Global Gridded Argo Dataset Based on Gradient-Dependent Optimal Interpolation

Round 1

Reviewer 1 Report

 

Minor Comments:

1 Abstract: Page 1, line 19: The words ""..useful and promising.." are vague, it should be better to present some quantitative measures such as the succession ratio or error percentage for objectivity,

 

2) Page 2, lines 64-66: Is there any study to extend vertical measurements at different elevations so that the sea surface values can be obtained even approximately,

 

3) Page 3, line 125-126: Arithmetic average is valid if the data has a symmetrical histogram; otherwise the use of mode value is recommended,

 

4) Page 4, line 138: Does "innovation" mean random, if so please write in parenthesis random innovation (random) value,

 

5) Page 4 lines 161-162: If a graph for a Gaussian exponential function is given, the readers can perceive the meaning better than equations,

 

6) Page 7, line 243: The values (0.02 and 0.01) are not explained as to whether they are percentages or attached to RMSE. This point needs brief explanation. If RMSE, what the practically acceptable limit is?

 

7) Page 8, Lines 280-281: What is the criterion for confidence level? Is it based on Gaussian probability distribution function? Please explain briefly,

8) Page 10, line 331: What is the benefit of T/S difference, because they are different variables. Is it not enough to talk about their relationships? What is the unit of the difference, because they are in different units?

Author Response

• 1) Abstract: Page 1, line 19: The words ""..useful and promising.." are vague, it should be better to present some quantitative measures such as the succession ratio or error percentage for objectivity,

 Response: We appreciate the comments from this reviewer. The quantitative measures of the theoretical verifications have been added. The sentence " The results show that the GDCSM-Argo dataset developed in this study is a useful and promising addition to currently available Argo datasets " has been rewritten as "The results show that the maximum mean RMSEs are 0.8 ^oC for temperature and 0.1 for salinity, and more than 90% of the analysis results are reliable under the statistical probability of 95%" in line 18-20.

 

• 2) Page 2, lines 64-66: Is there any study to extend vertical measurements at different elevations so that the sea surface values can be obtained even approximately,

 Response: As for the Argo surface observation, hundreds of Argo floats with non-pumped CTD sensors were launched to observe the temperature and salinity value of 0-5m between since October 2008. However, the deployment of this kind of floats is discontinuous. The sea surface observation obtained by them are very sparse. The data quality control is also immature. Therefore, most Argo gridded dataset adopted the satellite observation or merged sea surface measurements obtained by traditional means (e.g., XBT, CTD, and TAO) to construct the surface data.

As for the statistical method of extending vertical measurements, there are few studies to deduce surface information from deep observation in our humble opinion. Several statistical models have been developed to estimate the subsurface information from satellite observation before the rapid growth of Argo profiles. Referring to a model of previous research (Chu et al.,2000), we proposed an approach to estimate approximately the sea surface temperature and salinity corresponding to the scatter Argo profiles (referred to the reference 30). This scheme was used to construct the sea surface data in this study. More information could be found from the references:

Chu, P.C., Fan, C.W., Liu, W.T., 2000. Determination of vertical thermal structure from sea surface temperature. Journal of Atmospheric and Oceanic Technology Meteorological Society. 17, 971-979. https://doi.org/10.1175/1520-0426(2000)017<0971:DOVTSF>2.0.CO;2

Zhang Chunling*, Zhang Mengli, Wang Z-F, Hu Song, Wang Danyang, Yang Shenglong*. Thermocline model for estimating Argo surface temperature. Sustainable Marine Structure. 2022, 4(1):1-12. http://dx.doi.org/10.36956/sms.v4i1.474

The above is our rough understanding about this question. If there is anything inappropriate, please not hesitate to comment and correct it.

