Review Reports - Comparison of Satellite-Derived Sea Surface Temperature and Sea Surface Salinity Gradients Using the Saildrone California/Baja and North Atlantic Gulf Stream Deployments

Round 1

Reviewer 1 Report

The manuscript investigates the consistency between SST/SSS gradients from in-situ data collected during two Saildrone campaigns and satellite products, and compares also such consistency with that of the SST/SSS full fields. I found the manuscript well written and interesting, and I recommend its publication after adding a discussion/analysis (see below) and addressing a couple of very minor issues.

The paper could do a better job in understanding why there exists disagreement between the Saildrone derived and satellite derived SST gradients. As the authors themselves state (L115-117), the L4 products include some form of statistical mapping (optimal interp., variational scheme or others, depending on the product), which imply smoothing through horizontal correlations. The characteristics of the horizontal correlation scales will depend on the specific product, but this means for all products that the resolution of the L4 SST signal (e.g. >= 50 km) is much coarser than nominal resolution of the L4 grid (0.1-0.2°). The authors should be able to verify this point, for instance by increasing the “d_s” length parameter used in the colocation procedure, and checking if the consistency between the two types of data increases (e.g. showing also a p.d.f. of the gradients for a certain L4 product in comparison with Saildrone). I guess this could not be done for the entire Saildrone dataset due to limited spatial coverage but for some points, and using as illustration just one L4 product. Alternatively, low-pass filtering the Saildrone derived gradient track with cutoff spatial scale consistent with that of the L4 product (see their references) could be done.

Minor points

1) In abstract and Intro, it is said drifting/moored buoys and Argo cannot easily be used for the verification of satellite SST/SSS gradients. I dont fully agree for drifters. In principle, when they are equipped with SST and/or conductivity sensors and have, as usual, a 10/15 minutes sampling rate, they could be used as well. The authors also mention the high sampling rate of Saildrone as and advantage for sub-mesoscale variability studies, but then L4 data are basically daily, so the reference to the former seems misleading to me.

2) Please include statistical significance tests at least for the correlations (e.g. at 95% confidence level via bootstrapping or t-distribution). In the two tables, significant correlations may be written with bold font. It would be nice adding this values also to Figure 7 (e.g. thick line corresponding to the minimum significance correlation), and performing the significance test also on the RMSE (squared departures behave as Normal pdf, and the 0 mean hypothesis can be tested). Although values of correlations are relatively low (0.3-0.4) they are probably still statistically significant, for most pairs of data.

L27 “lower” → lower signal-to-noise ratios than those on SST (?)

Author Response

Thank you for your thoughtful and positive comments.

Thank you to the reviewers for the comments. In terms of “nominal resolution” of L4 products, this depends on the product itself, as can be seen in the supplemental files (gradient time series). One cannot assume that the grid resolution (as you mentioned) equals the feature resolution. Although the MUR product has a grid resolution of 1 km, previous work has shown that the feature resolution is around 10 km. Of course, as with all the products, this will depend on the availability of the infrared data. Thus, the situation is more complex as the feature resolution is also not constant in time. One of the points of the manuscript is to determine how well the different L4 GHRSST data sets represent the scales associated with the actual ocean dynamics (as derived from Saildrone), we wanted to work with the grid instead of feature resolution.

While the ds parameter could be increased for a better match between the “feature resolution” captured by L4 and Saildrone, this may not be an interesting comparison from a user perspective. Most of the user community is still unaware of this “feature vs. grid” resolution difference, and the selection of L4 products is based solely on the grid resolution. We have also found that the overall statistics were robust with respect to the application of a different co-location methodology.

Minor points

In abstract and Intro, it is said drifting/moored buoys and Argo cannot easily be used for the verification of satellite SST/SSS gradients. I dont fully agree for drifters. In principle, when they are equipped with SST and/or conductivity sensors and have, as usual, a 10/15 minutes sampling rate, they could be used as well. The authors also mention the high sampling rate of Saildrone as and advantage for sub-mesoscale variability studies, but then L4 data are basically daily, so the reference to the former seems misleading to me.

