# Summary

This is a review of “A New Global Gridded Sea Surface Temperature Data Product Based on Multisource Data” by Mengmeng Cao1, Kebiao Mao, Yibo Yan1, Jiancheng Shi, Han Wang, Tongren Xu, Shu Fang, and Zijin Yuan. The authors have merged the sea surface temperature (SST) data from multiple sources to create a new high-resolution global dataset of monthly SST. The new data product does not contain any missing values and has been shown to be more accurate than the unmerged datasets. However, I have some concerns about the usefullness of the new dataset and the assessment of accuracy (see general comments). Moreover, the authors may need to emphasize the novelty and the uniqueness of the methods used to create the new data set.

# General comments

1. The conclusion of the better performance of the new dataset is drawn based on the smaller differences between the new data set and the in-situ observations (iQuam) than between the original datasets and iQuam. Should it simply because the iQuam data is used to create the new dataset? If iQuam is supposed to be the closest to the truth (as it is used as reference to validate other datasets), then why researcher do not use iQuam but use the new dataset created by the authors? I think author needs to better justify it. Missing values generally do not have a significant impact on the statistical analysis of climate. Moreover, the missing values can be obtained by interpolating other data using simple interpolation methods without losing much accuracy.
2. I wonder how accurate the new dataset is compared with global reanalysis such as ERA5. The global reanalysis may not have such a high-resolution as the new dataset. But is it possible to compare the new dataset with some regional high-resolution reanalysis? The reanalysis is created using both information from observations and model simulations. Therefore, should one expect a reanalysis product to be more accurate than the new dataset generated using only observations?

# Specific comments

1. Line 109-110: Why need to correct the observation time difference if the new dataset is monthly data. Isn’t the difference measured in days?
2. Section 2.4: Suggest separating the section of data and method.
3. Line 224: Figure caption in wrong place.
4. Figure 6: The scatter points for B looks strange (different behavior than the pdf and the box plot). Are the box plots already enough to demonstrate what you want show?
5. Eq. (8): Suggest using bold font for vector and matrix.
6. Eq. (9): X_{(t)} looks a bit uncommon, suggest X_{t} or X(t). But it is the author's choice.
7. Eq. (10): I don’t really understand this equation, if Eq. (10) holds, then Eq. (11) shouldn’t be correct.
8. Eq. (12): Should it be (P+R). Additionally, they are matrices and should be write in such a way P(P+R)^{-1}.
9. Fig. 15: the original SST data seems quite interesting, as it has a lower bound of around 9 Celsius. 

I apology that I haven't checked the quality of data set. I don't have necessary tools to access the data with the given format on my laptop.

The manuscript describes the methodology and procedure adapted to produce a new global dataset with 0.041° spatial resolution of monthly SST fields. The authors use MODIS SST data as benchmark and many other supplementary and complementary data sets including in situ observations and those retrieved from AVHRR infrared sensors, and AMSR and Windsat microwave sensors are utilized for obtaining a fusion.  Essentially, the values in the blank or missing and low quality pixels are replaced by values of in situ observations and those derived or interpolated through the processes of Optimal Interpolation and Kalman Filter. The missing and low quality pixel problems arise due essentially to three reasons: (1) Cloudiness, fog, sea ice and proximity to shore influence the SST measurement. (2) Different sensors have different responses, that is, different sensors observe a pixel at different hours of the day and sense the temperature at different depths. (3) Latitudinal position of the pixel and the angle of sight from the sensor. The improvements in the new dataset, in comparison with earlier dataset, are statistically quantified.

A reader who is not highly specialized in the fields (of remote sensing and statistical manipulation of geophysical data) finds the manuscript difficult to read and assimilate.  There is a certain amount of repetition in the description which did not contribute to clarity.

The methodology section should be improved to offer more clarity.  In many places, a lot of empiricism is found about parameters that cannot be measured directly or easily or with sufficient accuracy. To make the understanding easier, they should provide units for the variables in the equations, tables and figures. Also the figure legends need to be more complete.

For Climatologists and Oceanographers who wish to use the SST, without bothering to go into miniscule details of the elaborate processing procedure, the present product provides a more accurate dataset. The quality control statistics presented shows substantial improvements in the new SST product.

One fundamental question over monthly time scales: Do we require a spatial resolution of 4,1 km, especially in the open oceans? This high resolution SST perhaps helps coastal studies like upwelling and estuary biology.

On the whole the authors did a good and useful job. Some specific points to be considered are:

1. How can GOTM produce accurate values while utilizing several variables, such as 2 m temperature, 10 m wind, sensible heat flux, latent heat flux, which have poor accuracy? Does the ECMWF reanalysis of these variables present the accuracy needed?
2. Fig. 7: Why the non-null pixel frequency is low off the Peru coast, exactly in the ENSO signal region? Because, the region is covered by low cloud almost all the time. You can see that the ITCZ region and other tropical oceanic regions west of the continents also present low non-null frequency due to clouds, high as well as low clouds. Inclusion of these comments may enrich your manuscript.
3. Make the difference between skin temperature and surface temperature clear.
4. I agree that the differences in time and depth of observation have to be compensated. What guarantees that Eqs. 1 and 2 can fix these problems? l is empirical, m is the frictional velocity in the water. These parameters may introduce uncertainties. What is the sanctity of the formula in Eq. 3? As you said in the Discussion section, the procedure relies on the performance of GOTM.
5. Why can’t you use the diurnal variability from in situ observations, at different places and in different months, instead of relying on a model?
6. Lines 428-429: You say “I” is identity matrix and immediately after you say “H” is an identity matrix.
7. The procedure described in lines 430 through 458 needs some more clearer explanation.
8. Tables: You better provide in the text expressions for the statistical metrics shown the tables.
9. Figs. 12 and 13 call for aithors’ comments. Fig. 16: what do blank circles and filled circles represent? Fig. 17: Indian_R and Indian_E. Tell what they represent in the legend.
10. L 683: By the expression “different surface depths” you mean “different depths in the surface layer”?
11. Many uncertainties you mentioned in your discussion will remain uncertain for a long time to come. Rewrite the last sentence, L 700. At many other places too the write-up needs improvement. Some repetitions can be suppressed while more explanation and comments are needed in some places. The conclusion section can be merged with discussion section and some repetitions can be avoided.

Please pay attention to comment 7 above.