• 3) Page 3, line 125-126: Arithmetic average is valid if the data has a symmetrical histogram; otherwise the use of mode value is recommended,

 Response:  Thanks for giving a new idea! Taking mode value instead of arithmetic average may be a better choice when the data has asymmetric structure. However, the recalculation of the dataset needs about 3-5d. Maybe the articles should be rewritten due to the revision of the results. Meanwhile, the arithmetic average was used only to obtain the first guess by merging the 2,191,788 profiles into 1^o× 1^o gridded boxes. Then the traditional optimal interpolation was used to obtain the climatological background. Based on the background, the gradient-dependent optimal interpolation was used to construct the climatological analysis results. After several interpolation, the error caused by arithmetic average will decrease little by little. Therefore, this article still remains the previous results without effecting the accuracy seriously. The use of mode value will be considered and tested in the update of GDCSM-Argo.

 

• 4) Page 4, line 138: Does "innovation" mean random, if so please write in parenthesis random innovation (random) value,

  Response:  We are sorry the word "innovation" has brought some ambiguity. That "innovation value" means the observational increments, , that is a real time observation information relative to the background. The "innovation value" has been revised by "the observational increments" in line 144.

• 5) Page 4 lines 161-162: If a graph for a Gaussian exponential function is given, the readers can perceive the meaning better than equations,

   Response:  This is a good point. The graphs of the principle and Gaussian exponential function shown here have been illustrated in detail our previous study (referred to reference 17). In view of this, the schematic diagrams are no longer added in this manuscript, but the corresponding reference is supplemented in line 168.

Figure 1. Schematic of fishery points (red squares), irregularly distributed observations (hollow circles and solid circles), and the radius of influence around fishery point i (blue star), marked with a black circle. The solid circles denote the available observations on the same day as the analysis point i.

 (Figure 1 is cited from Zhang et al., 2021)

Zhang, C.L.; Wang, Z.F.; Liu,Y. An Argo-based experiment providing near-real-time subsurface oceanic environmental information for fishery data. Fish. Oceanogr. 2021, 30, 85–98.

 

 

Figure 2. Scatter of the background error correlation  (indicated by blue circle) and the optimal weight  (indicated by blue circle) as a function of the distance from i

(Figure 2 is cited from Zhang et al., 2021)

Zhang, C.L.; Wang, Z.F.; Liu,Y. An Argo-based experiment providing near-real-time subsurface oceanic environmental information for fishery data. Fish. Oceanogr. 2021, 30, 85–98.

 

• 6) Page 7, line 243: The values (0.02 and 0.01) are not explained as to whether they are percentages or attached to RMSE. This point needs brief explanation. If RMSE, what the practically acceptable limit is?

  Response: The values (0.02 and 0.01) indicate the RMSE of salinity with the unit PSU (a dimensionless unit) that often is omitted. We omitted the unit of salinity in our manuscript. In order to be better understanded, the sentence has been rewritten by "At depths greater than 1000 m, the RMSEs of temperature were smaller than 0.2 ^oC with mean values of 0.05 ^oC, the RMSEs of salinity were smaller than 0.02 with mean values of 0.01." in line 251-252.

The RMSE represents how the analysis results match the observation values. Theoretically, the RMSEs should not exceed the confidence interval determined by mean value plus or minus three times the variance(referred to the reference 32). As for the Argo observation, we take the observation errors as the true mean value. The maximum absolute confident values of the monthly RMSEs are about 2.9^oC for temperature and 0.63 for salinity. In practice, because the true values of observation error are different, the RMSEs in different research results do not have regular limits. In order to confirm the validity of our results, the theoretical limits have been added in line 257-258.

• 7) Page 8, Lines 280-281: What is the criterion for confidence level? Is it based on Gaussian probability distribution function? Please explain briefly.

 Response: We appreciate the comments from this reviewer. The confidence level refers to the probability of analysis values falling within a certain area of the sample statistical values (confidence interval). The larger the confidence interval, the higher the confidence level. It is based on the normal distribution. As for Argo observations, the sample sequence available at most analysis points follows an approximate normal distribution when the confidence level is 95% (with a confidence of 0.05). The explanation has been added in line 290-296.

• 8) Page 10, line 331: What is the benefit of T/S difference, because they are different variables. Is it not enough to talk about their relationships? What is the unit of the difference, because they are in different units?