Thank you for pointing this out. In fact, drifting buoys cannot be used for gradients, that has been removed from the manuscript. In terms of the reference to the high sampling rate, we wanted to focus on deriving statistics representing how well the L4 products, with their inherent smoothing and temporal resolution, represented the gradients.

Please include statistical significance tests at least for the correlations (e.g. at 95% confidence level via bootstrapping or t-distribution). In the two tables, significant correlations may be written with bold font. It would be nice adding this values also to Figure 7 (e.g. thick line corresponding to the minimum significance correlation), and performing the significance test also on the RMSE (squared departures behave as Normal pdf, and the 0 mean hypothesis can be tested). Although values of correlations are relatively low (0.3-0.4) they are probably still statistically significant, for most pairs of data.

We clarified the statistical significance of the correlations in the manuscript.

Correlation of SST gradients is statistically significant at the 95% confidence level, except for K10 in the Gulf Stream region (p-value > 0.2). For both JPLSMAP and RSS40km, the statistical correlation of SSS gradients for both campaigns was not statistically significant.

L27 “lower” → lower signal-to-noise ratios than those on SST (?)

Corrected

Submission Date

20 April 2020

Date of this review

10 May 2020 19:16:23

Bottom of Form

Reviewer 2 Report

Dear authors:

Lines 26-28: Must include the reference about the effect in the accuracy evidence the land contamination into the North Pacific/Atlantic, it will help us understand the importance of the SSS validation and future application in coastal areas.

Lines 117-121: The supplementary material (animation maps) showed more clearly information temporal evolution SST and SSS gradient. Could be recommended mention the supplementary material.

Line 187-196: I strongly recommended a rewrite of the key idea in that conclusion, because of that lines discuss the main results. Reconsider to put that lines in the results.

Line 222: Include the year in the publication.

Author Response

Dear authors:

The following reference was added:

Meissner, T.; Wentz, F.J.; Le Vine, D.M. The Salinity Retrieval Algorithms for the NASA Aquarius Version 5 and SMAP Version. Remote Sensing 2018, 10.

Lines 117-121: The supplementary material (animation maps) showed more clearly information temporal evolution SST and SSS gradient. Could be recommended mention the supplementary material.

Thank you. We have mentioned the animations in the conclusion section.

Line 187-196: I strongly recommended a rewrite of the key idea in that conclusion, because of that lines discuss the main results. Reconsider to put that lines in the results.

Thanks. We have removed these ideas from the conclusion and included them in the results section.

Line 222: Include the year in the publication.

As of the writing of the manuscript, the article had not yet been officially published. We have now updated it to 2020.

Submission Date

20 April 2020

Date of this review

29 Apr 2020 14:45:47

Bottom of Form

Author Response File: Author Response.doc

Reviewer 3 Report

The manuscript titled “Validation of satellite-derived sea surface temperature and sea surface salinity gradients: Comparisons with the Saildrone California/Baja and North Atlantic Gulf Stream deployments” is very interesting work.

However, work is not fully valorized because too many “issues” mentioned in the previous papers are taken for granted (e.g., the characteristics of the Saildrone instrument, data and the areas of study). The authors should rewrite the paper to be self-contained.

In additional, it would be very interesting to better investigate the “meaning” of the differences between "Statistics of SST/SSS" (i.e., Bias, RMSE, and Correlations) and the ones of SST/SSS gradients. A discussion section could help to better understand these differences and the meaning of every table and figure.

Moreover, please add the meaning of the gradient, and check the values “4 locations” of “the Satellite/Saildrone collocated observations” (pag.3). The authors should read the article carefully because there are some typing errors (e.g., line 32).

