Review of
The benefit of using sea ice concentration satellite data products with uncertainty estimates in summer sea ice data assimilation - Revision 01
by
Yang, Q., et al.
This is revision 1 of this manuscript. Hence I am not describing one more time the general content of the paper.
I would like to repeat, though, that this paper is an important contribution to current knowledge and the paper is a good example for the usage of sea ice products WITH uncertainty estimates.
The paper received two reviews and the authors tried to reply to the comments of the two reviewers and tried to change their manuscript accordingly.
However, from reading the revised manuscript I would say that the authors did not succeed completely. Many details are still open and/or not convincingly answered as I will detail in my general comments (still) and my detailed comments. I still consider the revisions which are needed to be done as major before the manuscript is acceptable for publication in "The Cryosphere".
General comments:
1) The authors tried to describe the data sets they have used better. Still it first reads as if they are using SICCI SIC data based on SSM/I data before it becomes clear that they use in fact SICCI SIC based on AMSR-E data.
In addition to that the authors seem not to have looked to careful at their results here and there because I have the feeling they are misinterpreting their results and/or are showing wrong figures to the things they are discussing (see lines 250-263)
2) The authors are too global in the description of their main results. The comparison regarding the SIC data and their impact in the assimilation focuses on SIC < 35% while this is not reflected in the abstract or conclusions.
3) The authors are still not critical enough with the limited value the two BGEP mooring data points have in terms of their spatial representativity. I was hoping that they would discuss this a bit more - in particular in the light that there are more moorings than these two. I would have hoped to see a discussion of the potential impact of ice types and of ice drift on the differing observations. The advantage of the BGEP mooring data lies in the fact that they provide a full seasonal cycle of sea ice draft and hence sea ice thickness estimates. A more in-depth discussion of the potential uncertainties that could be involved with such point measurements would be required to underline the authors' hypothesis that with an increased minimum SIC uncertainty one can reduce the bias in the grid-cell mean sea ice thickness.
4) The authors argument that the SIC (most likely) does underestimate SIC in the central Arctic in summer because of the melt ponds. They refer to their example on July 16. I would have been very good to take the respective MODIS melt pond cover data set (also from ICDC) into account and show a mal of the melt pond fraction along with a map of the SICCI SIC for that particular date to underline their statement (see lines 265-283)
In the following I abbreviate line with L and do NOT refer to the page.
Detailed comments:
L15
I would start the abstract in a more general way and re-organize the sentences:
"Data assimilation experiments that aim at ... in the Arctic are carried out. The data assimilation system used is based on the MIT ... filter. The effect ... different experiments using sea ice concentration data of the European ... SICCI) which are provided with a per-grid cell physically based SIC uncertainty estimate."
L20: I would recommend to start here first with the "regular" SICCI uncertainty estimate, then with the modified one with an elevated minimum uncertainty and then, at third place, mention the constant uncertainty.
L23/24: See general comment 2)
L24: "ice thickness is not affected in a systematic way" I would say that your work is not detailed and not critically discussed enough to write this global statement.
L25-28: "Further ... in ponds)" This reads as if you have shown this. But this is not the case. You did not (yet) present this - as you say - fundamental mismatch. I therefore recommend that you delete this sentence or re-formulate it.
L42: local --> regional because local is on much smaller scales.
L70: You could add here the NOAA SIC CDR (Peng et al., Earth System Science Data, 5, 2013)
L64-L73:
This reads as if you want to use the SSM/I based data and not the AMSR-E ones. This needs to be changed. I was about to suggest to remove the sentence about AMSR-E because I considered it irrelevant until I learned later that you used the AMSR-E data. (see general comment 1)
L75-L82: This paragraph still leaves to open question why the local SEIK (LSEIK) is / needs to be used.
L97-L102: From this I still don't understand what was so particularly interesting in the year 2010 compared to e.g. 2007 or 2011 or even 2012?
L100: "(data from NSIDC)" is not enough. I guess you refer to the sea ice extent values. The source and the date when you accessed these would be required.
L102: "concentration was also much" ... Why "also"?
L123: "The SEIK analysis is performed locally ..." Is this LSEIK then?
L130-143: This paragraph causes several question marks.
1) You use NASA-Team SIC based on SSM/IS data and SICCI SIC based on AMSR-E. These data are independent - not as written by the authors.