 Response: We are sorry the abbreviation "T/S difference" give the inaccurate statement. T/S difference indicate temperature deviation and salinity deviation respectively. Their units are degree and PSU（or omitted salinity unit）. The "T/S difference" has been replaced by"temperature and salinity deviation" in the whole article.

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear colleagues! Thank you very much for your paper. You have made a great job. For me it is interesting to see how ARGO project is developing last 20 years. And your product GDSCM-Argo dataset looks like a next step forward in developing of this system for obtaining new knowledges about in situ characterstics of World Ocean.

Nevertheless, as a reviewer I should point your attention to the weak points of your manuscript (in my humble opinion).

 

1. Everybody who concerned with the Argo-floaters knows that lifetime of each drifter con not be more than two or three years. Most of them has equipped with the pressure, temperature and conductivity sensors of Sea Beard company. And it is known problem that during the lifetime they have a drift of their measurement precision. More information could be found here:

https://argo.ucsd.edu/data/data-faq/#sbepsal

and from this references:

Barker, P. M., J. R. Dunn, C. M. Domingues, and S. E. Wijffels, 2011: Pressure Sensor Drifts in Argo and Their Impacts. Journal of Atmospheric and Oceanic Technology, 28, 1036-1049, http://dx.doi.org/10.1175/2011JTECHO831.1

Abraham, J. P., M. Baringer, N. L. Bindoff, T. Boyer, L. J. Cheng, J. A. Church, J. L. Conroy, C. M. Domingues, J. T. Fasullo, J. Gilson, G. Goni, S. A. Good, J. M. Gorman, V. Gouretski, M. Ishii, G. C. Johnson, S. Kizu, J. M. Lyman, A. M. Macdonald, W. J. Minkowycz, S. E. Moffitt, M. D. Palmer, A. R. Piola, F. Reseghetti, K. Schuckmann, K. E. Trenberth, I. Velicogna, and J. K. Willis, 2013: A review of global ocean temperature observations: Implications for ocean heat content estimates and climate change. Reviews of Geophysics, 51, 450-483, http://dx.doi.org/10.1002/rog.20022

Wong, A. P. S., S. E. Wijffels, S. C. Riser, S. Pouliquen, S. Hosoda, D. Roemmich, et al, 2020: Argo Data 1999–2019: Two Million Temperature-Salinity Profiles and Subsurface Velocity Observations From a Global Array of Profiling Floats. Frontiers in Marine Science, 7, https://doi.org/10.1016/j.dynatmoce.2020.101131

 

So the researchers who faced with this problem is always wonder, how the people who created this kind of datasets solve this problem. So I think it is good idea to show how this problem have been solved during the creation of GDSCM-Argo dataset.

 

2. In chapter “Objective Analysis System for Argo” (Background settings) you highlight not only temperature and salinity, but also emphasize analysis of the speed of sound from Argo. My question is: why? You have from Argo only the temperature and salinity directly measured. Sound speed is the artificial parameter, calculated from equation of sea water state (different methods – latest the most popular is TEOS-10 https://www.teos-10.org/ ). So the final question here is why, and which method you have used?

 

3. In abstract you wrote that GDCSM-Argo “can also retain more meso–micro-scale features than other gridded Argo datasets.” From my understanding from your manuscript GDCSM-Argo dataset has a resolution about 1x1 degree. If we translate degree to kilometers we will have about 113 km near equator area. Hence the statement about meso-micro-scale may look doubtful. The smallest mesoscale eddy size observed in the marginal seas of the North-East Asia is near to 20 km and only near the Oyashio or Kuroshio currents we can observed so large-scale eddies. Could you clarify, what do you mean by that?

 

4. According with the chapter 2.2.3 and next one you introduce your calculations of the surface temperature and salinity layer. And compare results with other models. It is reasonable. But why you did not show results (if any is exists) on comparing between model results and measurements from satellite (NOAA and CMEMS COPERNICUS provide correspondent products, based on altimetry and modelling with the resolution of 0.25x0.25 degree and higher)? For me it is interesting how far modelled results from ARGO datasets and satellite observed data (probably a lot of people would be happy to see the answer).