Author Response

Thank you to the reviewer for their comments. We have added some sentences and discussion that hopefully clarify the issues alluded to by the reviewer. Characteristics of the Saildrone instrument have been defined in other papers, including a recent publication in BAMS (cited in the paper) as well as the previous paper published by the authors using the same data from the Baja deployment. The following statement was added to summarize the results in the Gentemann et al. (2020) paper.

“On Saildrone, the SST and SSS Conductivity Temperature Depth Profilers (CTD), are only two of the multiple instruments onboard. Other sensors include a fluorometer, as well as an Acoustic Doppler Current Profiler (ADCP). For a complete description of the Saildrone instrumentation and known accuracies, please see [3]. “

We cite both of those papers to avoid having to go over previously published results. The paper [1] focused on the validation of the actual GHRSST L4 products as well as SMAP salinity in the coastal region of California and Baja. This paper takes things a step further by comparing gradients while adding the critical Saildrone deployment in the Gulf Stream, representing a Western Boundary Region.

of every table and figure.

We want to clarify that our paper does not aim to explain why gradients from Saildrone differ from those in Level 4 products. Instead, it aims to provide the user and producer communities a new methodology to compare satellite-based gradients at Level 4 with those captured by Sailrone.

An in-depth investigation of these differences is beyond the scope of this paper and would require an analysis of SST/SSS products at all levels (2, 3, and 4). Several factors contribute to these differences, even at Level 2. For SST, for example, retrieval algorithm, the accuracy of cloud masking, sensor noise are all factors that could lead to gradients at Level 2 being quite different from one product to the other.

Once at Level 4, several Level 2 datasets from microwave and infrared radiometers are ingested and smoothed in space and time, with smoothing scales that differ from one satellite product to another. In our opinion, results reported in Tables 1 and 2 contain clear information (so far unknown to most of the user community) that validation statistics are good for SST/SSS but not for corresponding gradients.

A full description of the meaning of the differences between the SST and SSS gradients is left to future research and investigation.

Thanks for pointing this out.

The last observation for the collocated observation (i,j-1) has be modified to (i, j+1).

Author Response File: Author Response.doc

Reviewer 4 Report

Summary

SST and SSS data were collected during two Saildrone campaigns, one in the California Baja region and one in the North Atlantic Gulf Stream, and were compared to several satellite-derived SST and SSS data products. Gradients in SST and SSS are useful for identifying important oceanographic features such as fronts, and the high temporal and spatial resolution of the Saildrone measurements can resolve these features. The gradients in SST and SSS were compared between the two types of sensor platforms, but the resulting correlation coefficients and statistical analyses were not strong enough to support the Saildrone measurements as validation for the satellite-derived gradient products.

Recommendations

Some grammatical errors throughout, a few are pointed out in the comments below. I recommend a careful re-reading of the manuscript for edits. Introduction is a bit thin and could use more background information and references. Methods section needs more specific details about data collection by the Saildrone. Discussion should include more scientific information/analysis of the fronts and oceanographic features identified by the Saildrone during the two campaigns. I don’t think the animations in the Supplementary Section add any crucial support to major findings of the paper. Upon reading Vazques-Cuervo, J. et al., 2019, I found everything I felt was lacking in this paper.

Comments

Line 1: Your conclusions find that Saildrone-derived gradients are not satisfactory validations for satellite gradient products. I might recommend changing ‘validation’ to ‘comparison’ in your title or just removing ‘validation’.

Line 30: satellite-derived fluxes in what? Be specific please.

Line 32. Need to add a few more supporting references at the end of this sentence.

Lines 47 - 48: Give a number here, like differences were up to __% or by _ degrees, etc.

Line 56: You must mention that SST from satellite-derived data measures only the top micrometer of the ocean surface. Mixing processes, especially in seasonally stratified waters, can lead to under/overestimates at warmer temperatures. If the sensor on the Saildrone is a few cm below the skin surface it may be measuring cooler water than the satellite sensors. The 2019 paper notes that the sensor is 0.6m below the surface, you must also mention that here.

