beginning of section 3.3

Line 343:  SD1038 -> 1037

Done

 

Line 348/9 vis 270:  Here (later) it is said not adding noise, whereas on line 270 it implies more noise – please clarify for consistency.

The following statement was added in the Discussion  to be more explicit about the noise. Based on the spectra, the RSS40km product has increased noise at higher wavelengths. This would be consistent with values closer to the coast and increased land contamination.

 

Line 377:  the -2.27 appears to be -2.26 elsewhere

Thank you. Was changed to -2.26

 

Line 377 vs 385:  What slope is really “best associated”?  There appears to be reference to different numbers

Thank you. We did not want to identify one slope as being the best. The reference articles that determine spectral slopes of -2 approximate slopes associated with the mesoscale-submessocale variability.

 

Line 451:  a conclusion is that retrievals should be improved, but what specific aspect did this paper imply needs to be improved?  This connection is seemingly missing to me and would be a very beneficial aspect to close the paper.

 

The following statement was added at the end of the Conclusions and Summary. “ Future work needs to explore further differences in the SMAP products, inclusive of the large freshening even observed in the JPLSSS at day 170;  This should include future exploration of differences in SSS close to land and possible issues with the SMAP instrument malfunction.

“

 

Author Response File: Author Response.docx

Round 2

Reviewer 3 Report

Reading the new version of these paper and responses to my comments, I find that the authors only did a half-hearted effort in addressing my comments. Apart from the problem with data gaps in the SMAS RSS product (that was also raised by the other reviewer), only the easiest of my suggested corrections have been took into account, and all my suggestions for figures improvement have been rejected, often based on argumentation that I do not find very solid. I still think the paper should be published, but suggestions must be took into account more seriously. Reviewing was not helped by the line numbering that is not fully readable.

 

I appreciate that the problem of data gap in SMAS RSS product around the second freshening is now discussed, although I still have a few comments regarding this point :

- L232 : How can the author see this freshening event at day 175 (170+5) in the RSS SMAP products although they are not available at this time ?

- The authors did not care to address my recommandation to relate the second coastal freshening to a particular runoff so I did the job for them : it is probably attributable to outflow from the Kasegaluk lagoon, which receives waters from 3 rivers and communicate with the ocean through several passes (see reference https://www.jstor.org/stable/40511357). This should be added to provide physical interpretation.

- L258 : I note he authors that there is a second data gap in the SMAP RSS products around days 110-120, when SMAP JPL see another strong freshening. This event, if it is coastal, would strengthen the argument on land mask/coastal flagging. I was interested to check where this freshening occured, but unfortunately there is not enough information on the saildrone deployment given in the paper to know that, and I found that the report [11] referenced for this cruise in fact corresponds to another deployment off California. Please correct the reference and discuss this data gap and freshening to exploit more fully the set of data you have.

 

I insist on switching the « day of year » label to month and day for the x-axis for figure 3 (and in the text accordingly), as it is the usual date convention, much more intuitive to get a sense of seasonality, and the authors’ argument that the deployments were over a time range of less than 200 days does not hold up as they used a month/day label in their previous 2019 paper, also comparing satellite and saildrone data over a shorter 2-month period.

 

I do not understand the justification for merging figures 5 & 6 / 7 & 8 that they are derived over a different range for observations and model. Obs and model are both collocated along the same saildrone trajectories and the axes ranges for plots are the same (not exactly, actually, for the wavelength axis of figures 5 and 6, which I recommend to homogenize). The authors also want to keep the figures separated to easily identify the differences but, on the contrary, I think the differences between plots are more obvious when they are compared in the same figure. They could also reuse the differents colors attributed in Figure 4 for each product in figures 5/6 and 7/8 for easier identification of products.

 

Other minor comments :

 

Please correct title : from the SMAP satellite and from Ocean Models

 

L41 where low salinity waters exits from (rather than « is fluxed from »)

L188 Not sure if you compare separately the RMSD between RSS products, and between JPL products, or between all satellite products, and which criteria you use for estimating the significance of differences in RMSD. From 0.73 to 1.34, there is a 83 % increase in RMSD that , I guess, is significant. Moreover you try to explain these differences later (L258), which makes sense if they are significant only. Please precise.

 

 

Author Response

Reading the new version of these paper and responses to my comments, I find that the authors only did a half-hearted effort in addressing my comments. Apart from the problem with data gaps in the SMAS RSS product (that was also raised by the other reviewer), only the easiest of my suggested corrections have been took into account, and all my suggestions for figures improvement have been rejected, often based on argumentation that I do not find very solid. I still think the paper should be published, but suggestions must be took into account more seriously. Reviewing was not helped by the line numbering that is not fully readable.

We apologize and do thank the reviewer for the comments. There was no intent to do a halfhearted effort, but simply to honestly address the issues in a way to improve the manuscript. The issue of the figures is addressed below in a way we hope is satisfactory.