Anyway I think your paper is almost ready for publication. Just need to polish some small details for better understanding.

Thank you for interesting reading.

Looking forward to see your paper printed ASAP.

 

With best wishes,

your reviewer.

Author Response

 

Dear colleagues! Thank you very much for your paper. You have made a great job. For me it is interesting to see how ARGO project is developing last 20 years. And your product GDSCM-Argo dataset looks like a next step forward in developing of this system for obtaining new knowledges about in situ characterstics of World Ocean.

Nevertheless, as a reviewer I should point your attention to the weak points of your manuscript (in my humble opinion).

 

• Everybody who concerned with the Argo-floaters knows that lifetime of each drifter con not be more than two or three years. Most of them has equipped with the pressure, temperature and conductivity sensors of Sea Beard company. And it is known problem that during the lifetime they have a drift of their measurement precision. More information could be found here:

https://argo.ucsd.edu/data/data-faq/#sbepsal

and from this references:

Barker, P. M., J. R. Dunn, C. M. Domingues, and S. E. Wijffels, 2011: Pressure Sensor Drifts in Argo and Their Impacts. Journal of Atmospheric and Oceanic Technology, 28, 1036-1049, http://dx.doi.org/10.1175/2011JTECHO831.1

Abraham, J. P., M. Baringer, N. L. Bindoff, T. Boyer, L. J. Cheng, J. A. Church, J. L. Conroy, C. M. Domingues, J. T. Fasullo, J. Gilson, G. Goni, S. A. Good, J. M. Gorman, V. Gouretski, M. Ishii, G. C. Johnson, S. Kizu, J. M. Lyman, A. M. Macdonald, W. J. Minkowycz, S. E. Moffitt, M. D. Palmer, A. R. Piola, F. Reseghetti, K. Schuckmann, K. E. Trenberth, I. Velicogna, and J. K. Willis, 2013: A review of global ocean temperature observations: Implications for ocean heat content estimates and climate change. Reviews of Geophysics, 51, 450-483, http://dx.doi.org/10.1002/rog.20022

Wong, A. P. S., S. E. Wijffels, S. C. Riser, S. Pouliquen, S. Hosoda, D. Roemmich, et al, 2020: Argo Data 1999–2019: Two Million Temperature-Salinity Profiles and Subsurface Velocity Observations From a Global Array of Profiling Floats. Frontiers in Marine Science, 7, https://doi.org/10.1016/j.dynatmoce.2020.101131

 

So the researchers who faced with this problem is always wonder, how the people who created this kind of datasets solve this problem. So I think it is good idea to show how this problem have been solved during the creation of GDSCM-Argo dataset.

 Response: We are very pleased to receive these positive comments. Thanks for pointing this critical question. The quality control of observation profile is the premise of data assimilation or analysis. We also carried out quality control processing before the gridded analysis. As mentioned in Section 2.2.1, the GDCSM-Argo is based on the Argo T/S profiles provided by CARDC. The CARDC Argo profiles were downloaded from the Global Argo Assembly Center, in which more than 60% of the T/S observations had been delayed-mode quality controlled (adjust salinity drift by historical CTD reference data set) by each national Argo assembly center. For those data haven’t been delayed-mode quality controlled, CARDC conducts a climatological test using high-quality climatology T/S dataset to detect T/S sensor drift (referred to References 12 & 22), if T/S profiles fail this test (out of 5×±STDs of the climatology), the data will be discarded or re-flagged.

Liu, Z.H.; Li, Z.Q.; Lu, S.L.; et al. Scattered Dataset of Global Ocean Temperature and Salinity Profiles from the International Argo Program. Journal of Global Change Data & Discovery 2021, 5(4), 22–31.

 

Li, Z.Q.; Liu, Z.H.; Lu, S.L. Global Argo data fast receiving and post-quality-control-system. IOP Conf. Series: Earth and Environmental Science. 2020.