Lines 60 - 61: Right, atmospheric corrections should be localized and can be very accurate when tuned for local conditions.

Lines 17 – 61: It might be nice to break up this paragraph into two: one that introduces SST and one that introduces SSS.

Lines 66 - 67: I would mention the novelty of the Saildrone here, why does it provide better or more measurements that traditional floats,buoys, etc.?

Lines 68 - 69: The concluding sentence is unnecessary, you can remove.

Line 70 (72?): Change “measurements derived from Argo floats, drifting/moored buoys” to “measurements derived from Argo floats and drifting/moored buoys”

Line 70(73?): Change “In fact, gradients estimated from satellite observations” to “Gradients estimated from satellite observations”

Line 70 (74?): Change “in situ data are typically associated to one particular geographical location” to “in situ data are collected at one particular geographical location.”

Line 70 (91-92?): Can you reiterate why this paper requires a different collocation strategy? What was the strategy in [1]?

Lines 85: How did you scale down or average the Saildrone data measurements over time and space to compare with the much larger temporal and spatial scale measurements of satellite SSS? Please be specifc and mention this earlier in the paper to avoid confusion.

Lines 101 - 103: I would put this sentence in the Methods Section.

Lines 109: Can you include a figure of the timeseries in the Supplement? It might be nice to see this temporal variability.

Lines 126 - 128: Good discussion Point about land contamination. I would also discuss the large spatial scale of the satellite data and how that might affect bias towards certain high or low measurements.

Line 182: Please cite the studies that have evaluated satellite estimations of ocean fronts

Line 189: land can cause contamination in coastal SST imagery also.

Lines 199 – 201: I would go into more detail here, how would the availability of higher resolution satellite products affect the statistics presented here with the low-resolution data?

Line 203: Change “As future Saildrone campaigns are conducted in the future, the” to “When future Saildrone campaigns are conducted, the”

Line 205: are you talking here about only future SST and SSS gradient products? Or additional measurements? Be specific.

Line 207: I think you’re missing a word here, should be “estimates of surface temperature and salinity”

Author Response

Summary

Thank you for your comments. We respectfully disagree with the statement: "resulting correlation coefficients and statistical analysis were not strong enough to support the Saildrone measurements as validation for satellite-derived gradients products." The correlation coefficients reflect the limitations of the application of SST and SSS remote sensing products in coastal regions. Many reasons exist for the reduced gradient correlations, including insufficient resolution of the satellite-derived products to resolve the mesoscale and submesoscale variability associated with coastal upwelling and western boundary current dynamics, as well as land contamination.

It should be noted that Saildrone data is used here as a reference, similar to what is done with drifters, Argo floats, and mooring buoys for the “validation” of satellite SST.

To avoid any confusion, “validation” in the title has been replaced by “comparison”.

Recommendations

In fact, the manuscript appears to be short on references which due to a lack of papers in the literature that have attempted to compare SST/SSS gradients (as opposed to SST/SSS) from satellite products with in situ data. Please also note the relatively new Saildrone technology.

Methods section needs more specific details about data collection by the Saildrone.

A description of Saildrone data acquisition is beyond the scope of this paper. A full description of Saildrone instrumentation is provided in the referred paper [3].

Discussion should include more scientific information/analysis of the fronts and oceanographic features identified by the Saildrone during the two campaigns.

This paper is not about fronts or features but gradients. In fact, there's no universal “mathematical” definition of a front in the physical oceanography literature. In all published papers, this requires the selection of a subjective threshold to identify/locate fronts. This makes any study based on fronts/features dependent on the very selection of such thresholds. Hence our comparison was based on gradients, which can be computed without subjectively selecting any parameter.

Using Saildrone/satellite SST to identify fronts and ocean features is not the scope of our work, which focus is mainly about a methodology that can be used to compare gradients from satellite products with those from Saildrone from current and future campaigns.