2) Later its turns out that you compare the MITgcm output with both SICCI SIC and NSIDC SIC. However, you do this particularly for the low sea ice concentrations, i.e. < 35%, and you see differences (Fig. 3). You might need to discuss that the SIC data sets used here have two main differences: One is that these are on different Earth grids (polarstereographic versus EASE). The second is that AMSR-E has a finer native spatial resolution than SSM/IS. Therefore, even though both products are provided on a 25 km grid the SSM/IS based ones will show less details and will appear more smoothed than those based on AMSR-E. You could show a set of maps with a zoom of the SIC close to the ice edge for a compact, clearly defined ice edge, and an open ice edge to demonstrate that perhaps the difference between the two data sets is not that large and that by taking into account the 126 km radius (line 124) any difference due to gridding and different native resolution smears out.
3) I wonder whether the cited Cavalieri et al (2012) paper is the correct one for citing the data set. Please check.
4) L137: delete "also"
5) L139/140: Why do you stress "different algorithm (NASA Team)"? A different algorithm than in the NSDIC data set?
L145-146: While this sentence is correct you can be more specific by looking in the respective literature. Instead of citing the two papers cited I would refer e.g. to Zygmuntowska et al., The Cryosphere, 8, 2014; and to Laxon et al., Geophysical Research Letters, 40, 2013, and to Kwok et al., Journal of Geophysical Research, 114, 2009 to underline that there are ICESat data (limited periods) and CryoSat-2 - starting 10/2010 and that there are simply NO satellite-based sea ice thickness data for the Arctic Ocean for the time and year of your investigation. Then you could make a comment that it seems to be generally impossible to retrieve sea ice thickness from either laser or radar altimetry during summer melt conditions due to wet snow conditions and or clouds.
L146: Is ULS a direct measurement or is this perhaps also a form of remote sensing?
L147: "Experiment" --> "Exploration"
L151-153: This is something I wouldn't write in this form. First of all there are at least two sources of ice type information one could have used (why didn't you do that?). Secondly, is has been shown by e.g. Laxon et al., 2013, and Kern et al. (2015) that including ice type and hence ice density information into satellite radar altimetry is recommended. If you ignore the ice type then you might wish to argue for that. One point could be to say that the ratio between draft and thickness is much close to 1 than the ratio between freeboard (as measured by the altimeters) and thickness ... Or you might find another paper which has dealt with the conversion of draft into thickness from ULS Data (you might look for the NPI ULS data in the Fram Strait and look what they did in terms of different ice types. You can solve this issue with 2-3 sentences but the way it is written here it is not satisfying.
Why you didn't consider sea ice thickness from Operation Ice Bridge (OIB) flights would also be one sentence to give as OIB data are very valuable.
L156: Did you already give a motivation why you have chosen 0.25?
L158-159: I don't understand this sentence. What do you mean by "the measurement error to account for a representation error" ? How do you know what the measurement error is?
L163: Perhaps I did not get it but did you give a motivation why you have chosen 0.1 (or 10%) for the sea ice concentration uncertainty?
L166: You have chosen an uncertainty of 0.3 (30%) for the north pole hole. What is the movitation for such a high uncertainty? In the uncertainty maps shown in Figure 2 the uncertainty values around the pole hole barely exceed 0.1.
L186: You write further down that you have chosen to show the RMSE only for SIC < 35% because of the likely impact of melt ponds at higher ice concentrations. May I ask
i) Do you show all SIC values, starting at 0%? In this case I need to ask whether both images in Figure 3 (a and b) show the same number of data points because the SICCI SIC has spurious SIC over open water due to non-corrected weather influence while the NSIDC data set does not have these.
ii) How would the results shown in Fig. 3 look for 25% or 50%? I find the chosen threshold a bit arbitrary and I would like to see how this would change ... unless you can give a convincing motivation for choosing 35%.
L188: "Cyan color on Fig. 1", please change on --> in. Also: where are the cyan area in the central arctic you are referring to?