 

• In chapter “Objective Analysis System for Argo” (Background settings) you highlight not only temperature and salinity, but also emphasize analysis of the speed of sound from Argo. My question is: why? You have from Argo only the temperature and salinity directly measured. Sound speed is the artificial parameter, calculated from equation of sea water state (different methods – latest the most popular is TEOS-10 https://www.teos-10.org/ ). So the final question here is why, and which method you have used?

  Response: We produce the sound velocity because some researchers have requested them from us. Yes, sound velocity can be calculated directly from the gridded temperature and salinity, but we think it would be better or more appropriate if we calculate sound velocity from T/S profiles and then input our objective analysis system. The method we used is based on the UNESCO 1983 algorithm that has been mentioned briefly in line 100 (referred to the Reference 24).

Fofonoff, P. and Millard, R.C. Jr. Unesco 1983. Algorithms for computation of fundamental properties of seawater, 1983. _Unesco Tech. Pap. in Mar. Sci._, No. 44, 53 pp.

• In abstract you wrote that GDCSM-Argo “can also retain more meso–micro-scale features than other gridded Argo datasets.” From my understanding from your manuscript GDCSM-Argo dataset has a resolution about 1x1 degree. If we translate degree to kilometers we will have about 113 km near equator area. Hence the statement about meso-micro-scale may look doubtful. The smallest mesoscale eddy size observed in the marginal seas of the North-East Asia is near to 20 km and only near the Oyashio or Kuroshio currents we can observed so large-scale eddies. Could you clarify, what do you mean by that?

  Response: That is right! GDCSM-Argo dataset has a resolution of 1^o × 1^o. It is not very accurate for the statement about microscale. The corresponding statements have been corrected in line 21, 521, and 556. The advantage of the Gradient-dependent Correlation Scale Method (GDCSM) is that it can extract more real-time information of observation data by adjusting the relevant scale automatically. Limited to the number of Argo profiles, we developed the monthly gridded data set with the spatial resolution of 1^o × 1^o. That may filter out some microscale information. Even though the meso scale (~50 km) information can be well represented. More detail verification could be found in the reference 16. The similar conclusions are also displayed in the EOF results (Figure.13). The microscale information has not been reflected obviously in the current dataset. We believe that more detailed observation information can be represented in the update GDCSM-Argo with the increase of Argoprofiles.

Zhang, C.; Xu, J.; Bao, X.; et al. An effective method for improving the accuracy of Argo objective analysis. Acta Oceanologica Sinica, 2013, 32(7), 66-77.  

 

• According with the chapter 2.2.3 and next one you introduce your calculations of the surface temperature and salinity layer. And compare results with other models. It is reasonable. But why you did not show results (if any is exists) on comparing between model results and measurements from satellite (NOAA and CMEMS COPERNICUS provide correspondent products, based on altimetry and modelling with the resolution of 0.25x0.25 degree and higher)? For me it is interesting how far modelled results from ARGO datasets and satellite observed data (probably a lot of people would be happy to see the answer).

Response:  We appreciate the comments from this reviewer. Firstly, the surface temperature obtained by statistical model are close to the measurements at 2–5 m. They have non-negligibly different from the SST from satellite (Lu et al., 2014) due to their calculation method based on Argo subsurface observation and thermocline parameters. We compared the T/S results with the WOA18 reanalysis surface temperature and salinity that merged multiple traditional observation data. Secondly, the effective combination of Argo and satellite measurements is a very promising work. The bilinear interpolation is usually used for simple combination in our application due to space-time matching degree. The more accurate fusion model needs to be tried in future research.

Lu S，Xu J，Liu Z. Analysis of the differences between microwave remote sensing SST and Argo NST in the Southern Hemisphere. Marine Forecasts, 2014, 31(1), 1-8.

 

Anyway I think your paper is almost ready for publication. Just need to polish some small details for better understanding. 

Thank you for interesting reading.

Looking forward to see your paper printed ASAP.

 

With best wishes,

your reviewer.

 

Response:  Thanks for the affirmation of our study again. We’d like to go deep into discussion of any questions about GDCSM-Argo.

Author Response File: Author Response.pdf