Please also note that we are currently submitting the manuscript to a special edition of Remote Sensing titled “Advances in Retrieval, Operationalization, Monitoring and Application of Sea Surface Temperature “. An analysis of fronts/features would be more suited for a purely physical oceanography-oriented journal.

I don’t think the animations in the Supplementary Section add any crucial support to major findings of the paper. Upon reading Vazques-Cuervo, J. et al., 2019, I found everything I felt was lacking in this paper.

We feel the animations are essential as they visually show the differences in gradients between the data sets and the cloud issue. One of the reviewers asked for more statements highlighting the animations. For example, it is evident in the MUR animation that one transition between cloud-free and cloudy conditions, when only the microwave data is available. The gradients become much smoother/blurred. This point is critical for understanding the correlations between the gradients and why they differ from the absolute correlations between the products themselves

Comments

Our conclusion is that gradients from satellite SST/SSS products do not match very well those from Saildrone. The conclusion does not indicate that Saildrone is not satisfactory. When strong biases between satellite SST and Argo floats are observed, these discrepancies are attributed to limitations in the satellite data (retrieval algorithm, undetected clouds...) and not issues with Argo floats. Saildrone measurements are, after all, conducted in-situ, have been validated with other in-situ instruments (see [3]), and can, therefore, be used as a reference.

Per our recommendation the title has been changed to avoid any confusion.

Line 30: satellite-derived fluxes in what? Be specific please.

We have now reworded the sentence to explicitly say “sensible heat flux”.

Line 32. Need to add a few more supporting references at the end of this sentence.

We have added the following references to introduce the importance of SST gradients:

Hou, A.; Bahr, A.; Schmidt, S.; Strebl, C.; Albuquerque, A.L.; Chiessi, C.M.; Friedrich, O. Forcing of Western Tropical South Atlantic Sea Surface Temperature Across Three Glacial-Interglacial Cycles. Global and Planetary Change 2020, 188. DOI: 1016/j.gloplacha.2020.103150

Yang, H.; Gao, Q.Q.; Ji, H.F.; He, P.D.; Zhu, T.M. Sea surface temperature data from coastal observation stations: quality control and semidiurnal characteristics. Acta Oceanologica Sinica 2019, 38. DOI: 1007/s13131-019-1496-1

Lines 47 - 48: Give a number here, like differences were up to __% or by _ degrees, etc.

The sentence was modified to “They concluded that differences as large as 0.02°C/km

between SST gradient magnitudes derived from the two algorithms...”

Thank you. We added a sentence, but with the satellite products, it is more complicated.

For example, the NLSST algorithm regresses against in-situ data. Thus, although it is technically correct that satellites measure skin temperature, the applications of the different algorithms can tune the temperatures to a bulk temperature. MODIS is tuned towards a skin temperature using the MEARI instrument, which is a shipboard infrared radiometer. MUR is a foundation temperature as it does remove biases based on the incorporation of in-situ data. Additionally, it only uses nighttime data to reflect an accurate foundation temperature.

The following sentence has been added:

“These biases may also be because satellite-based SST is an estimate of the top layer of the ocean surface (~1 micrometer) whereas the Saildrone sensor, being placed below the surface, measures cooler or warmer temperatures depending on the mixing processes involved in a given region or during a specific season.”

Lines 60 - 61: Right, atmospheric corrections should be localized and can be very accurate when tuned for local conditions.

Thank you.

Lines 17 – 61: It might be nice to break up this paragraph into two: one that introduces SST and one that introduces SSS.

Thank you. The section from 17 to 28 has been broken up into two paragraphs.

Lines 66 - 67: I would mention the novelty of the Saildrone here, why does it provide better or more measurements that traditional floats, buoys, etc.?

Thank you. We have included the following statement: “The unique ability of Saildrone to sample at high spatio-temporal resolutions over extended periods (i.e., several months) allows for validation of both SST, SSS, and their corresponding gradients. “

Lines 68 - 69: The concluding sentence is unnecessary, you can remove.

Removed.

