The Sentinel-3 SLSTR Atmospheric Motion Vectors Product at EUMETSAT

Round 1

Reviewer 1 Report

MINOR COMMENTS

M.1: Section 2.1.1, “the risk of sometimes deriving ground-level null vectors”. Are these later filtered out to avoid including them in the product?

M.2: Section 2.1.2. “The quality of AMVs, relative to the ECMWF forecast model, is higher for S3B than for S3A (as explained in Section 3).” I would suggest removing this statement from this part of the report, as when taken in isolation it implies that S3B is somehow “better” than S3A, when (as later explained) the difference is actually due to the coverage.

M.3. Section 3.1: “The RMS and RMSVD are slightly better for S3B. The higher time gap between the images may explain that behaviour”. Isn’t this contradicted by the next paragraph which shows that when the coverage is limited to the same area, then the statistics between A and B are similar. Hence, the dominant effect in the overall behaviour is coming from the different coverage. It would be handy to have the stats for S3A polewards of 50 degrees included in Table 1.

M4. Could figure 6 include a profile of the number of winds to show the vertical distribution?

M5. Figure 8 would be easier to interpret if it was separated by height bands (e.g. high-level, mid-level, low-level), rather than data at all levels which can lead to biases at different levels averaging out. Could this be done?

M6. Table 3. Should it be clarified that these are for Dual AVHRR?

M7. Section 3.4. “Considering that the biases are high in the tropical regions”. Should clarify that we mean biases are high against the model.

 

TECHNICAL COMMENTS

T.1: Section 2.1.2: “induces a too low overlap”. Could be changed to “induces too small an overlap”

T.2: Section2.1.2: “successive overpasses of a same satellite”, change to “successive overpasses of the same satellite”

Author Response

Dear Reviewer,

thank you for reading our paper and providing us with detailed comments. Here are our answers:

M.1) At the moment there is no post-processing to remove potential null AMVs. However this could be easily implemented if the NWP community wishes so.

M.2) With your approval, we would like to keep this comment. As you mentioned, part of the explanation is the difference of coverage between S3A and S3B. However, there is another factor, as explained in the manuscript: considering that the tracking error (in pixels) is similar between S3A and S3B, dividing it by a bigger time gap (in the case of S3B) results in a smaller error. We have computed the statistics for S3A winds polewards of 50 degrees as you suggested, and they still show a bigger RMSVD than S3B winds. Hence, with your approval, we would like to keep this statement.
However, you are right to point out that we say later in the text that error profiles align. We actually meant that the bias profiles become similar, so for this reason, we are changing:

"When limiting the S3A products to AMVs derived polewards of 50° [...], AMV statistics from S3A and S3B are similar at all altitudes"
to
"When limiting the S3A products to AMVs derived polewards of 50° [...], AMV biases from S3A and S3B are similar at all altitudes"

M.3) This comment is partly answered in our answer to comment M.2. Additionally, we have included, in Table 1, the statistics for S3A for AMVs polewards of 50°, and the corresponding scatter plot in Figure 5.

M.4) As suggested, we have included the vertical distribution of winds in all subfigures of Figure 6.

M.5) The Figure representing the map of biases was changed to four subfigures: the map for low-level (1000 hPa < pressure <= 700 hPa) winds, the map for mid-level (700 hPa < pressure <= 400 hPa) winds, the map for high-level (400 hPa < pressure <= 100 hPa) winds and the map including all winds.

M.6) The caption of Table 3 has been changed to:

Properties of SLSTR and AVHRR, and the associated settings for the derivation of AMVs in dual-satellite mode.

M.7) The clarification "biases against the model" has been added to the text.

T.1) T.2) Thank you for your comments. This part of the text was edited, following the comments of another reviewer, to:

"The 1420 km swath width of SLSTR has little overlap between successive overpasses of the same satellite"

We hope this answers both comment T.1 and comment T.2 at the same time.

We thank you again for your work on this review.

Best regards.

Reviewer 2 Report

See attached file

Comments for author File: Comments.pdf

Author Response

Dear Reviewer,

thank you for reading our article and providing detailed comments. Below are our answers:

1) The sentence

"Any new source of AMVs is of interest for the end-users of AMVs, in particular from LEO platforms, as they can bridge the gap in AMV coverage at high latitudes, that sensors on board geostationnary satellites cannot see."

was replaced with the following two sentences:

"AMV producers like EUMETSAT are always encouraged by end-users to propose AMV products from new sensors, to increase the coverage and density of wind observations. In particular, LEO satellite sensors allow observing atmospheric motion at high latitudes, which are not visible for geostationary satellite sensors."

We hope this formulation is easier to read and understand.

2) We would like to use the replacement sentence that you proposed:

"Any pixels declared clouds by any masks within the L1B product are considered clouds in the AMV processing here."

without further modification.

3) Regarding Section 2.1.2: we added the sentence you proposed, together with:

"Furthermore, the temporal resolution of the observations is lower in a single platform approach (101 minutes) compared to a dual-platform approach (61.5 minutes or 39.5 minutes)."

4) The number of minutes (101) between successive overpasses was added to the caption of Figure 1a.

5) Regarding Section 2.2: in our opinion, the bullets 2 and 3 express two different things. The first one is: if a cloud is present in both images, but is cut off from one because it is not in the frame of the other, this cloud is not tracked. The third bullet, however, refers to features that could be tracked if we added some scans from the previous image of the same orbit. To clarify these two concepts, we would like to rephrase Bullet 3 as follows:

"- This approach uses only one PDU from each orbit. However, successive SLSTR scans may be put in different PDUs, resulting in the loss of potential features lying in these scans."