L194/195: The mentioned explicit weather correction does not correct for cloud liquid water and (of course) cannot eliminate all weather influences on the SIC. In contrast the weather filter used for the NSIDC SIC cuts-off SIC at various values (see Ivanova et al., 2015 [Figure 5])
L195/197: Coming back to the gridding issue. The point is that the interpolation of the SIC derived on a footprint by footprint basis into a grid at the Earth surface does not cause smearing uncertainty for SIC = 0% and SIC = 100%. However, for all other SIC values suffiently away from these two end points of the SIC distribution the smearing error can be quite large and eventually dominate the total SIC retrieval uncertainty. It could be that the uncertainty which is assimilated in the grid cells which have that low SIC as shown in Fig 3 is particularly dominated by the smearing uncertainty and not by the retrieval uncertainty. Since the SICCI SIC data set explicitely gives this information it might have been worth (and easy) to check this - especially in the context of focussing on SIC < 35% when showing the results.
L207: I suggest to write "with the originally SICCI-provided"
L208: I suggest to write "produces an ensemble ... concentration which agrees better with the two SIC data sets chosen when the full range of uncertainties provided by the SICCI satellite observation is used."
The final comment I have to this is that this is a results from just one summer and it might be different in other summers.
L214: I suggest to write "ULS-NSIDC grid-cell mean"
L219: "after late July" Do you have an idea why this does not start earlier?
L225: This is true for BGEP2009A but not for 2009D. So you have a 50:50 situation and this needs to be discussed (see my general comment 3)
L234: "for sea ice concentration in summer." You did not show this yet, because Fig. 3 concentrates on SIC < 35%.
L236: "validation" --> "inter-comparison". This is not a validation.
L237: "grid-cell mean sea ice thickness" instead of "mean ice thickness"
L238: "which is not real" I suggest to avoid such a statement in your text because it puts in question your approach to use the satellite SIC to compute a grid-cell mean SIC.
L241-243: I cannot see this main message from Figure 3. Perhaps the authors can elaborate a bit on this?
L245-249: I don't understand what you want to say with this sentence. Where can we see this statement in your results? What are "teams" in this context?
General question to the difference of LSEIK-1 to the other two choices: What would have happened if you would have used 0.05 instead of 0.25?
L250-263: This paragraph needs to be re-written because I have the feeling that it is referring to the wrong date. Fig 5c (and in general Figure 5 and also Figure 6 seem not to show results from August 30 (compare to Figure 2). I have the feeling that Figure 5 and 6 are instead referring to a much earlier sea ice state - perhaps even of June 1.
In that case, as we are talking about a 24 h forecase, I would not be that negative about a STD which is 0.01 or 0.03. At this time of the year the SIC won't change by more than 1-2% within 24 hours unless directly at the ice edge where ice drift has a major impact. So it is perhaps not that bad? The authors might look at maps of SIC difference (in SICCI or NSIDC) around that date (June 1). I guess they will figure out that the overall day-to-day difference in SIC is quite small and on average < 5% if not even less.
The same argument applies to the sea ice thickness standard deviations shown in Figure 6. If this is really June 1 (which I assume) then the changes shown seem not unrealistic to me.
Finally, the last sentence in this paragraph is fine for one of the BGEP data sets but not for both (see general comment 3)
L253: "similar" to what?
L253/254: What is a "spread distribution pattern"?
L265-283: Here my general comment 4) applies. Showing more of the data and of additional data (melt pond fraction) would underline the statements made in these two paragraphs.
L275 "are" needs to be shown.
L290: This seems to be true, yes, but it does increase the RMSE in the ice edge region.
L293-297: This is a too global description of the results.
L299-303: You did not show this in your paper.
L303: I suggest to write: "do not reflect" instead of "are not enlarged to accommodate"
L309-315: Please check once again whether you have given the respective web pages and sources where the data can be obtained from in the paper. You might want to repeat these in a condensed form here in the acknowledgements.
References:
Schweiger et al., 2011 is missing
Please exchange the Ivanova et al., reference by the full citation now; the paper is out.
Table 1: In the table you use 2010, in the text 2009. What is correct?
Figure 3 and 4: red and magenta lines are difficult to discriminate
Figure 3 contains an superfluous "(a)" annotation.
The free MITgcm runs differ between a) and b) in Figure 3) and Figure 4). You might wish to comment on this in the text.
Figure 4: The black and the blue lines are difficult to discriminate.
Caption: I suggest to write "grid-cell mean sea ice thicknes"
Figure 5 & 6: Compared to Figure 2 this seems not to be August 30.
Typos / editoral comments:
L38: can be --> has been
L156: e.i. --> i.e.
L172: region --> regions
L185: reports --> shows
L186: Both location --> locations, and --> an
L196: grids --> grid |
- Printer-friendly version
- Supplement