Line 70 (72?): Change “measurements derived from Argo floats, drifting/moored buoys” to “measurements derived from Argo floats and drifting/moored buoys”

Changed to “measurements derived from Argo floats and tropical/coastal moored buoys

“.

Line 70(73?): Change “In fact, gradients estimated from satellite observations” to “Gradients estimated from satellite observations”

Corrected.

Line 70 (74?): Change “in situ data are typically associated to one particular geographical location” to “in situ data are collected at one particular geographical location.”

Corrected.

Line 70 (91-92?): Can you reiterate why this paper requires a different collocation strategy? What was the strategy in [1]?

In [1], the collocation is done by using all grid points in the Level4 that are within a given distance to the Saildrone lat/lon position. To exploit as much as possible the high resolution of Saildrone, that distance should be lower than the grid resolution of the Level 4 product. This leads to the typical configuration illustrated in Figure 1. As can be seen, NaNs with such a collocation approach do not allow for the estimation of gradients via central differences (i.e., NaN - number = NaN).

Hence the new collocation approach sees gradients as differences of successive measurements of Saildrone and “looks” for the corresponding location of these measurements in the Level 4 grid.

We believe this is explained briefly but clearly in the paragraph right before equation (2).

“...all Saildrone measurements acquired between latitudes i-ds and i+ds and longitudes j-ds and j+ds are averaged”

“The temporal window used for the collocation is the temporal resolution of the Level 4 datasets”

Please note that this collocation strategy differs significantly from previous ones and the collocation in space is adaptive/nonlinear (hence the denominator term in equation 2).

Lines 101 - 103: I would put this sentence in the Methods Section.

Done.

Lines 109: Can you include a figure of the timeseries in the Supplement? It might be nice to see this temporal variability.

Netcdf files of the California/Baja campaign containing the time series have been added to the supplementary files.

The following sentence was added: “...due to land contamination as well as the larger spatial scale of passive microwave sensors, which affects...”

Line 182: Please cite the studies that have evaluated satellite estimations of ocean fronts

We added the following reference:

“Chang, Y.; Cornillon, P. A comparison of satellite-derived sea surface temperature fronts using two edge detection algorithms. Deep-Sea Research Part II. Topical Studies in Oceanography 2015, 119. DOI: 10.1016/j.dsr2.2013.12.001”

Please note that there is a difference between fronts (binary measure defined using a subjective threshold on gradients or for the identification of bimodal histograms) and gradients (measure in R+ defined without any parameter).

To the best of our knowledge, this is the first study that aims to compare gradients. Thus, the references are few.

Line 189: land can cause contamination in coastal SST imagery also.

Yes, but to a significantly lesser extent than SSS, especially when the infrared data is available.

Lines 199 – 201: I would go into more detail here, how would the availability of higher resolution satellite products affect the statistics presented here with the low-resolution data?

We have added /modified the following sentence: “However, persistent cloud coverage or misclassification of fronts as clouds in infrared observations, and the relatively short duration of Saildrone campaigns (1-2 months) results in a significantly low amount of collocated points to derive reliable statistics.

The availability of higher resolution observations, as well as improved cloud masking at Level 2, is expected to improve the accuracy of gradients significantly, especially in coastal regions and western boundary currents, where the mesoscale to submesoscale dynamics dominate.”

Line 203: Change “As future Saildrone campaigns are conducted in the future, the” to “When future Saildrone campaigns are conducted, the”

Corrected

Line 205: are you talking here about only future SST and SSS gradient products? Or additional measurements? Be specific.

Clarified. “...in both Level 2 and Level 4 satellite products that include SST, SSS and other ocean parameters.”

Line 207: I think you’re missing a word here, should be “estimates of surface temperature and salinity”

Corrected

Submission Date

20 April 2020

Date of this review

08 May 2020 07:13:09

Bottom of Form

Author Response File: Author Response.doc

Round 2

Reviewer 4 Report

The authors have address all of my comments. I recommend this manuscript be accepted.