Regarding Bullet 2, we would like to implement your proposition, but with a small modification, considering that the tracking is performed backwards (as mentioned in Section 2.1.2):

"Clouds in both images which moved from outside of the frame (in the second image) to the inside of the frame cannot be tracked." 

6) “…window of size 32 pixel” changed to “32-pixel sized window” as proposed.

7) The threshold of cloud pixels (25% of the target) and the contrast (0.0) were added in Section 2.3.1. Since the threshold chosen is 0.0, we simplified the sentence to:

"Have a stricly positive contrast."

8) Regarding the convolution mentioned in Section 2.3.2: we realized that the term "convolution" is misleading. Actually, the Gaussian kernel has the same size as the correlation surface, and we multiply the two element-wise.. To avoid confusion, we propose to change the terms "convolution" and "convolved" to "multiplication" and "multiplied", respectively. This also applies to the captions of Figure 4.

9) Regarding the BUFR format, we added a citation to Karhila & Hernandez-Carrascal (2010).

10) We added the missing parenthesis after "(RMSVD" in Section 3.1.

11) We expanded the acronym QI in Section 3.1 an added a citation to Holmlund (1998).

12) Our comment on the time gap is actually very pragmatic, and relates to the computation of the AMVs and not the physical meaning behind them. We observe lower RMS and RMSVD for S3B AMVs. After investigation, we have found that the error (in pixels) made on the tracking is quite comparable between S3A and S3B. However, the RMS and RMSVD are relative to the speed, that is, the observed displacement divided by the time gap between the observations. Considering that the error on the displacement is similar between S3A and S3B, it makes sense that the AMVs derived from the satellite with the longer time gap (S3B with 61.5 minutes) present a smaller speed RMS and RMSVD than those derived from S3A (time gap: 39.5 minutes).

13) Section 3.2 and Figure 9: indeed, we project on a grid with coarser resolution while keeping the same target size of 24 (in pixels), resulting in the consideration of a broader geographical area. However, our intent was simply to study the impact of the projecton resolution on the AMVs, regardless of the other parameters (especially the target size). This Section actually meant to show that the change to a 2 km resolution did not provide any significant benefit, although we track "bigger" features. Nevertheless, we can provide answers to your comments. Dividing the target size by 2 would result in too little information within the target box to operate cross-correlation reliably. Conversely, keeping the higher spatial resolution and doubling the size of the target box presents two problems: (a) the computational cost is higher and (b) the correlation surface are smoother, resulting in more uncertainty of the precise location of the feature in the new image. Although these are small problems, in our experiments, the increase in quality of the AMVs was not significant enough to justify the loss in number of AMVs implied by the use of a bigger target.

14) In Section 3.3, “…15 millions of AMVs of QI > 60.” was changed to “15 million AMVs with QI > 60” as proposed.

We thank you again for your work on this review and hope our answers and changes meet your expectations.

Best regards.

Reviewer 3 Report

This was a very well-written and comprehensive paper. The text was easy to follow, the organization was superb, and the readability of the figures was excellent. I have no major comment. This manuscript is suitable for publication in the present form.

Below are very minor suggested edits that may improve readability:

• Defining S3A and S3B for the reader earlier in Section 2.1.2. Possibly move the sentence in Section 2.1.2 from paragraph 2 to the beginning of paragraph 1 would help define S3A and S3B for the reader: " The two Sentinel-3 satellites are 140 degrees apart, implying a difference in the acquisition time gap for the two dual products: 39.5 minutes for S3A products, and 61.5 minutes for S3B products."

• Replace the comma with a period at the end of the first sentence in Section 2.1.2: "... successive overpasses of a same satellite (see Figure 1a), Consequently, only ..."

• Remove extra period in Section 2.3.2 sentence 4: "by Borde et Garcia Perreda. [10]."
• Define QI acronym in Section 3.1. Paragraph 1 last sentence.

Author Response

Dear Reviewer,

we would like to thank you for the time you took to read and comment our article. Below are our answers to your comments:

1) Section 2.1.2 now starts with:

"SLSTR AMVs are derived using L1B data from both Sentinel-3 satellites, S3A and S3B. The orbits of S3A and S3B are identical, but S3B flies with a phase of 140 degrees compared to S3B."

The sentence you quoted later in the second paragraph was slightly edited to be:

"The phase between the two Sentinel-3 satellites implies a difference in the acquisition time gap for the two dual products: 39.5 minutes for S3A products, and 61.5 minutes for S3B products."

If you agree, we would avoid expanding the acronyms "S3A" and "S3B" again, as this is already done in Section 1.

2) The comma you noticed in Section 2.1.2 was replaced with a period.

3) The extra period in Section 2.3.2 was removed.

4) The QI acronym was defined in Section 3.1. Paragraph 1.

Thank you again for your review.

Best regards.

Reviewer 4 Report

My compliments for a well designed and constructed manuscript.  You have covered all of the points and provided comparisons to similar AMVs which I would expect.  NWP Centers should find the necessary information required to consider the use of your product. 

My only suggestion, I would not have used the adverb nowadays.  It is not needed.

Author Response

Dear Reviewer,

we appreciate that you took time to read our article, and would like to thank you for your review. Following your comment, the sentence

"the time gap between images is nowadays small"

was replaced by

"the time gap between images is small"

Best regards.