I appreciate that the problem of data gap in SMAS RSS product around the second freshening is now discussed, although I still have a few comments regarding this point :

- L232 : How can the author see this freshening event at day 175 (170+5) in the RSS SMAP products although they are not available at this time ?

The statement at line 232 is:

This freshening is not seen in the RSS products. This also occurs before SMAP experienced a “data gap”. The freshening is not seen in the RSS products as there is no data for that time period.  As we mention the difference in the two products I most likely due to different quality control procedures. We clearly state this requires more research beyond the scope of this paper.  To clarify we have changed the statement to:

This freshening is not seen in the RSS products due to no data. This also occurs before SMAP experienced a data gap.

- The authors did not care to address my recommandation to relate the second coastal freshening to a particular runoff so I did the job for them : it is probably attributable to outflow from the Kasegaluk lagoon, which receives waters from 3 rivers and communicate with the ocean through several passes (see reference https://www.jstor.org/stable/40511357). This should be added to provide physical interpretation.

Thank you very much. We have added that statement as a possible explanation for the freshening event seen at day 170.

- L258 : I note he authors that there is a second data gap in the SMAP RSS products around days 110-120, when SMAP JPL see another strong freshening. This event, if it is coastal, would strengthen the argument on land mask/coastal flagging. I was interested to check where this freshening occured, but unfortunately there is not enough information on the saildrone deployment given in the paper to know that, and I found that the report [11] referenced for this cruise in fact corresponds to another deployment off California. Please correct the reference and discuss this data gap and freshening to exploit more fully the set of data you have.

 

Thank you for pointing this out. I assume the reviewer is referring to day 210-220 not 110-120?

We completely agree. As that event does not appear in the Saildrone data. Another possible explanation (as the deployments at those times are further North) is possible issue with ice. This would certainly require future work, as the ice flagging is another area of major research.

The wording at line 113 has been modified. Reference 11 is not a cruise report, but journal article in BAMS describing the Saildrone program and characteristics. Validation statistics are also reported for the California deployment. The following statements were added:

The data for this cruise is distributed through the PO. DAAC (https://podaac.jpl.nasa.gov/dataset/SAILDRONE_ARCTIC?ids=&values=&search=Saildrone). Details and characteristics of the Arctic deployment may be found at the above. For more details on the Saildrone vehicle, and validation results, please see [11].

I insist on switching the « day of year » label to month and day for the x-axis for figure 3 (and in the text accordingly), as it is the usual date convention, much more intuitive to get a sense of seasonality, and the authors’ argument that the deployments were over a time range of less than 200 days does not hold up as they used a month/day label in their previous 2019 paper, also comparing satellite and saildrone data over a shorter 2-month period.

Figure 3 has now been replaced with the month and day added. Please see below for the redone figure.

 

 

I do not understand the justification for merging figures 5 & 6 / 7 & 8 that they are derived over a different range for observations and model. Obs and model are both collocated along the same saildrone trajectories and the axes ranges for plots are the same (not exactly, actually, for the wavelength axis of figures 5 and 6, which I recommend to homogenize). The authors also want to keep the figures separated to easily identify the differences but, on the contrary, I think the differences between plots are more obvious when they are compared in the same figure. They could also reuse the differents colors attributed in Figure 4 for each product in figures 5/6 and 7/8 for easier identification of products.

We have now taken the reviewer’s advice and updated the figures. For the spectra we have combined all the SMAP/Saildrone spectra in one Figure 5. We combined the model spectra in Figure 6. We have kept these separated for the following reason. The Saildrone spectra for SMAP colocations close to the being the same due to SMAP products being gridded on the same 0l25 degree grid. However, the two models use different grids and Saildrone spectra would be different for the two models. Thus, to keep things clean and easily comparable with the correct Saildrone spectra we maintained the two figures, one for SMAP products and one for the models. The new figure is below. The coherence plots we would like to keep the same because combining all of them on one figure would make it difficult to distinguish the error bars associated with each spectrum. An example of the problem is shown in the coherence figure below.

The images are in the attached document.

We do thank the reviewer for the suggestion as we feel the combined spectra do facilitate greatly the comparison.

 

 

Other minor comments :

 

Please correct title : from the SMAP satellite and from Ocean Models

Done. Thank you.

 

L41 where low salinity waters exits from (rather than « is fluxed from »)

Done. Thank you.

L188 Not sure if you compare separately the RMSD between RSS products, and between JPL products, or between all satellite products, and which criteria you use for estimating the significance of differences in RMSD. From 0.73 to 1.34, there is a 83 % increase in RMSD that , I guess, is significant. Moreover you try to explain these differences later (L258), which makes sense if they are significant only. Please precise.

 

Thank you. At line 192 we have now added the following sentence:

RMSD differences between the RSS products and JPL vary between approximately 1.2 PSU and 0.7 PSU

 

Author Response File: Author Response.docx